CONTRIBUTIONS TO THE EPIDEMIOLOGY OF HEROIN USE 

By  

Samantha Jo Bauer 

 

 

 

 

 

 

 

 

A DISSERTATION 

Michigan State University 

in partial fulfillment of the requirements  

Submitted to  

for the degree of 

Epidemiology—Doctor of Philosophy 

2019  

 

 

 
 
 

 

ABSTRACT 

CONTRIBUTIONS TO THE EPIDEMIOLOGY OF HEROIN USE 

 
By 

Samantha Jo Bauer 

Background & Aims: Despite the marketing of heroin in 1898 and its history as a public health 

concern, heroin use has not been a major topic of epidemiologic research (1). Little is known 

about the problems and experiences (PEs) during the earliest months following newly incident 

heroin use. These PEs after heroin use onset might arise and co-occur in pairwise or other 

multivariate association patterns that are consistent with the idea that there is a dimension of 

heroin use disorder (HUD). It might be assumed that this HUD dimension is comparable across 

subgroups, such as sex and history of other opioid use, but there is little evidence of this. Lastly, 

consequences of heroin use have not been fully explored in relation to an emerging public health 

concern about increased rates of stimulant related overdose concurrent with heroin related 

overdose. Thus, this dissertation aims: 1) To estimate the pairwise association between PEs of 

HUD.  2) To assess measurement equivalence of HUD, adopting a single dimension idea, across 

subgroups defined by sex and extra-medical prescription opioid use (EMPOU) history. 3) To 

investigate the degree to which onset of heroin use may precipitate excess risk of onset of 

stimulant use.  

Methods: The National Surveys on Drug Use and Health (NSDUH), 2002-2016, target 

population includes non-institutionalized United States (US) civilians age 12 years and older, 

including inhabitants of households, group quarters, and homeless shelters. Multi-stage area 

probability sampling identified 837,326 participants, 896 of whom were newly incident heroin 

users (NIHUs). Time intervals stratified NIHUs by elapsed time between first heroin use (i.e., 

within 12 months of assessment) and survey assessment. Pairwise combinations of PEs of HUD 

 

 

have been studied; odds ratios (ORs) have been estimated. The PEs were also combined into 10 

diagnostic criteria indicators for an evaluation of measurement equivalence. Lastly, a case 

crossover design and analysis estimated the degree to which NIHU may lead to months of 

increased risk for starting extra-medical use of prescription stimulants or cocaine. Two-month 

hazard and control intervals were specified a priori.  

Results: The majority of PE to PE ORs suggested robust and statistically significant associations 

throughout time intervals from first use to up to 12 months after first use. Measurement 

equivalence evaluation indicated no difference between males and females, but suggested lack of 

measurement equivalence across EMPOU subgroups. Case crossover discordant pair odds ratios 

show no statistically significant increased risk of newly incident stimulant use within two months 

of NIHU. A post-estimation exploration of alternatives suggests use of one-month intervals.  

Discussion: The current study is novel as a US nation-scale community epidemiology study of 

the natural history of HUD. Other facets of novelty involve its fine-grain stratification of elapsed 

time since first heroin use, and its focus on individuals and aired PEs that develop over time. 

This dissertation also draws attention to the need for more precise HUD assessment considering 

EMPOU. There may be a linkage between NIHU and extra-medical stimulant use, but this topic 

deserves more attention in future research. 

Conclusions: First, heroin PEs now begin to coalesce into a HUD syndrome within 90-120 days 

after first heroin use. Second, female-male contrasts in level of HUD can now be made, but this 

is not the case for prescription opioid subgroups. Third, this study’s evidence does not confirm a 

suspected triggering of stimulant use by heroin use.  

 

 

 

 

ACKNOWLEDGMENTS 

 

I would like to acknowledge Dr. James C. Anthony for his devoted mentorship 

throughout my PhD education and for him graciously accepting me into his T32 training 

program. He has been invaluable in my growth as a scientist and his National Institute on Drug 

Abuse award has made my research and pre-doctoral education possible (T32DA021129). His 

investment in my education has been an incredible gift. 

 

In addition, I would like to acknowledge my guidance committee Dr. Brian Ahmedani, 

Dr. Ahnalee Brincks, and Dr. Ruben Parra-Cardona. Their valuable help and encouragement 

throughout the dissertation process gave me the confidence and the subject matter insight to 

accomplish my goal. I would like to acknowledge Dr. David Barondess for his dedicated 

guidance throughout my graduate career. His ability to see the forest for the trees, in both 

education and in matters of the heart, has made him a remarkable mentor and friend. I would also 

like to thank the entire Department of Epidemiology and Biostatistics for being my academic 

home over the past years. Together the faculty, staff, and students have irreplaceably balanced 

education, camaraderie, and care. 

Lastly, I must acknowledge my mom, dad, and brothers – Brennan, Baron, Bradly, and 

Bronson. The journey up to the completion of this doctoral degree started long before my formal 

education. I could not have set out on this venture if it were not for your love and unwavering 

support. The most important knowledge I have carried with me has always been that I have a 

home with all of you to come back to.  

 

 

iv 

TABLE OF CONTENTS 

 
LIST OF TABLES ...................................................................................................................... vii 
 
LIST OF FIGURES ................................................................................................................... xiii 
 
KEY TO ABBREVIATIONS .................................................................................................... xiv 
 
CHAPTER 1 INTRODUCTION ................................................................................................. 1 
1.0 Introduction ........................................................................................................................... 1 
1.1 Specific Aims ........................................................................................................................ 2 

 
CHAPTER 2 BACKGROUND AND SIGNIFICANCE ........................................................... 4 
2.0 Heroin .................................................................................................................................... 4 
2.0.1 General Introduction ...................................................................................................... 4 
2.0.2 Heroin Pharmacology & Physiology ............................................................................. 4 
2.1 Opioids Overview ................................................................................................................. 6 
2.2 History ................................................................................................................................... 7 
2.1.1 From Opium Cultivation to Heroin Synthesis ............................................................... 7 
2.2.2 The History of Heroin in the United States ................................................................... 8 
2.2.3 America’s Opioids Crisis ............................................................................................. 12 
2.3 Heroin Use Disorder ........................................................................................................... 14 
2.4 The Main Rubrics of Epidemiology Applied to Heroin Use .............................................. 16 
2.5 Significance and Purpose .................................................................................................... 22 

 
CHAPTER 3 MATERIALS – NATIONAL SURVEYS ON DRUG USE AND HEALTH . 24 
3.0 Chapter Overview ............................................................................................................... 24 
3.1 The Population Under Study ............................................................................................... 24 
3.2 Cross-Sectional Survey Design and Probability Sampling Approach ................................ 26 
3.3 Human Subjects Protection ................................................................................................. 26 
3.4 Participation Rates and Assessment Protocol ..................................................................... 27 
3.5 Statistical Analysis .............................................................................................................. 30 

 
CHAPTER 4 MANUSCRIPT 1 – THE NATURAL HISTORY OF HEROIN USE 
DISORDER EXPLAINED BY PROBLEMS AND EXPERIENCES FROM FIRST 
MONTHS OF USE TO UP TO A YEAR AFTER ONSET .................................................... 32 
4.0 Abstract ............................................................................................................................... 32 
4.1 Introduction ......................................................................................................................... 33 
4.2 Methods ............................................................................................................................... 38 
4.3 Results ................................................................................................................................. 46 
4.4 Discussion ........................................................................................................................... 86 
4.5 Conclusions ......................................................................................................................... 91 

 
CHAPTER 5 MANUSCRIPT 2 – EVALUATING MEASUREMENT EQUIVALENCE OF 
HEROIN USE DISORDER ACROSS HISTORY OF EXTRA-MEDICAL 
PRESCRIPTION OPIOID USE AND SEX ............................................................................. 93 
5.0 Abstract ............................................................................................................................... 93 

 

v 

5.1 Introduction ......................................................................................................................... 95 
5.2  Methods ....................................................................................................................... 101 
Results ......................................................................................................................... 113 
5.3 
Discussion ................................................................................................................... 114 
5.4 
5.5 
Conclusions ................................................................................................................. 116 

 
CHAPTER 6 MANUSCRIPT 3 – WILL THE OPIOIDS CRISIS PRECIPITATE EXCESS 
RISK OF A STIMULANT CRISIS? ....................................................................................... 119 
6.0 Abstract ............................................................................................................................. 119 
6.1 Introduction ....................................................................................................................... 121 
6.2 Methods ............................................................................................................................. 124 
6.3 Results ............................................................................................................................... 130 
6.4 Discussion ......................................................................................................................... 131 
6.5 Conclusions ....................................................................................................................... 135 

 
CHAPTER 7 DISCUSSION AND CONCLUSIONS ............................................................. 138 
7.0 Summary of Findings ........................................................................................................ 138 
7.1 Strengths and Limitations ................................................................................................. 140 
7.2 Public Health Implications and Next Steps ...................................................................... 141 
7.3 Conclusions ....................................................................................................................... 145 

 
APPENDICES ........................................................................................................................... 147 
APPENDIX A – IRB DETERMINATION ......................................................................... 148 
APPENDIX B – MANUSCRIPT 1 ...................................................................................... 150 
APPENDIX C – MANUSCRIPT 2 ...................................................................................... 185 
APPENDIX D – MANUSCRIPT 3 ...................................................................................... 188 

 
BIBLIOGRAPHY ..................................................................................................................... 190 
 
 
 
 

 

vi 

LIST OF TABLES 

 
 
Table 1. Diagnostic Criteria for Substance Use Disorders Defined by the Diagnostic and 
Statistical Manual. ......................................................................................................................... 15 
 
Table 2. Population Projections of Newly Incident Heroin Users and Dependence .................... 17 
 
Table 3. United States Overdose Deaths by Drug. ....................................................................... 18 
 
Table 4. Contrast of Heroin Knowledge Based on Lee Robins’ Study of Vietnam Veterans 
Return to the United States. .......................................................................................................... 36 
 
Table 5. Description of Problems and Experiences of Heroin Use Disorder Included in the 
Diagnostic and Statistical Manual. ............................................................................................... 37 
 
Table 6. Description of Lag-Time Intervals Indicating Elapsed Time Between Newly Incident 
Heroin Use and Survey Assessment. ............................................................................................ 44 
 
Table 7. Characteristics of Overall Sample and Newly Incident Heroin Use Analytic Sample. .. 45 
 
Table 8.  Description of Problems and Experiences of Heroin Use Disorder. ............................. 52 
 
Table 9. Weighted Odds Ratios of Problem and Experience Pairs Within 0-90 Days of First 
Using Heroin. ................................................................................................................................ 53 
 
Table 10 Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 0-
90 Days of First Using Heroin. ..................................................................................................... 54 
 
Table 11. Weighted Odds Ratios of Problem and Experience Pairs Within 1-120 Days of First 
Using Heroin. ................................................................................................................................ 55 
 
Table 12. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 1-
120 Days of First Using Heroin. ................................................................................................... 56 
 
Table 13. Weighted Odds Ratios of Problem and Experience Pairs Within 30-150 Days of First 
Using Heroin. ................................................................................................................................ 57 
 
Table 14. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 90 
Days of First Using Heroin. .......................................................................................................... 58 
 
Table 15. Weighted Odds Ratios of Problem and Experience Pairs Within 60-180 Days of First 
Using Heroin. ................................................................................................................................ 59 
 
Table 16. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
60-180 Days of First Using Heroin. .............................................................................................. 60 
 

 

vii 

Table 17. Weighted Odds Ratios of Problem and Experience Pairs Within 150 Days of First 
Using Heroin. ................................................................................................................................ 61 
 
Table 18. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
150 Days of First Using Heroin. ................................................................................................... 62 
 
Table 19. Weighted Odds Ratios of Problem and Experience Pairs Within 120-240 Days of First 
Using Heroin. ................................................................................................................................ 63 
 
Table 20. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
120-240 Days of First Using Heroin. ............................................................................................ 64 
 
Table 21. Weighted Odds Ratios of Problem and Experience Pairs Within 150-270 Days of First 
Using Heroin. ................................................................................................................................ 65 
 
Table 22. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
150-270 Days of First Using Heroin. ............................................................................................ 66 
 
Table 23. Weighted Odds Ratios of Problem and Experience Pairs Within 180-300 Days of First 
Using Heroin. ................................................................................................................................ 67 
 
Table 24. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
180-300 Days of First Using Heroin. ............................................................................................ 68 
 
Table 25. Weighted Odds Ratios of Problem and Experience Pairs Within 210-330 Days of First 
Using Heroin. ................................................................................................................................ 69 
 
Table 26. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
210-330 Days of First Using Heroin. ............................................................................................ 70 
 
Table 27. Weighted Odds Ratios of Problem and Experience Pairs Within 240-360 Days of First 
Using Heroin. ................................................................................................................................ 71 
 
Table 28.  Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
240-360 Days of First Using Heroin. ............................................................................................ 72 
 
Table 29. Weighted Odds Ratios of Problem and Experience Pairs Within 270-390 Days of First 
Using Heroin. ................................................................................................................................ 73 
 
Table 30. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
270-390 Days of First Using Heroin. ............................................................................................ 74 
 
Table 31. Weighted Odds Ratios of Problem and Experience Pairs Within 300-420 Days of First 
Using Heroin. ................................................................................................................................ 75 
 
Table 32. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
300-420 Days of First Using Heroin. ............................................................................................ 76 
 

 

viii 

Table 33. Weighted Odds Ratios of Problem and Experience Pairs Within 330-450 Days of First 
Using Heroin. ................................................................................................................................ 77 
 
Table 34. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
330-450 Days of First Using Heroin. ............................................................................................ 78 
 
Table 35. Sensitivity Analysis: Weighted Odds Ratios of Problem and Experience Pairs Within 
0-150 Days of First Using Heroin. ................................................................................................ 80 
 
Table 36. Sensitivity Analysis: Weighted Odds Ratio Confidence Intervals of Problem and 
Experience Pairs Within 0-150 Days of First Using Heroin. ........................................................ 81 
 
Table 37. Sensitivity Analysis: Weighted Odds Ratios of Problem and Experience Pairs Within 
150-300 Days of First Using Heroin. ............................................................................................ 82 
 
Table 38. Sensitivity Analysis: Weighted Odds Ratio Confidence Intervals of Problem and 
Experience Pairs Within 150-300 Days of First Using Heroin. .................................................... 83 
 
Table 39. Sensitivity Analysis: Weighted Odds Ratios of Problem and Experience Pairs Within 
300-390 Days of First Using Heroin. ............................................................................................ 84 
 
Table 40. Sensitivity Analysis: Weighted Odds Ratio Confidence Intervals of Problem and 
Experience Pairs Within 300-390 Days of First Using Heroin. .................................................... 85 
 
Table 41. Problems and Experiences of Heroin Use Based on Diagnostic Criteria. .................... 99 
 
Table 42. DSM-5 Use Disorder Criteria and Corresponding Heroin Use Disorder Problems and 
Experiences. ................................................................................................................................ 100 
 
Table 43. Description of Confirmatory Factor Analysis Measurement Equivalence ................. 100 
 
Table 44. Characteristics of Newly Incident Heroin Use Sample. ............................................. 106 
 
Table 45. Description of Problems and Experiences involving Heroin Use Disorder. .............. 107 
 
Table 46. Description of DSM-5 Criteria for Heroin Use Disorder. .......................................... 108 
 
Table 47.  Goodness of Fit Indices Evaluating Measurement Equivalence of Heroin Use Disorder 
Level Across Sex Among Newly Incident Heroin Users (Without Analysis Weights). ............ 109 
 
Table 48. Goodness of Fit Indices Evaluating Measurement Equivalence of Heroin Use Disorder 
Level Across Sex Among Newly Incident Heroin Users with Analysis Weights. ..................... 110 
 
Table 49. Goodness of Fit Indices Evaluating Measurement Equivalence of DSM-5 Criteria for 
Heroin Use Disorder Level Across History of Extra-Medical Prescription Opioid Use Among 
Newly Incident Heroin Users (Without Analysis Weights). ...................................................... 111 
 

 

ix 

Table 50. Goodness of Fit Indices Evaluating Measurement Equivalence of DSM-5 Criteria for 
Heroin Use Disorder Level Across History of Extra-Medical Prescription Opioid Use Among 
Newly Incident Heroin Users with Analysis Weights. ............................................................... 112 
 
Table 51. Characteristics of Newly Incident Stimulant Use Sample. ......................................... 126 
 
Table 52. Counts and Discordant Pair Odds Ratios of Newly Incident Heroin Use Preceding 
Newly Incident Stimulant Use. ................................................................................................... 127 
 
Table 53. Post Hoc Interval Width Exploration: Unweighted Discordant Pair Odds Ratios of 
Newly Incident Heroin Use Preceding Newly Incident Stimulant Use. ..................................... 128 
 
Table 54. Year-Specific Numbers of NISUs Stratified by Analysis Weight Quartiles and 
Interval-Specific Exposure Status. .............................................................................................. 129 
 
Table 55. Year-Specific Odds Ratios of NISUs Stratified by Analysis Weight Quartiles and 
Interval-Specific Exposure Status ............................................................................................... 130 
 
Table 56. Manuscript 1 - Stata Code .......................................................................................... 151 
 
Table 57. Meta-Analysis of Newly Incident Heroin Use to Dependence Probabilities. ............ 158 
 
Table 58. Unweighted Odds Ratios of Problem and Experience Pairs Within 0-90 Days of First 
Using Heroin. .............................................................................................................................. 159 
 
Table 59. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
0-90 Days of First Using Heroin. ................................................................................................ 160 
 
Table 60. Unweighted Odds Ratios of Problem and Experience Pairs Within 1-120 Days of First 
Using Heroin. .............................................................................................................................. 161 
 
Table 61. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
1-120 Days of First Using Heroin. .............................................................................................. 162 
 
Table 62. Unweighted Odds Ratios of Problem and Experience Pairs Within 30-150 Days of 
First Using Heroin. ...................................................................................................................... 163 
 
Table 63. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
30-150 Days of First Using Heroin. ............................................................................................ 164 
 
Table 64. Unweighted Odds Ratios of Problem and Experience Pairs Within 60-180 Days of 
First Using Heroin. ...................................................................................................................... 165 
 
Table 65. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
60-120 Days of First Using Heroin. ............................................................................................ 166 
 
Table 66. Unweighted Odds Ratios of Problem and Experience Pairs Within 90-210 Days of 
First Using Heroin. ...................................................................................................................... 167 

 

x 

Table 67. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
90-210 Days of First Using Heroin. ............................................................................................ 168 
 
Table 68. Unweighted Odds Ratios of Problem and Experience Pairs Within 120-240 Days of 
First Using Heroin. ...................................................................................................................... 169 
 
Table 69. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
120-240 Days of First Using Heroin. .......................................................................................... 170 
 
Table 70. Unweighted Odds Ratios of Problem and Experience Pairs Within 150-210 Days of 
First Using Heroin. ...................................................................................................................... 171 
 
Table 71. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
150-270 Days of First Using Heroin. .......................................................................................... 172 
 
Table 72. Unweighted Odds Ratios of Problem and Experience Pairs Within 180-300 Days of 
First Using Heroin. ...................................................................................................................... 173 
 
Table 73. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
180-300 Days of First Using Heroin. .......................................................................................... 174 
 
Table 74. Unweighted Odds Ratios of Problem and Experience Pairs Within 210-330 Days of 
First Using Heroin. ...................................................................................................................... 175 
 
Table 75. Unweighted Odds Ratios of Problem and Experience Pairs Within 210-330 Days of 
First Using Heroin. ...................................................................................................................... 176 
 
Table 76. Unweighted Odds Ratios of Problem and Experience Pairs Within 240-360 Days of 
First Using Heroin. ...................................................................................................................... 177 
 
Table 77. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
240-360 Days of First Using Heroin. .......................................................................................... 178 
 
Table 78. Unweighted Odds Ratios of Problem and Experience Pairs Within 270-390 Days of 
First Using Heroin. ...................................................................................................................... 179 
 
Table 79. Unweighted Odds Ratios of Problem and Experience Pairs Within 270-390 Days of 
First Using Heroin. ...................................................................................................................... 180 
 
Table 80. Unweighted Odds Ratios of Problem and Experience Pairs Within 300-420 Days of 
First Using Heroin. ...................................................................................................................... 181 
 
Table 81. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
300-420 Days of First Using Heroin. .......................................................................................... 182 
 
Table 82. Unweighted Odds Ratios of Problem and Experience Pairs Within 330-450 Days of 
First Using Heroin. ...................................................................................................................... 183 
 

 

xi 

Table 83. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 
330-450 Days of First Using Heroin. .......................................................................................... 184 
 
Table 84. MPlus Code to Evaluate Measurement Equivalence for Heroin Use Disorder Among 
Newly Incident Heroin Users. ..................................................................................................... 186 
 
Table 85. Stata Code Used to Yield Case Crossover Discordant Pair Odds Ratios Displayed 
within Manuscript 3. ................................................................................................................... 189 

 

xii 

LIST OF FIGURES 

 
Figure 1. Chemical Structure of Heroin .......................................................................................... 5 
 
Figure 2. Centers for Disease Control and Prevention: 3 Waves of the Rise in Opioid Overdose 
Deaths ........................................................................................................................................... 14 
 
Figure 3. Five Main Rubrics of Epidemiology as Applied to Drug Dependence ......................... 16 
 
Figure 4. Description of Heroin Use Disorder in the Past Year among Individuals Aged 12 or 
Older, by Age Group: Percentages. .............................................................................................. 19 
 
Figure 5. Probability of Transition from Heroin First Use to Dependence within One Year. ...... 21 
 
Figure 6. Declining Response Rates of Federally-Funded National Surveys ............................... 29 
 
Figure 7. Declining Response Rates - National Surveys on Drug Use and Health ...................... 30 
 
Figure 8. Flow of Participants in Newly Incident Heroin Use Case Ascertainment. ................... 43 
 
Figure 9. Odds Ratios with 95% Confidence Intervals of Problem and Experience Pairs Stratified 
by Analysis Weight Quartiles. ...................................................................................................... 79 
 
Figure 10. Diagram of Confirmatory Factor Analysis Model Testing Measurement Equivalence 
of Problems and Experiences of Heroin Use Disorder Among Newly Incident Heroin Users. . 101 
 
Figure 11. Flow of Participants in Newly Incident Heroin Use Case Ascertainment. ............... 105 
 
Figure 12. Diagram of Case Crossover Design Studying Newly Incident Heroin Use and Excess 
Risk of Newly Incident Stimulant Use. ...................................................................................... 123 
 
Figure 13. An epidemic curve for acute-onset disease following a point exposure. .................. 135 
 
Figure 15. Infographic Exerts for Citizen Science-Based Initiative ........................................... 143 
 
Figure 16. Flow of Heroin Use Disorder Questionnaire Based on DSM-IV Criteria ................. 156 
 

 

 

xiii 

KEY TO ABBREVIATIONS 

 
 

CFA – Confirmatory Factor Analysis  

CI – Confidence Intervals  

DC – District of Columbia  

DSM – Diagnostic and Statistical Manual  

EMPOU – Extra-Medical Prescription Opioid Use 

FI – Field Interviewer 

 

HUD – Heroin Use Disorder 

NIHU – Newly Incident Heroin Use 

NISU – Newly Incident Stimulant Use 

NSDUH – National Surveys on Drug Use and Health  

OR – Odds Ratio 

PE – Problems and Experiences 

PPR – Prescription Pain Relievers 

SC – Psychostimulants including Cocaine 

US – United States 

 

 

 

xiv 

CHAPTER 1 INTRODUCTION 
 
1.0 Introduction 
 

The use of heroin and other opioids has become a major public health concern, in the US and 

in countries around the world. In many countries, individuals who use opioids experience higher 

crude mortality rates and a higher risk of death than the general population (2)(3)(4). 

Epidemiologic research from an international systematic review suggests all cause crude 

mortality among regular or dependent users of heroin to be 2.09 per 100 person-years (95% 

CI=1.93, 2.26), with a standardized mortality rate at 14.66 (95% CI=12.82, 16.50) (2).  

The position of heroin use as an important health concern is further highlighted by research 

on the disability-adjusted life years (DALYs) for opioid use disorders, recently estimated 

9,152,000 (95% CI=7,066,000, 11,443,000)(5). Motivated by the morbidity and mortality among 

heroin users, this dissertation will delineate implications of heroin use and use disorder.   

Over the past two decades the United States (US) has and continues to experience opioids 

crises in that the use of prescription opioids, heroin, and synthetic opioids have sequentially 

increased in nearly all demographic sub-groups. Evidence of these opioid crises is seen by the 

nearly five-fold increase in opioid overdose deaths from 1999 to 2016 (3). Heroin is a prominent 

driver of the mortality increase, as heroin-related overdose deaths alone have increased five-fold 

from 2010 to 2016 (6). 

In response to the opioids crises, public health has targeted the healthcare sector to 

promote responsible prescribing practices of opioids and has increased awareness among 

communities and the public safety sector. Even so, public health has yet to alleviate the heroin 

use crisis. A strategic effort to understand and intervene on heroin use is imperative for several 

reasons: 1) alone heroin was responsible for roughly 15,500 deaths in 2016 (6), 2) heroin use 

disorder, infection, psychiatric comorbidities, and poor quality of life are just some of the 

 

1 

outcomes associated with heroin use, and 3) heroin is increasingly being laced with synthetic 

opioids (i.e., fentanyl and its derivatives), which increases risk of overdose. At present, the 

presence of adulterants such as fentanyl is may or may not be known to users.  

Through this dissertation, natural history of heroin use disorder (HUD) is investigated 

from the first months of use to the development of use disorder with consideration of heroin use 

disorder dimensionalities and drug use trajectories. The overarching goal is to understand the 

HUD natural history from the most granular perspective possible to provide targeted public 

health primary prevention efforts. Additional implications may also lend themselves to 

clinicians, health professionals, and communities by providing early tell-tale signs of HUD 

forming in individuals.  

Rationale for HUD investigation is evident from the sparse epidemiologic literature on its 

natural history, measurement equivalence, and trajectories. Past studies have often been limited 

by the reliance on treatment-seeking samples, often with left-censoring problems and selection 

bias, or by HUD case definitions or lifetime history recall bias (2)(7)(5). Given these study 

design limitations and the dynamic and unprecedented landscape of the nation’s current heroin 

and other opioid crises, new HUD research is needed. 

1.1 Specific Aims  

Specific Aim 1. To estimate the pairwise association between problems and experiences of 

heroin use disorder, interval by interval, forward across the first months of use.  

Specific Aim 2. To assess measurement equivalence of heroin use disorder, assuming a 

single dimension, with evaluation of non-equivalence for sex and extra-medical prescription 

opioid use subgroups.  

Specific Aim 3. To investigate the degree to which onset of heroin use might precipitate 

onset of extra-medical stimulant use.  

 

2 

At this stage, four abbreviations can be introduced. HUD is heroin use disorder. NIHU is 

newly incident heroin user. EMPPR and EMSC refer to extra-medical use of prescription pain 

relievers (PPR) or psychostimulants including cocaine (SC).  

 

 

 

3 

CHAPTER 2 BACKGROUND AND SIGNIFICANCE  

2.0 Heroin 

 

2.0.1  General Introduction  

Heroin is a semi-synthetic opioid derived from the opium poppy, Papaver somniferum, 

which is typically grown in warm, dry climates such as southern Asia and Latin America (8). 

From the poppy’s seed pod, morphine is extracted and then converted to diacetylmorphine (i.e., 

heroin), which may be sold as a white or brown powder or a sticky black tar (8)(9). Individuals 

can then smoke, snort, inject, or orally administer the product (9). Notably, “on the streets” 

heroin is frequently “cut” with products ranging from sugar or powdered milk to synthetic 

opioids such as fentanyl (9). “Street” names include but are not limited to black tar, horse, 

smack, thunder, big H, and hell dust (10). In addition, the term “narcotic” from the Greek word 

for “stupor”, also refers to opioids.   

The US Drug Enforcement Administration’s (DEA) Controlled Substances Act (CSA) 

categorizes drugs into five schedules based on the drug’s accepted medical use and potential for 

abuse or dependence. Schedule I drugs have the highest potential for dependence and no 

accepted medical use in the US, whereas Schedule V drugs have the relatively lowest potential 

for dependence and may be used medically (10). Often Schedule V drugs are sold over the 

counter without prescription, as is the case of cough syrup containing small doses of opioids. 

Heroin is classified as Schedule I and is considered illegal in the US (9)(10).  

2.0.2  Heroin Pharmacology & Physiology 

The chemical structure of heroin, also known as diamorphine or diacetylmorphine, is 

similar to that of morphine, but can be distinguished by the addition of two acetyl groups (Figure 

1) (11)(12). The lipid solubility  and quick onset of effects distinguish heroin and might promote 

its having a higher probability for dependence (11). In the body heroin is quickly metabolized to 

 

4 

morphine, since heroin’s half-life is roughly two minutes (13). Thus, the physiological effects of 

heroin and morphine are comparable. The lipid solubility and quick onset of effects distinguish 

heroin as having a higher probability for dependence (13). Furthermore, the probability of 

dependence is exacerbated when heroin is administered intravenously when these effects are felt 

even sooner than other routes of administration (14).  

Figure 1. Chemical Structure of Heroin 

Link: 
https://pubchem.ncbi.nlm.nih.gov/image/imagefly.cgi?cid=5462328&width=300&height=300 
Heroin: C21H23NO5, also known as diamorphine or diacetylmorphine. 
Source: Pub Chem: Open Chemistry Database, 2018 (12). 

 

 

Opioid receptors are found throughout the central and peripheral nervous systems and are 

intended to be stimulated by endogenous opioids (15). That is, the identification of opioid 

receptors in the brain led scientists to research what in the brain could be a natural activator of 

the receptors. In the process, enkephalins and endorphins were identified as neurotransmitters 

with actions similar to morphine (11). These endogenous (e.g., endorphins) and exogenous (e.g., 

heroin) opioids may act on four different classes of opioid receptors, mu, kappa, delta, sigma, 

categorized by their prototype agonist (15). All classes are G-coupled protein receptors and can 

bind agonists, partial agonists, and antagonists (14). 

More specifically, the mu-opioid receptor agonists include morphine and heroin, which 

primarily produce rewarding analgesic effects and with locations that include brainstem and 

 

5 

medial thalamus (16). Other functions of mu-opioid receptors include respiratory depression, 

sedation, decreased gastrointestinal motility, physical dependence, and rewarding effects (e.g., 

subjective high, euphoria) (15). Furthermore, the binding of heroin to mu-opioid receptors in the 

central nervous system activates the release of dopamine (i.e., neurotransmitter related to desire) 

in the nucleus accumbens, a region that is important in the brain’s reward circuit (17)(18). 

Additional brain regions important in the physiological effects of heroin include the locus 

coeruleus, amygdala, dorsal raphe nucleus, and ventral tegmental area (16). Altogether, these 

areas are involved in the body’s stress response, reward circuit, emotions and motivation, 

serotonin activation (i.e., neurotransmitter related to satiety and inhibition), and cognition 

(16)(17)(18). 

Chronic use of exogenous opioids (e.g., heroin) affects the brain differently in that the 

pulsatile and phasic activation of mu-opioid receptors can be disrupted and may lead to the 

dysregulation of receptor recycling and abnormal cell desensitization (16). These neuronal 

adaptations, among others, may result in HUD, tolerance, and a withdrawal syndrome, although 

a single definitive mechanism has not been confirmed (16). 

Common functional consequences of heroin use include: liver disease, gastrointestinal 

disruption (e.g., constipation), impaired visual acuity, sclerosed veins (i.e., commonly referred to 

as “tracks”) peripheral edema, infections from unsafe injection practices (e.g., HIV, hepatitis, 

bacterial endocarditis). Disrupted sexual function and reproduction (e.g., irregular menses) also 

have been observed (19). 

2.1 Opioids Overview 

 
Opioids include natural, semi-synthetic, and synthetic compounds that bind to opioid 

receptors in the brain, with effects such as pain relief, as well as anti-diarrheal and anti-tussive 

effects (11). Natural opioids include codeine and morphine, which are naturally occurring 

 

6 

compounds from the opium poppy. Semi-synthetic opioids, such as heroin, are those derived 

from opium poppy compounds. Both natural and semi-synthetic opioids are often referred to as 

opiates. More specifically, opiates are the naturally occurring alkaloids found in the opioid 

poppy (15). Synthetic opioids are not derived from the opium poppy but bind to opioid receptors 

and produce similar physiological effects (11). Common synonym terms for opioids include: 

narcotics, pain relievers, and analgesics, although some drugs in these drug sub-types are not 

opioids. 

2.2 History  

2.1.1 From Opium Cultivation to Heroin Synthesis 
 

The date of opium poppy cultivation is uncertain, but may be seen as early as the 9th 

century BC in Homer’s description of Helen, daughter of Zeus, preparing a drug to forget grief in 

“The Odyssey” (1). While authors have debated over whether ambiguous ancient writings have 

described opium, there is evidence of opium poppy cultivation by the Sumerians in third century 

BC in the vicinity of contemporary Iraq (1). Opium and the poppy were referred to as “gil” and 

“hul gil” meaning “joy” and “plant of joy”, respectively (1). During this time, opium was 

primarily was used in religious rituals and was considered to be healing (1).  

Earlier, in 1500 BC, the Ebers Papyrus has been interpreted with a reference to opium for 

its ability to prevent children from crying and for surgical pain relief, although opium was not 

widely used. Thereafter, opium was spread to India, China, Asia Minor, and all of Europe 

throughout the 8th-13th centuries AD. It was not until the 16th century though, that drug abuse and 

tolerance became mentioned in texts aout modern Turkey, Egypt, Germany, and England. Later, 

smoking opium was made popular by China’s ban on smoking tobacco during the 17th century 

(11)(1). Despite efforts to decrease the sale and use of opium, trade became popular with the 

British and French (1).  

 

7 

In 1806, F. Sertürner discovered morphine as an active ingredient in opium, which was 

then named from the Greek god of dreams, Morpheus (11). Not long after in 1832, codeine was 

isolated form opium by Pierre Jean Robiquet and named from the Greek word “kodeia” which 

translated to “poppy head” (11). Use of these compounds increased in prevalence with the 

invention of the hypodermic syringe and hollow needle in the early 19th century. After the Civil 

War in 1861-1865, opioid use greatly increased as the hypodermic needle was available 

wholesale (20). At this time opiates could be used for surgery, postoperative and chronic pain, 

along with general anesthetics (1). In addition, the various medical preparations of opium, such 

as cough syrups and pain-killers, further helped to increase opium and morphine’s popularity 

(20).  

Via chemical experimentation heroin was synthesized in 1874 by Charles Romley Alder 

Wright, but initially there was no practical use identified (21). It was not until 1897 that Felix 

Hoffmann made heroin in effort to develop a safer, non-addictive opiate. Hoffmann, a chemist 

for the Bayer Company in Germany, made way for Bayer to introduce the compound in 1889 as 

heroin, named for the German word “heroisch”, meaning “hero” (1)(21). At this time, heroin was 

believed to be more effective than morphine without the potential to develop dependence (11). In 

fact, heroin was recommended for the treatment of morphine dependence  (also termed 

‘morphinomania’) and was generally regarded as an overall safe alternative included in cough 

suppressants and various medical preparations (20).  

2.2.2 The History of Heroin in the United States 
 

Despite the initial belief that heroin was a safe alternative to morphine, by the 20th 

century the development of tolerance, dependence, and intense withdrawal syndrome were 

eventually realized (11)(21). Heroin’s eventual rise in popularity was set in the context the US’ 

increasingly opioid dependent population. Nearly one percent of Americans were dependent on 

 

8 

opioids given the lack of government regulations, drug availability, inexpensiveness, 

effectiveness, socially acceptable environment, and the freedom of individuals to self-medicate 

(20). Notably, at the time it was not heroin that was driving prevalence of opioid dependence, but 

rather morphine and the oral-intake and smoking of opium (11). Many of the chronic opioid 

users were women who became dependent via self-medication with over the counter opioids or 

had received legal physical prescriptions (11). 

As a result of growing medical concerns and advance legislation at the local and state 

level, in 1906 the Pure Food and Drugs Act was passed which helped increase consumer 

protection by banning adulterated and mislabeled products (20). The warnings of the 

indiscriminate use of opium slowly increased and physicians were largely thought to be 

responsible for the population’s opioid dependence (11). Physicians’ trust of opioids possibly is 

understandable, given the pharmaceutical recommendation that morphine was an advantageous 

treatment option for alcohol dependence. Morphine being inexpensive and available for self-

medicating only exacerbated the population’s opium problem (11)(20).   

In 1914 the Harrison Act was implemented and served to regulate and tax the production, 

importation, and distribution of opiates and coca products, as well as their derivatives and 

preparations (11) (20). The act led to a significant shift in the environment from one that allowed 

opioids to be purchased at local pharmacies or from physicians to one that created black markets 

with illegal drug dealers (11). Physicians were still able to prescribe opioids to maintain an 

individual’s dependence, yet apparently few would do so given the drastic change in attitudes 

(11). Under these circumstances, heroin became the most popular opioid that pleased both 

dealers and users for its potency and easy concealment. Furthermore, the large price increase 

from change in markets (i.e., from pharmacies to illegal dealers) led users to inject heroin to 

experience the drug’s greatest effect (11).The Harrison Act essentially criminalized drug 

 

9 

dependence rather than treat it as a health issue and led to the public’s stigmatization of 

individuals dependent on drugs (11)(20). Soon after, in 1924, the US Congress banned the 

importation of opium for the purpose of manufacturing heroin in 1924 (21).  

Between World War I and World War II (WWII), that is around 1918-1939, heroin use in 

the US declined among women and others in the general public, and became more limited to 

subgroups considered “outsiders”, such as musicians and ethnic minorities (22)(23). WWII 

interrupted trade routes from opium-producing countries and caused a five year hiatus of heroin 

use in the US, but was followed by an increased market around 1947-1951 due to a renaissance 

of heroin that was of high quality and low cost (23). Moreover, the decades following WWII saw 

widespread use of various drugs to the point that high school youth became highly affected (22). 

Consequently, the increased demand for heroin resulted in the price increasing.  

In response, during this time, the US strengthened its federal controls over drug use, 

partially in response to the public’s heightened awareness of criminal activity associated with 

heroin use (24). Social cohesion amongst users and organized crime that sold drugs disintegrated 

and the market opened up for independent drug dealers. Thereafter, during the late 1960s, the 

nation faced a decline in heroin supply which drove up the price (23).  

Looking forward to 1971, the Vietnam War of the 1960s helped fuel US experience with 

heroin. An estimated 10-15% of American troops deployed in Vietnam became dependent on 

heroin after consumption of opioids made available at low cost, sometimes in the form of opium 

to be smoked in a traditional pipe, but also in the form of heroin, also often smoked (11). The 

purity, availability, and inexpensive-nature of heroin contributed to it being considered the “main 

drug” of almost 10% of those using (25). Notably, heroin use is usually paired with use of other 

drugs and alcohol. Nevertheless, in response to the troops’ heroin use the US Department of 

Defense relied on Dr. Jerome Jaffe’s urine-screening program to test for the presence of heroin in 

 

10 

individuals about to return home to the US. Troops that tested positive for heroin were briefly 

detained for treatment and sent home upon a negative screen.  

Following the screening effort, Dr. Lee Robins’ follow-up study estimated that only one 

to two percent of veterans used opioids eight to 12 months after returning home to the US, which 

was roughly the same percentage of individuals using opioids at the start of military service 

(11)(25). Contrary to popular belief, follow-up showed that continued and compulsive heroin use 

and dependence are not inevitable in the natural history of heroin use (25)(26).  

The 1970s remained an influential time in heroin’s history as criminalization of drug use 

continued with the implementation of the Drug Enforcement Administration’s Controlled 

Substances Act in 1970. This classified heroin as a Schedule I drug, among the five drug 

schedules, making it illegal to produce or consume (10). Also at this time, Turkey became a 

major source of opium, which was then converted to heroin in the southern part of France (i.e., 

known as “the French connection”) (11). To taper American’s use of heroin, in 1972 the US paid 

Turkey $35 million to ban opium cultivation and simultaneously partnered with the French to 

ultimately reduce the US’ heroin supply.  

To recap this era, President Nixon took office in 1969 and diverted much attention to the 

“War on Drugs” which was declared in 1971 (24). In response to the public’s concern for 

increased drug use and the US military’s heroin use during the Vietnam War, Nixon headed 

important initiatives to diminish the US heroin epidemic of the 1960s/1970s: 1) The elimination 

of Turkey as the primary source of US heroin and attacking the “French Connection” where 

opium was converted to heroin, 2) The creation of federal institutions focused on drug use (i.e., 

Special Action Office for Drug Abuse Prevention (SAODAP) and National Institute on Drug 

Abuse), and 3) Overall afforded resources for drug issue by increasing federal spending from 

 

11 

$83 million in 1969 to $730 million in 1973 (24). All and all, this time impacted the nation’s 

relationship with drugs which is still felt today in the 21st century.  

The reduction in US heroin use was short lived and Mexico became the leading US 

supplier of heroin in 1975. Mexico’s heroin was commonly referred to as Mexican Brown or 

Black Tar and was characterized by its high purity and inexpensive nature. Although the US did 

work with Mexico to eradicate opium production, overall the 1970s heroin use epidemic was 

slow to decline. Following the 1970s the majority of US heroin came from Southwest Asia, 

Mexico, and South America. By the end of the 20th century, heroin use was considered rare with 

only a fraction of the population using (11).  

2.2.3 America’s Opioids Crisis 
 

Today the US is experiencing an opioids epidemic that has roots starting in the late 1990s 

when the call for improved pain management strengthened among various organizations and 

state medical boards (27)(28). For example, organizations such as the Institute of Medicine, the 

American Pain Society, and the US Veterans Health Administration advocated a message of 

“Pain: The Fifth Vital Sign”. Likewise, the Joint Commission on the Accreditation of Health 

Care Organizations supported this movement by compiling examples of standards (e.g., the 

Institute of Medicine’s “Pain: The Fifth Vital Sign”) and emphasized the need for pain awareness 

(28)(29). Importantly, the Joint Commission states that they never implemented the standard of 

pain as a fifth vital sign (28). These pain awareness efforts contributed to a large increase in 

opioid prescribing for non-cancer pain. 

Concurrently, the pharmaceutical intensified opioid prescribing by aggressively 

marketing opioids as highly effective, safe, and without adverse effect when directed by a 

physician (27). Purdue Pharma has received much of the public blame for its approaches to 

marketing and increasing market share of prescription opioids, but it was not the only vendor of 

 

12 

prescription opioids to take advantage of the opportunities created in the evolving American 

scene. In addition, there were processes such as prescribers writing prescription orders for 10-14 

days of post-surgery pain relief when 1-3 days might have sufficed. The result seems to have 

included an accumulation of unused prescription opioid dosage units in American homes, and 

made available for diversion toward extra-medical use (30). 

Importantly, the rise in opioid prescribing was followed by increases in opioid overdose 

deaths and dependence. By 2008, the US experienced one opioid-related death every 36 minutes 

(27). In response to prescription opioid concern, stricter prescribing practices and non-

pharmacological strategies for pain management were endorsed by the Joint Commission and 

various organizations (29). 

Despite the curtail of opioid prescriptions, an unintended failure of the success (e.g., 

responsible prescribing practices) was the second wave of the opioids epidemic in 2010 (Figure 

2) (31). This wave developed in the context of an already high prevalence opioid dependence so, 

to maintain opioid dependence and avoid withdrawal many turned to using heroin. The heroin 

use epidemic between 2010-2016 saw a five-fold increase in heroin overdose deaths (31). In 

2013, a third wave of the opioids epidemic emerged with the increased use of synthetic opioids 

(e.g., fentanyl). By 2016, overdose deaths due to synthetic opioids (e.g., fentanyl) surmounted 

that of heroin (31). Though given that the presence of synthetic opioids is often hidden within 

heroin and other drugs to “cut” the supply, heroin use remains a substantial threat today.  

 

13 

Figure 2. Centers for Disease Control and Prevention: 3 Waves of the Rise in Opioid Overdose 
Deaths 

 

Link: https://www.cdc.gov/drugoverdose/images/epidemic/3-waves-of-the-rise-in-opioid-
overdose-deaths.JPG 
Source: CDC, National Vital Statistics System Mortality File, 2017 (31). 

 

2.3 Heroin Use Disorder  

The American Psychiatric Association’s Diagnostic and Statistical Manual of Mental 

Disorders (DSM) outlines criteria for drug use disorders with 10 separate drug classes: alcohol, 

caffeine, cannabis, hallucinogens, inhalants, opioids, sedatives, hypnotics, anxiolytics, 

stimulants, tobacco, and other known drugs. DSM-IV differentiated ‘drug use’ and ‘drug 

dependence’ with abuse criteria focused on the negative consequences of drug use (e.g., 

problems with family), while dependence criteria involved experiences of tolerance, withdrawal, 

and uncontrolled use (17). The most current version of the DSM, DSM-5, has shifted to 

diagnosing use disorders as a single disorder rather than differentiated abuse from dependence as 

separate diagnostic categories. DSM-5 also encouraged a more dimensional approach, with mild, 

moderate, and severe sub-classifications. However, a majority of DSM-5 criteria overlap with 

those of DSM-IV (17). 

 

14 

The DSM-5 criteria for use disorders are categorized into four groupings: impaired control, 

social impairment, risky use, and pharmacological, and there are 11 criteria corresponding to use 

disorder symptoms (Table 1). Diagnoses are determined based on the number of symptoms 

present, with 0-1 being no disorder, 2-3 being mild, 4-5 being moderate, and 6 or more being 

severe use disorder (17)(19). Importantly, the change from DSM-IV to DSM-5 diagnoses 

signifies a departure from a dichotomized framework of dependence (i.e., an individual either 

has or does not have the use disorder) to a diagnostic gradation that acknowledges a potentially 

slow development of use disorder in attempt to identify disease early on (17)(19). In addition, 

tolerance and withdrawal are no longer deciding factors of dependence which allows for even 

earlier diagnosis and allows use disorder diagnoses for drugs that may not produce a withdrawal 

syndrome (17). In addition, while the presence of biomarkers is not necessary to diagnosis HUD, 

urine toxicology screening for heroin will remain positive for 12-36 hours after administration 

(17). 

Table 1. Diagnostic Criteria for Substance Use Disorders Defined by the Diagnostic and 
Statistical Manual. 

Grouping 

Impaired control 

Criterion 

Social impairment 

# 
1 
2 
3 
4 
5 
6 
7 
8 
9 
10 
11 
Number of symptoms present: 0-1 = no disorder, 2-3 = mild, 4-5 = moderate, and 6 or more = 
severe  
Source: American Psychiatric Association, 2013 (19). 

Using in larger amounts or for longer than intended 
Wanting to cut down or stop using, but not managing to 
Spending a lot of time to get, use, or recover from use 
Craving 
Inability to manage commitments due to use 
Continuing to use, even when it causes problems in relationships 
Giving up important activities because of use 
Continuing to use, even when it puts you in danger 
Continuing to use, even when physical or psychological problems 
may be made worse by use 
Increasing tolerance 
Withdrawal symptoms 

Risky use 

Pharmacological 

15 

 

 

2.4 The Main Rubrics of Epidemiology Applied to Heroin Use 

Adapted from Morris’ 1957 “seven uses of epidemiology”, Anthony’s 2002 “five main 

rubrics of epidemiology” detail the general research questions that must be answered to progress 

epidemiology and specifically that of drug dependence (Figure 3) (32)(33). The following 

section uses the five rubrics of epidemiology to organize the current state of heroin use in the 

United States. 

Figure 3. Five Main Rubrics of Epidemiology as Applied to Drug Dependence 

Link: https://acnp.org/wp-content/uploads/2017/11/C109_1557-1574.pdf 
Source: Anthony, 2002 (32).  

 

 

Quantity: The prevalence of heroin has increased worldwide over recent years. The 

crudest evidence of the increased quantity of heroin is the increased cultivation of the opium 

poppy and opium production (i.e., respectively, approximately 400,000 hectares and over 10,000 

tons in 2017) (30). Likewise, seizures of heroin surmounted to 91 tons globally in 2016 (4). 

Opioids in general remain the most threatening drug as 34 million and 19 million individuals 

used opioids and opiates in the past year in 2016, respectively (4).  

As mentioned previously, all-cause mortality among individuals who regularly use heroin or 

have HUD is estimated to be 2.09 per 100-person years (95% CI=1.93, 2.26) and the 

standardized mortality rate to be 14.66 (95% CI=12.82, 16.50) (2). Worldwide, the greatest 

prevalence of opioid use disorder is seen in South Asia (n=4,331,000) (5). Age standardized 

 

16 

prevalence estimates suggest the greatest burden to be Australia at 0.46% (95% CI=0.41, 0.53) 

(5). 

In the US, the Centers for Disease Control and Prevention details that 948,000 

individuals ages 12 and older used heroin in the last year (i.e., 0.4/100 persons) (6). Between 

2002 to 2016 the number of active cases of heroin dependence has increased steadily with 2015-

2016 reaching 574,000 individuals with dependence (Table 2) (34). In 2016, there were nearly 

15,500 fatal heroin-related overdose deaths (i.e., 5/100,000 American) (Table 3).  From 2015 to 

2016 alone, the heroin overdose death rate increased approximately 20% (6).  

Table 2. Population Projections of Newly Incident Heroin Users and Dependence 

Year Pair 

Newly Incident Heroin 

NIHUs Transitioned 

to Dependence 

Users 
104,000 
2002-2003 
114,000 
2004-2005 
98,000 
2006-2007 
150,000 
2008-2009 
162,000 
2010-2011 
163,000 
2012-2013 
149,000 
2015-2016 
Link: https://rdas.samhsa.gov/#/ 
Source: NSDUH Restricted Use Dataset.  
Newly incident heroin users: first use within 12 months of survey assessment.  

26,000 
17,000 
26,000 
39,000 
63,000 
57,000 
52,000 

Dependence 
Prevalence 
179,000  
213,000 
244,000 
288,000 
349,000 
433,000 
574,000 

 

 

 

 

 

 

 

17 

Table 3. United States Overdose Deaths by Drug.  
Data from United States Center for Disease Control and Prevention, CDC Wonder.  

Link: https://www.drugabuse.gov/related-topics/trends-statistics/overdose-death-rates 
Source: National Institute on Drug Abuse, 2018. 
Drugs Involved in U.S. Overdose Deaths* - Among the more than 72,000 drug overdose 
deaths estimated in 2017*, the sharpest increase occurred among deaths related to fentanyl and 
fentanyl analogs (synthetic opioids) with nearly 30,000 overdose deaths. Source: CDC 
WONDER. 

 

 

Location: According to the National Surveys on Drug Use and Health (US), which are 

intended to be nationally representative samples, about 3,000 individuals ages 12 to 17 and 

88,000 individuals ages 18 to 25 used heroin recently in 2016 (35). In the same year, 383,000 

individuals ages 26 and older used heroin recently (35). Heroin use disorder (HUD) trends mirror 

the between age group comparisons with 1,000, 152,000, and 473,000 individuals having HUD 

in the respective age groups 12-17, 18-25, and 26 and older (35). HUD age group trends from 

2002 to 2016 are depicted in Figure 4 below. 

 

 

 

 

 

18 

Figure 4. Description of Heroin Use Disorder in the Past Year among Individuals Aged 12 or 
Older, by Age Group: Percentages. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

Link: https://www.samhsa.gov/data/sites/default/files/NSDUH-FFR1-2016/NSDUH-FFR1-
2016.pdf 
Source: Key Substance Use and Mental Health Indicators in the United States (35).   

 

 

Past epidemiologic research has suggested drug use is a phenomenon predominantly 

among males (22)(36)(37), yet as of late, studies have suggested otherwise for several drugs 

(e.g., alcohol) (38)(39)(40). Male and female prevalence of heroin use once agape, has narrowed 

respectively from 2.4 and 0.8 persons/1,000 in 2002-2004 to 3.6 and 1.6 persons/1,000 in 2011-

2013 (41). While both males and females have experienced an increase in heroin use, as of 2016 

males age 25-44 experience the greatest burden of demographic subgroups with the highest death 

rate of 15.5/100,000 individuals (6). 

Additional location variations in heroin use are seen by urbanicity with larger more urban 

counties experiencing higher age-adjusted heroin overdose death rates. That is, in 2016 large 

metropolitan counties with populations of one million or more individuals experienced 5.3 and 

 

19 

6.1 deaths/100,000 in central and fringe areas, repetitively (6). Whereas micropolitan counties 

with populations of 10,000 to 50,000 individuals experienced 3.6 deaths/100,000 (6). Likewise, 

metropolitan and non-metropolitan counties of various populations fall within a respective 

gradient. Variations in heroin overdose death rates by ethnic self-identification depict non-

Hispanic white individuals as experiencing the largest death rates (42). While individuals of all 

ethnicities have experienced increases in heroin use and related outcomes, the largest increases 

have been for non-Hispanic whites (6)(42). 

Causes & Mechanisms: To study heroin use and related outcomes (e.g., dependence, 

overdose) causal agents, relatively distal correlates, and the circumstances and processes must be 

considered. By far, the strongest risk predictor of heroin use identified by epidemiologic research 

and subsequently emphasized by the media is prior prescription opioid use and the like (e.g., 

opioid use disorder) (6)(43)(44). Other suggested risk predictors include poly-drug use or 

dependence – excluding opioid pain relievers, compared to no past year use or drug use disorder, 

past year cocaine use or use disorder has the strongest association to heroin use disorder with an 

adjusted odds ratio of 14.7 (95% CI = 7.4, 29.2)(44). In other words, cannabis, alcohol, and 

psychotherapeutic drugs increase the risk of heroin use and use disorder, but have a lower 

associated risk than cocaine and prescription opioids (44). Related yet more distal etiologic 

correlates of heroin use may include psychiatric comorbidities such as depression and anxiety 

disorders, chronic pain, childhood trauma or adverse experiences, and neighborhood poverty 

(45)(46)(47)(48) (49).   

The natural history of heroin use disorder has not been thoroughly detailed based on the 

current study’s literature review. Among all drug use disorders, that of heroin has the highest 

burden of disease with mortality rates comparable to that of the elderly (50). Notably, the risk of 

heroin use disorder within 12 months of first using the drug has been estimated to be 

 

20 

approximately 30.2% (95% CI = 22.9, 37.5) (Figure 5) (51), which is almost as large as 

corresponding risk estimates for tobacco products. This large HUD probability for heroin has a 

parallel in animal models. Across several different strains for rat models, the animals will self-

administer heroin, as is the case for non-human primates (52).  

Figure 5. Probability of Transition from Heroin First Use to Dependence within One Year. 
Data from United States National Surveys on Drug Use and Health.  

Link: https://www.drugabuse.gov/related-topics/trends-statistics/overdose-death-rates 
Source: JAMA Psychiatry, 2018.  

 

 

Prevention & Control: As the US continues to experience the current opioids crisis, 

prevention and control continue to be at the forefront of public health initiatives. Medication 

assisted treatment such as methadone, buprenorphine, and suboxone play an important role in 

treating heroin use disorder, along with naloxone for emergency overdose reversal (53). 

Additionally, medication assisted treatment should be coupled with counseling (e.g., cognitive 

behavioral therapy) to increase the likelihood of better health outcomes (41)(53). 

 

21 

To shift to a more nuanced review of public health in the face of the opioids crisis, the 

US federal government is a crucial structural component of the HUD treatment. Syringe services 

programs are aimed to control infectious disease (e.g., human immunodeficiency virus) spread 

among injection drug users. As of 2016, the federal ban on such harm reduction programs has 

been lessened and funds have been directed towards the opioids crisis (54). Yet, there is still a 

tangible need for more money to be directed to community-based harm reduction programs.   

Distal prevention and control methods have also included stricter opioid prescribing policies, the 

elimination of pain as a vital sign, and improved chronic pain management (28). Public health 

should consider increasing citizen science programs, which aims to connect researchers and 

citizens for crowdsourcing, program implementation, and community-tailored interventions (55).   

2.5 Significance and Purpose 

Upon literature review, there is currently no available epidemiologic work detailing HUD as 

disaggregated indicators of the latent construct to assess the pairwise associations and 

measurement equivalence across subgroups. This study is also the first to investigate whether 

heroin use onset precipitates excess risk of stimulant use onset. Conclusions drawn from this 

work will provide a more in-depth understanding of HUD as a syndrome (via paired 

combinations of potential symptoms) and as a dimension (via the latent structure analysis). It 

will illuminate whether a DSM-IV-based instrument for measuring HUD is comparable for both 

sexes and for those with a history of EMPOU. Additionally, conclusions drawn from estimating 

risk of stimulant onset based on heroin onset will be useful in beginning to identify 

vulnerabilities associated with heroin use. Said information will shed light on the natural history 

of HUD for more informed public health secondary and tertiary intervention strategies.  

The current study is targeted at understanding the “how?” or mechanisms of HUD 

epidemiology – that is, what sequence of conditions or processes lead to the development of 

 

22 

HUD (32). The opioids crisis continues to be on an upward trend and has become more severe 

with the increasing presence of heroin adulterated with synthetic opioids such as fentanyl. 

Increased mortality and morbidity are concerning and will continue to be public health concerns 

until interventions can effectively prevent and control the epidemic. 

 

23 

CHAPTER 3 MATERIALS – NATIONAL SURVEYS ON DRUG USE AND HEALTH 
 
3.0 Chapter Overview 
 

This chapter is organized into a standard set of sections for epidemiology research 

reports, beginning with a specification of a study population in conceptual terms (as opposed to 

strictly operational or sampling terms). Next, the cross-sectional design and multi-stage area 

probability sampling approach are described briefly, with a citation on more detailed online 

methods monographs. A section on human subjects’ protection procedures describes consent of 

the participants (and assent/consent for minors), with a paragraph on implications for 

participation levels. The next sections address levels of participation, year by year, and an 

apparent decline in those levels over time, which deserves discussion in later sections, as well as 

assessment procedures for the main constructs under study. The final section of the chapter 

provides an introduction to the use of analysis weights that account for both selection 

probabilities and also post-stratification adjustments to US Census distributions. It then describes 

issue of variance estimation and the several approaches used to account for non-independence of 

observations in complex survey designs of the type used here. Then, a final set of paragraphs 

provide an overview of the statistical approaches in the form of cross-classification and 

contingency table analyses, the analysis of survey data for case-crossover research when there 

are analysis weights involved, and the multivariate statistics required for study of dimensionally 

distributed levels of heroin problems when there is uncertainty about measurement equivalence 

across subgroups (e.g., males versus females). A final section of the chapter described the 

analysis software used for this project and provides appropriate citations. 

3.1 The Population Under Study 
 

Chapter 2 has described global dimensions of heroin’s distribution across boundaries of 

multiple countries. Nonetheless, access to epidemiological data on heroin is more constrained 

 

24 

when parameters of data quality are taken into account (56). For this reason, this dissertation 

research project specifies a study population that consists of all United States dwelling unit 

residents of the 50 states and the District of Columbia, excluding those who are residents of 

institutional group quarters such as prisons and long-stay hospitals, as well as infants and 

children under age 12 years. This study population of non-institutionalized US residents can be 

sampled more or less rigorously by using multi-stage area probability sampling methods that 

start with state-level sampling frames. Institutionalized residents have been omitted for a 

combination of logistical reasons. These reasons include difficulty accessing prison and long-

stay hospital patient populations, informed consent constraints, and other impediments to 

institutional sample surveys of the type described in connection with the Epidemiologic 

Catchment Area (ECA) research on institutional and non-institutional populations in five US 

metropolitan areas, conducted more than 25 years ago (53).  

Infants and children have not been included due to the relatively low attack rates for self-

administration of heroin and other drugs at these ages. If bioassays are used, there might be a 

rationale for including young people (e.g., newborns exposed in uterus). In this project, no 

bioassays have been used. 

Others left out of these samples of national population surveys include migrants and others 

described as ‘homeless’ persons who are not regular residents of non-institutional group quarters 

such as the homeless shelters. Here again, there are logistical difficulties when one seeks to 

integrate data on these ‘no conventional residence’ population members with data on residents of 

conventional group quarters, as described when the ECA surveys studied ‘homeless’ individuals 

without a fixed abode (57). 

Additional information about issues of this type are described in the next section of this 

chapter, on sampling procedures. In addition, the exclusion of heroin users who are in the US 

 

25 

study population but not included in the multi-stage area probability samples is a topic for the 

Discussion sections of this project report. These sections follow presentation of the dissertation 

estimates and represent examples of limitations of the dissertation research project that require 

‘mixed methods’ research approaches in future investigations (58).  

3.2 Cross-Sectional Survey Design and Probability Sampling Approach 
 

The specified US population has been studied using a cross-sectional survey design with 

sampling and assessment procedures that make it possible to secure time sequences of events and 

processes that ordinarily are not considered in cross-sectional survey analyses. The Assessment 

section of this chapter provides more information about these time sequences and how they have 

been used for this project. 

The sampling approach and survey procedures for data collection have been designed by 

the sampling statisticians, epidemiologists, and other scientists employed by the US federal 

government for completion of the US National Surveys on Drug Use and Health (NSDUH). The 

NSDUH team also forged data sharing agreements and produced analysis-ready datasets that 

have made this dissertation research project possible. A detailed methods report on sampling is 

available online. Here, in brief, it can be summarized as follows, but interested readers will find 

useful details in the online reports (59):  https://www.datafiles.samhsa.gov/study-

publication/nsduh-2002-methodological-resource-book-mrb-nid14370) 

3.3 Human Subjects Protection  
 

Field interviewers (FIs) are IRB trained on objectives such as ethics, human subjects 

research, the IRB’s role, and the role of FIs in protecting participants’ rights. Additional FI 

training includes Confidential Information Protection and Statistical Efficiency Act (CIPSEA), 

records management training, and bilingual training for FIs who are bilingual. After one month 

of field work, FIs continue training to understand sampling procedures in order to protect 

 

26 

sampling integrity, challenging field situations, screening and interview procedures, how to 

answer respondent questions, and an overall quarterly review of project procedures and 

protocols. All new FIs must pass certification. FIs are trained to identify unlisted dwelling units, 

which are then entered into computer and selected for participation (60). When FIs struggled to 

obtain participation of a dwelling unit, refusal conversion procedures were in place which 

centered on professional answering of questions, conversion letters, returning to the dwelling 

unit at another time.  

FI assess drug use and dependence using DSM-IV criteria via institutional review board 

approved protocols for recruitment and standardized computer-assisted self-interviews The US 

Department of Health and Human Services Center for Behavioral Health Statistics and Quality 

generated the NSDUH open access analysis files with de-identified data after applying disclosure 

analyses designed to prevent re-identification of participants. Given the circumstances of open 

access and no contact with participants, the Michigan State University institutional review board 

ruled that plans to analyze these data qualified for the federal category “not human subjects 

research.” 

3.4 Participation Rates and Assessment Protocol 
 

Screening response rates were calculated based on a numerator of completed screenings 

and a denominator of total eligible dwelling units (Figure 7. Declining Response Rates - National 

Surveys on Drug Use and Health 6). Conditional interview response rates were calculated with 

numerators of completed interviews and denominators of the number of eligible respondents 

chosen through screening. Ineligible respondents were subtracted from the total. The 

unconditional interview response rate was calculated based on numerators of completed 

interviews and denominators of those that would have been eligible to be interview if all 

dwelling units had been screened properly. Overall response rates are based on the screening 

 

27 

response rate multiplied by the interview response rate. To maximize response rates participants 

receive $30 for completing the survey. Participation levels are generally between 65-70%.  

Year by year, response rates have declined yet compared to other federally funded national 

surveys rates for NSDUH are relatively large (Figure 7) (61). Non-response can be problematic 

if individuals who are newly incident heroin users are more likely to not participate. Research 

conducted in 1999 suggested that populations with high non-response rates do not always have 

high rates of drug use (61). While this research has not been replicated in recent years, it is 

hypothesized that sources of bias most likely cancel each other out.  

Possible explanations for declining response rates may include increasing number of two-

worker households, longer commuting times, increased prevalence of caller ID, and increased 

number of federal surveys. Another potential explanation is decreased survey effort due to 

increased survey costs (61). 

 

 

 

 

28 

Figure 6. Declining Response Rates of Federally-Funded National Surveys 

Source: “Declining response rates in Federal Surveys: Trends and Implications” (June 2016), 
Mathematic Policy Research – Final Report Volume 1. (61) 
Link: https://aspe.hhs.gov/system/files/pdf/255531/Decliningresponserates.pdf 

 

  

29 

 
 
 
 

 

 

 

 

 

 

 

 
 
 

 

Figure 7. Declining Response Rates - National Surveys on Drug Use and Health 

Conditional Interview  Unconditional Interview 

72.0 
71.1 

 

57.6 
55.5 

Survey Year 

2015a 
2016 a 

Screening 

80.0 
78.1 

Source: “Declining response rates in Federal Surveys: Trends and Implications” (June 2016), 
Mathematic Policy Research – Final Report Volume 1. (61) 
Link: https://aspe.hhs.gov/system/files/pdf/255531/Decliningresponserates.pdf 
aSource: National Surveys on Drug Use and Health: Methodological Resource Book Section 8. 
Link: https://www.samhsa.gov/data/sites/default/files/NSDUHmrbDCFR2016.pdf 

 
3.5 Statistical Analysis  
 

NSDUH does not conduct a census of the population, thus sampling error is possible. To 

combat this possibility, large samples and probability sampling methods are utilized. Taylor 

Series Linearization is used for calculus-based variance and standard error estimation based on 

the complex survey design. A SUDAAN software for Statistical Analysis of Correlated Data 

package is used for such calculations.  

 

30 

To ensure collected data reflect the target population analysis weights were introduced. 

Analysis weights are based on the inverse probability of selection and account for the multi-stage 

design which included steps below. Related post-stratification adjustment corrected non-

response, known population control totals, and extreme weights (62).   

(1) Adjustment of household weights for nonresponse at the screener level. (2) Post 

stratification  of  household  weights  to  meet  population  controls  for  various 

household-level demographics by state. 

(3) Adjustment of household weights for extremes. 

(4) Post stratification of selected person weights. 

(5) Adjustment of responding person weights for nonresponse at the questionnaire 

level.  

(6) Post stratification of responding person weights. 

(7) Adjustment of responding person weights for extremes. 

 

 

 

31 

CHAPTER 4 MANUSCRIPT 1 – THE NATURAL HISTORY OF HEROIN USE 

DISORDER EXPLAINED BY PROBLEMS AND EXPERIENCES FROM FIRST 

MONTHS OF USE TO UP TO A YEAR AFTER ONSET  

4.0 Abstract 
 
Background & Aim: Despite the United States’ history and continued experience with heroin 

use and related outcomes, little is known about the natural history of heroin use disorder (HUD). 

The current study aims to estimate the pairwise association between problems and experiences of 

heroin use disorder interval by interval from the first months of use. 

Methods: National Surveys on Drug Use and Health (NSDUH) 2002-2016 aggregated files 

include 837,326 participants with 876 newly incident heroin users (NIHUs) identified. The cross-

sectional study was sliced to indicate time based on intervals between first months of heroin use 

to survey assessment. Problems and experiences (PE) to PE odds ratios were estimated using 

generalized linear model logistic regression by time intervals.  

Results: Within 0-90 days of first using heroin a robust association is seen between spending 

less time doing important activities and continued use despite problems with family or friends 

(OR=936.4, 95% CI=66.6, 13,164.9). By 330-390 days after first using heroin the strongest 

association is seen between spending a lot of time getting or using and continued use despite 

problems with family or friends (OR=441.3, 95% CI=43.4, 4,487.1).  

Discussion: Likely explanations for the robust point estimates and wide confidence intervals 

may reflect: 1) issues of sample size and small cell counts and 2) exchangeability between PE 

pairs. 

Conclusions: Soon after first using heroin there appears to be strong associations related to 

problems with personal relationships. Public health may benefit from engaging citizens in family 

prevention and intervention initiatives.    

 

32 

4.1 Introduction  
 

Heroin use is a global public health concern. In 2016 alone, 97 tons of heroin were seized 

globally (4). Specific to the US heroin, an estimated 0.4 per 100 persons have used in the past 

year in 2016 (6). In 2015 heroin-related overdoses reached about 26 per 100,000 persons and 

overdose deaths increased by five-fold from 2010 to 2016 (6). The opioid drug market has 

changed from its initial stages beginning in 1999. While extra-medical use of prescription 

opioids led to increased heroin use, and now the significant rise in synthetic opioids, there has 

still yet to be a departure from this upward trend (6).  

From a public health perspective, the increased use of heroin is troubling far beyond the 

direct mortality from heroin-related overdose. Heroin use disorder is associated with an 

estimated 40 years of life lost (50). Furthermore, other heroin use-related outcomes may include 

but are not limited to heroin use disorder (HUD), infections (e.g., Human Immunodeficiency 

Virus, Hepatitis C virus), psychiatric co-morbidities (e.g., depression), and low quality of life 

circumstances (e.g., homelessness) (63)(64). In tandem with the opioids crisis, the incidence of 

hepatitis C and other infections have increased (65)(66). In 2015 the economic burden of HUD 

was estimated to be roughly $51.2 billion which spanned crime, incarceration, chronic infectious 

diseases, treatment, neonatal abstinence syndrome, lost productivity, and overdose deaths (67).  

To meet the DSM-5 case definition for HUD only two criteria are necessary of the eleven 

possible: 1) using longer than intended or in larger amounts, 2) trying to cut down, 3) spending a 

lot of time getting/using/recovering, 4) unable to keep important commitments, 5) use affects 

personal relationships, 6) unable to do important activities, 7) use puts individual in physical 

danger, 8) physical or emotional problems, 9) tolerance develops, 10) withdrawal symptoms, 11) 

craving. 

 

33 

The current opioids epidemic is not the first of its kind in the US. After the passage of the 

1914 Harrison Act, which put federal restrictions on opiate and coca products, heroin use became 

popular as individuals using once using widely available narcotics turned to illegal means of 

obtaining opioids (11). During the 1950s heroin was relatively inexpensive which led to an 

increase in use; naturally the price inflated with demand (11). Again the US struggled with 

heroin use associated with the US soldiers’ return from Vietnam in the early 1970s (25)(26).  

When considering all of the epidemiological studies on the natural history of heroin use 

disorders, one of the most influential studies was completed during the mid-1970s by a research 

team at the Washington University at Saint Louis. Led by the late Professor Lee N. Robins, the 

team was commissioned to investigate the fate of veterans who had returned from Vietnam and 

to estimate how often they had returned to use of heroin or other opioids in their home 

communities. This influential research dispelled commonly accepted beliefs about the natural 

history of heroin use and ‘addiction’ to heroin. Its evidence indicated an important fact: there 

was very low recurrence of heroin use or heroin use disorders among the veterans who had tried 

heroin in Southeast Asia.  Instead, heterogeneity of post-return outcomes was found, often linked 

with individual-level susceptibilities, and sometimes with characteristics of communities in 

which the veterans were living (Table 4). Prominent among the predictors of recurrent heroin use 

were: (1)  a history of non-opioid drug use, (2) peers’ use of drugs, and (3) long-standing 

histories of early ‘deviant’ or other socially maladaptive behaviors often grouped under the DSM 

heading of ‘anti-social personality disorder’ (68).  

Most importantly, what is known about heroin use is that everyone who uses will not go 

on to experience HUD. The most recent HUD research suggests 23 to 38% of newly incident 

heroin users develop HUD within about a year of onset (69). Based on literature review for the 

 

34 

current study, little is known about the progression of HUD. That is, after first use when HUD is 

possible but is still subclinical, what are the first clinical features that become apparent?  

The general goal of the current study is to investigate the development of heroin use 

disorder shortly after an individual starts to use heroin. Specifically, in addition to estimating 

occurrence rates for individual heroin problems and experiences (PE), the aim is to estimate the 

degree to which paired combinations of these heroin PE co-occur, interval by interval during the 

first months after first heroin use (Table 5)1. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

                                                
1 Withdrawal symptoms included in questionnaire:1) feeling kind of blue or down, 2) vomiting or feeling nauseous, 
3) having cramps or muscle aches, 4) having teary eyes or a runny nose, 5) feeling sweaty, having enlarged pupils, 
or having body hair standing up on your skin, 6) having diarrhea, 7) yawning, 8) having a fever, and 9) having 
trouble sleeping. 

 

35 

Table 4. Contrast of Heroin Knowledge Based on Lee Robins’ Study of Vietnam Veterans 
Return to the United States. 

Contradictory Evidence  

-  After returning to the US, 8% of veterans 

used heroin and less than half of them used 
more than once a week for a month or more.  
-  Heroin purchased in the US did not lead to a 

use disorder any more than cannabis or 
amphetamine did.  

-  Heroin was considered to be the “main drug” 

-  The majority of heroin users uses other drugs 

of 10% of users.  

and alcohol. 

Before L. Robins’ Study 

1.  Heroin use rapidly transitions to 

HUD. 

2.  Heroin use replaces the use of 

other drugs.  

3.  HUD is likely permanent without 

prolonged treatment.  

4.  Recovery from HUD requires 

abstinence.  

5.  Heroin use constitutes a major 

social problem.  

-  After returning to the US, 1% experienced 

HUD during the first year back from 
Vietnam; 2% after the second or third year 
back.  

-  Since returning, 12% of veterans experienced 

HUD relapse. 

-  Less than half of the veterans used heroin 

upon returning and 1/8th experienced HUD.  

-  Veterans who used heroin did experience 

more social problems (e.g., unemployment) 
than the rest of the sample, but given the 
majority used more drugs than heroin it 
cannot be concluded heroin is responsible for 
the problems.  

Source: “Vietnam Veterans Three Years after Vietnam: How Our Study Changed Our View of 
Heroin”, Robins et al., 2010. Link:  
Of the 14,000 enlisted men that returned from Vietnam in 1971, a random sample totaling 617 
were followed for three years. Those who tested positive for drugs were oversampled. 

36 

 
 

 

Table 5. Description of Problems and Experiences of Heroin Use Disorder Included in the 
Diagnostic and Statistical Manual. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

work, or school?  

have put you in physical danger?  

using heroin?  

wanted?  

by your use of heroin?  

 “During the past 12 months… “ 

Problems & Experiences 

were probably caused or made worse by your use of heroin?  

important activities (spending time with friends & family)?  

have problems with your emotions, nerves, or mental health?  

PE 
 # 
1  Did you try to set limits on how often or how much heroin you used?  
2  Did you have any problems with your emotions, nerves, or mental health that 
3  Was there a month or more when you spent a lot of your time getting or 
4  Did you need to use more heroin than you used in order to get the effect you 
5  Did you have any problems with family or friends that were probably caused 
6  Did using heroin cause you to give up or spend less time doing these types of 
7  Did using heroin cause you to have serious problems like this either at home, 
8  Did you cut down or stop using heroin at least one time?  
9  Did you continue to use heroin even though you thought it was causing you to 
10  Did you regularly use heroin and then do something where using heroin might 
11  Did you continue to use heroin even though you thought it caused problems 
12  Were you able to keep to the limits you set, or did you often use heroin more 
13  Were you able to cut down or stop using heroin every time you wanted to or 
14  Did you have 3 or more of these symptoms after you cut back or stopped 
15  Did using heroin cause you to do things that repeatedly got you in trouble with 
16  Did you have 3 or more of these symptoms at the same time that lasted for 
17  Did you notice that using the same amount of heroin has less effect on you 
18  Was there a month or more when you spent a lot of your time getting over 
19  Did you have any physical problems that were probably caused or made 
20  Did you continue to use heroin even though you thought it was causing you to 
NA  Craving 
NSDUH Downloadable Public Use Dataset. 
*See footnote above for list of symptoms. 

longer than a day after you cut back or stopped using heroin?* 

tried to?  

using heroin?* 

the law?  

with family or friends? 

than you intended to?  

worse by using heroin?  

have physical problems?  

that it used to?  

the effects of the heroin you used? 

DSM 
IV  5 
x 
x 
x 
x 

x 

x 

x 

x 

x 
x 
x 

x 

x 

x 

x 

x 

x 

x 

x 

x 

x 

x 
 

x 

x 

x 

x 

x 
x 
x 

x 

x 

x 

x 

x 

x 

 

x 

x 

x 

 
x 

 

37 

4.2 Methods 
 

Chapter 3 provides an overview of materials and methods for this sub-project of the 

dissertation research. Interested readers are referred to that chapter for information about the 

study population for the US National Surveys on Drug Use and Health, 2002-2016, sampling and 

recruitment, assessments, and basic details about statistical analyses. 

Sample: Aggregate NSDUH files included 837,326 participants, with 896 identified as 

newly incident heroin users (NIHUs) from survey years 2002-2016. All NIHUs were assessed 

one to 12 months after first heroin use, excluding all past users. Observations with missing 

month of first use were dropped resulting in 876 NIHUs (described further in Statistical Analyses 

section) (Figure 8).  

The analytic sample of 876 NIHU was made up of 465 males with the majority of NIHUs 

within 18-25 years of age, and white ethnic self-identification (Table 7). Most NIHUs has a 

history of extra-medical use of prescription pain relievers (i.e., prescription opioids) within 12 

months of survey assessment (n=667).   

Statistical Analyses: Time between first use and survey assessment was calculated based 

on available month and year of first use minus the year quarter of survey assessment. 

Participants’ month of survey assessment is not available NSDUH public use files, so the year 

quarter yields three months of uncertainty in elapsed time between first use and assessment (i.e., 

possible 30-90-day discrepancy in time between first use and assessment). To aid in 

interpretation, imposed assumptions include: 1) survey assessment occurred on the 15th of the 

mid-quarter month and 2) day of first occurred on the 15th of the month. This elapsed time was 

categorized as ‘lag-time’ intervals, coded a categorical variable (Table 6).  

To account for the multi-stage complex sampling of the NSDUH, estimates used 

analysis-weights with variances calculated using Taylor Series linearization. Variance estimation 

 

38 

was set to single unit centered to produce standard errors based on the grand mean for the most 

conservative estimation. Analyses were repeated without analysis-weights and are available in 

the Appendix B – Manuscript 1. Stata SE 12.1 was used to carry out all statistical analyses.  

HUD was assessed using 20 questions about heroin problems and experiences (PE) as 

described by each newly incident heroin user – i.e., these are self-characterized problems and 

experiences with no information from parents, spouses, partners, employers, or other key 

informants. Some of the PEs would qualify as HUD ‘symptoms’ if a diagnosis of HUD could be 

established and confirmed. However, it would be a mistake to characterize each PE as an ‘HUD 

symptom’ in the absence of a diagnosed HUD because some experiences do not have the quality 

of a ‘symptom.’ A good example involves the experience of spending a lot of time getting 

heroin. Consider a newly incident user who has tried heroin one time, found the experience to be 

reinforcing, and had to spend a lot of time tracking down and getting a second dose. This NIHU 

might answer ‘Yes’ to the NSDUH question about spending a lot of time ‘getting’ heroin, but at 

this stage of progression beyond heroin onset, the context is not one of a heroin use disorder 

experience that is being manifest as a symptom. Rather, it is just an ‘experience’ in the life of the 

user, and the user might or might not consider it to be a ‘problem.’ 

Other PE most likely would qualify as ‘symptoms’ when they occur. For example, any 

NIHU who has experienced three or more withdrawal symptoms most likely is describing a clear 

symptomatic manifestation of the underlying pathological processes described in the early 

descriptions of ‘heroin addiction’ of the late 19th century. 

PE-specific proportions are estimates of PE attack rates with an implicit time dimension 

based on the lag-time interval. These proportions have been estimated based on ratios. 

Numerators for the ratio are participants who have experienced with the designated PE. The 

ratio’s denominator is the total number of all NIHUs, interval by interval (Table 4). 

 

39 

The NSDUH survey design relies on “gated” questions to minimize participants’ time 

needed to complete the survey, which results in some participants not being asked all 20 PE 

questions (see figure in Appendix B – Manuscript 1). To stabilize the denominator of PE-specific 

proportions, gated questions were re-coded to allow a stable denominator of all NIHU for each 

PE (n=876). Responses including “Don’t Know” or “Refused” were coded as “No”. As an 

example, 876 NIHUs were asked a primary question of: “Did you try to set limits on how often 

or how much heroin you would use?” The 147 NIHUs that responded “yes” were asked a 

secondary question: “Were you able to keep to the limits you set, or did you often use heroin 

more than you intended to?” Thus, the secondary question was re-coded to use a denominator of 

the 876 NIHUs that were asked the primary question.  

In order to quantify the degree of association linking PE-PE pairs, a generalized linear 

model with the logistic link function was used (i.e., logistic regression). Via these logistic 

regression models, the strength of association linking each PE pair might be inverse (i.e., with 

estimated regression beta < 0.0, exponentiated for an OR < 1.0) or might be positive (i.e., with 

estimated regression beta >0.0, exponentiated for an OR > 1.0). The OR estimator has been used 

because it is not margin-dependent. Other statistical indices of agreement and association such as 

kappa or correlation coefficients would depend upon the marginal frequencies of the PE. A 

margin-dependent estimator for degree of association or agreement is not useful in a study of 

natural history across intervals of elapsed time from an infection or exposure in the direction of 

sub-clinical and clinical disease processes because it is typical for manifestations of the 

pathological process to increase in their frequency across units of elapsed time. In consequence, 

an increasing correlation coefficient for a PE-PE pair might reflect a shift in the marginal 

frequency of each PE, with no shift in the degree of association. 

 

40 

The decision about which PE to place on the Y-side of the regression equation versus on 

the X-side of the regression equation is irrelevant in this context. It does not matter. The 

estimated logistic regression beta coefficient (and OR = exponentiated beta) will be the same, 

irrespective of which PE is on the Y-side or the X-side of the equations. 

The work of the dissertation research project included writing software programs to 

derive the estimated odds ratios, with due attention to the analysis weights and to the use of 

Taylor series linearization for variance estimation and derivation of 95% confidence intervals 

and tests of statistical significance via p-values. The programming was facilitated via Stata 

software ‘svy subpop’ commands. These commands were devised to produce PE to PE ORs for 

each of the 13 lag-time intervals. 

In addition, sensitivity analyses were completed using pseudo-lag-time intervals created 

by collapsing lag-time intervals one through three, seven through eight, and 12 through 13, with 

all other intervals considered ‘missing’ in order to produce non-overlapping elapsed time 

intervals. This additional set of analysis steps is helpful because the result is a set of non-

overlapping NIHU (i.e., each NIHU is assigned to one and only one interval bin), and there is no 

covariance issue to be addressed when one seeks to compare estimates for one interval versus 

another. 

A note about the size of the odds ratio estimates may be in order. Epidemiologists 

studying exposure-disease relationships typically encounter OR estimates across a range from 

0.10 through 10.0, with 0.1 for a strong inverse association under 1.0 and with 10.0 being a 

strong association above 1.0. In this study of natural history and the process of becoming a case 

of HUD, the relationships of interest are not exposure-disease relationships. Rather, they pertain 

to inter-relationships of individual features of the process that leads from first heroin use toward 

becoming a case of heroin use disorder. In this process, some of the PEs satisfy a condition of 

 

41 

exchangeability of the type that is seen when education researchers construct tests of arithmetic 

ability, with one item Q1 asking “2+2=?” and another item Q2 asking “1+1=?). Most students 

with decent arithmetic ability will answer both Q1 and Q2 correctly, and the students who 

correctly answer just one item might have been guessing when they answered. Students who do 

not have arithmetic ability will give incorrect answers to both Q1 and Q2. The result will be a 

very large association in the Q1-Q2 contingency table, with an odds ratio most likely well above 

100. 

A similar phenomenon can be observed when a diagnostic test for a drug use disorder 

includes exchangeable items as in the case of ‘getting less of an effect when using the same 

dose’ cross-classified with ‘needing a larger dose to get the same effect.’ Both responses to PE 

items on these experiences are manifestations of subjectively-felt tolerance, and the estimated 

OR for these two items should be quite large, in a reflection of their exchangeability. For this 

reason, the use of the OR in this study produces some quite large OR estimates that are not 

typically seen in exposure-disease investigations, but that are quite often seen when 

epidemiologists study patterns of interdependent responses to exposure. As shown below, some 

of the estimated ORs are far beyond the values expected in studies of exposure-disease 

associations, but they are not exceptional in studies of interdependent manifestations of an 

observed disease process. 

 
 
 
 
 
 
 

 

42 

Figure 8. Flow of Participants in Newly Incident Heroin Use Case Ascertainment. 
Data from US, National Surveys on Drug Use and Health, 2002-2016.  

NSDUH – Year-Specific Files 

2003 

2004 

2005 

2006 

2007 

2008 

2009 

n=54,079  n=55,230  n=55,602  n=55,905  n=55,279  n=55,435  n=55,110  n=55,234 

2012 

n=55,268 

2013 

n=55,160 

2014 

n=55,271 

2015 

n=57,146 

2016 

n=56,897 

2002 

2010 

2011 

n= 57,313 

n= 58,397 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

2002-2016 Aggregate Files 

NSDUH  

n= 837,326 

Never Used Heroin 

n=825,420 

Past Onset Heroin 

n=9,153 

Newly Incident Heroin Users 

n=896 

Newly Incident Heroin Users 
within Lag-Time Intervals 1-13 

n=876 

Missing Month of First Use 

n=20 

NSDUH Downloadable Public Use Dataset.  
Unweighted counts (n). 
Newly incident heroin use: first heroin use within 12 months of survey assessment; past onset 
users excluded.  
Lag-time intervals indicate elapsed time between first use and survey assessment. 
Participants included in incident use due to year of first use information but missing month of 
first use were dropped from analytic sample due to inability to define participant’s lag-time 
interval between first use and survey assessment.  

43 

 
 
 
 
 
 
 
 
 
 

 

Table 6. Description of Lag-Time Intervals Indicating Elapsed Time Between Newly Incident 
Heroin Use and Survey Assessment.  
Data from United States National Surveys on Drug Use and Health 2002-2016. 

 

Lag-Time Interval  

Elapsed Time (days) 

Assumeda 

Possible 

0-90 
1-120 
30-150 
60-180 
90-210 
120-240 
150-270 
180-300 
210-330 
240-360 
270-390 
300-420 
330-450 

30 
60 
90 
120 
150 
180 
210 
240 
270 
300 
330 
360 
390 

Without Overlapb  

0-150 

 

 
 
 
 

150-300 

 
 
 

300-450 

NSDUH Downloadable Public Use Dataset.  
aAssumptions include 1) survey assessment occurred on the 15th of the mid-quarter month and 
2) day of first occurred on the 15th of the month. 
bIntervals without overlap drop observations in intervals 4-6 and 9-11 to increase certainty of 
time elapsed for sensitivity analysis.  

1 
2 
3 
4 
5 
6 
7 
8 
9 
10 
11 
12 
13 

 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

44 

Table 7. Characteristics of Overall Sample and Newly Incident Heroin Use Analytic Sample. 
Data from US, National Surveys on Drug Use and Health, 2002-2016. 

Aggregate NSDUH 

2002-2016 
n=837,326 

n (%) 

NIHUs 
n=896 
n (%) 

NIHUs 
Intervals* 

n=876 
n (%) 

 
 
Demographic Characteristics 
Sex 

Age Groups (Years) 

Male 

12-17 
18-25 
26-34 
35-49 
50-64 
65+ 

401,270 (48) 
 
258,309 (10) 
263,258 (13) 
93,449 (14) 
128,684 (25) 
56,998 (22) 
36,628 (16) 
 
532,503 (67) 
106,106 (12) 
12,284 (<1) 
4,042 (<1) 
30,449 (5) 
26,309 (1) 
134,633 (14) 

476 (62) 
 
227 (15) 
539 (47) 
82 (22) 
44 (14) 
4 (2) 
0 
 
715 (81) 
21 (4) 
20 (<1) 
8 (1) 
5 (1) 
31 (1) 
96 (10) 

Ethnic Self-Identification 

White 
Black/African American 
Native American/Alaskan Native 
Native Hawaiian /Other Pacific Island. 
Asian 
>1 Ethnicity 
Hispanic 
Extra-Medical Prescription Opioid History 
Lifetime Use (>12 Mon.*) 
Within 12 Mon.* 
Never Used 
Missing Data 
NSDUH Downloadable Public Use Data. 
NIHUs: Newly incident heroin users first used heroin within 12 months of survey assessment. 
Past onset users excluded. 
Ethnic self-identification: all ethnicities other than Hispanic are specified as non-Hispanic per 
survey assessment. 
*More than or within 12 months of survey assessment. 
*within lag-time intervals 
Unweighted n; weighted %. 

89,390 (11) 
88, 649 (9) 
608,112 (74) 
51,175 (5) 

93 (10) 
677 (76) 
112 (13) 
15 (2) 

465 (62) 
 
221 (15) 
528 (48) 
82 (23) 
42 (13) 
3 (1) 
0 
 
699 (81) 
20 (3) 
19 (<1) 
8 (1) 
5 (1) 
30 (2) 
95 (10) 

90 (9) 
667 (77) 
105 (11) 
14 (2) 

 
 
 
 
 
 

 

45 

4.3 Results 
 
 

PE-Specific Attack Rates: Each estimated proportion shown in Table 8 can be 

conceptualized as an ‘attack rate’ with the time dimension implicit in the formulation of the rate, 

as would be the case in the estimation of food-specific attack rates for diarrhea, vomiting, fever, 

and other facets of food-borne illness investigated in the aftermath of a food-borne illness 

outbreak. That is, in these attack rates, the time dimension typically is implicit, and is not 

quantified in terms of person-days, person-weeks, or person-months. 

 

Table 8 shows PE-specific proportions across a broad range of estimates. For 

example, an estimated 40% of NIHUs had tried to set limits on heroin use (PE1), whereas no 

more than about one percent had the experience of ‘continuing to use despite physical problems.’ 

A quarter or larger of NIHUs experienced PE2-problems with emotions (31%), PE3-spending a 

lot of time getting or using (28%), PE4-needing more to get the same effect (27%), PE5-

problems with family or friends (26%), or PE5-spending less time doing important activities 

(25%).  

 

PE to PE Odds Ratios by Lag-Time Intervals: As shown in Tables 9-10, within 

0-90 days of first using heroin, some of the strongest associations between PE pairs include: 

PR6/PE11- spending less time doing important activities/continued to use despite problems with 

family or friends (OR=936.4, 95%CI= 66.6, 12,511.5), PE6/PE10-spending less time doing 

important activities/put in physical danger (OR= 873.7, 95% CI=70.8, 10,785.4), PE10/PE11- in 

physical danger/continued to use despite problems with family or friends (OR=871, 

95%CI=60.6, 12,511.5), and PE3/PE6-spent a lot of time getting or using heroin/spending less 

time doing important activities (OR=556.2, 95% CI=38.5, 8,027.5).  

 

Within 1-120 days of first using heroin, some of the strongest associations between 

PE pairs include: PE9/PE13-continued to use despite having problems with emotions/was not 

 

46 

able to cut down (OR=551.7 95% CI=45.4, 6,700.4), PE9/PE14-cotinued to use despite having 

problems with emotions/had three or more withdrawal symptoms (OR=551.7 95% CI=45.4, 

6,700.4), PE12/PE16-unable to keep limits/had three or more withdrawal symptoms at the same 

time (OR=514.8, 95% CI=36.7, 7220.3), and PE3/PE16-spent a lot of time getting or using 

heroin/had three or more withdrawal symptoms at the same time (OR=299.1, 95% CI=23.0, 

3883.9) (Tables 11, 12). 

 

Within 30-150 days of first using heroin, some of the strongest associations 

between PE pairs include: PE17/PE18-same amount had less effect/spent a lot of time getting 

over the effects (OR=320.6, 95% CI=15.7, 6,536.7), PE9/PE17-continued to use despite having 

problems with emotions/same amount had less effect (OR=213.4, 95% CI= 16.3, 2,796.4), 

PE3/PE12-spent a lot of time getting or using/unable to keep limits (OR=166.4, 95% CI= 18.8, 

1,472.5), PE15/PE19-trouble with the law/had physical problems (OR= 150.7, 95% CI= 6.6, 

3,425.3), and PE15/PE20-trouble with the law/continued to use despite having physical problems 

(OR= 150.7, 95% CI= 6.6, 3,425.3) (Tables 13, 14). 

 

Within 60-180 days of first using heroin, some of the strongest associations 

between PE pairs include: PE3/PE7-spent a lot of time getting or using/caused you to have 

serious problems (OR=1,181.8, 95% CI=120.6, 11,582.3), PE7/PE15-caused you to have serious 

problems/trouble with the law (OR= 258.5, 95% CI= 33.7, 3,907.8), PE6/PE7-spent less time 

doing important activities (OR=258.5, 95% CI= 32.3, 2,066.4), and PE7/PE11-continued to use 

despite having problems with family or friends (OR=232.2, 95% CI= 24.3, 2.219.0) (Tables 15, 

16). 

 

Within 90-210 days of first using heroin, some of the strongest associations 

between PE pairs include: PE9/PE10-cotinued to use despite having problems with 

emotions/physical danger (OR=331.9, 95% CI=35.7, 3,088.8), PE3/PE14-spent a lot time getting 

 

47 

or using/had three or more withdrawal symptoms (OR=196.5, 95% CI=19.2, 2,002.1), and 

PE4/PE6-needed more to get the same effect/spent less time doing important activities 

(OR=147.3, 95% CI= 22.5, 966.3) (Tables 17, 18).  

 

Within 120-240 days of first using heroin, some of the strongest associations 

between PE pairs include: PE4/PE13-needed more to get the same effect/not able to cut down 

(OR=319.8, 95% CI=30.3, 3371.0), PE4/PE14-needed more to get the same effect/had three or 

more withdrawal symptoms (OR=319.8, 95% CI=30.3, 3371.0), PE4/PE16-needed more to get 

the same effect/had three or more withdrawal symptoms at the same time, PE12/PE13-not able to 

set limits/not able to cut down (OR=301.0, 95% CI=28.0, 3,238.5), PE12/PE14-unable to set 

limits/had three or more withdrawal symptoms (OR=301.0, 95% CI=28.0, 3,238.5), and 

PE12/PE16-not able to set limits/had three or more withdrawal symptoms at the same time 

(OR=301.0, 95% CI=28.0, 3,238.5) (Tables 19, 20). 

 

Within 150-270 days of first using heroin, some of the strongest associations 

between PE pairs include: PE3/PE4-spent a lot of time getting or using/needed more to get the 

same effect (OR=1,008.0, 95% CI= 99.4, 10,219.5), PE11/PE15-continued to use despite having 

problems with family or friends/trouble with the law (OR=454.8, 95% CI=41.2, 5,019.9), and 

PE4/PE6-needed more to get the same effect/spent less time doing important activities 

(OR=50.0, 2,731.6) (Tables 21, 22).  

 

Within 180-300 days of first using heroin, some of the strongest associations 

between PE pairs include: PE11/PE12-continued to use despite having problems with family or 

friends/unable to set limits (OR=2,396.0, 95% CI=178.7, 32,117.4), PE11/PE13-continued to use 

despite having problems with family or friends/unable to cut down (OR=1,301.5, 95% CI=144.5, 

11,721.8 ), and PE6/PE7-spent less time doing important activities/caused you to have serious 

problems (OR=1,293.5, 95% CI=86.6, 19,326.3) (Tables 23, 24).  

 

48 

 

Within 210-330 days of first using heroin, some of the strongest associations 

between PE pairs include: PE9/PE11-continued to use despite having problems with 

emotions/continued to use despite having problems with family or friends (OR=2,111.3, 95% 

CI=102.8, 43,376.5), PE2/PE5-had problems with emotions/had problems with family or friends 

(OR=342.0, 95% CI= 29.5, 3,964.4), PE2/PE4-had problems with emotions/needed more to get 

the same effect (OR=240.7, 95% CI=19.0, 1,030.4), and PE2/PE11-had problems with 

emotions/continued to use despite having problems with family or friends (OR= 238.4, 95% CI= 

18.1, 3,145.3) (Tables 25, 26).  

 

Within 240-360 days of first using heroin, some of the strongest associations 

between PE pairs include: PE6/PE7-spent less time doing important activities/caused you to have 

serious problems (OR=514.8, 95% CI=52.5, 5,052.5), PE3/PE12-spent a lot of time getting or 

using/unable to keep limits (OR=401.5, 95% CI=39.5, 4,084.5), PE17/PE18-the same amount 

had less effect/spent a lot of time getting over the effects (OR=236.0, 95% CI=8.2, 6,776.6), 

PE6/PE15-spent less time doing important activities/trouble with the law (OR=155.1, 95% CI= 

13.1, 1,839.0), PE7/PE15-caused you to have serious problems/trouble with the law (OR=147.7, 

95% CI=12.4, 1,75.5), and PE11/PE15-continued to use despite problems with family or 

friends/trouble with the law (OR=133.0, 95% CI=11.2, 1,581.5) (Tables 27, 28).  

 

Within 270-390 days of first using heroin, some of the strongest associations 

between PE pairs include: PE6/PE7-spent less time doing important activities/caused you to have 

serious problems (OR=221.7, 95% CI=16.7, 2,940.9), PE11/PE13=continued to use despite 

having problems with family or friends/unable to cut down (OR=110.1, 95% CI=13.1, 925.9), 

and PE4/PE6-needed more to have the same effect/spent less time doing important activities 

(OR=99.5, 95% CI=11.3, 877.2) (Tables 29, 30).  

 

49 

 

Within 300-420 days of first using heroin, some of the strongest associations 

between PE pairs include: PE6/PE7-spent less time doing important activities/caused you to have 

serious problems (OR=261.8, 95% CI=28.3, 2,425.0), PE2/PE6-had problems with 

emotions/spent less time doing important activities (OR=217.6, 95% CI=25.4, 1,864.5), and 

PE6/PE10-spent less time doing important activities/physical danger (OR=115.1, 95% CI=8.3, 

1,597.0) (Tables 31, 32).  

 

Within 390-450 days of first using heroin, some of the strongest associations 

between PE pairs include: PE3/PE11-spent a lot of time getting or using/continued to use despite 

having trouble with family or friends (OR=441.3, 95% CI=43.4, 4,487.1), PE1/PE2=tried to set 

limits/had problems with emotions (OR=377.7, 95%CI=46.7, 3,053.7), PE2/PE6-had problems 

with emotions/spent less time doing important activities (OR=371.1, 95% CI=32.7, 4,206.4), 

PE6/PE7-spent less time doing important activities/caused you to have serious problems 

(OR=286.1, 95% CI=26.0, 3,149.1), and PE3/PE6-spent a lot of time getting or using/spent less 

time doing important activities (OR=235.2, 95% CI=21.5, 2,567.2) (Tables 33, 34). 

 

Stratified by Analysis Weight Quartiles: In this study, an investigation of 

variation in the OR estimates across quantiles of analysis weights is needed to show what might 

have been observed if analysis weights had been ignored. For this reason, the analyses were 

repeated for each quartile of analysis weight. As graphically displayed, the PE to PE ORs 

stratified by analysis weight quartiles do not differ appreciably. The distribution of OR strength 

appears to be similar in each quartile. 

 

Sensitivity Analysis: Within 0-150 days of first using heroin, some of the strongest 

associations between PE pairs include: PE8/PE13-tried to cut down or stop using at least 

once/unable to cut down (OR=480.6, 95% CI=70.3, 3,286.2), PE12/PE16-unable to set 

limits/had three or more withdrawal symptoms at the same time (OR=416.7, 95% CI=35.8, 

 

50 

4,845.8), PE3/PE16-spent a lot of time getting or using/had three or more withdrawal symptoms 

at the same time (OR=127.4, 95% CI=18.1, 897.2), and PE2/PE16-had problems with 

emotions/had three or more withdrawal symptoms at the same time (OR=107.9, 95% CI=10.2, 

1,140.8) (Tables 35, 36). 

 

Within 150-300 days of first using heroin some of the strongest associations 

between PE pairs include: PE5/PE7-had problems with family or friends/caused serious 

problems (OR=663.0, 95% CI=147.5, 2,979.8), PE3/PE11-spent a lot of time getting or 

using/continued to use despite having problems with family or friends (OR=523.2, 95% CI=56.0, 

4,888.9), PE6/PE12-spent less time doing important activities/unable to set limits (OR=381.9, 

95% CI=), PE7/PE11-caused serious problems/continued to use despite problems with family or 

friends (OR=289.5, 95% CI=58.4, 1,434.7), PE11/PE15-continued to use despite problems with 

family or friends (OR=242.0, 95% CI=25.9, 2,260.7), and PE6/PE7-spent less time doing 

important activities/caused serious problems (OR=208.1, 95% CI=37.9, 1,143.1) (Tables 37, 38).  

 

Within 300-450 days of first using heroin, some of the strongest associations 

between PE pairs include: PE3/PE11-spent a lot of time getting or using/continued to use despite 

having problems with family or friends (OR=558.3, 95% CI=64.9, 4,803.9), PE2/PE6-had 

problems with emotions/spent less time doing important activities (OR=273.5, 95% CI=57.4, 

1,303.3), PE2/PE10-had problems with emotions/physical danger (OR=279.6, 95% CI=279.6, 

95% CI=31.2, 2,503.0), and PE2/PE12-had problems with emotions/unable to set limits 

(OR=154.4, 95% CI=16.0, 1,487.0) (Tables 39, 40). 

 

 

51 

caused by your use of heroin?  

home, work, or school?  

or using heroin?  

effect you wanted?  

might have put you in physical danger?  

40  Did you try to set limits on how often or how much heroin you used?  
31  Did you have any problems with your emotions, nerves, or mental 
health that were probably caused or made worse by your use of heroin?  
28  Was there a month or more when you spent a lot of your time getting 
27  Did you need to use more heroin than you used in order to get the 
26  Did you have any problems with family or friends that were probably 
25  Did using heroin cause you to give up or spend less time doing these 
types of important activities (spending time with friends & family)?  
23  Did using heroin cause you to have serious problems like this either at 
22  Did you cut down or stop using heroin at least one time?  
23  Did you continue to use heroin even though you thought it was causing 
you to have problems with your emotions, nerves, or mental health?  
21  Did you regularly use heroin and then do something where using heroin 
18  Did you continue to use heroin even though you thought it caused 
17  Were you able to keep to the limits you set, or did you often use heroin 
12  Were you able to cut down or stop using heroin every time you wanted 
9  Did you have 3 or more of these symptoms after you cut back or 
8  Did using heroin cause you to do things that repeatedly got you in 
8  Did you have 3 or more of these symptoms at the same time that 
lasted for longer than a day after you cut back or stopped using heroin?* 
6  Did you notice that using the same amount of heroin has less effect on 
6  Was there a month or more when you spent a lot of your time getting 
3  Did you have any physical problems that were probably caused or 
1  Did you continue to use heroin even though you thought it was causing 

over the effects of the heroin you used? 

problems with family or friends? 

stopped using heroin?* 

trouble with the law?  

made worse by using heroin?  

you to have physical problems?  

more than you intended to?  

to or tried to?  

you that it used to?  

238 

228 

231 

195 

192 
185 
170 

163 

148 

135 

105 

76 

68 

67 

39 

38 

25 

11 

876 

876 

876 

876 

876 
876 
876 

876 

876 

876 

876 

876 

876 

876 

876 

876 

876 

876 

Table 8.  Description of Problems and Experiences of Heroin Use Disorder. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

Problems & Experiences 
 “During the past 12 months…” 

Proportionsa 
n. 
356 
246 

d.  % 
876 
876 

NSDUH Downloadable Public Use Dataset. 
aUnweighted n: numerator & d: denominator; % weighted proportions. 
*See footnote above for list of symptoms.  

 

52 

Table 9. Weighted Odds Ratios of Problem and Experience Pairs Within 0-90 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

 
NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.  
Lag-time interval 1: possible 0-90 days, assumed 30 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

53 

 
 
 
 

 

Table 10 Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 0-90 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 1: possible 0-90 days, assumed 30 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation.  

 

 

54 

 
 
 
 
 

 

Table 11. Weighted Odds Ratios of Problem and Experience Pairs Within 1-120 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.  
Lag-time interval 2: possible 1-120 days, assumed 60 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

55 

 
 
 
 
 

 

Table 12. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 1-120 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 2: possible 1-120 days, assumed 60 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

 

56 

 

 

Table 13. Weighted Odds Ratios of Problem and Experience Pairs Within 30-150 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets. 
*Statistically significant: 95% confidence interval does not contain 1.0.   
Values of 1.0 indicate null associations or inability to calculate OR.  
Lag-time interval 3: possible 30-150 days, assumed 90 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

57 

 
 
 
 

 

Table 14. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 90 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 3: possible 30-150 days, assumed 90 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

58 

 
 
 
 
 
 
 

 

Table 15. Weighted Odds Ratios of Problem and Experience Pairs Within 60-180 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.  
Lag-time interval 4: possible 60-180 days, assumed 120 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

59 

 
 
 
 
 

 

Table 16. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 60-180 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 4: possible 60-180 days, assumed 120 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

60 

 
 
 
 
 
 
 

 

Table 17. Weighted Odds Ratios of Problem and Experience Pairs Within 150 Days of First Using Heroin.  
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.   
Values of 1.0 indicate null associations or inability to calculate OR.  
Lag-time interval 5: possible 90-210 days, assumed 150 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

61 

 
 
 
 
 

 

 

 
 
 
 
 
 

 

Table 18. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 150 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 5: possible 90-210 days, assumed 150 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

62 

Table 19. Weighted Odds Ratios of Problem and Experience Pairs Within 120-240 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0. 
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 6: possible 120-240, assumed 180 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

63 

 
 
 
 
 

 

Table 20. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 120-240 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 6: possible 120-240, assumed 180 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

64 

 
 
 
 
 
 

 

Table 21. Weighted Odds Ratios of Problem and Experience Pairs Within 150-270 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0. 
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 7: possible 150-270, assumed 210 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

65 

 
 
 
 
 

 

Table 22. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 150-270 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 7: possible 150-270, assumed 210 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

66 

 
 
 
 
 
 
 

 

Table 23. Weighted Odds Ratios of Problem and Experience Pairs Within 180-300 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 8: possible 180-300, assumed 240 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

67 

 
 
 
 
 

 

Table 24. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 180-300 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 8: possible 180-300, assumed 240 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

68 

 
 
 
 
 
 
 

 

Table 25. Weighted Odds Ratios of Problem and Experience Pairs Within 210-330 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 9: possible 210-330, assumed 270 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

69 

 
 
 
 
 

 

Table 26. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 210-330 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 9: possible 210-330, assumed 270 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

70 

 
 
 
 
 
 
 

 

Table 27. Weighted Odds Ratios of Problem and Experience Pairs Within 240-360 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 10: possible 240-360, assumed 300 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

71 

 
 
 
 
 

 

Table 28.  Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 240-360 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 10: possible 240-360, assumed 300 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

72 

 
 
 
 
 

 

Table 29. Weighted Odds Ratios of Problem and Experience Pairs Within 270-390 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 11: possible 270-390, assumed 330 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

73 

 
 
 
 
 

 

Table 30. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 270-390 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 11: possible 270-390, assumed 330 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

74 

 
 
 
 
 
 
 

 

Table 31. Weighted Odds Ratios of Problem and Experience Pairs Within 300-420 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 12: possible 300-420, assumed 360 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

75 

 
 
 
 

 

Table 32. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 300-420 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 12: possible 300-420, assumed 360 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

76 

 
 
 
 
 

 

Table 33. Weighted Odds Ratios of Problem and Experience Pairs Within 330-450 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time interval 13: possible 330-450, assumed 390 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

77 

 
 
 
 
 

 

Table 34. Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 330-450 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 13: possible 330-450, assumed 390 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

 

78 

Figure 9. Odds Ratios with 95% Confidence Intervals of Problem and Experience Pairs 
Stratified by Analysis Weight Quartiles. 
Data from United States National Surveys on Drug Use and Health, 2002-2016 

A.  First Quartile (0 > x £ 61.8) 

B.  Second Quartile (61.8 > x £ 134.1) 

2,52,62,7
2,102,11
2,14
2,16
2,4

2,3
2,13
2,12

2,15

2,8
2,18
2,9
2,192,20
2,17

3,4
3,6
3,11
3,16
3,14

3,123,13
3,9
3,7
3,5
3,10
3,15

3,8
3,193,20
3,173,18

4,6
4,9
4,114,12
4,16
4,5
4,14

4,13
4,7
4,10

4,15

4,8
4,20
4,19
4,17
4,18

5,65,7
5,10

5,15

5,16

5,9
5,12

5,14
5,13

5,20

5,8
5,185,19
5,17
5,11

6,7
6,96,106,11
6,15

7,107,11
7,15

8,13

9,11
9,14
9,16

10,11

6,16
6,12

6,14

6,13

6,196,20
6,8
6,17
6,18

9,13

9,12
9,10

9,15

9,189,199,20
9,17

11,15

11,12

11,16

11,14
11,13

11,20

11,17
11,19
11,18

10,15

10,12

10,13
10,16

10,14
10,20

10,19
10,18
10,17

7,12
7,9
7,19

7,16

7,137,14
7,20

7,18
7,8
7,17

8,9
8,12
8,11
8,15
8,10
8,168,178,188,198,20
8,14

1,16
1,4
1,13
1,3
1,6
1,14
1,91,101,11
1,2
1,15
1,5
1,7
1,20
1,171,181,19
1,12
1,8

12,16

12,13

12,14

12,15

13,15

18,19

15,19

15,16
15,20
15,18
15,17

14,15

14,19
14,18
14,16
14,17
14,20

12,20
12,19
12,17
12,18

13,19
13,20
13,14
13,16
13,17
13,18

16,18
16,17
16,19
16,20

17,18
17,19
17,20

18,20

19,20

1
C.  Third Quartile (134.1 > x £ 337.2) 

6
Discordant Pair Odds Ratios by PE

16

11

21

 

1,13

1,3

1,14
1,4
1,16

1,6
1,15
1,9
1,2
1,101,11
1,7
1,5

1,171,18
1,8
1,12
1,191,20

2,10
2,3
2,6
2,16
2,11
2,4

2,7
2,14
2,12
2,5

2,13

2,15
2,8
2,18
2,17
2,9
2,192,20

3,4
3,63,7
3,93,10
3,123,133,14
3,16

3,11

3,5

3,15

3,8

3,17
3,19
3,18
3,20

4,64,7
4,9
4,114,124,134,14
4,16

4,10
4,5

4,15

4,8
4,19

4,17
4,20
4,18

5,65,7
5,95,10
5,15

5,16
5,13

5,12
5,14

5,8

5,195,20
5,11
5,175,18

6,7
6,96,106,116,12
6,16
6,15

6,14

6,13

6,8

6,20
6,176,186,19

7,9
7,11

7,15
7,10

7,12

7,16

7,14

7,13

7,8

7,20
7,177,187,19

8,13

9,109,11
9,16

9,14
9,12

10,11
10,13
10,14
10,16
10,12

12,13
12,14
12,16

11,12
11,13
11,16

11,14

9,13

10,15

11,15

13,20

8,10
8,118,12
8,9

8,15
8,20
8,14
8,168,178,188,19

9,15

9,179,189,199,20

12,15

12,19

12,18
12,20
12,17

13,19
13,15

13,14
13,16
13,18
13,17

14,20
14,15

14,19
14,17
14,16
14,18

10,20
10,17
10,18
10,19

11,20
11,17
11,18
11,19

17,18

15,19
15,20
15,16

15,17
15,18

16,17
16,18
16,19
16,20

17,19
17,20

18,19
18,20

19,20

1,16

1,131,14

1,3

1,4

1,6
1,18
1,5
1,9
1,2
1,7
1,101,11
1,15
1,19
1,8
1,12
1,17
1,20

2,6

2,5
2,11

2,10

2,3
2,4
2,12
2,7
2,13
2,16
2,14

2,18

2,15
2,8
2,17
2,9
2,192,20

3,4
3,113,123,133,14

3,6

3,16
3,9

3,10

3,5

3,15
3,7

3,8

3,19
3,173,18
3,20

4,13
4,114,12

4,14
4,6

4,5
4,9
4,16

4,7
4,10

4,15

4,8
4,18
4,19
4,17
4,20

5,65,7
5,10

5,9

5,12

5,18
5,13

5,16
5,14
5,19
5,17
5,8
5,11
5,15
5,20

6,7
6,9
6,11

6,12
6,10

6,15
6,20
6,13

6,16
6,14

6,8
6,186,19
6,17

7,11

8,13

9,11

10,11
10,15

11,15

12,13
12,14
12,16

7,15

7,10

7,9
7,20
7,12

7,19

7,18
7,13
7,17
7,14
7,16
7,8

9,10

9,12
9,16
9,13
9,14

9,15

9,179,189,199,20

8,12
8,15
8,9
8,11
8,20
8,10
8,168,178,18
8,14
8,19

10,20

10,12

10,14
10,16
10,13
10,19
10,18
10,17

12,20

12,15
12,17

12,19
12,18

11,13
11,12

11,19
11,16
11,14

11,18

11,20
11,17

15,20

15,19

15,16
15,18

15,17

14,15

14,18
14,17
14,16
14,19
14,20

13,15

13,18
13,14
13,16
13,19
13,20
13,17

17,18

18,19

16,18

16,17
16,19
16,20

17,19
17,20

18,20

19,20

1
D.  Fourth Quartile (x > 337.2) 

6
Discordant Pair Odds Ratios by PE

11

16

21

 

1,3

1,11
1,5

1,2
1,10
1,14
1,61,7
1,4
1,16
1,9
1,13

1,19
1,171,18

1,12
1,15
1,20
1,8

2,42,52,62,7
2,132,14
2,16
2,18
2,11
2,3
2,10
2,15

2,17

2,8
2,9
2,12
2,192,20

3,43,53,63,7
3,113,12
3,15

4,6

5,65,7
5,10

6,7
6,11
6,15

7,11

8,13

9,119,129,139,14
9,16

10,15

4,11

4,7
4,5

4,9
4,15
4,12
4,10

4,13
4,16
4,14
4,19

4,174,18
4,20
4,8

3,9

3,10
3,13
3,16

3,14

3,17
3,19

3,8
3,18
3,20

6,9

6,13
6,10

6,14

6,16

6,12

6,20
6,176,18
6,8
6,19

5,16

5,13
5,9

5,15
5,12
5,14

5,19

5,17

5,8
5,18
5,11
5,20

7,10

7,13
7,9

7,14
7,16
7,12

7,20
7,17
7,8
7,187,19
7,15

10,11

10,16

10,14
10,13

10,12

10,19
10,18
10,17
10,20

9,18

9,10
9,17
9,15

9,199,20

8,17
8,188,198,20
8,9
8,118,12
8,10
8,148,158,16

11,13

11,16

11,14

11,15

11,12

11,19
11,17

11,20
11,18

12,16

12,14
12,13

13,20

13,18
13,17
13,19
13,14
13,16
13,15

12,15
12,17

12,18
12,19
12,20

14,20

14,18
14,17
14,19

14,15
14,16

15,17
15,18
15,16
15,19
15,20

16,20

16,17
16,19
16,18

17,18

17,19
17,20

18,19
18,20

19,20

 

21

1

16

6
Discordant Pair Odds Ratios by PE

6
Discordant Pair Odds Ratios by PE

11
NSDUH Public Use Dataset. 
Odds ratios and 95% confidence intervals produced from generalized linear model logistic 
regression. 
X-axis: PE pairs. Y-axis: OR and 95% CI values. 
Lag-intervals 1-13 collapsed. 
Analysis weights use Taylor Series Linearization for calculus-based variance estimation. 
Upper-bound confidence intervals ³20 capped at 20; corresponding point estimates capped at 
20.  

16

11

21

 

40

30

20

10

0

40

30

20

10

0

1

 
 

 

40

30

20

10

0

40

30

20

10

0

79 

Table 35. Sensitivity Analysis: Weighted Odds Ratios of Problem and Experience Pairs Within 0-150 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time intervals 1-3: possible 0-150 days. 
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

 

80 

 
 
 
 

 

Table 36. Sensitivity Analysis: Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 0-150 Days of 
First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time intervals 1-3: possible 0-150 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation.  

 

 

81 

 
 
 
 

 

Table 37. Sensitivity Analysis: Weighted Odds Ratios of Problem and Experience Pairs Within 150-300 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time intervals 7-8: possible 150-300 days. 
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

 

82 

 
 
 
 
 

 

Table 38. Sensitivity Analysis: Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 150-300 Days 
of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 7-8: possible 150-300 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation.  

 

83 

 
 
 
 
 
 

 

Table 39. Sensitivity Analysis: Weighted Odds Ratios of Problem and Experience Pairs Within 300-390 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
*Statistically significant: 95% confidence interval does not include 1.0.  
Values of 1.0 indicate null associations or inability to calculate OR.   
Lag-time intervals 12-13: possible 300-390 days. 
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

84 

 
 
 
 
 

 

Table 40. Sensitivity Analysis: Weighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 300-390 Days 
of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 12-13: possible 300-390.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation.  
  
 
 

 

 

85 

4.4 Discussion  
 

There is tangible evidence that heroin use disorder syndrome begins to form within 

months of first using heroin. This is made evident by the incidence of PE pairs which qualify as a 

mild case of DSM HUD. What is unique to this study is the slicing of cross-sectional survey 

assessment data into increments of increasing lengths of time participants would have had to 

develop HUD. The natural history of HUD is illustrated by PE pair associations that show 

clinical features’ coalescence over time.  

Within the first months of using heroin for the first time the most prominent associations 

relate to having use affect personal relationships, physical danger, and ability to do important 

activities. Any two of these PEs alone are enough to make a DSM-5 diagnosis of HUD. Having a 

PE of spending less time doing important activities is associated with continuing use despite 

problems with family or friends (OR=936.4, 95% CI=66.6, 13,164.9). A PE of spending less 

time doing important activities is associated with physical danger (OR=873, 95% CI=70.8, 

10,785.4) and physical danger is associated with continuing to use despite problems with family 

or friends (OR=871, 95% CI=60.6, 12,511.5). Point estimates are robust and 95% confidence 

intervals are statistically significant.  

Within 1-120 days of first using heroin the HUD syndrome appears to form through the 

coalescence of continued use despite emotional problems, being unable to cut down, being 

unable to set limits, and having withdrawal syndrome symptoms. Pair-wise combinations tend to 

have robust associations:  Continued use despite problems with emotions/unable to cut down 

(OR=551.7, 95% CI=45.4, 6,700.2); continued use despite problems with emotions/having three 

or more withdrawal symptoms (OR=551.7, 95% CI=45.4, 6,700.2); unable to set limits/having 

three or more withdrawal symptoms (OR=514.8, 95% CI=36.7, 7,220.3). 

 

86 

From 30-150 days of first heroin use prominent associations are seen among PE such as 

signs of tolerance, continued use despite problems with emotions, withdrawal symptoms, unable 

to set limits, and physical problems. The strongest combination is between the same amount 

having less effect (i.e., signs of tolerance) and spending a lot of time getting over the effect 

(OR=320.6, 95% CI=15.7, 6,536.7). Additionally, continuing to use despite problems with 

emotions and the same amount having less effect (OR=213.4, 95% CI=16.3, 2,796.4), spending a 

lot of time getting or using and unable to set limits (OR=166.4, 95% CI= 18.8, 1,472.5), and 

trouble with the law and physical problems (OR=150.7, 95% CI=6.6., 3,425.3) have a high 

chance of presenting together.  

Time intervals included about a 90-day uncertainty regarding the amount of time 

participants had to develop HUD from first use. The sensitivity analysis rid intervals of this err in 

exchange for a less fine grain natural history assessment. The strongest PE to PE coalescence 

among participants with 0-150 days since first heroin use to assessment were trying to cut down 

at least once and being unable to cut down – PEs within the same DSM-5 HUD criteria. Being 

unable to set limits and withdrawal syndrome symptoms were strongly associated with each 

other (OR=416.7, 95% CI=35.8, 4,845.8) and qualify as a HUD diagnosis. Other strong 

associations were among spending a lot of time getting or using heroin and withdrawal 

symptoms (OR=127.4, 95% CI=18.1, 897.2) and having problems with emotions along with 

withdrawal symptoms (OR=107.9, 95% CI=10.2, 1,140.8). Both PE pairs qualify as a HUD 

diagnosis.   

Between 150-270 days after first using heroin results suggest characteristic PEs include 

spending a lot of time getting or using, signs of tolerance, and spending less time doing 

important activities. A robust association is seen between diagnosable HUD PEs spending a lot 

of time getting or using and needing more to get the same effect (OR=1,008.0, 95% CI=99.4, 

 

87 

10,219.5). Additionally, spending less time doing important activities and spending a lot of time 

getting/using or needing more to get the same effect also have an increased chance of presenting 

together (OR=258.3, 95% CI=41.3, 1,615.2; OR=369.7, 95% CI=50.0, 2,731.6, respectively). 

Sensitivity analyses roughly corresponding to this timeframe (i.e., 150-300 days) suggest 

problems with family or friends increases the risk of also having serious problems or trouble 

with the law (OR=663.0, 95% CI=147.5, 2,979.8; OR=242.0, 95% CI=25.9, 2,260.7, 

respectively). Spending a lot of time getting or using and spending less time doing important 

activities also appear to have a strong coalescence during this time.  

Within 300-420 days of first using heroin, HUD is evident from robust associations 

between spending less time doing important activities and use causing serious problems 

(OR=261.8, 95% CI=28.3, 2,425.0 ), having problems with emotions and spending less time 

doing important activities (OR=217.6, 95% CI=25.4, 1,864.5), and spending less time doing 

important activities and physical danger (OR=115.1, 95% CI=8.3, 1,597.0). Each of these 

combinations qualify as a DSM-5 HUD diagnosis because PE pairs include two distinct DSM 

criteria. Within 330-450 days of first using, HUD is again evident from large associations 

between spending a lot of time getting or using, having problems with emotion, spending less 

time doing important activities, unable to set limits, and continued use despite problems with 

family or friends.  

Sensitivity analysis confirm the time-frame of 300-450 days of first using heroin strongly 

correlates with spending a lot of time getting or using heroin, having problems with emotions, 

continued use despite problems with family or friends, physical danger, and spending less time 

doing important activities. The strongest coalescence is seen between spending a lot of time 

getting or using and continued use despite having problems with family or friends (OR=558.3, 

 

88 

95% CI=64.9, 4,803.9). Not only does this PE pair qualify as a HUD diagnosis, but it has been 

shown to be a prominent combination throughout lag-time intervals.  

Problems with family or friends is a prominent PE associated with the development of 

HUD. Shane Darke’s work within the National Drug and Alcohol Research Centre in Australia 

describes early-life parent/family discord such as parental drug use, separation/absence, 

psychopathology, and child abuse (50). Much of those with HUD as adults experienced 

childhood psychopathology, specifically antisocial behavioral disorders and 40-80% met 

diagnostic criteria for childhood conduct disorder (50). This connection between personal 

relationships and heroin use may point to: 1) a feedback loop between family relationships and 

psychiatric comorbidities with HUD 2) the degree to which heroin corresponds to 

neuroadaptation responsible for personal relationships, and/or 3) HUD intervention points during 

latency period(s).  

Entangled within a strong correlation between spending a lot of time getting or using 

heroin and continued use despite problems with family or friends is a likely feedback loop. 

Animal model research has suggested that drug seeking behavior can be precipitated by stressful 

stimuli (70). Equally, PE to PE associations between stressful stimuli, for example physical 

danger, trouble with the law, having serious problems, may drive NIHUs to spend more time 

getting/using heroin. Based on that premise, HUD could possibly develop in a manner 

interdependent with the consequences of the initial heroin use.  

Strengths and Limitations: The current study relies on annual cross-sectional surveys 

intended to yield nationally representative samples. There are three limitations intrinsic to this 

study design: 1) Left truncation and left censoring may underestimate the number of NIHU. To 

directly combat challenges of left truncation and censoring, the newly incident heroin use case 

definition (i.e., within 12 months of assessment) restricts the amount of time participants have to 

 

89 

develop HUD-related experiences that would keep them from being sampled or completing the 

interview (e.g., incarceration, fatal overdose). Even so, considering the ratio of newly incident 

heroin users to those who die of heroin overdose in the first year of use, potential left truncation 

is negligible. The same is true of left censoring when reflecting on those incarcerated within the 

first year of use. 2) The study’s reliance on self-report merits attention, yet there is currently no 

feasible alternative for surveys assessing history of drug use.  3) Cross-sectional surveys are 

often criticized for limited temporal sequencing. Notably, the current study circumvents this 

issue given participants cannot develop heroin use disorder without first having tried the drug.  

Likely explanations for the robust point estimates and wide confidence intervals may 

reflect: 1) There may be issues of sample size and small cell counts. Despite the rise in heroin 

use incidence, it is still a relatively rare phenomenon. Analytically, this poses challenges. Thinly 

slicing cross-sectional data to ascertain NIHUs within 12 months of assessment and further 

stratifications yield small sample sizes within each lag-time interval and involve zero cells when 

calculating PE to PE ORs. As a result, OR point estimates are notably large and the associated 

95% confidence intervals are wide. 2) There may also be exchangeability between PE pairs. 

Exchangeability is seen when two cross tabulated binary variables that may be measuring the 

same underlying trait. PEs within the same DSM-5 HUD criteria (e.g., spending a lot of time 

getting or using/spending a lot of time getting over the effects) readily speak to the concept of 

exchangeability in that the two PEs are measuring the same behavior. Large ORs of PEs that are 

not in the same diagnostic criteria may reflect the two PEs picking up on the same construct, 

HUD.  

Via the logistic regression models, the strength of association linking each PE pair might 

be inverse or might be positive. The OR estimator has been used because it is not margin-

 

90 

dependent. Other statistical indices of agreement and association such as kappa or correlation 

coefficients would depend upon the marginal frequencies of the PE.  

4.5 Conclusions  
 

Among all drug use disorder, that of heroin has the highest burden of disease with 

mortality rates comparable to that of the elderly (50). That combined with the continued upward 

trend of opioid mortality in the US should be concerning to researchers, public health officials, 

healthcare teams, and all as citizens. Prevention and intervention efforts have already curbed 

prescription opioids by implementing stricter physician prescribing recommendations and pill 

return centers. In contrast, heroin use interventions are lagging behind in prevalence and 

efficacy.  

With this study suggesting heroin-associated emotions and interpersonal relationships as 

important manifestations of HUD development, intervention efforts might consider a focus on 

that arena. For instance, clean needle exchanges could make available group therapy between an 

individual user and their non-drug using support system. This could be modified for primary 

prevention by providing therapy for prescription pain relievers since they are more vulnerable to 

heroin use later. The group counseling would be between the individual receiving the medication 

and others living in their household and/or close family and friends. To promote adherence, 

therapy via telephone could be an option. Including individuals within the patient’s already 

established social network would call upon Valente’s work on social diffusion (71). That is, it 

may be beneficial to work on controlling the diffusion of opioid use to different social networks 

and help keep a strong support system within the existing network (71).  

In general, intervention efforts geared toward individuals who use heroin should 

purposefully include close friend and family of the user in order to work within a social network. 

This may include but is not limited to education efforts, naloxone training, surveillance, 

 

91 

rehabilitation, and physical activity-based interventions. Next steps for the field include adapting 

research to randomized control trials which would allow immediate intervention and inferences 

that may more readily speak to causal interpretations.  

 
 
 
 
 

 

92 

CHAPTER 5 MANUSCRIPT 2 – EVALUATING MEASUREMENT EQUIVALENCE OF 

HEROIN USE DISORDER ACROSS HISTORY OF EXTRA-MEDICAL 

PRESCRIPTION OPIOID USE AND SEX 

5.0 Abstract 
 
Aim: The current study aims to assess measurement equivalence of NSDUH audio computer 

assisted self-interview module as a measure of heroin use disorder (HUD), assuming a single 

dimension to evaluate against sex and extra-medical prescription opioid use history.  

Methods: National Surveys on Drug Use and Health (NSDUH) 2002-2016 aggregated files 

include 837,326 participants with 896 newly incident heroin users (NIHUs) identified. Problems 

and experiences (PE) of HUD were recoded into 10 diagnostic criteria dichotomous indicators. A 

single factor model was assumed for HUD. Chi-square tests for model differences compared 

nested models to assess metric and scalar equivalence across sex and extra-medical prescription 

opioid use history (EMPOU).  

Results: Assessing HUD across sex, models constraining loadings and thresholds fail to reject 

the null hypotheses of equal model fit. Constrained loadings vs free model: c2 = 10.907 (p>0.05); 

constrained loadings and threshold compared to free model: c2=26.730 (p>0.05). Assessing 

HUD across EMPOU, models constraining reject the null hypotheses of equal model fit (c2 = 

10.271, p<0.001).  

Discussion: With respect to HUD measurement equivalence, the estimates indicate loadings and 

thresholds are not appreciably different when males-female contrasts are made; the null 

hypothesis (equivalent values) cannot be rejected. Conversely, measurement equivalence was not 

confirmed for HUD across EMPOU subgroups.  

Conclusions: The study supports continued study of male-female variations in HUD levels 

without calibration of measurements, but the study of HUD levels across EMPOU subgroups 

 

93 

will require more research with calibration of the current tools used to measure HUD as a 

dimensional construct. 

 

94 

5.1 Introduction 
 

HUD can be studied via algorithmic combinations of 20 problems and experiences (PE) 

items designed to measure 10 criteria designated by the Diagnostic and Statistical Manual 52 

(Table 41, 42)3. Alternately, HUD can be studied as a dimensional latent construct for which 

NSDUH provides no tangible ‘gold-standard’ diagnosis. Since HUD, as a latent construct, is not 

directly observable in these data, there is uncertainty about the appropriate parameters of the 

dimensional measurement model. More specifically, when one combines PE experiences for 

comparison of HUD levels across subgroups of newly incident heroin users (e.g., females versus 

males; prior opioid users versus never users), is there validity of the comparisons when an 

uncalibrated HUD measurement is used? Research has yet to answer this question as it relates to 

HUD levels and potential sex-differences or exposure to extra-medical prescription opioid use.  

Measurement Equivalence: Measurement equivalence (or invariance) involves 

elaborated structural equation modeling used to compare variation or differences of latent 

constructs between different groups or time points (72). In simpler terms, it is a statistical 

property meant to demonstrate that the construct of study can be measured without calibration 

challenges across subgroups or across time (73). While latent constructs are by definition not 

directly observable (e.g., depression or HUD level) there are observable indicators of the 

construct that are measurable (e.g., sad or depressed mood; heroin withdrawal). A confirmatory 

factor analysis (CFA) can be used for testing the construct validity in which there is already an 

                                                
2 The National Surveys on Drug Use and Health, 2002-2016, base survey questionnaire on DSM-IV criteria which is 
different from DSM-5 in that it includes the problem/experience of having trouble with the law. DSM-IV does not 
include the problem/experience of craving which is included in DSM-5.  
3 Symptoms listed when asking about withdrawal experience: 1) feeling kind of blue or down, 2) vomiting or feeling 
nauseous, 3) having cramps or muscle aches, 4) having teary eyes or a runny nose, 5) feeling sweaty, having 
enlarged pupils, or having body hair standing up on your skin, 6) having diarrhea, 7) yawning, 8) having a fever, and 
9) having trouble sleeping. 

 

95 

established conceptual and measurement framework (73). Using CFA, evaluation of three forms 

of measurement equivalence can be achieved: configural, metric, and scalar (Table 43). 

The current study evaluates the continuous latent construct HUD based on ordered 

categorical indicators across multiple groups. Such conditions require unique identification 

conditions in order to estimate the model parameters (74)(75). Namely, the scale (i.e., mean and 

variance) of ordered categorical indicators is unknown, which requires additional model 

constraints (74). There are various recommendations on ways to constrain said model. Resulting 

parameter estimates may vary, but the standardized parameter estimates and fit indices will not 

be affected (74).   

There are two parameterizations that can be used for multiple group analysis - the delta 

approach and the theta approach (76). This study will use the theta approach in which the 

residual variances are standardized to equal one in the overall group and are free for subgroups 

(76). For this reason, the stepwise constraints to assess levels of measurement equivalence will 

not include constraints on residual variances. Additional identification conditions, suggested by 

Millsap and Tien (2004), that are used in this study are outlined within the methods section.  

Sex Differences: Both males and females have experienced an increase in heroin use, 

with aggregate data suggesting males age 25-44 to be the largest using demographic subgroup 

(6). Likewise, past epidemiologic research has shown that drug use is a phenomenon 

predominantly among males (22)(36)(37), yet as of late, studies have suggested otherwise for 

several drugs (e.g., alcohol) (38)(39)(40). Male and female prevalence of heroin use once 

suggested a large disparity, has narrowed respectively from 2.4 and 0.8 persons/1,000 in 2002-

2004 to 3.6 and 1.6 persons/1,000 in 2011-2013 (41). In simpler words, over time sex differences 

in the prevalence of heroin use have diminished over time.  

 

96 

There may be biological differences between males and females that influence sex-

specific reactions to drugs. Ovarian hormones or metabolites may influence females’ experience 

with drugs as studies have shown that the subjective high felt can fluctuate with the female 

menstrual cycle (77). Along the same lines, some studies indicate for females a lessened stress 

response after drug use. These differences seem to be unique to female drug use and in theory 

could possibly influence susceptibility to HUD. There is evidence that females more rapidly 

progress to a use disorder after drug use onset and have experienced more severe job-related 

impairment related to opioid use disorder (77), but measurement equivalence issues have not 

been resolved. Thus, before making male-female comparison of HUD levels, there is reason to 

investigate sex differences that might create measurement problems when measuring HUD levels 

or in the progress on HUD development.  

Extra-Medical Prescription Opioid Use: It is well established that there is an 

association between extra-medical prescription opioid use and later progress to heroin use 

(78)(79). Pharmacologically, heroin and prescription opioids are very similar in their interactions 

with the endogenous opioid system in the body. Both involve mu-opioid receptor signaling and 

influence dopamine availability in brain regions such as the nucleus accumbens (78). However 

there are differences across heroin and prescription opioids can be seen in various parameters: 

lipophilicity, mu-receptor binding affinities, and pharmacokinetic properties (78). This leaves 

uncertainty as to whether having used prescription opioids extra-medically in the past may 

differentially influence measurement of HUD as a latent dimensional construct. It may be that 

EMPOU history is associated with a more rapid progression across HUD levels after initiation, 

but measurement calibration might be required before measurement equivalence is achieved. 

Overview: Research has yet to evaluate whether HUD level as a latent construct is 

comparable between males and females. Based on the fact that females tend to start heroin use 

 

97 

younger than males, it may be that likewise females outpace males in HUD development (38). 

Additionally, HUD diagnoses do not account for previous exposure to opioids, which may in fact 

prime individuals in terms of the various indicators used to measure HUD (e.g., spending a lot of 

time getting or using; tolerance). The current study aims to assess measurement equivalence of 

heroin use disorder, assuming a single dimension to evaluate against sex and extra-medical 

prescription opioid use history. 

A priori hypotheses about new heroin users were: (1) Males and females are not comparable in 

HUD measurement; (2) Subgroups defined by extra-medical use of prescription opioids, past 

year, lifetime, and never users are not comparable in HUD measurement. 

 

 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

98 

by your use of heroin?  

work, or school?  

have put you in physical danger?  

 “During the past 12 months…” 

Problems & Experiences 

using heroin?  

wanted?  

were probably caused or made worse by your use of heroin?  

important activities (spending time with friends & family)?  

have problems with your emotions, nerves, or mental health?  

Table 41. Problems and Experiences of Heroin Use Based on Diagnostic Criteria. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
PE 
 # 
1  Did you try to set limits on how often or how much heroin you used?  
2  Did you have any problems with your emotions, nerves, or mental health that 
3  Was there a month or more when you spent a lot of your time getting or 
4  Did you need to use more heroin than you used in order to get the effect you 
5  Did you have any problems with family or friends that were probably caused 
6  Did using heroin cause you to give up or spend less time doing these types of 
7  Did using heroin cause you to have serious problems like this either at home, 
8  Did you cut down or stop using heroin at least one time?  
9  Did you continue to use heroin even though you thought it was causing you to 
10  Did you regularly use heroin and then do something where using heroin might 
11  Did you continue to use heroin even though you thought it caused problems 
12  Were you able to keep to the limits you set, or did you often use heroin more 
13  Were you able to cut down or stop using heroin every time you wanted to or 
14  Did you have 3 or more of these symptoms after you cut back or stopped 
15  Did you have 3 or more of these symptoms at the same time that lasted for 
16  Did using heroin cause you to do things that repeatedly got you in trouble with 
17  Did you notice that using the same amount of heroin has less effect on you 
18  Was there a month or more when you spent a lot of your time getting over the 
19  Did you have any physical problems that were probably caused or made 
20  Did you continue to use heroin even though you thought it was causing you to 
NA  Craving 
NSDUH Downloadable Public Use Dataset. 

longer than a day after you cut back or stopped using heroin? 

the law?  

that it used to?  

with family or friends? 

than you intended to?  

effects of the heroin you used? 

worse by using heroin?  

have physical problems?  

tried to?  

using heroin? 

DSM 
IV  5 
x 
x 
x 
x 

x 

x 

x 

x 

x 
x 
x 

x 

x 

x 

x 

x 

x 

x 

x 

x 

x 

x 
 

x 

x 

x 

x 

x 
x 
x 

x 

x 

x 

x 

x 

x 

 

x 

x 

x 

 
x 

99 

 
 

 

Inability to manage commitments due to use 

Table 42. DSM-5 Use Disorder Criteria and Corresponding Heroin Use Disorder Problems and 
Experiences. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
#  Criterion 
1  Using in larger amounts or for longer than intended 
2  Wanting to cut down or stop using, but not managing to 
3  Spending a lot of time to get, use, or recover from use 
--  Craving 
4 
5  Continuing to use, even when it causes problems in relationships 
6  Giving up important activities because of use 
7  Continuing to use, even when it puts you in danger 
8  Continuing to use, even when physical or psychological problems 
9 
10  Withdrawal symptoms 
NSDUH Public Use Dataset.  
Source: American Psychiatric Association, 2013 (19). 
PE16 (problems with the law) not included in DSM-5 criteria. Craving not included in 
NSDUH survey assessment. 

Corresponding PE 
1, 12 
8, 13 
3, 18 
N/A 
7 
5, 11 
6 
10 
2, 9, 19, 20  
4, 17 
14, 15 

may be made worse by use 
Increasing tolerance 

Table 43. Description of Confirmatory Factor Analysis Measurement Equivalence 

Levels  

Description 

Configural  Null model. 

 

 

Metric 

Scalar 

Factor loading, thresholds (i.e., 
intercepts), and residual 
variances are free. 
 
Factor loadings constrained. 
 

Factor loadings and thresholds 
constrained. 
 

Fit Indices 
Overall fit 

 

 

Chi-Square 

Chi-Square 
RMSEA 

CFI/TLI 

Good Fit Standards 

 

 

The smaller the better. 
 
The smaller the better. 
Unacceptable >0.10 
Good <0.06 
Acceptable >0.90 
Good >0.95 

 

 
Table based on Claremont Evaluation Center. (73) 
Link available: 
http://comm.eval.org/HigherLogic/System/DownloadDocumentFile.ashx?DocumentFileKey=
63758fed-a490-43f2-8862-2de0217a08b8 

 

 

100 

 
 

 
 
 

 

Figure 10. Diagram of Confirmatory Factor Analysis Model Testing Measurement 
Equivalence of Problems and Experiences of Heroin Use Disorder Among Newly Incident 
Heroin Users. 

 
 
 
Heroin 
 
Use 
 
ll 

Disorder 

l1 

l2 

l20 
l19 

PE19 

PE20 

  

 

e19 

e20 

 

                                                                            
 

Latent variable: heroin use disorder.  
Indicators: Problems and Experiences (PE) numbers 1 through 20.  
Residual error: e 
Factor loadings: l  
CFA diagram with for DSM-5 criteria replacing problems and experiences (PE) indicators.  

PE1 

PE2 

  

e1 

e2 

 
5.2 Methods 

Sample: The study’s population for the US National Surveys on Drug Use and Health 

(NSDUH), 2006-16, consists of non-institutionalized US civilians in all 50 states and D.C., 

restricted to those 12 to 21 years of age. Multi-stage area probability sampling and institutional 

review board approved protocols for recruitment and standardized computer-assisted self-

interviews were used. As a result, aggregate files included 837,326 participants, with 896 

identified as newly incident heroin users (NIHUs) from survey years 2002-2016. NIHUs were 

assessed within 12 months after first use, with all past users excluded (Figure 11).  

The US Department of Health and Human Services Center for Behavioral Health Statistics 

and Quality created NSDUH open access analysis files with de-identified data after applying 

disclosure analyses designed to prevent re-identification of participants. Given the circumstances 

 

101 

of open access and no contact with participants, the Michigan State University institutional 

review board ruled that plans to analyze these data qualified for the federal category “not human 

subjects research.” 

Measurements: Observable indicators of HUD levels included 20 questions on problems 

and experiences (PE) based on the Diagnostic and Statistical Manual IV criterion for case 

ascertainment (Table 45). PE indicators were coded as binary response variables. NSDUH 

survey questions use ‘gating’ (e.g., based on NSDUH measurement assumptions or on logic); 

hence, not all 896 NIHUs are asked all 20 questions. To have PE-specific proportions with the 

same denominator, 896, a skipped item is coded 0 and reflects the measurement assumption). All 

PE variables were recoded into 10 binary indicators based on DSM-5 HUD case ascertainment 

criteria (Table 46). Among 20 PEs, a DSM-IV ‘legal troubles’ PE was dropped to match DSM-5 

HUD criteria. (Note: NSDUH 2002-2016 did not assess DSM-5-specified ‘craving.’) 

Subgrouping of new heroin users involved sex (i.e., male, female) and extra-medical 

prescription opioid use history (EMPOU). Subgroups were: 1) EMPO use <=12 months of 

assessment, 2) lifetime EMPOU history >12 months before assessment, and 3) never-EMPOU.  

Statistical Analyses: PE and criteria binary variables were recoded using Stata SE 12. 

Corresponding PE and criteria-specific proportions (‘attack rates’) were estimated with 

numerators based on new heroin users ‘positive’ for the corresponding PE/criterion. 

Measurement equivalence was assessed via MPlus version 7.3 software.  

HUD levels were measured with the 10 criterion indicator variables that combined 20 PE 

item responses. A single-factor model in which each observed indicator loads on only one factor 

is identified as a congeneric factor structure. In this case, binary indicators imply that only one 

threshold parameter is needed for each observed indicator. This congeneric model with 

 

102 

dichotomous indicators requires an alternative set of constraints relative to polytomous indicators 

(75): 

1.  All thresholds are free to vary, except for a subset that are constrained to be invariant 

across subgroups.  

2.  Common factor matrices vary freely. Common factor means in the overall group are 

fixed at zero, but subgroup means freely vary. 

3.  All unique factor matrices are fixed overall; within subgroups, they vary freely. 

4.  The thresholds are fixed to be equal for each subgroup. 

5.  For all subgroups besides overall group, diagonal elements of sum unit values are 

fixed.   

This congeneric model does not allow residual variances to be estimated. Three models 

were estimated, two of which were nested within the baseline model for comparison. Model 1 

allowed both thresholds and factor loadings for all subgroups to be freely estimated. Model 2 

constrained factor loadings with chi-square used to test model differences between Model 1. 

Model 3 constrained thresholds and factor loadings with chi-square used to test model 

differences between Model 1. Configural equivalence was assumed to allow all indicators to load 

onto one ‘Level of HUD’ factor. Metric equivalence was assessed in Model 2. Scalar 

equivalence was assessed in Model 3.  

Goodness of fit statistics such as RMSEA, CFI, and TLI were also used. RMSEA values 

less than 0.05 are generally considered to be of good fit (73). CFI and TLI values greater than 

0.95 are generally regarded as indicative of good model fit (73).  

Initial unweighted analyses were repeated with analysis weights using Taylor Series 

Linearization for calculus-based variance estimation. MPlus specifications included complex 

analysis type and weight, stratification, and cluster commands.  

 

103 

The dichotomous congeneric model does not allow residual variances to be estimated. 

Three models were estimated, two of which were nested within the baseline model for 

comparison. Model 1 allowed both thresholds and factor loadings for all subgroups to be freely 

estimated. Model 2 constrained factor loadings with chi-square used to test model differences 

between Model 1. Model 3 constrained thresholds and factor loadings with chi-square used to 

test model differences between Model 1. Configural equivalence was assumed to allow all 

indicators to load onto one factor for HUD assessment. Metric equivalence was assessed in 

Model 2. Scalar equivalence was assessed in Model 3.  

Goodness of fit statistics such as RMSEA, CFI, and TLI were also used. RMSEA values 

less than 0.06 are considered to be of good fit (73). CFI and TLI values greater than 0.95 are 

considered to be of good model fit (73).  

Initial unweighted analyses were repeated with analysis weights using Taylor Series 

Linearization for calculus-based variance estimation. MPlus specifications included complex 

analysis type and weight, stratification, and cluster commands.  

 

 
 

 

104 

2003 

2004 

2005 

2006 

2007 

2008 

2009 

n=54,079  n=55,230  n=55,602  n=55,905  n=55,279  n=55,435  n=55,110  n=55,234 

2012 

n=55,268 

2013 

n=55,160 

2014 

n=55,271 

2015 

n=57,146 

2016 

n=56,897 

2002 

2010 

2011 

n= 57,313 

n= 58,397 

Figure 11. Flow of Participants in Newly Incident Heroin Use Case Ascertainment. 
Data from United States National Surveys on Drug Use and Health, 2002-2016.  

NSDUH – Year-Specific Files 

NSDUH  

n= 837,326 

2002-2016 Aggregate Files 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
NSDUH Downloadable Public Use Dataset.  
Unweighted counts (n).  
Newly incident heroin use defined as first heroin use within 12 months of survey assessment; 
past onset users excluded.  

Newly Incident Heroin Users 

Never Used Heroin 

Past Onset Heroin 

n=825,420 

n=9,153 

n=896 

105 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

Table 44. Characteristics of Newly Incident Heroin Use Sample. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

Aggregated NSDUH 

2002-2016 
n=837,326 

n (%) 

Newly Incident  
Heroin Users 

n=896 
n (%) 

Demographic Characteristics  

 

 

Sex 

Age Groups (Years) 

Male 

12-17 
18-25 
26-34 
35-49 
50-64 
65+ 

401,270 (48) 
 
258,309 (10) 
263,258 (13) 
93,449 (14) 
128,684 (25) 
56,998 (22) 
36,628 (16) 
 
532,503 (67) 
106,106 (12) 
12,284 (<1) 
4,042 (<1) 
30,449 (5) 
26,309 (1) 
134,633 (14) 

476 (62) 
 
227 (15) 
539 (47) 
82 (22) 
44 (14) 
4 (2) 
0 
 
715 (81) 
21 (4) 
20 (<1) 
8 (1) 
5 (1) 
31 (1) 
96 (10) 

Ethnic Self-Identification 

White 
Black/African American 
Native American/Alaskan  
Native Hawaiian /Other Pacific Island. 
Asian 
>1 Ethnicity 
Hispanic 

Extra-Medical Prescription Opioid History 

Lifetime Use (>12 Months*) 
Within 12 Months* 
Never Used 
Missing Data 

89,390 (11) 
88, 649 (9) 
608,112 (74) 
51,175 (5) 

93 (10) 
677 (76) 
112 (13) 
15 (2) 

NSDUH Downloadable Public Use Data. 
NIHUs: Newly incident heroin users first used heroin within 12 months of survey assessment. 
Past onset users excluded. 
Ethnic self-identification: all ethnicities other than Hispanic are specified as non-Hispanic per 
survey assessment. 
*More than or within 12 months of survey assessment. 
Unweighted n; weighted % 

106 

 
 
 
 
 
 
 
 
 
 
 
 

 

249 

896 

242 

896 

232 

896 

216 

896 

197 

896 

193 
189 

173 

896 
896 

896 

164 

896 

150 

896 

136 

896 

108 

896 

79 

70 

68 

41 

40 

27 

896 

896 

896 

896 

896 

896 

home, work, or school?  

caused by your use of heroin?  

using heroin?  

you wanted?  

40.7  Did you try to set limits on how often or how much heroin you used?  
31.4  Did you have any problems with your emotions, nerves, or mental 
health that were probably caused or made worse by your use of heroin?  
27.6  Was there a month or more when you spent a lot of your time getting or 
27.9  Did you need to use more heroin than you used in order to get the effect 
25.6  Did you have any problems with family or friends that were probably 
24.4  Did using heroin cause you to give up or spend less time doing these 
types of important activities (spending time with friends & family)?  
22.5  Did using heroin cause you to have serious problems like this either at 
22.7  Did you cut down or stop using heroin at least one time?  
23.3  Did you continue to use heroin even though you thought it was causing 
you to have problems with your emotions, nerves, or mental health?  
20.8  Did you regularly use heroin and then do something where using heroin 
17.8  Did you continue to use heroin even though you thought it caused 
17.3  Were you able to keep to the limits you set, or did you often use heroin 
12.4  Were you able to cut down or stop using heroin every time you wanted 
9.7  Did you have 3 or more of these symptoms after you cut back or 
9.0  Did you have 3 or more of these symptoms at the same time that lasted 
8.2  Did using heroin cause you to do things that repeatedly got you in trouble 
5.7  Did you notice that using the same amount of heroin has less effect on 
6.9  Was there a month or more when you spent a lot of your time getting 
3.2  Did you have any physical problems that were probably caused or made 
1.5  Did you continue to use heroin even though you thought it was causing 

for longer than a day after you cut back or stopped using heroin?* 

might have put you in physical danger?  

over the effects of the heroin you used? 

problems with family or friends? 

with the law?  

you that it used to?  

to or tried to?  

stopped using heroin?* 

more than you intended to?  

Table 45. Description of Problems and Experiences involving Heroin Use Disorder. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

 “During the past 12 months… “ 

Problems & Experiences 

Counts  Wt. 
n. 
d.  % 
896 
361 

worse by using heroin?  

896 

13 
NSDUH Downloadable Public Use Dataset. 
Unweighted n: numerator & d: denominator; % weighted percentages.  

you to have physical problems?  

 

107 

 

 
 
 

 

Table 46. Description of DSM-5 Criteria for Heroin Use Disorder. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
Counts  Wt. 
n. 
361  896  40.7  Using in larger amounts or for longer than intended. 
209  896  24.8  Wanting to cut down or stop using, but not managing to. 
282  896  34.5  Spending a lot of time to get, use, or recover from use. 

d.  %  DSM-5 Criteria 

 

-- 

-- 

--  Craving. 

 

PE #s 
1, 12 
8, 13 
3, 18 
N/A 
7 
5, 11 
6 
10 
2, 9, 
19, 20 
4, 17 
14, 15 

193  896  22.5  Inability to manage commitments due to use. 
216  896  25.6  Continuing to use, even when it causes problems in relationships. 
197  896  24.4  Giving up important activities because of use. 
164  896  20.8  Continuing to use, even when it puts you in danger. 

Continuing to use, even when physical or psychological problems 
may be made worse by use. 

276  896  34.6 
273  896  33.6  Increasing tolerance. 
130  896  16.0  Withdrawal symptoms. 
NSDUH Public Use Dataset.  
DSM-5 Criterion Source: American Psychiatric Association, 2013 (19). 
PE16 (problems with the law) not included in DSM-5 criteria. Craving is in DSM-5, but is not 
included in NSDUH survey assessment of heroin use disorder.  
Counts: unweighted n(numerator) and d(denominator); % weighted percentages rounded to 
thousandths place. 

108 

Table 47.  Goodness of Fit Indices Evaluating Measurement Equivalence of Heroin Use Disorder Level Across Sex Among Newly 
Incident Heroin Users (Without Analysis Weights). 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
 

df 

df 
70 
 
80 

c2 

-- 
 
10.907 

298.293 
 
181.790 

p-value 
<0.001 
 
<0.001 

Chi-Square Model Fit 
c2 

Chi-Square for Difference Testing 
p-value 
-- 
 
0.3648 

Models 
M1 – free model 
 
M2 – constrain loadings 
M1 vs M2 
 
M3 – constrain botha 
M1 vs M3 
 
NSDUH Downloadable Public Use Dataset.  
Sex: male/female. RMSEA: root mean square error of approximation 
Unweighted estimates.  
--Null model M1 is the full model used to test differences compared M2-M3 constrained models.   
aloadings and thresholds constrained. 

-- 
 
10 

 
20 

 

 
<0.001 

 
26.730 

 
198.667 

 
0.1430 

 

 

 

 
90 

 

 

Comparative 
CFI 
TLI 
0.982 
 
0.993 

0.986 
 
0.994 

Absolute 
RMSEA 
0.085 
 
0.053 

 
0.993 

 
0.993 

 

 

 
0.052 

 

109 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

Table 48. Goodness of Fit Indices Evaluating Measurement Equivalence of Heroin Use Disorder Level Across Sex Among Newly 
Incident Heroin Users with Analysis Weights. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
 

df 

df 
70 
 
80 

c2 

-- 
 
15.850 

139.526 
 
137.391 

p-value 
<0.001 
 
<0.001 

Chi-Square Model Fit 
c2 

Chi-Square for Difference Testing 
p-value 
-- 
 
0.1040 

Models 
M1 – free model 
 
M2 – constrain loadings 
M1 vs M2 
 
M3 – constrain botha 
M1 vs M3 
 
NSDUH Downloadable Public Use Dataset.  
Sex: male/female. RMSEA: root mean square error of approximation 
Analysis weights applied.  
--Null model M1 is the full model used to test differences compared M2-M3 constrained models.   
aloadings and thresholds constrained. 

-- 
 
10 

 
20 

 

 
<0.001 

 
25.501 

 
142.734 

 
0.1829 

 

 

 

 
90 

 

 

Comparative 
CFI 
TLI 
0.991 
 
0.993 

0.993 
 
0.994 

Absolute 
RMSEA 
0.047 
 
0.040 

 
0.995 

 
0.995 

 

 

 
0.036 

 

110 

 
 
 
 
 
 
 
 
 
 
 
 
 

 

Table 49. Goodness of Fit Indices Evaluating Measurement Equivalence of DSM-5 Criteria for Heroin Use Disorder Level Across 
History of Extra-Medical Prescription Opioid Use Among Newly Incident Heroin Users (Without Analysis Weights). 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
 

Comparative 
CFI 
TLI 
0.987 
 

0.990 
 

Absolute 
RMSEA 
0.075 
 

0.991 
 

0.989 
 

0.990 
 

0.990 
 

0.064 
 

0.067 
 

c2 

df 

-- 
 

df 
105 
 

p-value 
<0.001 
 

Chi-Square Model Fit 
c2 

Chi-Square for Difference Testing 
p-value 
-- 
 

Models 
M1 – free model 
 
M2 – constrain loadings 
M1 vs M2 
 
M3 – constrain botha 
M1 vs M3 
 
NSDUH Downloadable Public Use Dataset.  
Sex: male/female. RMSEA: root mean square error of approximation 
Unweighted estimates. 
--Null model M1 is the full model used to test differences compared M2-M3 constrained models.  
aloadings and thresholds constrained.  

108.807 
 

125 
 

145 
 

-- 
 

20 
 

40 
 

<0.001 
 

<0.001 
 

<0.001 
 

<0.001 
 

52.271 
 

280.090 
 

275.839 
 

335.108 
 

111 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

Table 50. Goodness of Fit Indices Evaluating Measurement Equivalence of DSM-5 Criteria for Heroin Use Disorder Level Across 
History of Extra-Medical Prescription Opioid Use Among Newly Incident Heroin Users with Analysis Weights. 
 Data from United States National Surveys on Drug Use and Health, 2002-2016. 

 

df 

165.564 
 
216.660 

 
246.127 

c2 

-- 
 
55.020 

df 
105 
 
125 

 
145 

p-value 
0.002 
 
<0.001 

Chi-Square Model Fit 
c2 

Chi-Square for Difference Testing 
p-value 
-- 
 
<0.001 

Models 
M1 – free model 
 
M2 – constrain loadings 
M1 vs M2 
 
M3 – constrain botha 
M1 vs M3 
 
NSDUH Downloadable Public Use Dataset.  
Sex: male/female. RMSEA: root mean square error of approximation 
Analysis weights applied.  
--Null model M1 is the full model used to test differences compared M2-M3 constrained models. 
aloadings and thresholds constrained.   

-- 
 
20 

 
40 

 

 
<0.001 

 
<0.001 

 
92.529 

 

 

 

 

 

Comparative 
CFI 
TLI 
0.990 
 
0.987 

0.992 
 
0.988 

Absolute 
RMSEA 
0.044 
 
0.050 

 
0.987 

 
0.988 

 

 

 
0.049 

 

 

112 

5.3 Results 

 

Testing measurement equivalence across sex and EMPOU using 20 PE items as 

indicators, rather than 10 criterion indicator variables, showed that some items were highly 

correlated with one another and should be combined for better estimation. Results on the 20 

items do not include PE measurement equivalence assessment due to this poor model estimation. 

Ten criteria were specified as observed indicator variables for the ME modeling steps. 

Sex: Model 1 comparative fit indices generate CFI=0.986 and TLI=0.982. Model 1 

absolute index RMSEA=0.085. Model 1 with loadings and thresholds free to vary compared to 

Model 2 with constrained factor loadings yields c2 = 10.907 (p>0.05) (Table 47). Chi-square test 

for difference testing fails to reject the null hypothesis, suggesting no need for measurement 

calibration. Comparative indices indicate good model fit (CFI=0.994, TLI=0.993). Absolute 

index suggests near-adequate fit (RMSEA=0.053). Model 1 compared to Model 3 with 

constrained loadings and thresholds fails to reject the null hypothesis, suggesting no calibration 

need (c2=26.730 (p>0.05). Model 3 comparative indices show good model fit (CFI=0.993, 

TLI=0.993). Absolute index for Model 3 suggests near-adequate fit (RMSEA=0.052). Model 

comparisons with analysis weights show Model 1 versus Model 2 yields c2=15.870 (p>0.05) and 

Model 1 versus Model 3 yields c2=25.501 (p>0.05) (Table 48). Both chi-square tests for 

differences fail to reject the null hypothesis, suggesting no need for measurement calibration 

Extra-Medical Prescription Opioid Use: Model 1 comparative fit indices generate 

CFI=0.990 and TLI=0.987. Absolute fit index RMSEA= 0.075. Model 1 with loadings and 

thresholds free to vary compared to Model 2 with constrained factor loadings yields c2 = 

10.52.271 (p<0.001) (Table 49). Null hypothesis of equal fit between Model 1 and Model 2 is 

rejected, indicative of need for measurement calibration. Comparative model fit indices for 

Model 2 suggest good model fit (CFI=0.991, TLI=0.990). Absolute fit index suggests near-

 

113 

adequate fit (RMSEA=0.064). Model 1 compared to Model 3 with constrained loadings and 

thresholds yields c2=108.807, (p<0.001). Comparative fit indices generate CFI=0.989 TLI=0.990 

and absolute fit index generate RMSEA=0.067 suggesting acceptable model fit. Model 

comparisons with analysis weights show Model 1 compared to Model 2 yields c2=55.020 

(p<0.001) and Model 1 compared to Model 2 yields c2=92.529 (p<0.001) (Table 50). Both chi-

square tests for differences reject the null hypothesis of equal fit; measurement calibration is 

needed. 

5.4 Discussion  

 
Among newly incident heroin users, assessing measurement equivalence of HUD across sex 

yields results that lend support to inferences about both metric and scalar equivalence. The 

completely free model (M1) and the constrained loadings model (M2) are not significantly 

different and the more parsimonious model is not rejected. Without calibration, NSDUH items 

can be used to study sex differences in HUD levels among newly incident heroin users. The 

addition of analysis weights yields the same conclusion of metric equivalence. The implication is 

that male-female HUD levels can be compared without additional measurement calibration steps. 

Likewise, the free model (M1) and the constrained loadings and threshold model (M3) 

are not significantly different, meaning the more parsimonious model is not rejected. This is 

evidence that no measurement calibration is needed. The addition of analysis weights yields the 

same conclusion of scalar equivalence. Thus, HUD levels of males and females could be 

compared based on their responses captured by these criterion indicator variables. 

Among newly incident heroin users, there is not measurement equivalence of HUD 

across extra-medical prescription opioid use history subgroups. Measurement calibration work 

will be needed if this comparison is to be made validly. 

 

114 

Models with constrained loadings yield better model fit than the completely free model, 

meaning the parsimonious model of measurement equivalence must be rejected. Similarly, 

constraining both loadings and thresholds produces better model fit than the completely free 

model. Therefore, EMPOU history designates NIHU subgroups for whom the meaning of HUD 

items might differ, or for whom the items perform differently as measurements of HUD levels. 

Further research and measurement calibration will be needed if we are to make valid 

comparisons of EMPOU subgroups in relation to HUD levels among  newly incident heroin 

users. (One potential intermediate approach would be a measurement equivalence assessment 

with EMPOU dichotomized to ever and never users.) 

Strengths and Limitations: The current study relies on annual cross-sectional surveys 

intended to yield nationally representative samples. There are three limitations intrinsic to this 

study design: 1) Processes akin to left truncation and left censoring in follow-up studies may lead 

to under-enumeration of some NIHU in the study population. To directly combat challenges of 

this type, the newly incident heroin use definition might be shortened in order to restrict the 

amount of time participants have to develop HUD-related experiences that would keep them 

from being included in the sample or from completing the interview (e.g., incarceration, fatal 

overdose). Even so, considering the ratio of newly incident heroin users to those who die of 

heroin overdose in the first year of use, left -truncation-like processes might be negligible. The 

same is true of left censoring when reflecting on those sampled but incarcerated or overdosing 

before assessments are complete. 2) The study’s reliance on self-report merits attention, yet there 

is currently no feasible alternative for nation-level surveys assessing history of drug use. 3) 

Cross-sectional surveys are often criticized for limited temporal sequencing issues. Notably, the 

current study circumvents this issue given participants cannot develop heroin use disorder 

without first having tried heroin.  

 

115 

Additionally, to evaluate measurement equivalence the chi-square test of differences 

between models was used. The reliance on this test has limitations. When evaluating 

measurement equivalence between three groups for EMPOU history, rejecting the null 

hypothesis suggests there are subgroup differences and that there is not measurement 

equivalence. Yet, it is still unknown where the source of the differences might be found. It is 

unknown if all three groups are not comparable to each other or if there are two groups that 

might be collapsed.  

Counter-balancing strengths include a sample of newly incident heroin users identified 

within samples and under standardized survey conditions that are designed to produce nationally 

representative samples with controlled selection probabilities and with proper attention to 

analysis weights that reflect those selection probabilities. The DSM-IV criteria are assessed, 

often with multiple items that have been combined here in order to strengthen the assessment of 

DSM-5 HUD diagnostic criteria. Notwithstanding the omission of DSM-5 ‘craving’ items, the 

resulting covariation of criterion indicator variables suggests coherence of the measurement of 

HUD levels in these newly incident users. A clear basis for making comparisons between males 

and females can be seen. More measurement calibration work is needed before it is possible to 

compare HUD levels across subgroups defined by histories of extra-medical use of prescription 

opioid pain relieving compounds. 

5.5 Conclusions 

 

This study’s focus is on the measurement models used to evaluate HUD levels among 

newly incident heroin users. This methods contribution creates a more solid basis for comparing 

males and females in relation to their HUD levels as observed soon after heroin onset. Some 

additional measurement calibration work might be done to address the vexing issue of residual 

variances, but solutions might not less achievable when the PE and criterion response indicators 

 

116 

are binary (0/1) as compared with measurement models for indicators with Gaussian or 

normalized response distributions. The same cannot be said for comparisons of NIHUs across 

subgroups defined by prior histories of opioid pain relievers. Additional measurement modeling 

will be required, and calibration of the HUD level measurement seems indicated at this point in 

the investigation. 

In measurement model work of this type, one potential source of non-equivalence 

involves how the respondents interpret the meaning of assessment items, and whether there are 

variations in interpretation or understanding about what was expected by those who designed the 

measurements. In measurement models for levels of drug use disorders, the source of variation 

might be traced to subgroup variations in pharmacokinetics (biotransformation, metabolism 

rates), involving issues much different from the just-mentioned ‘variation in meaning’ issues. For 

example, the EMPOU with recent opioids use might already be opioids-tolerant such that 

tolerance to heroin become manifest quickly rather than slowly. From a measurement modeling 

perspective, the PE ‘tolerance’ items might be ‘easier’ (with a lower threshold) and possibly less 

discriminating (with a shallower ‘loading slope’) across EMPOU subgroups. 

Origins of measurement non-equivalence also can be found when experiences and 

context vary. Social sharing of heroin and social isolation of heroin users might differ across sex 

or EMPOU subgroups (80). If so, if female new heroin users experience more sharing of heroin 

or more social connectedness within heroin-using groups, they might not experience having to 

spend a lot of time getting heroin, as compared with male counterparts, or vice versa. 

There are implications for clinical practice of note. First, the methodological results 

highlight that measurement tools for diagnostic workup are often assumed to function 

equivalently across populations. However, this may not be the case. During a consultation with 

young patient who says heroin use just started, the non-specialist clinician or counselor might not 

 

117 

immediately think about whether prior histories of opioids or other drug use might be important 

to evaluate during the course of asking questions about heroin experiences and trying to decide 

whether referral to a specialist is needed. Second, diagnostic criteria for heroin use disorder have 

shifted downward (i.e., with ‘mild’ implying just two criteria have been fulfilled). The result 

might be greater sensitivity of the assessment – i.e., to identify more mild cases of HUD with the 

presence of only two diagnostic criteria. In public health work one can give consideration to 

improvement of HUD assessment to better identify cases for early referral for evaluation and 

treatment by specialists. Thus, this methods investigation might have some potential importance 

in efforts to control HUD and attenuate heroin-related overdose mortality.  

 

 

 

118 

CHAPTER 6 MANUSCRIPT 3 – WILL THE OPIOIDS CRISIS PRECIPITATE EXCESS 

RISK OF A STIMULANT CRISIS?  

6.0 Abstract 
 
Aim: This study aims to investigate the degree to which onset of heroin use may precipitate 

excess risk of onset of stimulant use. 

Methods: The study population consisted of non-institutionalized civilian residents of 50 states 

and DC, sampled for the US National Surveys on Drug Use and Health (NSDUH), 2002-2016 

(n=837,326 participants). Among them, newly incident extra-medical stimulant users (NISUs) 

were identified (including prescription stimulants (e.g., amphetamines) or cocaine). A case 

crossover design was used to study newly incident heroin use (NIHU) prior to NISU with 2-

month hazard and control intervals specified a priori. Analysis weights and Taylor Series 

Linearization are used (e.g., for calculus-based variance estimation).  

Results: Discordant pair case crossover odds ratios show no association (OR=1.0, 95% CI=0.8, 

1.2) both with and without analysis weights. Post hoc exploration shows a moderate association 

between NIHU and NISU with 1-month hazard and control intervals (OR=1.3, 95% CI=1.0, 1.7).  

Discussion: Two months after NIHU does not appear to be a time of excess risk for NISU. The 

post hoc specification for one-month hazard and control intervals identify event minus 1 as a 

month of excess risk of starting to use stimulants extra-medically. Possible explanations for these 

findings might be methods-related (e.g., correlated recall error) or substantively important (e.g., 

environments conducive to new opportunities to try drugs).  

Conclusions: Whereas the ‘a priori’ specification indicates a null relationship, the ‘post hoc’ 1-

month specification can be used to guide future investigations of this type. Closely following 

drug onset an individual may have relatively more opportunities to try other drugs given their 

environment during that interval. In this way, heroin onset might be quickly followed by 

 

119 

psychostimulant onset. Possible prevention methods may include virtual reality simulations or 

roleplaying to prepare individuals for situations that can occur as a result of initial drug use.   

 

 

120 

6.1 Introduction 
 

Trends in drug use have shown some cycles of note. The US experience with the heroin 

use epidemic of the late 1960s and early 1970s was followed by an epidemic cocaine and other 

psychostimulant use (e.g., methamphetamine ‘ice’) during the next two decades. Likewise, 

epidemiologic evidence suggests stimulant use now is returning toward epidemic proportions as 

the current US opioids crisis ebbs. 

Linkages between opioids and psychostimulants suggest interesting patterns of co-

occurrence or sequencing. For example, opioids include prescription pain relievers, heroin, and 

synthetic opioids, all of which can induce drowsiness (among other ‘downers’ in the 

pharmacopeia). On this basis, the concept of opioids as ‘narcotics’) originates from Greek 

narcoun, meaning to benumb, make unconscious (80). In contrast, stimulants including 

amphetamine, methamphetamine, cocaine, and crack cocaine promote wakefulness (among other 

‘uppers’ in the pharmacopeia) (11). Some prescription stimulants are even used to directly 

combat sleep for disorders such as narcolepsy (11).  

Given the difference in subjective effects of opioids and stimulants, the intuitive notion 

may be that those who tend to use these different drugs represent distinct subgroups. Yet, one 

could speculate that the widespread use of opioids may lead individuals to seek remedies for 

their opioid-induced drowsiness and fatigue. Thus, it is plausible that opioid and stimulant users 

may be overlapping subgroups, at the very least. Evidence of an overlap has been shown in 

reports about the ‘speedball’ combination (e.g., heroin and cocaine), as well as in research on 

overdose deaths where both cocaine and some form of opioid have been detected (81).  

Several unanswered epidemiological questions about the ‘downers-uppers’ relationship 

prompt the current investigation about predicting stimulant use as might be triggered by opioid 

use. The epidemiologic case-crossover design lends itself to answering such questions. Rather 

 

121 

than traditionally answering “Why this outcome?” the case crossover answers “Why now?” More 

specifically, the case-crossover design was developed in order to study time-variant exposures 

associated with what might be experienced as a transient increased risk for some health outcome 

(82). The leading strength of this design is that time-invariant confounders are controlled for 

since cases serve as their own controls (e.g., genetic and other individual-level susceptibility 

traits that pre-date the intervals of experiences under study). Equally, when focused on issues of 

drug dependence, the case crossover minimizes selection bias often introduced by the tendency 

to rely on non-drug using controls (82). As a result, this design has been popular within drug 

dependence epidemiology (83)(84)(85)(86).  

Case crossover methodology relies on pre-specified ‘hazard’ intervals (i.e., excess odds 

of an exposure) versus ‘control’ intervals (i.e., exposure odds at expected values), generally of 

equal duration in time (t) and aligned in relation to a hypothesized latency period (87). In respect 

to this study, t represents the time passing – e.g., after an interval of newly incident heroin use 

when individuals may be at an increased risk to begin newly incident stimulant use. 

There is scant literature on an estimated latency period between heroin use onset and 

stimulant use onset, but there are hints that there might be short latency. On this basis, and 

guided by research suggesting that prolonged opioid use can result in sleep changes in as little as 

six weeks (88), the hazard interval was set to two months just prior to the month of starting to 

use psychostimulants extra-medically, where ‘extra-medical’ refers to using a stimulant outside 

the boundaries of prescribed indications (e.g., for feeling states such as ‘to get high’). Given that 

sleep disturbances contribute to feelings of fatigue, the hypothesized intervals were chosen based 

on the premise that heroin users may be more likely to increase energy levels with a stimulant 

within two months after starting to use heroin. 

 

122 

As noted below, it is important to specify the hypothesized latency period in advance of 

inspecting the temporal sequences in the study data. However, it has become customary to 

conduct ‘post hoc’ sensitivity analyses with other intervals in order to examine how robust or 

fragile the estimated associations might be across a range of alternative specifications. For this 

reason, in a post hoc analysis clearly labeled as an exercise of post-estimation exploration of 

alternatives, the 1-month hazard and control interval was considered in this study, as are other 

alternatives. 

Akin to prior case-crossover research on the cannabis-cocaine relationship (83), this case 

crossover study spans two distinct drug subtypes in order to understand possible drug-using 

trajectories. In this instance, extra-medical psychostimulant use can be considered as a 

consequence of NIHU, concurrent with consideration of NIHU as a trigger for new onsets of 

extra-medical stimulant use.  This study aims to investigate the degree to which onset of heroin 

use may precipitate excess risk of onset of stimulant use. 

Figure 12. Diagram of Case Crossover Design Studying Newly Incident Heroin Use and 
Excess Risk of Newly Incident Stimulant Use.  
 
 
 
 

Newly Incident  
Heroin Use? 
Hazard Interval 

Newly Incident  
Heroin Use? 

Control Interval 

Exposure 

 
 
 
 

 

Outcome 

Newly Incident 
Stimulant Use 

 

 

 

 

1 

2 

3 

4 

Time (months) 
0 
*Study design proposed by Maclure (1991) to study transient effects on the risk of acute 
events. Working backward from the month ‘t’ of onset of extra-medical stimulant use to the 
two-month hazard interval and the two-month control interval.  
Estimation based on how often heroin onset has occurred in the hazard interval but not in the 
control interval (evidence that favors a non-null ‘triggering’ hypothesis) versus how often 
heroin onset has occurred in the control interval and not in the hazard interval (evidence that 
favors a ‘null’ hypothesis of no triggering. 
Link available: https://www.ncbi.nlm.nih.gov/pubmed/1985444 
Stimulants include prescription stimulants (e.g., amphetamines) and cocaine, as well as ‘ice’ 
and other forms of non-prescription methamphetamines. 

123 

 

 

6.2 Methods 
 

Study Population: The study population for this epidemiological research consists of 

non-institutionalized civilian residents of the 50 States and the District of Columbia, age 12 years 

and older, living within the United States (US) during 2002-2016.  

Sample: Each year, research staff for the US National Surveys on Drug Use and Health 

secured a new probability sample of the study population and administered computer-assisted 

self-interviews to assess health topics. Staff created de-identified public use files (PUF) for each 

year. All work complied with institutional review board–approved protocols. 

The US Department of Health and Human Services Center for Behavioral Health 

Statistics and Quality created NSDUH open access analysis files with de-identified data after 

applying disclosure analyses designed to prevent re-identification of participants. Given the 

circumstances of open access and no contact with participants, the Michigan State University 

institutional review board ruled that plans to analyze these data qualified for the federal category 

“not human subjects research.” 

Aggregate NSDUH files included 837,326 participants, including those identified as 

extra-medical users who started newly incident stimulant use (NISUs) during the 12 months 

prior to survey assessment. Here, stimulant refers to prescription stimulants (e.g., amphetamines, 

methylphenidate), cocaine, and derivative compounds. NISUs were assessed within 12 months 

after first use, excluding all past users. For case-crossover estimation purposes, the informative 

NISUs who contribute to the odds ratio estimation of relative risk are those with newly incident 

heroin use (NIHU) during either the hazard interval or the control interval (i.e., within one to 

four months of NISU onset, extra-medically). Participants with NIHU and NISU within the same 

month were dropped from the analysis to be more certain that heroin use preceded stimulant use.  

 

124 

Statistical Analysis: Prespecified control and hazard intervals were each set equal to two 

months a priori, as explained in the study introduction. ‘Evidence favoring’ NISUs were 

considered to be stimulant users who started to use heroin within the hazard interval but not the 

control interval. ‘Evidence not favoring’ NISUs were EM NISUs who started to use heroin 

within the control interval but not during the hazard interval. Counts of these ‘evidence favoring’ 

and ‘evidence not favoring’ NISUs were tabulated by study year and then were used to calculate 

discordant pair ‘exposure odds ratios’ (ORs) based on matched pair contingency tables. A Stata 

software command “mcci” produced estimated OR based on dividing the two counts as just 

described. Meta-analyses were conducted across the 15 NSDUH study years 2002-2016, which 

can be considered as independent replication samples of the US study population as 

characterized at the start of this section. Meta-analysis summary estimates required log 

transformations for asymmetrical confidence intervals (CI) and a random effects estimator was 

used to capture heterogeneity across year-specific OR estimates.  

Analyses were repeated with analysis-weighted counts using Taylor Series Linearization 

for variance estimation (i.e., Stata ‘svy’ command). Stata SE software (StataCorp, LLC) and 

‘svy’. Additionally, to explore the degree to which the unweighted OR estimates might have 

been influenced by factors controlled in the analysis weights (age, sex, race, probability of 

selection), a stratified analysis without weights was completed, with strata defined by quartiles of 

the analysis weights. The expectation was that the OR estimates would not show appreciable 

variation across these strata. 

Finally, as noted above, a post-estimation set of exploratory data analyses looked into the 

issue of specifying the latency period. These post hoc analyses estimated discordant pair odds 

ratios based on a range of hazard and control interval widths, t. Four alternative specifications 

were considered, involving intervals of one, three, four, and six months. 

 

125 

Table 51. Characteristics of Newly Incident Stimulant Use Sample. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

 

NISUs w/ 

 Lifetime Heroin 

Case Crossover 

Sample** 

NSDUH  
2002-2016 

n=837,326 
n 

(%) 

Use 

n=3,696 
n 

Male 

401,270 

Characteristics 
Sex 

Age Groups (Years) 

12-17 
18-25 
26-34 
35-49 
50-64 
65+ 
Ethnic Self-Identification* 
White 
Black/African American 
Native American/Alaskan  
Native Hawaiian/Pacific 
Island.  
Asian 
>1 Ethnicity 
Hispanic 

258,309 
263,258 
93,449 
128,684 
56,998 
36,628 

532,503 
106,106 
12,284 
4,042 

30,449 
26,309 
134,633 

Extra-Medical Prescription Opioid History 
Lifetime Use (>12 Mon.*) 
Within 12 Mon.* 
Never Used 
Missing Data 

67,677 
53,527 
614,389 
101,737 

(%) 

 
(69) 
 
(3) 
(27) 
(29) 
(26) 
(14) 
(0.6) 
 
(76) 
(10) 
(0.8) 
(0.3) 

n=579 
n 

 
396 
 
29  
116 
47 
155 
203 
29 
 
416 
66 
12 
4 

(%) 

 
(74) 
 
(0.8) 
(4) 
(6) 
(24) 
(60) 
(6) 
 
(69) 
(15) 
(0.7) 
(0.2) 

(0.7) 
(2) 
(11) 
 
(28) 
(54) 
(14) 
(3) 

6 
15 
60 
 
161 
200 
173 
45 

(2) 
(0.5) 
(13) 
 
(4) 
(8) 
(75) 
(13) 

 
2,248 
 
332 
1,979 
690 
528 
156 
11 
 
2,845 
192 
67 
20 

31 
149 
392 

888  
2,285 
430 
93 

 
(48) 
 
(10) 
(13) 
(14) 
(25) 
(22) 
(16) 

(67) 
(12) 
(0.5) 
(0.3) 

(5) 
(1) 
(14) 

(8) 
(4) 
(75) 
(13) 

126 

NSDUH Downloadable Public Use Data. 
NISUs: Newly incident stimulant users first used within 12 months of survey assessment. Past 
onset users excluded. Stimulants include amphetamine, methamphetamine, or cocaine.  
Unweighted n’s, weighted percentages. 
*Ethnic self-identification: all ethnicities other than Hispanic are specified as non-Hispanic per 
survey assessment. 
** NISUs who are discordantly informative for estimation of OR - participants either started 
heroin use in the hazard interval or in the control interval. All other NISUs excluded. 

 
 
 
 
 
 
 
 
 

 

Table 52. Counts and Discordant Pair Odds Ratios of Newly Incident Heroin Use Preceding 
Newly Incident Stimulant Use. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

Discordant Pair Odds Ratios 

 
 

Heroin Exposure Intervals 
Control Int. 
Hazard Int. 

n (%)a 

n (%)a 

Unweighted 
OR (95% CI) 

Weighted 

OR (95% CI) 

1 (47) 
3 (45) 
3 (100) 
29 (93) 
25 (40) 
23 (41) 
20 (54) 
28 (37) 
30 (61) 
17 (49) 
15 (27) 
21 (48) 
29 (63) 
21 (61) 
21 (43) 
289 (55) 
-- 

Year 
1 (53) 
2002 
0 (55) 
2003 
1 (0) 
2004 
37 (6) 
2005 
25 (60) 
2006 
22 (59) 
2007 
30 (46) 
2008 
17 (63) 
2009 
20 (39) 
2010 
37 (51) 
2011 
20 (73) 
2012 
15 (52) 
2013 
23 (37) 
2014 
24 (39) 
2015 
21 (57) 
2016 
293 (45) 
Pooled 
Metab 
-- 
NSDUH Downloadable Public Use Data. 
Outcome: stimulants: methamphetamines, amphetamines, or cocaine. Hazard and control 
intervals equal 2 months set a priori.  
aUnweighted n. Weighted percent. 
bMeta-analysis summary estimates use log transformation. Estimates rounded to tenths place.  
Unweighted: I2=10.6%, p>0.05, MA=1.001, 95% CI=0.840, 1.194 
Weighted: I2=82.6%, p<0.001, MA=1.048, 95% CI=0.788, 1.394. with random effects.  

<0.1, 78.5 
 -- 
0.2, 157.5 
0.5, 1.3 
0.6, 1.8 
0.6, 2.0 
0.4, 1.2 
0.9, 3.2 
0.8, 2.8 
0.2, 0.8 
0.4, 1.5 
0.7, 2.9 
0.7, 2.3 
0.5, 1.6 
0.5, 1.9 
0.8, 1.2 
0.8, 1.2 

0.6, 1.3 
0.5, 1.2 
-- 
6.9, 43.3 
0.4, 1.0 
0.5, 1.1 
0.8, 1.8 
0.4, 0.9 
1.0, 2.4 
0.6, 1.5 
0.2, 0.6 
0.6, 1.4 
1.1, 2.6 
1.0, 2.4 
0.5, 1.1 
0.8, 1.9 
0.8, 1.4 

0.9 
0.8 
-- 
15.5 
0.7 
0.7 
1.2 
0.6 
1.6 
1.0 
0.4 
0.9 
1.7 
1.6 
0.8 
1.2 
1.0 

1.0 
-- 
3.0 
0.8 
1.0 
1.0 
0.7 
1.6 
1.5 
0.5 
0.8 
1.4 
1.3 
0.9 
1.0 
1.0 
1.0 

127 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 

Table 53. Post Hoc Interval Width Exploration: Unweighted Discordant Pair Odds Ratios of Newly Incident Heroin Use Preceding 
Newly Incident Stimulant Use. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
  

6 months  

3 months 

4 months 

1 month  

Cases 
n=573 
OR 

Cases 
n=1,001 

OR 

Controls 
n=1,188 
95% CI 

0.4 
0.5 
0.9 
1.3 
1.2 
1.2 
1.1 
1.0 
0.9 
0.8 
0.7 
0.6 
0.7 
0.6 
0.6 
0.8 
0.9 

0.1, 1.6 
0.1, 1.6 
0.2, 3.0 
0.9, 1.7 
0.9, 1.6 
0.9, 1.7 
0.8, 1.5 
0.8, 1.4 
0.6, 1.2 
0.6, 1.1 
0.5, 0.9 
0.4, 0.8 
0.5, 0.9 
0.4, 0.8 
0.4, 0.9 
0.8, 0.9 
0.7, 1.0 

 
 

Cases 
n=457 
OR 

Cases  
n=168 
OR 

Controls 
n=119 
95% CI 

Controls 
n=568 
95% CI 
0.1, 118.0 
0.2, 18.0 
0.3, 22.1 
0.6, 1.5 
0.5, 1.1 
0.4, 0.9 
0.5, 1.2 
0.4, 0.9 
0.7, 1.7 
0.5, 1.2 
0.4, 1.0 
0.4, 1.2 
0.9, 2.2 
0.5, 1.4 
0.5, 1.5 
0.7, 0.9 
0.7, 0.9 

2.0 
1.5 
2.0 
1.0 
0.7 
0.6 
0.8 
0.6 
1.1 
0.8 
0.6 
0.7 
1.4 
0.9 
0.9 
0.8 
0.8 

-- 
0.5 
0.5 
1.6 
1.0 
2.8 
1.2 
1.5 
1.7 
1.4 
6.5 
1.1 
1.2 
0.8 
1.1 
1.4 
1.3 

Year 
2002 
-- 
2003 
<0.1, 9.6 
2004 
<0.1, 9.6 
2005 
0.7, 3.8 
2006 
0.5, 2.6 
2007 
1.1, 8.8 
2008 
0.5, 3.3 
2009 
0.7, 3.7 
2010 
0.8, 4.0 
2011 
0.5, 4.4 
2012 
1.5, 59.3 
2013 
0.4, 2.9 
2014 
0.6, 2.8 
2015 
0.3, 2.1 
2016 
0.4, 2.9 
Pooled 
1.1, 1.8 
Meta*  
1.0, 1.7 
NSDUH Downloadable Public Use Data. 
Post hoc analysis explored interval width. A priori interval width set to 2 months. 
Estimates are unweighted and rounded to the tenths place. 
-- due to zero cells. 
*Meta-analysis summary estimates use log transformation.  
1month: I2=8.78, p>0.05, MA=1.33, 95% CI=1.020, 1.735 
3month: I2=15.6%, p>0.05, MA=0.807, 95% CI=0.709, 0.919 
4month: I2=39.0%, p>0.05, MA=0688, 95% CI=0.614, 0.771 
6month: I2=61.5%, p<0.001, MA=0.853, 95% CI=0.729, 0.997 with random effects.  

1.0 
1.5 
1.0 
1.0 
0.6 
0.6 
0.6 
0.4 
0.8 
0.7 
0.5 
0.7 
0.9 
1.0 
0.8 
0.7 
0.7 

Controls 
n=826 
95% CI 
0.1, 13.8 
0.2, 18.0 
0.2, 5.4 
0.7, 1.5 
0.4, 0.9 
0.4, 0.9 
0.4, 0.9 
0.3, 0.6 
0.6, 1.2 
0.5, 1.0 
0.3, 0.7 
0.4, 1.1 
0.6, 1.3 
0.6, 1.5 
0.5, 1.2 
0.6, 0.8 
0.6, 0.8 

 

128 

Table 54. Year-Specific Numbers of NISUs Stratified by Analysis Weight Quartiles and 
Interval-Specific Exposure Status. 
 Data from United States National Surveys on Drug Use and Health, 2002-2016. 

 
 

Quartile 1 
0 > x £ 61.8 

Quartile 2 

Quartile 3 

61.8 > x £ 134.1 

134.1 > x £ 337.2 

Quartile 4 
> 337.2 

0 
0 
0 
11 
8 
4 
3 
6 
4 
2 
2 
5 
2 
3 
0 
50 

0 
0 
0 
9 
6 
4 
2 
3 
5 
7 
3 
4 
5 
4 
2 
54 

0 
1 
0 
4 
5 
4 
4 
5 
3 
3 
4 
8 
9 
2 
5 
57 

0 
0 
1 
2 
4 
4 
2 
3 
5 
2 
0 
1 
1 
2 
1 
28 

Year 
2002 
2003 
2004 
2005 
2006 
2007 
2008 
2009 
2010 
2011 
2012 
2013 
2014 
2015 
2016 
Totals 

Hazard  Control  Hazard  Control  Hazard  Control  Hazard  Control 
1 
0 
0 
19 
12 
10 
21 
7 
8 
22 
12 
8 
8 
16 
15 
159 

0 
0 
1 
6 
6 
3 
2 
6 
9 
3 
2 
2 
5 
1 
2 
48 
NSDUH Downloadable Public Use Data. 
Outcome: stimulants: methamphetamines, amphetamines, or cocaine. Hazard and control 
intervals equal 2 months set a priori. Control interval: heroin first use within 3 or 4 months of 
stimulant first use. Hazard interval: heroin first use within 1 or 2 months of stimulant first use.  
Estimates rounded to tenths place. 
Unweighted counts stratified by analysis weight quartiles indicate how many NISUs were 
exposed and had first heroin use in the hazard interval but not in the control interval (shown in 
the columns labeled ‘Hazard’ and how many NISUs were exposed and had first heroin use in 
the control interval but not in the hazard interval (shown in the columns labeled ‘Control’. 

1 
2 
2 
8 
6 
12 
11 
11 
14 
9 
7 
6 
13 
15 
14 
131 

| 
0 
0 
7 
3 
4 
5 
4 
2 
6 
5 
2 
9 
2 
3 
52 

129 

 
 
 
 
 
 
 
 
 
 

 

Table 55. Year-Specific Odds Ratios of NISUs Stratified by Analysis Weight Quartiles and 
Interval-Specific Exposure Status 
 Data from United States National Surveys on Drug Use and Health, 2002-2016. 

 
 

Quartile 1 
0 > x £ 61.8 

95% CI 

Quartile 2 

Quartile 3 

134.1 > x £ 337.2 
OR 

95% CI  OR 

OR 
-- 
-- 
-- 
5.5 
2.0 
1.0 
1.5 
2.0 
0.8 
1.0 
-- 
5.0 
2.0 
1.5 
-- 
1.8 
1.6 

61.8 > x £ 134.1 
OR 
95% CI 
Year 
-- 
-- 
2002 
-- 
-- 
2003 
-- 
-- 
2004 
0.2, 2.1 
0.7 
2005 
0.3 3.7 
1.0 
2006 
0.1 4.4 
0.8 
2007 
0.1, 13.8 
1.0 
2008 
0.4, 12.4 
2.0 
2009 
0.5, 6.8 
1.8 
2010 
0.1, 1.9 
0.4 
2011 
0.7 
0.1, 5.8 
2012 
0.5  <0.1, 3.5 
2013 
1.0 
0.2, 4.3 
2014 
0.3  <0.1, 2.5  
2015 
1.0 
0.1, 13.8 
2016 
0.6, 1.3 
0.9 
Pooled 
Meta* 
0.8 
0.5, 1.3 
NSDUH Downloadable Public Use Data. 
*Meta-analysis summary estimates use log transformation and random effects estimator. 
Estimates rounded to tenths place.  
-- represent zero cells.  

-- 
-- 
-- 
-- 
-- 
-- 
0.1, 2.2 
0.6 
1.7  0.3, 10.7 
0.2, 5.4 
1.0 
0.8 
0.2, 3.7 
0.8 
0.2, 3.7 
1.5  0.2, 18.0 
0.1, 2.3 
0.5 
0.8 
0.2, 3.7 
4.0  0.8, 38.7 
1.0 
0.4, 2.8 
1.0  0.1, 13.8 
1.7  0.3, 10.7 
0.7, 1.6 
1.1 
1.0 
0.6, 1.6 

 -- 
-- 
-- 
1.2, 51.1 
0.5, 9.1 
0.2, 5.4 
0.2, 18.0 
0.4, 12.4 
0.2, 3.7 
0.1, 13.8 
-- 
0.6, 236.5 
0.1, 118.0 
0.2, 18.0 
-- 
1.1, 2.9 
0.9, 3.0 

Quartile 4 
x > 337.2 

95% CI 
1.0  <0.1, 78.5 
-- 
-- 
-- 
-- 
0.2, 1.0 
0.4 
0.2, 1.4 
0.5 
0.5, 3.1 
1.2 
0.2, 1.1 
0.5 
1.6 
0.6, 4.8 
0.7, 4.8 
1.8 
0.2, 0.9 
0.4 
0.2, 1.6 
0.6 
0.8 
0.2, 2.5 
0.6, 4.5 
1.6 
0.4, 2.0 
0.9 
0.9 
0.4, 2.0 
0.6, 1.0 
0.8 
0.8 
0.6, 1.1 

 
6.3 Results 
 

Based on the discordant pairs within the NISU sample of informative 579 participants, 

the hazard interval included 289 individuals with heroin onset within the hazard interval (one or 

two months prior to NISU), but not in the control interval (Table 52, bottom rows). The control 

interval included 293 individuals with heroin onset within the control interval (three to four 

months prior to NISU) but not in the hazard interval. Year-specific estimates are not appreciably 

different, except in years with small numbers of informative NISUs (0 to 3 individuals).  

Year-specific unweighted ORs are not statistically significant and show both weakly 

positive and inverse associations (Table 52). Across study years 2002-2016, a meta-analysis and 

 

130 

a pooled analysis yielded ORs=1.0 (95% CIs=0.8, 1.2). Year-specific ORs with analysis weights 

applied also show both positive and inverse associations. ORs for study years 2005 and 2014 are 

statistically significant (OR=15.5, 95% CI=6.9, 43.3; OR=1.7, 95% CI=1.1, 2.6). Across study 

years 2002-2016, a pooled analysis yielded OR=1.2 (95% CI=0.8, 1.9). A meta-analysis yielded 

OR=1.0 (95% CI= 0.8, 1.4). 

Post hoc exploration suggests that a 1-month interval specification might be useful in 

future case-crossover research of this type. The exploratory OR based on the 1-month interval 

specification show a weakly positive association (OR=1.3, 95% CI=1.0, 1.7) (Table 53). 

Exploration of other interval specifications suggested null associations. Similarly, in stratified 

analyses with quartiles based on analysis-weights, the general pattern is one of a null case-

crossover relationship and evidence that does not favor the triggering hypothesis as proposed. 

6.4 Discussion 
 

The ‘a priori’ latency period was specified in relation to two successive two-month 

intervals working backward from the calendar month of first newly incident extra-medical use of 

stimulant drugs (i.e., t-1 and t-2 months as hazard interval, t-3 and t-4 months as control 

interval). This specification produced a null odds ratio estimate and the result does not support a 

hypothesis of heroin onset triggering extra-medical onset of stimulants drugs. 

  Looking over the post hoc exploratory data analysis results, it appears that a better 

choice for the duration of the latency interval is ‘one month’ in that the first month after NIHU 

appears to be a month of excess risk of newly incident extra-medical stimulant use (OR> 1.0; 

p<0.05). All other specifications suggest null (or possibly inverse) associations. 

Moreover, in the stratified analyses, individuals with the lowest possible analysis weight 

values seem to be of interest (i.e., NISUs who live alone and others in the first analysis weight 

 

131 

quartile). For this quartile, there is evidence of an OR estimate that departs from the null, unlike 

other analysis weight quartiles. 

It is important to consider limitations. For example, in research of this type, even with 

recall restricted to fairly memorable events that have occurred within 12 months prior to the 

assessment, it is not possible to rule out correlated recall error related to an individual’s 

remembered drug use history. A spurious association could result from systematic error in 

reporting hazard interval versus control interval information. For example, individuals who 

remember using heroin shortly before using stimulants may recall these events being in adjacent 

months (i.e., one-month interval). Differential misclassification of this sort can lead to 

conclusions for or against the null (87).  

Furthermore, a potential threat to validity with case crossover designs is related to mis-

specified hazard and control intervals. In theory, the interval duration should reflect the 

minimum induction period (Figure 14) (87). If this interval is incorrectly specified, the result can 

be non-differential exposure misclassification and a potentially attenuated OR estimate (87). 

A third limitation involves elimination of participants for whom a single calendar month 

was stated as the month of first extra-medical stimulant use and also the month of first heroin 

use. These participants were judged non-informative because the NSDUH does not ask about 

ordering of stimulant and heroin use when both onsets occur in the same calendar month. It is not 

possible to say whether the heroin use came first within the month, or vice versa. As such, it is 

possible for heroin onset to trigger extra-medical use of a stimulant with a very short latency or 

induction period (shorter than the 1 month period specified here as the lowest possible value). 

This limitation can be overcome only if very fine-grained data about temporal sequencing is 

available. For example, when the same calendar month is stated, the user might be asked 

 

132 

questions about which came first, as well as questions about the interval of time separating the 

two onsets. 

A fourth possibility is differential loss of the most seriously drug-involved users in 

probability samples of community populations. That is, non-participation might be a more 

serious threat to validity for individuals who rapidly progressed from heroin use to extra-medical 

stimulant onset, as compared with individuals who made no such progression. 

Lastly, among limitations, the current study relies on annual nation-scale cross-sectional 

field surveys. It is possible that the attempt to secure nationally representative samples of newly 

incident drug users might introduce study artifacts that are not faced when alternatives are 

considered such as ‘snowball’ or ‘respondent-driven’ sampling (RDS). Leaving aside any 

concerns about snowball or RDS approaches, in the context of national cross-sectional surveys, 

this facet of study design introduces concerns about artifacts akin to left truncation and left 

censoring processes in follow-up studies. These are processes that can constrain valid 

enumeration of NISUs and NIHUs in samples of this type, especially for those in which NIHU 

follows NISU quickly, or vice versa. Larger samples would make it possible to focus the ‘newly 

incident’ interval on persons observed within 0-90 days after first stimulant onset. This 

manipulation of study design can reduce these theoretical processes akin to left-truncation and 

left-censoring as observed in longitudinal time-to-event studies, and it should be noted that 

neither ‘snowball’ nor RDS approaches will avoid these issues and concerns. At present, drug 

dependence epidemiology, even when it has turned to snowball or RDS approaches, has 

produced no particularly useful estimates of the ‘instantaneous risk’ of dying from overdose, 

being incarcerated, or migrating quickly across community dwelling units. It would be these 

‘instantaneous risk’ estimates that would help to better understanding of these sources of bias in 

study estimates of the type reported here. 

 

133 

 Notwithstanding said limitations, there may be importance in the one-month interval 

specification, coupled with the non-null OR seen in exploratory analyses stratified by quartile of 

analysis weight. As described previously, one of the major determinants of the analysis weight is 

the number of eligible participants within the sampled dwelling unit, with the largest selection 

probability (1.0) assigned to those who live by themselves.  One might speculate that there is a 

feedback loop between living alone and rapid progressions from one drug to another – i.e., that 

the living alone followed drug use rather than preceded drug use. However, it also is true that 

living alone might influence density of chances to try a stimulant after first heroin use. Living 

alone might imply reduce social sharing of drugs and thus there might be a less rapid transition 

from heroin to stimulant user in this stratum. These are speculations that will require more 

empirical research before a complete understanding of the relationships linking the current 

opioids crisis with an emerging epidemic of extra-medical psychostimulant use in the US. 

The current study is the first of its kind and there was no prior literature on this topic, as 

might have been used to guide its specification of optimal latency period. Based on this work, an 

optimal choice might be the one-month interval. 

 

 

 

 

 

 

 

 

 

 

134 

Figure 13. An epidemic curve for acute-onset disease following a point exposure. 
 

 
Study design proposed by Maclure (1991) to study transient effects on the risk of acute events.  
Link available: https://www.ncbi.nlm.nih.gov/pubmed/1985444 

 

 
6.5 Conclusions 
 

It was hoped that the current study would shed new light on transitions from newly 

incident heroin use to newly incident extra-medical use of psychostimulant drugs. Likewise, this 

information could have helped to understand the current US opioids crisis and the emerging 

pattern of increased incidence rates for use and outcomes related to extra-medical use of 

psychostimulant drugs. Based on prior case-crossover research (cited above), it was possible to 

conceptualize a theoretical possibility that heroin onsets might trigger onsets of extra-medical 

use of these stimulant drug compounds, and to specify a hypothesized latency period of two 

months for the case-crossover hazard and control intervals in the life of each newly incident 

stimulant user. 

Conversely, there is not strong evidence supporting idea that heroin onset triggers onset 

of extra-medical use of stimulant drugs. The implication is that there is some other process that 

helps explain what happened in the epidemiology of heroin and stimulants use back in the 1960s 

 

135 

through 1980s and what now is happening in the epidemiology of heroin and stimulants use in 

the first decades of the 21st century. 

It is important to remember that studying etiological processes do not lend themselves to 

individual implications, such as clinician recommendations for patients. In the context of 

generally null findings, clinical practitioners and counselors should look elsewhere for insights 

as might guide intervention plans for their patients and clients. Having said that, based on this 

work, the imminent risk of starting extra-medical use of cocaine or amphetamine or one of the 

other psychostimulants is not something the practitioner should be especially concerned about. 

Instead, the implications of this research are for future epidemiological research into the 

heroin-stimulants relationship. First, if other case-crossover designs are contemplated, a starting 

point can be a 1-month interval as opposed to a two-month interval for the hypothesized latency 

period. Plus, there is reason to think through the possibility that the association can be seen most 

prominently among individuals who are living alone, or who have other weight-influencing 

characteristics that prompt their membership in the lowest quartile of NSDUH analysis weights. 

In this respect, the discovery of null OR estimates in this dissertation research project’s 

case-crossover investigation can be informative. As it pertains to epidemiology, more than a few 

steps forward have been taken. There is new evidence and methodological considerations for 

guidance of future research on the epidemiology of heroin use and its potential consequences. 

With respect to the title of this chapter, the null findings might suggest that the 

individual-level processes of starting extra-medical stimulant use soon after heroin onset are 

irrelevant to the epidemiological processes that can be seen in trends. The evidence from 

discordant pair odds ratio estimates of this study should not lead to any premature disruption 

about ideas that link an earlier opioids epidemic with a later stimulant epidemic. The evidence of 

this study pertains to highly individualized experiences of the newly incident stimulant users in 

 

136 

this nation-level sample. It does not exhaust all of the possibilities for either individual-level or 

macro-level processes that produce a sequence from an opioids epidemic to a stimulant 

epidemic.  

There is good reason to conceptualize heroin-stimulant sequences that are prompted by a 

common underlying susceptibility trait of the type that is completely controlled in the case-

crossover design (81-85). Alternately, there are supra-individual processes that operate on more 

of a macro-level of influence (e.g., drug regulatory and enforcement processes), some of which 

have been mentioned in the context of ideas about social sharing of drug compounds from one 

individual to another in socially connected groups. Other processes can be operating at a supra-

individual level, as might be the case in the hypothesis that a black-market stimulants vendor 

might study patterns of heroin/opioids overdose and use, community by community, and 

selectively target communities with emerging problems as manifest in opioids overdoses or 

arrest rates.  

Given the common practice of using geo-coded ‘Google Trends’ and other internet-

mediated information to target zones of susceptibility for consumer purchases or social sharing 

of information, it is not difficult to imagine that black-market vendors might start to use these 

new elements of information about geographical clustering of ‘opioids’ mentions in order to 

make decisions about marketing of stimulants. These are epidemiological and social processes 

that would tend to connect the opioids experiences of US civilians with the stimulant experiences 

of US civilians. A case-crossover study of this type would not disclose these processes of 

‘precipitation’ of a stimulants outbreak in the aftermath of an opioids outbreak, and would not 

disclose an opioids epidemic giving rise to a stimulants epidemic.  

 

 

137 

CHAPTER 7 DISCUSSION AND CONCLUSIONS  
 
7.0 Summary of Findings 
 
 

Aim 1: Aim 1 findings focus on estimation of the pairwise associations between 

problems and experiences of heroin use disorder interval by interval beginning from the first 

months of use. The most common PEs experienced among NIHUs are trying to set limits, having 

problems with emotions, spending a lot of time getting or using heroin, needing more to get the 

same effect, having problems with family or friends, and spending less time doing important 

activities. Throughout time intervals, the majority of PE to PE odds ratios are robust, positive, 

and are statistically significant departures from the null (OR=1.0). These large associations and 

wide confidence intervals likely reflect maldistribution within contingency tables and 

exchangeability between PEs. 

Notable exceptions within 0-90 days of first using heroin include inverse, but not 

statistically significant, associations between cutting down at least once and trying to set limits, 

spending a lot of time getting or using, having problems with family or friends, continuing to use 

despite problems with emotions, physical danger, continuing to use despite problems with family 

or friends, and being unable to keep limits. Inverse associations may indicate that the PE of 

cutting down at least once is not a signal of HUD but is a signal of being less dependent. 

Continuing to use despite having physical problems has no association with any PEs. Within 1-

120 days of first using, cutting down at least once does not have as many inverse associations – 

an inverse association between that and needing more for the same effect and physical danger. 

Having physical problems, continuing to use despite physical problems, trouble with the law, and 

spending a lot of time getting over the effects are not shown to be associated with any PEs.  

Within 30-150 days, more inverse associations are seen related to cutting down at least 

once, continued use despite problems with emotions, spending less time doing important 

 

138 

activities, having problems with family or friends, and spending a lot of time getting or using. A 

similar pattern of inverse association with the PE cutting down at least once is seen within 60-

180 days of first using. Having physical problems also inversely correlates with spending less 

time doing important activities, being unable to set limits, and spending a lot of time getting over 

the effects of heroin during this time frame. A possible explanation is heterogeneity in those with 

HUD in that physical problems is not at a severity where the associated PEs would be 

experienced.  

Within 90-120 days of first using heroin, spending a lot of time getting over the effects of 

heroin, having physical problems, and continuing to use despite having physical problems is not 

associated with any PEs. A similar pattern is seen within 120-240 days of first using. Notably, a 

departure from the pattern occurs during 150-270 days of first using. Continued use despite 

having physical problems is robustly and significantly associated with spending a lot of time 

getting or using, needing more to get the same effects, having problems with family or friends, 

spending less time doing important activities, physical danger, continuing to use despite 

problems with family or friends, and trouble with the law. The 150-270-day time interval reveals 

the first occurrence of large, statistically significant associations with continuing to use despite 

having physical problems. Yet, there are null associations with the PE within subsequent lag-

time intervals. Most likely this is due to small cell counts within continuing to use despite 

physical problems, which is also the least common PE. 

With 180-300 days of first using heroin, inverse associations are seen between PEs 

spending a lot of time getting over the effects, the same amount having less effect, having 

problems with family or friends, spending less time doing important activities, and continuing to 

use despite having problems with emotions. Within 210-330 days of first using heroin, inverse 

association appear between pairs related to use causing serious problems, having three or more 

 

139 

withdrawal symptoms and having them at the same time, and spending a lot of time getting over 

the effects and needing more to get the same effect. Likewise, there are inverse associations 

between cutting down at least once, trying to set limits, needing more to get the same effect, and 

the same amount having less effect within 240-360 days of incident use. These associations are 

not statistically significant though. Within 270-330 days of first using, there are several inverse 

associations related to trying to set limits. This may signal that trying to set limits has a meaning 

related to a less severe form of dependence or showing a departure from the use disorder. This 

trend is not seen in the intervals of 300-420 and 330-450 days within first use. 

Aim 2: The evaluation of measurement equivalence from aim 2 confirms 

methodologically important and interesting details about the NSDUH assessment of diagnostic 

criteria for HUD and evaluation of HUD levels. First, study of sex differences in HUD level 

seems to require no more calibration of the measurements under the specified models. Second, 

measurement calibration will be needed if the aim is to estimate HUD variation across types or 

levels of prior opioids-using experiences. 

 

Aim 3: Aim 3 investigating the degree to which onset of heroin use might precipitate 

excess risk of stimulant use onset, produced a null finding with the a priori specification of a 2-

month latency period. Post-estimation exploratory analyses indicate that future research might be 

improved by substituting a 1-month interval in place of the 2-month interval used in this study. 

7.1 Strengths and Limitations 
 

The current study relies on NSDUH annual cross-sectional surveys intended to yield 

nationally representative samples. There are three limitations intrinsic to this study design: 1) 

Left truncation and left censoring may underestimate the number of NIHU. To directly combat 

challenges of left truncation and censoring, the newly incident heroin use case definition (i.e., 

within 12 months of assessment) restricts the amount of time participants have to develop HUD-

 

140 

related experiences that would keep them from being sampled or completing the interview (e.g., 

incarceration, fatal overdose). Even so, considering the ratio of newly incident heroin users to 

those who die of heroin overdose in the first year of use, potential left truncation is negligible. 

The same is true of left censoring when reflecting on those incarcerated within the first year of 

use. 2) The study’s reliance on self-report merits attention, yet there is currently no feasible 

alternative for surveys assessing history of drug use. 3) Cross-sectional surveys are often 

criticized for limited temporal sequencing. Notably, the current study circumvents this issue 

given participants cannot develop heroin use disorder without first having tried the drug.  

Inferences relating to the natural history of HUD within aim 1 are limited in that there 

cannot be a causal interpretation of PE to PE relationships. A strong association between PE 

pairs may indicate that those PEs are measuring the same underlying construct such as HUD or 

HUD-related diagnostic criteria. Aim 1 does shed light on the coalescence of problems and 

experiences associated with HUD and how that may change as individuals have had more time to 

develop HUD. Causal inferences are also limited as they pertain to aim 3. It cannot be said that 

heroin use causes an individual to start using heroin. Likely, there are circumstantial experiences 

that may make a NIHUs more susceptible to start using another drug. Also, given prescription 

stimulants and cocaine were combined into one stimulant category, it is unknown if there is a 

particular drug driving some association or attenuating an association. Additionally, aim 2 can 

only offer the recommendation that EMPOU may differentially influence HUD measurement. 

More work needs to be done in order to understand where the subgroup differences are located 

and what the effects are.  

7.2 Public Health Implications and Next Steps 
 
 

The future of heroin-related research and public health implications are areas that would 

greatly benefit moving forward intertwined with each other. That is, I believe that ethically, there 

 

141 

should be no study to understand heroin use without simultaneously intervening since the heroin-

related overdose mortality rate is unprecedented and the epidemic proportions of use can be 

linked to highly morbid sequalae (e.g., HIV/AIDS). Moreover, providing participants with more 

than incentives for participation, but actual resources, knowledge, and care should be the new 

IRB-gold standard. 

 

An intriguing and new area of intervention research is citizen science. Often citizen 

science is in conjunction with crowdsourcing, which may be a useful way to gauge a given 

community’s severity of heroin use and use disorder. Important additives may then include 

community-based reaction to the gauged severity based on a matrix of options (Figure 15). An 

important goal of citizen science is recruiting community members not as participants, but as 

collaborators in science to allow sustained program dissemination. For example, a multi-step 

process may require citizen education and focus-group feedback. A D2L (desire to learn) module 

could walk community members through naloxone administration, clean needle exchange and 

proper disposal methods, and effective communication strategies, to name a few. Following 

education, researchers can meet with those citizens to assess their understanding and figure out 

how those tools can be implemented in their community. Michigan State University Extension 

throughout Michigan is a possible mechanism for this cycle. 

 

142 

Figure 14. Infographic Exerts for Citizen Science-Based Initiative 

Graphics from www.infogram.com 
   

 

Furthermore, citizen science could benefit from integrated harm reduction initiatives. 

Aside from clean needle exchange, “open air” or safe spaces for injection should be provided 

within a community. There, citizens trained in naloxone administration could support those using 

heroin and help provide resources to users such as clean needles, connections to treatment, and 

 

143 

aid in recovery from heroin use. A “safety kit” including drug test kits, fentanyl test strips, 

condoms, pregnancy tests, alcohol swabs, naloxone, and hand sanitizer could be available and 

would minimize morbidity and mortality.  

Relatively more distal points for intervention may consider intervening within families. 

Adverse childhood events have been shown to be linked to adverse health outcomes in 

adulthood, including non-fatal overdose among individuals who inject drugs (49). Thus, family 

intervention may mitigate future generations of drug use. Additionally, families with adverse 

events may include drug-using parents in need of intervention. Family interventions should help 

parents with resources and execution of effective parenting (e.g., parent-child communication, 

help with obtaining affordable childcare). Such efforts may not have direct aims in heroin use 

prevention but may indirectly influence heroin use through more well-adjusted child rearing and 

community engagement.  

Importantly, intervention strategies outlined above should be coupled with clinician 

supervised medication substitution or assisted treatment psychological counseling. Results from 

aim one directly speak to clinicians for evidence-based screening recommendations. That is, 

strong PE to PE odds ratios indicate which questions should be asked if a patient has experienced 

newly incident heroin use. Clinicians can identify cases of HUD by asking two questions based 

on individual’s experiences in order to determine if a psychiatric consultation is needed. 

Evidence-based screening methods such as this are not currently available to clinicians and will 

allow for more accurate and timely HUD identification.   

Likewise, as problems with emotions emerged as an early and robust correlate within PE 

to PE ORs, mood disturbances may need to be a more central focus for identifying newly 

incident heroin use. To dovetail possible evidence-based screening for clinicians, evidence-based 

interactive journal logs are a possibility. For example, individuals could download an anonymous 

 

144 

mobile phone application allowing them to log their health experiences including drug use. The 

app could provide feedback ranging from universal recommendations such as staying hydrated 

and getting sleep to recommendations to meditate during mood disturbances. Logging 

information regarding newly incident heroin use and subsequently logging of emotional 

problems could elicit the application to provide treatment options and encourage the individual 

to seek local counseling. 

Some of the aforementioned interventions have structural barriers which make them 

particularly difficult for this population, as individuals who use heroin may be homeless and 

without health insurance. Similarly, users in poverty many not be able to afford transportation 

needed to make it to appointments or users may be incarcerated due to criminalization of heroin 

possession. Changing structural barriers may include educating the judicial system by talking 

with judges about drug court programs. Counseling via telephone or online chat could be a 

potential option. Even further, peer-based counseling could mirror Alcoholics Anonymous’ 

sponsors by pairing users with educated citizen science representatives to provide emotional 

support. Additional creative approaches to overcoming structural barriers to HUD treatment 

should be explored.  

7.3 Conclusions  
 
 

Drug dependence epidemiology and specifically that of heroin have quickly gained 

public awareness at a national and global level. Heroin-related overdose and other opioids crises-

related morbidity and mortality continue to increase. Even so, scientific knowledge and effective 

public health prevention strategies of heroin use and use disorder are lacking. The current study 

may help to move the field forward given it is the first to illustrate the natural history of HUD 

through the fine-grain stratification of the problems and experiences that develop over time. This 

dissertation also draws attention to the need for more precise HUD assessment, given that extra-

 

145 

medical prescription opioid use likely influences the progression or interpretation of HUD. This 

study also suggests there may be a linkage between newly incident heroin use and stimulant use 

which deserves more attention in future research. Moving forward, it will be important for public 

health to anticipate drug use trajectories in order to end cycles of drug use and the subsequent 

morbidity and mortality.   

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

146 

APPENDICES 

 

147 

APPENDIX A  

 

IRB DETERMINATION 

148 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

149 

APPENDIX B  

 

MANUSCRIPT 1

 

150 

recode hermfu(1/3=1)(4/6=2)(7/9=3)(10/12=4)(else=.),gen(herquarter) 

replace recher=1 if recher2_b1==1 & quarter==1 & herquarter<5 & herquarter!=. 
replace recher=1 if recher2_b1==1 & quarter==2 & herquarter>1 & herquarter!=. 
replace recher=1 if recher2_b1==1 & quarter==3 & herquarter>2 & herquarter!=. 
replace recher=1 if recher2_b1==1 & quarter==4 & herquarter==4 & herquarter!=. 
replace recher=1 if recher1_b1==1 
drop recher1_b* 

Table 56. Manuscript 1 - Stata Code  
//SB usual code - most conservative definition; drop missing mfu  
/7 don't knows/refused end up in recher==1, yfu must match survey year 
gen recher1_b1=1 if year==irheryfu 
gen recher2_b1=1 if irheryfu==year-1 & irherrc<3 
 
gen recher=0 
 
 
 
 
 
 
gen nwincher=1 if herflag==0 | recher==1 
 
replace nwincher=0 if nwincher==. 
*Step2 lagtime 
gen hermy=ym(irheryfu, hermfu) 
gen intm=. 
replace intm=2 if quarter==1 
replace intm=5 if quarter==2 
replace intm=8 if quarter==3 
replace intm=11 if quarter==4 
gen intmy=ym(year, intm) 
gen lagtime=intmy-hermy 
recode lagtime (-1=1) (0=1) (14/30=.) 
*Step 3 PEs 
foreach k of varlist hergtovr herwd3sx herwdsmt herlottm herlimit herkplmt herndmor herlsefx 
hercutdn hercutev hercut1x hercutev heremopb heremctd herphlpb herphctd herlsact herserpb 
herpdang herlawtr herfmfpb herfmctd { 
 
} 
*set limit 
*herkplmt=2 (could not keep limits) 
recode herkplmt(99=.)(93=.)(91=.)(1 97 98 94=1)(2=2) 
recode herlimit (99=.)(93=.)(91=.)(2 97 98 94 85=2)(1=1) 
gen limit=. 
replace limit=1 if herlimit==1 & herkplmt==2 
replace limit=2 if herlimit==1 & herkplmt==1 
replace limit=2 if herlimit==2 
*cut down 
*hercutev=2 (could not cut down) 
recode hercutev (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
recode hercutdn (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
gen cutdown=. 
replace cutdown=1 if hercutdn==1 & hercutev==2 
replace cutdown=2 if hercutdn==1 & hercutev==1 
replace cutdown=2 if hercutdn==2 
Code adapted from James C. Anthony NIDA T32 Trainee Research.  

tab `k' if lagtime<=13 & recher==1 

 

151 

 

 

 

Table 56 (Cont’d). 
**cut down at least once  
*cut1x: asked Q if hercutev==2 OR hercutev==2 
*will have some hetcutdn yes's bc of cutdn/cutev gating  
recode hercut1x (93=.)(91=.) (2 99 94 97 98 85=2) (1=1) 
*withdrawal symptoms 
*herwdsmt=1 (sympt at same time) 
recode herwd3sx (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
recode herwdsmt (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
gen withdwl=.  
replace withdwl=1 if herwd3sx==1 & herwdsmt==1 
replace withdwl=2 if herwd3sx==1 & herwdsmt==2 
replace withdwl=2 if herwd3sx==2 | hercut1x==2 | cutdown==2 
replace herwd3sx=2 if hercutdn==2 | hercut1x==2 
*emotions problems 
recode heremctd (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
recode heremopb (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
gen emot=. 
replace emot=1 if heremopb==1 & heremctd==1 
replace emot=2 if heremopb==1 & heremctd==2 
replace emot=2 if heremopb==2 
*physical problems 
recode herphctd (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
recode herphlpb (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
gen phys=. 
replace phys=1 if herphlpb==1 & herphctd==1 
replace phys=2 if herphlpb==1 & herphctd==2 
replace phys=2 if herphlpb==2 | heremctd==1 | emot==1 | heremopb==1 
*physical 
replace herphlpb=2 if heremopb==1 | heremctd==1 
*fam/friends 
recode herfmctd (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
recode herfmfpb (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
gen fam=. 
replace fam=1 if herfmfpb==1 & herfmctd==1 
replace fam=2 if herfmfpb==1 & herfmctd==2 
replace fam=2 if herfmfpb==2 
*get over 
recode herlottm (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
recode hergtovr (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
gen gtovr=. 
replace gtovr=1 if herlottm==2 & hergtovr==1 
replace gtovr=2 if herlottm==2 & hergtovr==2 
replace gtovr=2 if herlottm==1  
Code adapted from James C. Anthony NIDA T32 Trainee Research.  

152 

recode `k' (1=1) (2 94 97 98 85=0) (else=.) 
tab `k' if lagtime<=13 & recher==1 
 

Table 56 (Cont’d).  
*less effect 
recode herndmor (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
recode herlsefx (99=.)(93=.)(91=.) (2 94 97 98 85=2) (1=1) 
replace herlsefx=1 if herlsefx==1 
replace herlsefx=2 if herndmor==1 | herlsefx==2 
//PE frequency 
* 94="Don't know", 97=Refused, 98=Blank/No Answer  
foreach k of varlist gtovr herwd3sx withdwl herlottm herlimit limit herndmor herlsefx 
cutdown hercut1x heremopb emot herphlpb phys herlsact herserpb herpdang herlawtr 
herfmfpb fam { 
 
 
} 
*Recode for looping through PEs 
*ordered by frequency (==1) 
gen pe1=herlimit 
gen pe2=heremopb 
gen pe3=herlottm 
gen pe4=herndmor 
gen pe5=herfmfpb 
gen pe6=herlsact 
gen pe7=herserpb 
gen pe8=hercut1x 
gen pe9=emot 
gen pe10=herpdang 
gen pe11=fam 
gen pe12=limit 
gen pe13=cutdown 
gen pe14=herwd3sx 
gen pe15=herlawtr 
gen pe16=withdwl 
gen pe17=herlsefx 
gen pe18=gtovr 
gen pe19=herphlpb 
gen pe20=phys 
Code adapted from James C. Anthony NIDA T32 Trainee Research.  

153 

 
 
 
 
 
 
 
 
 
 

 

forvalues j=`k'/20{ 
foreach s of numlist 1(1)13 { 
capture noisily insheet using "/Users/Samantha/Library 

Table 56 (Cont’d).  
//Survey Weights for Pooled Years 
gen double vestryr=vestr*10000 
gen verepyr=verep*10000+year 
gen nanal_wt=analwt_c/15 
svyset verepyr [pweight=nanal_wt], strata(vestryr) singleunit(centered) 
forvalues i=1/19{ 
local k=`i'+1 
 
 
 
//Analysis weigth quartiles 
pctile qweight = nanal_wt, nq(4) 
summarize nanal_wt, detail 
gen newvar=. 
replace newvar=1 if nanal_wt>0 & nanal_wt<=61.84667        
replace newvar=2 if nanal_wt>61.84667 & nanal_wt<=134.05                     
replace newvar=3 if nanal_wt>134.05 & nanal_wt<=337.194      
replace newvar=4 if nanal_wt>337.194 & nanal_wt!=. 
forvalues i=1/19{ 
local k=`i'+1 
 
 
 
 
 
plain onecell label 
}}} 
// or label var 
egen string pelab = concat(firstpe secondpe), punct(,) 
// graph or by quartile 
*gen nlb=0 if lb<0 
gen nub=ub 
replace nub=20 if nub>=20 
Code adapted from James C. Anthony NIDA T32 Trainee Research.  

forvalues j=`k'/20{ 
foreach s of numlist 1(1)4 { 
dis "pe`i' pe`j' newvar`s'" 
capture noisily logit pe`i' pe`j' if recher==1 & newvar==`s', or 
capture noisily esttab using pe`i'_pe`j'_newvar`s'.csv, b ci noconstant nostar eform 

154 

 

 

forvalues j=`k'/20{ 
foreach s of numlist 1(1)13 { 
dis "pe`i' pe`j' lag`s'" 
capture noisily svy, subpop(if lagtime==`s' & recher==1): logit pe`i' pe`j', or 
capture noisily esttab using pe`i'_pe`j'_lag`s'wt.csv, b ci noconstant nostar eform plain 

Table 56(Cont’d).  
*Pe to Pe 
forvalues i=1/19{ 
local k=`i'+1 
 
 
 
 
 
onecell label 
}}} 
gen nor=or 
replace nor=20 if nor>=20 & nub==20 
twoway scatter nor firstpe if qt==1, mlabel(pelab)  mlabsize(1.2) mlabposition (2) || rspike lb 
nub firstpe if qt==1 
twoway scatter nor firstpe if qt==2, mlabel(pelab)  mlabsize(1.2) mlabposition (2) || rspike lb 
nub firstpe if qt==2 
twoway scatter nor firstpe if qt==3, mlabel(pelab)  mlabsize(1.2) mlabposition (2) || rspike lb 
nub firstpe if qt==3 
twoway scatter nor firstpe if qt==4, mlabel(pelab)  mlabsize(1.2) mlabposition (2) || rspike lb 
nub firstpe if qt==4 
//sensitivity analysis  
*Create non-overlapping lagtimes 
gen senlag=. 
replace senlag=1 if lagtime>=1 & lagtime<=3 
replace senlag=2 if lagtime>=7 & lagtime<=8 
replace senlag=3 if lagtime>=12 & lagtime<=13 
*Check senlag (no overlap) 
tab lagtime senlag 
*Weighted PE-to-PE  
forvalues i=1/19{ 
local k=`i'+1 
 
 
 
capture noisily svy, subpop(if recher==1 & senlag==`s'): logit pe`i' pe`j', or 
capture noisily esttab using pe`i'_pe`j'_senlag`s'.csv, b ci noconstant nostar eform plain onecell 
label 
}}} 
}}} 
gen nlb=lb 
replace nlb = round(nlb, .1) 
gen nub=ub 
replace nub = round(nub, .1) 
egen string ci = concat(nlb nub), punct(", ") 
export excel using "senlag.output", firstrow(variables) 
Code adapted from James C. Anthony NIDA T32 Trainee Research.  

forvalues j=`k'/20{ 
foreach s of numlist 1(1)3 { 
dis "pe`i' pe`j' senlag`s'" 

155 

 

 

Figure 15. Flow of Heroin Use Disorder Questionnaire Based on DSM-IV Criteria 
Data from United States National Surveys on Drug Use and Health, 2002-2016.  

 

Sequence follows choice/arrows (i.e., gated questions) and questions without choice/arrows 
followed by question in next row (i.e., asked for all participants who used heroin within 12 
months of assessment). 

 
 

156 

 

 

Figure 15 (cont’d).  
 

 

 

Questionnaire used for heroin dependence case ascertainment in National Surveys on Drug 
Use and Health survey assessment, 2002-2016. 
Questions not followed by choice/arrows were asked for all those who used heroin within 12 
months of survey assessment. To follow questionnaire sequence, follow choice/arrows and 
when not given choice/arrows then return to next row down. 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

157 

 
 

 

Table 57. Meta-Analysis of Newly Incident Heroin Use to Dependence Probabilities. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 
 

NIHU 
(n) 

Transitioned 
to HUD (%) 

Meta 

Anlaysis 
Forest Plot 

 

95% CI 

(13.8, 41.1) 

(7.9, 27.1) 

(15.4, 42.1) 

(16.3, 38.5) 

(27.8, 51.1) 

(23.9, 48.5) 

(24.2, 47.5) 

Population 
Projections 

Effective 
Samplea 
 HUD 
Cases 
(n) 

HUD 
Cases 
(n) 

38 

56 

41 

59 

65 

56 

63 

26,000 

104,000 

17,000 

114,000 

26,000 

98,000 

39,000 

150,000 

63,000 

162,000 

57,000 

163,000 

52,000 

149,000 

25.0 

15.2 

26.7 

25.9 

38.8 

35.2 

35.0 

Year Pairs 

2002-2003 

2004-2005 

2006-2007 

2008-2009 

2010-2011 

2012-2013 

2015-2016 

Meta.b 

 

 

 
NSDUH Restricted Use Dataset. (web Nov. 2018) 
aEffective n based on back-calculation method with standard errors, Vsevolozhskaya & 
Anthony, 2014 https://www.ncbi.nlm.nih.gov/pubmed/25175545 
2014-2015 estimates not included due to covariance between year pairs. 
bMeta-analytic summary estimate: Log transformation and random effects estimator. 
Heterogeneity chi-squared =   9.82 (d.f. = 6) p = 0.133.  

(24.5, 36.7) 

30.0 

 

158 

 

 

 

 

 
 
 
 
 
 
 

 

Table 58. Unweighted Odds Ratios of Problem and Experience Pairs Within 0-90 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 1: possible 0-90 days, assumed 30 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  

 

159 

Table 59. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 0-90 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 1: possible 0-90 days, assumed 30 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation.  

 

 

160 

 
 

 

Table 60. Unweighted Odds Ratios of Problem and Experience Pairs Within 1-120 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 2: possible 1-120 days, assumed 60 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

161 

 
 
 
 
 
 
 

 

Table 61. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 1-120 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 2: possible 1-120 days, assumed 60 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

 

162 

 

 

Table 62. Unweighted Odds Ratios of Problem and Experience Pairs Within 30-150 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 3: possible 30-150 days, assumed 90 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

163 

 
 
 
 
 
 
 

 

Table 63. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 30-150 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 3: possible 30-150 days, assumed 90 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

164 

 
 
 
 
 

 

Table 64. Unweighted Odds Ratios of Problem and Experience Pairs Within 60-180 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 4: possible 60-180 days, assumed 120 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

165 

 
 
 
 
 
 
 

 

Table 65. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 60-120 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 4: possible 60-180 days, assumed 120 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

166 

 
 
 
 
 

 

Table 66. Unweighted Odds Ratios of Problem and Experience Pairs Within 90-210 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 5: possible 90-210 days, assumed 150 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

167 

 
 
 
 
 
 
 

 

Table 67. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 90-210 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 5: possible 90-210 days, assumed 150 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

168 

 
 
 
 
 
 

 

Table 68. Unweighted Odds Ratios of Problem and Experience Pairs Within 120-240 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 6: possible 120-240, assumed 180 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

169 

 
 
 
 
 
 
 
 

 

Table 69. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 120-240 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 6: possible 120-240, assumed 180 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

 

170 

 
 
 
 
 
 
 

 

Table 70. Unweighted Odds Ratios of Problem and Experience Pairs Within 150-210 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 7: possible 150-270, assumed 210 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

171 

 
 
 
 
 
 
 
 

 

Table 71. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 150-270 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 7: possible 150-270, assumed 210 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

172 

 
 
 
 
 
 
 

 

Table 72. Unweighted Odds Ratios of Problem and Experience Pairs Within 180-300 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 8: possible 180-300, assumed 240 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

173 

 
 
 
 
 
 
 
 

 

Table 73. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 180-300 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 8: possible 180-300, assumed 240 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

174 

 
 
 
 
 
 
 

 

Table 74. Unweighted Odds Ratios of Problem and Experience Pairs Within 210-330 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 9: possible 210-330, assumed 270 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

175 

 
 
 
 
 
 
 
 

 

Table 75. Unweighted Odds Ratios of Problem and Experience Pairs Within 210-330 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 9: possible 210-330, assumed 270 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

176 

 
 
 
 
 
 
 

 

Table 76. Unweighted Odds Ratios of Problem and Experience Pairs Within 240-360 Days of First Using Heroin.  
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 10: possible 240-360, assumed 300 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

177 

 
 
 
 

 

Table 77. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 240-360 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 10: possible 240-360, assumed 300 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

178 

 
 
 
 
 
 

 

Table 78. Unweighted Odds Ratios of Problem and Experience Pairs Within 270-390 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 11: possible 270-390, assumed 330 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

179 

 
 
 
 
 
 
 

 

Table 79. Unweighted Odds Ratios of Problem and Experience Pairs Within 270-390 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 11: possible 270-390, assumed 330 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

180 

 
 
 
 
 
 
 

 

Table 80. Unweighted Odds Ratios of Problem and Experience Pairs Within 300-420 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 12: possible 300-420, assumed 360 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

181 

 
 
 
 
 
 
 
 

 

Table 81. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 300-420 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 12: possible 300-420, assumed 360 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

182 

 
 
 
 
 

 

Table 82. Unweighted Odds Ratios of Problem and Experience Pairs Within 330-450 Days of First Using Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 13: possible 330-450, assumed 390 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

183 

 
 
 
 
 
 
 

 

Table 83. Unweighted Odds Ratio Confidence Intervals of Problem and Experience Pairs Within 330-450 Days of First Using 
Heroin. 
Data from United States National Surveys on Drug Use and Health, 2002-2016. 

NSDUH Downloadable Public Use Datasets.  
Lag-time interval 12: possible 330-450, assumed 390 days.  
Odds ratios produced from generalized linear model logistic regression, rounded to the tenths place.  
Analysis-weighted 95% confidence intervals use Taylor Series Linearization for calculus-based variance estimation. 

 

184 

 
 
 
 
 
 

 

APPENDIX C  

 

MANUSCRIPT 2 

185 

 

 

 

Table 84. MPlus Code to Evaluate Measurement Equivalence for Heroin Use Disorder Among 
Newly Incident Heroin Users. 

Variable: 
  Names are  
     age2 irsex year recher lagtime pe1 pe2 pe3 pe4 pe5 pe6 pe7 pe8 pe9 
     pe10 pe11 pe12 pe13 pe14 pe15 pe16 pe17 pe18 pe19 pe20 vestryr verepyr 
     nanal_wt c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 opioid; 
  Missing are all (-9999) ;  
  Usevariables are c1-c10 opioid vestryr verepyr nanal_wt; 
  Categorical are c1-c10; 
  Grouping is opioid (1=twelve 2 = life 3 = never); 
  Weight = nanal_wt; 
  Stratification = vestryr;  
  Cluster = verepyr;  
ANALYSIS: ESTIMATOR IS WLSMV; 
          !DIFFTEST is deriv_.dat; 
          Type = complex; 
MODEL: !Latent variable HUD - the * after y1 asks Mplus to estimate that loading 
!the (L1-L10) names each loading so that we can constrain them equal for women; 
    HUD BY c1* c2 c3 c4 c5 c6 c7  c8  c9  c10 (L1-L10); 
    HUD@1;        !variance of latent variable = 1 
    [HUD@0];     !mean of latent variable = 0 
    {c1-c10@1};  !fixed for each item = 1 scale factor 
!thresholds     
    [c1$1] (T1); 
    [c2$1] (T2); 
    [c3$1] (T3); 
    [c4$1] (T4); 
    [c5$1] (T5); 
    [c6$1] (T6); 
    [c7$1] (T7); 
    [c8$1] (T8); 
    [c9$1] (T9); 
    [c10$1] (T10);       
Code citation: Dr. Ahnalee Brincks, 2018. 
Code adapted for weighted vs. unweighted analyses, comparisons across sex, and various 
model constraints.   

186 

 
 
 
 
 
 
 
 
 
 

 

Table 84 (cont’d).  
Model life: 
    !Removing the (L1-L10) here frees the loadings to be different from males 
    HUD BY c1 c2 c3 c4 c5 c6 c7  c8  c9  c10 (L1-L10); 
    !HUD*;       !asks Mplus to estimate the variance of HUD 
    [HUD@0];    !fixes mean of HUD = 0 
    {c1-c10@1}; !fixed for each item = 1 scale factor - required 
!thresholds     
    [c1$1] (T1); 
    [c2$1] (T2); 
    [c3$1] (T3); 
    [c4$1] (T4); 
    [c5$1] (T5); 
    [c6$1] (T6); 
    [c7$1] (T7); 
    [c8$1] (T8); 
    [c9$1] (T9); 
    [c10$1] (T10);      
Model twelve: 
    !Removing the (L1-L10) here frees the loadings to be different from males 
    HUD BY c1 c2 c3 c4 c5 c6 c7  c8  c9  c10 (L1-L10); 
    !HUD*;       !asks Mplus to estimate the variance of HUD 
    [HUD@0];    !fixes mean of HUD = 0 
    {c1-c10@1}; !fixed for each item = 1 scale factor - required 
!thresholds     
    [c1$1] (T1); 
    [c2$1] (T2); 
    [c3$1] (T3); 
    [c4$1] (T4); 
    [c5$1] (T5); 
    [c6$1] (T6); 
    [c7$1] (T7); 
    [c8$1] (T8); 
    [c9$1] (T9); 
    [c10$1] (T10);  
  Output: 
    modindices(3.84); 
    stdy; 
    residual; 
Code citation: Dr. Ahnalee Brincks, 2018. 
Code adapted for weighted vs. unweighted analyses, comparisons across sex, and various 
model constraints.   

187 

 

 

 

 

 

 

APPENDIX D  

MANUSCRIPT 3 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

188 

Table 85. Stata Code Used to Yield Case Crossover Discordant Pair Odds Ratios Displayed 
within Manuscript 3. 
set seed 618 
*Outcome 
egen mystim = group(irstmyfu stimmfu) 
egen mystimnm = group(irstmnmyfu stmnmmfu) 
egen mycoc = group(ircocyfu cocmfu) 
*Exposure 
egen myher = group(irheryfu hermfu) 
*Time 
gen time=. 
replace time=mystim-myher 
gen time2=. 
replace time2=mystimnm-myher 
gen time3=. 
replace time3=mycoc-myher 
*2 Month Window 
gen cco=. 
replace cco=1 if (time<=2 & time>=1 | time2<=2 & time2>=1 | time3<=2 & time3>=1 ) 
replace cco=0 if (time<=4 & time>=3 | time2<=4 & time2>=3 | time3<=4 & time3>=3)  
replace cco=. if herflag==0 
*Weights 
gen double vestryr=vestr*10000 
gen verepyr=verep*10000+year 
gen nanal_wt=analwt_c/15 
svyset verepyr [pweight=nanal_wt], strata(vestryr) singleunit(centered) 
*Cross Tabs 
*Unweighted Estimates 
foreach y of numlist 2002(1)2016 { 
 
* 0 case control 10 (case/control) 
*2002-2016 estimation… 
*Meta-Analyses 
replace luwest = log(uwest) 
replace lulb = log(uwlb) 
replace luub = log(uwub) 
metan luwest lulb luub if luwest>0, eform random lcols(year) 
//Weighted Estimates  
svy: tab cco 
foreach y of numlist 2002(1)2016 { 
 
* 0 case control 10 (case/control) 
*2002-2016 estimation… 
replace lwest = log(west) 
replace lwlb = log(wlb) 
replace lwub = log(wub) 
metan lwest lwlb lwub if lwest>0, eform random lcols(year) 
Code adapted from James C. Anthony NIDA T32 Trainee Research.  

svy, subpop(if year==`y'): tab cco} 

tab cco if year==`y' } 

 

189 

BIBLIOGRAPHY

 

190 

BIBLIOGRAPHY 

 

1.   Brownstein MJ. A brief history of opiates, opioid peptides, and opioid receptors. Proc Natl 

Acad Sci U S A. 1993 Jun 15;90(12):5391–3.  

2.   Degenhardt L, Bucello C, Mathers B, Briegleb C, Ali H, Hickman M, et al. Mortality 
among regular or dependent users of heroin and other opioids: a systematic review and 
meta-analysis of cohort studies: Mortality among opioid users. Addiction. 2011 
Jan;106(1):32–51.  

3.   Opioid Overdose | Drug Overdose | CDC Injury Center [Internet]. 2018 [cited 2018 May 1]. 

Available from: https://www.cdc.gov/drugoverdose/index.html 

4.   UNODC Executive Summary [Internet]. [cited 2018 Oct 26]. Available from: 
https://www.unodc.org/wdr2018/prelaunch/WDR18_Booklet_1_EXSUM.pdf 

5.   Degenhardt L, Whiteford HA, Ferrari AJ, Baxter AJ, Charlson FJ, Hall WD, et al. Global 
burden of disease attributable to illicit drug use and dependence: findings from the Global 
Burden of Disease Study 2010. The Lancet. 2013 Nov 9;382(9904):1564–74.  

6.   Heroin Overdose Data | Drug Overdose | CDC Injury Center [Internet]. 2018 [cited 2018 

Jun 20]. Available from: https://www.cdc.gov/drugoverdose/data/heroin.html 

7.   Degenhardt L, Charlson F, Mathers B, Hall WD, Flaxman AD, Johns N, et al. The global 
epidemiology and burden of opioid dependence: results from the global burden of disease 
2010 study. Addiction. 2014 Aug 1;109(8):1320–33.  

8.   Transforming Opium Poppies Into Heroin | FRONTLINE | PBS [Internet]. [cited 2018 Jul 
23]. Available from: https://www.pbs.org/wgbh/pages/frontline/shows/heroin/transform/ 

9.   Abuse NI on D. What is heroin and how is it used? [Internet]. [cited 2018 Jul 23]. Available 

from: https://www.drugabuse.gov/publications/research-reports/heroin/what-heroin 

10.   DEA / Drug Scheduling [Internet]. [cited 2018 Jul 23]. Available from: 

https://www.dea.gov/druginfo/ds.shtml 

11.   Hart, C.L & Ksir, C. Drugs, Society, and Human Behavior. 15th ed. McGraw Hill; 2013.  

12.   Pubchem. Diamorphine [Internet]. [cited 2018 Jul 30]. Available from: 

https://pubchem.ncbi.nlm.nih.gov/compound/5462328 

13.   Forensic Drug Testing for Opiates [Internet]. [cited 2018 Jul 30]. Available from: 

https://watermark.silverchair.com 

14.   Pathan H, Williams J. Basic opioid pharmacology: an update. Br J Pain. 2012 Feb;6(1):11–

6.  

 

191 

15.   Trescot AM, Datta S, Lee M, Hansen H. Opioid pharmacology. Pain Physician. 2008 

Mar;11(2 Suppl):S133-153.  

16.   Waldhoer M, Bartlett SE, Whistler JL. Opioid Receptors. Annu Rev Biochem. 2004 

Jun;73(1):953–90.  

17.   US Department of Health and Human Services (HHS). Facing Addiction in America: The 

Surgeon General’s Report on Alcohol, Drugs, and Health. Washington, DC: HHS: Office of 
the Surgeon General.; 2016.  

18.   Koob GF. Neural Mechanisms of Drug Reinforcementa. Ann N Y Acad Sci. 1992 Jun 

1;654(1):171–91.  

19.   American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders 

[Internet]. Fifth Edition. American Psychiatric Association; 2013. Available from: 
https://psychiatryonline.org/doi/book/10.1176/appi.books.9780890425596 

20.   Terry CE. The opium problem / [Internet]. New York,; 1928. Available from: 

http://hdl.handle.net/2027/uc1.$b56190 

21.   Heroin History and Statistics [Internet]. DrugAbuse.com. 2015 [cited 2018 Jul 30]. 

Available from: https://drugabuse.com/library/heroin-history-and-statistics/ 

22.   Robins LN. The natural history of drug abuse. Acta Psychiatr Scand. 1980 Aug 

1;62(s284):7–20.  

23.   Edward Preble, John J. Casey, Jr. Taking Care of Business - The Heroin User’s Life on the 

Street. Int J Addict. 1969;4(1):1–24.  

24.   Agar M, Reisinger HS. A Tale of Two Policies: The French Connection, Methadone, and 

Heroin Epidemics. Cult Med Psychiatry. 2002 Sep 1;26(3):371–96.  

25.   Vietnam Veterans Three Years after Vietnam: How Our Study Changed Our View of 
Heroin - Robins - 2010 - The American Journal on Addictions - Wiley Online Library 
[Internet]. [cited 2018 Jul 30]. Available from: 
https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1521-0391.2010.00046.x 

26.   Robins LN. The sixth Thomas James Okey Memorial Lecture. Vietnam veterans’ rapid 
recovery from heroin addiction: a fluke or normal expectation? Addict Abingdon Engl. 
1993 Aug;88(8):1041–54.  

27.   Opioid Epidemic in the United States [Internet]. [cited 2018 Jul 31]. Available from: 
http://www.painphysicianjournal.com/current/pdf?article=MTcwNA==&journal=68 

28.   Joint Commission Statement on Pain Management [Internet]. [cited 2018 Jul 31]. Available 

from: 
http://www.jointcommission.org/joint_commission_statement_on_pain_management/ 

29.   Pain_Std_History_Web_Version_05122017.pdf [Internet]. [cited 2018 Jul 31]. Available 

from: 

 

192 

https://www.jointcommission.org/assets/1/6/Pain_Std_History_Web_Version_05122017.pd
f 

30.   Van Zee A. The Promotion and Marketing of OxyContin: Commercial Triumph, Public 

Health Tragedy. Am J Public Health. 2009 Feb;99(2):221–7.  

31.   Understanding the Epidemic | Drug Overdose | CDC Injury Center [Internet]. 2017 [cited 

2017 Dec 4]. Available from: https://www.cdc.gov/drugoverdose/epidemic/index.html 

32.   Epidemiology of Drug Dependence [Internet]. [cited 2018 Apr 30]. Available from: 

https://acnp.org/wp-content/uploads/2017/11/C109_1557-1574.pdf 

33.   Morris JN. Uses of Epidemiology. Uses Epidemiol [Internet]. 1957 [cited 2018 Jul 31]; 

Available from: https://www.cabdirect.org/cabdirect/abstract/19582701461 

34.   SAMHDA [Internet]. [cited 2018 Oct 26]. Available from: https://rdas.samhsa.gov/#/ 

35.   Substance Abuse and Mental Health Services Administration. Key substance use and 

mental health inidcators in the United States: Results from the 2016 National Survey on 
Drug Use and Health. HHS Pulication No SMA 17-5044 NSDUH Ser H-52.  

36.   Ball JC, Rosen L, Flueck JA, NURCO DN. Lifetime Criminality of Heroin Addicts In The 

United States. J Drug Issues Tallahass Fla. 1982 Fall;12:225–239.  

37.   De Alarcón R, Rathod NH. Prevalence and early detection of heroin abuse. Br Med J. 1968 

Jun 1;2(5604):549–53.  

38.   Sex and Gender Differences in Substance Use | National Institute on Drug Abuse (NIDA) 

[Internet]. [cited 2018 Mar 29]. Available from: 
https://www.drugabuse.gov/publications/research-reports/substance-use-in-women/sex-
gender-differences-in-substance-use 

39.   Cheng HG, Cantave MD, Anthony JC. Taking the first full drink: Epidemiological evidence 

on male-female differences in the United States. Alcohol Clin Exp Res. 2016 
Apr;40(4):816–25.  

40.   Ann.Lynsen. Age- and Gender-Based Populations [Internet]. 2014 [cited 2018 May 7]. 

Available from: https://www.samhsa.gov/specific-populations/age-gender-based 

41.   Jones CM, Campopiano M, Baldwin G, McCance-Katz E. National and State Treatment 

Need and Capacity for Opioid Agonist Medication-Assisted Treatment. Am J Public 
Health. 2015 Aug;105(8):e55–63.  

42.   Increases in Heroin Overdose Deaths — 28 States, 2010 to 2012 [Internet]. [cited 2018 Oct 

27]. Available from: https://www.cdc.gov/mmwr/preview/mmwrhtml/mm6339a1.htm 

43.   Abuse NI on D. Prescription opioids and heroin have similar effects, different risk factors 

[Internet]. [cited 2018 Oct 27]. Available from: 
https://www.drugabuse.gov/publications/research-reports/relationship-between-

 

193 

prescription-drug-abuse-heroin-use/rx-opioids-heroin-have-similar-effects-different-risk-
factors 

44.   Jones CM, Logan J, Gladden RM, Bohm MK. Vital Signs: Demographic and Substance 
Use Trends Among Heroin Users — United States, 2002–2013. MMWR Morb Mortal 
Wkly Rep. 2015 Jul 10;64(26):719–25.  

45.   Huang B, Dawson DA, Stinson FS, Hasin DS, Ruan WJ, Saha TD, et al. Prevalence, 

Correlates, and Comorbidity of Nonmedical Prescription Drug Use and Drug Use Disorders 
in the United States. J Clin Psychiatry. 2006 Jul 15;67:1062–73.  

46.   Darke S, Ross J. Polydrug dependence and psychiatric comorbidity among heroin injectors. 

Drug Alcohol Depend. 1997 Nov 25;48(2):135–41.  

47.   Abuse NI on D. Introduction [Internet]. [cited 2018 Oct 27]. Available from: 

https://www.drugabuse.gov/publications/research-reports/relationship-between-
prescription-drug-abuse-heroin-use/introduction 

48.   Williams CT, Latkin CA. Neighborhood Socioeconomic Status, Personal Network 

Attributes, and Use of Heroin and Cocaine. Am J Prev Med. 2007 Jun;32(6 Suppl):S203–
10.  

49.   Lake S, Hayashi K, Milloy M-J, Wood E, Dong H, Montaner J, et al. Associations between 

childhood trauma and non-fatal overdose among people who inject drugs. Addict Behav. 
2015 Apr;43:83–8.  

50.   Darke S, Torok M, Ross J. Developmental trajectories to heroin dependence: Theoretical 

and clinical issues. J Appl Soc Psychol. 2017 Mar 1;47(3):165–71.  

51.   Rivera OJS, Havens JR, Parker MA, Anthony JC. Risk of Heroin Dependence in Newly 

Incident Heroin Users. JAMA Psychiatry. 2018 Aug 1;75(8):863–4.  

52.   The Predictive Validity of the Rat Self-Administration Model for Abuse Liability [Internet]. 

[cited 2018 Jul 23].  

53.   Kosten TR, George TP. The Neurobiology of Opioid Dependence: Implications for 

Treatment. Sci Pract Perspect. 2002 Jul;1(1):13–20.  

54.   Support Federal Funding for Harm Reduction Programs - Harm Reduction Coalition 

[Internet]. [cited 2018 Oct 27]. Available from: https://harmreduction.org/action-
center/syringe-access/ 

55.   Abuse NI on D. Mobilizing Citizen Science to Address the Overdose Epidemic [Internet]. 

2017 [cited 2018 Oct 27]. Available from: https://www.drugabuse.gov/about-nida/noras-
blog/2017/11/mobilizing-citizen-science-to-address-overdose-epidemic 

56.   WDR_2017_Methodology.pdf [Internet]. [cited 2018 Nov 20]. Available from: 

https://www.unodc.org/wdr2017/field/WDR_2017_Methodology.pdf 

 

194 

57.   Fischer PJ, Shapiro S, Breakey WR, Anthony JC, Kramer M. Mental health and social 

characteristics of the homeless: a survey of mission users. Am J Public Health. 1986 
May;76(5):519–24.  

58.   Creswell JW, Clark VLP. Designing and Conducting Mixed Methods Research. SAGE 

Publications; 2017. 521 p.  

59.   NSDUH 2002 Methodological Resource Book (MRB) | SAMHDA [Internet]. [cited 2018 
Nov 20]. Available from: https://www.datafiles.samhsa.gov/study-publication/nsduh-2002-
methodological-resource-book-mrb-nid14370 

60.   NSDUH-MethodSummDefs-2016.pdf [Internet]. [cited 2018 Sep 14]. Available from: 

https://www.samhsa.gov/data/sites/default/files/NSDUH-MethodSummDefs-
2016/NSDUH-MethodSummDefs-2016.pdf 

61.   Czajka JL, Beyler A. Background Paper Declining Response Rates in Federal Surveys: 

Trends and Implications. Math Policy Res. 2016;1:1–86.  

62.   Deville J-C, Särndal C-E. Calibration Estimators in Survey Sampling. J Am Stat Assoc. 

1992 Jun;87(418):376–82.  

63.   Khalsa JH, Treisman G, McCance-Katz E, Tedaldi E. Medical Consequences of Drug 
Abuse and Co-occurring Infections. Subst Abuse Off Publ Assoc Med Educ Res Subst 
Abuse. 2008;29(3):5–16.  

64.   Abuse NI on D. Introduction [Internet]. [cited 2018 May 22]. Available from: 
https://www.drugabuse.gov/publications/research-reports/relationship-between-
prescription-drug-abuse-heroin-use/introduction 

65.   Increase in hepatitis C infections linked to worsening opioid epidemic | 2017 | Newsroom | 

NCHHSTP | CDC [Internet]. 2018 [cited 2018 Nov 10]. Available from: 
https://www.cdc.gov/nchhstp/newsroom/2017/hepatitis-c-and-opioid-injection-press-
release.html 

66.   Hospitalizations for heart infection related to drug injection rising across the US [Internet]. 

ScienceDaily. [cited 2018 Nov 10]. Available from: 
https://www.sciencedaily.com/releases/2016/09/160901092818.htm 

67.   Jiang R, Lee I, Lee TA, Pickard AS. The societal cost of heroin use disorder in the United 

States. PLoS ONE [Internet]. 2017 May 30;12(5). Available from: 
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5448739/ 

68.   Robins LN, Slobodyan S. Post-Vietnam heroin use and injection by returning US veterans: 

clues to preventing injection today. Addict Abingdon Engl. 2003 Aug;98(8):1053–60.  

69.   Santiago Rivera OJ, Havens JR, Anthony JC. Heroin dependece transition rates for the 

United States before and during the current heroin epidemic: 2005 through 2013. Reveiw. 
2016;  

 

195 

70.   De Vries TJ, Schoffelmeer ANM, Binnekade R, Raaso H, Vanderschuren LJMJ. Relapse to 

Cocaine- and Heroin-Seeking Behavior Mediated by Dopamine D2 Receptors Is Time-
Dependent and Associated with Behavioral Sensitization. Neuropsychopharmacology. 2002 
Jan;26(1):18–26.  

71.   Valente TW, Dyal SR, Chu K-H, Wipfli H, Fujimoto K. Diffusion of innovations theory 

applied to global tobacco control treaty ratification. Soc Sci Med. 2015 Nov 
1;145(Supplement C):89–97.  

72.   Hirschfeld G, von Brachel R. Multiple-Group Confirmatory Factor Analysis in R--A 

Tutorial in Measurement Invariance with Continuous and Ordinal Indicators. Pract Assess 
Res Eval. 2014 Jul;19(7).  

73.   Do our Measures Measure Up? The Critical Role of Measurement Invariance. [Internet]. 

[cited 2018 Oct 8].  

74.   Wu H, Estabrook R. Identification of Confirmatory Factor Analysis Models of Different 

Levels of Invariance for Ordered Categorical Outcomes. Psychometrika. 2016 
Dec;81(4):1014–45.  

75.   Millsap RE, Yun-Tein J. Assessing Factorial Invariance in Ordered-Categorical Measures. 

Multivar Behav Res. 2004 Jul 1;39(3):479–515.  

76.   Latent Variable Analysis With Categorical Outcomes: Multiple-Group and Growth 

Modeling in Mplus [Internet]. [cited 2018 Nov 15]. Available from: 
https://www.statmodel.com/download/webnotes/CatMGLong.pdf 

77.   Greenfield SF, Back SE, Lawson K, Brady KT. Substance Abuse in Women. Psychiatr Clin 

North Am. 2010 Jun;33(2):339–55.  

78.   Compton WM, Jones CM, Baldwin GT. Relationship between Nonmedical Prescription-

Opioid Use and Heroin Use. N Engl J Med. 2016 Jan 14;374(2):154–63.  

79.   Cerdá M, Santaella J, Marshall BDL, Kim JH, Martins SS. Nonmedical Prescription Opioid 

Use in Childhood and Early Adolescence Predicts Transitions to Heroin Use in Young 
Adulthood: A National Study. J Pediatr. 2015 Sep;167(3):605-612.e1-2.  

80.   Narcotic | Origin and meaning of narcotic by Online Etymology Dictionary [Internet]. 

[cited 2018 Nov 16]. Available from: https://www.etymonline.com/word/narcotic 

81.   Jones CM, Baldwin GT, Compton WM. Recent Increases in Cocaine-Related Overdose 

Deaths and the Role of Opioids. Am J Public Health Wash. 2017 Mar;107(3):430–2.  

82.   Wu LT, Anthony JC. The use of the case-crossover design in studying illicit drug use. Subst 

Use Misuse. 2000 Jun;35(6–8):1035–50.  

83.   O’Brien MS, Comment LA, Liang KY, Anthony JC. Does cannabis onset trigger cocaine 
onset? A case-crossover approach. Int J Methods Psychiatr Res. 2012 Mar;21(1):66–75.  

 

196 

84.   Does Starting to Smoke Cigars Trigger Onset of Cannabis Blunt Smoking? - PubMed - 

NCBI [Internet]. [cited 2018 Nov 16]. Available from: 
https://www.ncbi.nlm.nih.gov/pubmed/28339595 

85.   O’Brien MS, Wu L-T, Anthony JC. Cocaine use and the occurrence of panic attacks in the 

community: a case-crossover approach. Subst Use Misuse. 2005;40(3):285–97.  

86.   Yeh H-H. Contribution to Epidemiological Research on the Use of Heroin and Other 

Opioid Compounds [Internet] [Ph.D.]. [United States -- Michigan]: Michigan State 
University; 2017. Available from: https://search-proquest-
com.proxy2.cl.msu.edu/pqdtglobal/docview/1952261929/abstract/43FB2F7FA84CE7PQ/1 

87.   Maclure M. The Case-Crossover Design: A Method for Studying Transient Effects on the 

Risk of Acute Events. Am J Epidemiol. 1991 Jan 15;133(2):144–53.  

88.   Sleep Profile in Opioid Dependence: A Polysomnographic Case–Control Study | Ovid 

[Internet]. [cited 2018 Aug 21]. Available from: https://oce-ovid-
com.proxy2.cl.msu.edu/article/00004691-201412000-00002/HTML 

 

 

197