CONTRIBUTION TO EPIDEMIOLOGICAL RESEARCH ON THE USE OF HEROIN
AND OTHER OPIOID COMPOUNDS
By
Hsueh-Han Yeh

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Epidemiology—Doctor of Philosophy
2017

ABSTRACT
CONTRIBUTION TO EPIDEMIOLOGICAL RESEARCH ON THE USE OF HEROIN
AND OTHER OPIOID COMPOUNDS
By
Hsueh-Han Yeh
Patterns of heroin use and extra-medical (EM) use of other opioid compounds have resurfaced as important public health issues in the United States (US) and in many countries.
Midway through a 21st century epidemic of death due to heroin overdose in the US, there are
many gaps in epidemiological evidence. For example, experts have already have concluded that
heroin often becomes an alternative or substitute for other prescription opioids because an unit
dose of heroin has become less expensive and now is more readily available than diverted
prescription opioids (Cicero et al., 2012; Jones 2013; Muhuri et al., 2013). Nonetheless, the
evidence on this issue largely is based on case reports, and provides little basis for causal
inference. This problem is compounded by the lack of published population-level estimates on
heroin incidence rates, akin to estimates derived by Hunt & Chambers (1976), based upon
treatment admissions during the US heroin epidemic of the 1960s and 1970s.
The first aim of this dissertation addresses this incidence rate gap in epidemiological
evidence for the US. It seeks to estimate heroin incidence for 1992-2012 using the HuntChambers method based on opioid treatment admissions. The second dissertation aim is
concentrated on a causal inference about the process of becoming a newly incident heroin user,
and whether this process might be triggered by prior EM use of OxyContinÂŽ, a specific
prescription opioid. Thus, the second specific aim is to conduct a case-crossover study to
investigate a “triggering hypothesis” that links antecedent EM OxyContin® use with later heroin

onset in the US. The third study of this dissertation aims to predict the probability of
transitioning from EM prescription pain reliever use to heroin onset using survival analysis
models, with attention to hypothesized subgroup variation (i.e., population density).
Based on Hunt’s model via re-calibration approaches using Treatment Episode Data Set Admissions datasets (TEDS-A), heroin onset increased 160% from 2000 to its peak in 2010; the
incidence rate in 2012 is similar to the first heroin epidemic in 1969. As for opioids other than
heroin (e.g., prescription opioids), the incidence increased more than 250% from 2000 to 2010.
The second study investigates whether EM OxyContinÂŽ use might have triggered heroin onset
among 12-25 year olds in the period of 2004-2014. The excess risk of newly incident heroin use
is seen in a four-month interval right after onset of EM OxyContinÂŽ use (case-crossover risk
ratio = 1.9). Post-estimation exploratory analyses suggest no excess risk for EM users of other
prescription pain relievers, and indicate no excess risk correlated with new formulations of
OxyContinÂŽ per se. In the third study, the peak risk for transitioning from onset of extra-medical
prescription pain reliever use (EMPPR) to heroin onset within 10 years emerged at the third year
since first EMPPR use. Approximately 5% of participants initiated heroin use in 10 years since
EMPPR onset. The estimates of these subgroups remain similar regardless of population density.
EMPPR users who are male, White, and with early EMPPR onset have an increased risk of
initiating heroin use.
Heroin outbreaks and epidemics revisited the US in the 2010s after the first US heroin
epidemic of the 1970s. If successful, this project’s new evidence on heroin in the US population
in recent epidemic years should improve our understanding of heroin epidemiology, and may be
an aid to new public health responses for primary prevention, outreach, and treatment resources
with respect to heroin in the current epidemic and during future epidemics.

Copyright by
HSUEH-HAN YEH
2017

This dissertation is dedicated to my parents.
Thank you for always believing in me,
always encouraging me,
and loving me unconditionally.

v

ACKNOWLEDGMENTS

I would like to acknowledge Professor Jim Anthony for his mentorship during the entire
process of this dissertation, and my dissertation committee members, Professors David
Barondess, Qing Lu, and Chuan-Yu Chen. This dissertation would not have been possible
without their guidance, valuable insight, and constructive advice. I also want to thank the fellows
of T32 program and faculty in the Department of Epidemiology and Biostatistics for their great
helps and knowledge sharing.
During this dissertation research, I have had the advantage of working with Professor Jim
Anthony while he was supported by his United States National Institutes of Health (NIH) and
National Institute on Drug Abuse (NIDA) K05 Senior Scientist and Mentorship award
(K05DA015799) and he allowed me to participate in the weekly workgroup mentoring sessions
for his NIDA T32 fellows. Those weekly workgroup sessions, supported by the NIH/NIDA grant
award (T32DA021129), did not give me a trainee stipend, but still helped me to understand
many facets of NIDA epidemiology and prevention research that I otherwise might not have
mastered. I also want to gratefully acknowledge funding supports for my doctoral study from
Professor RubĂŠn Parra-Cardona (K01DA036747, NIDA), Professor Gretchen Birbeck (The Bill
& Melinda Gates Foundation for the Global Burden of Disease Study of Neurologic Conditions),
and Professor Mangala Sadasivan (College of Osteopathic Medicine, Michigan State University).
Special thanks to all colleagues in the International Neurologic and Psychiatric Epidemiology
Program (INPEP, Michigan State University) for their tremendous support, and College of

vi

Human Medicine for awarding me with Dissertation Fellowship to support me completing
dissertation during the last year of my doctoral study.
I would like to give my deepest and greatest gratitude to my family, my parents and my
brother, for their unconditional support and love. I also would like to thank all friends for their
encouragement and support during these years.

vii

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................................... xi
LIST OF FIGURES .................................................................................................................... xii
KEY TO ABBREVIATIONS .................................................................................................... xv
CHAPTER 1 OVERVIEW AND SPECIFIC AIMS................................................................. 1
1.1 OVERVIEW ................................................................................................................. 1
1.2 SPECIFIC AIMS .......................................................................................................... 3
1.2.1 SPECIFIC AIM 1: To estimate the incidence of heroin use
considering lag-time between onset and admission to a treatment facility,
and to compare treatment-based estimates with field survey estimates. ................ 3
1.2.2 SPECIFIC AIM 2: To estimate the triggering effect from
OxyContinÂŽ use specifically and from prescription opioids generally to
heroin use via the case-crossover approach. ........................................................... 3
1.2.3 SPECIFIC AIM 3: To estimate the probability of transition from
extra-medical EMPPR onset to heroin onset with using survival analysis
models, with attention to hypothesized subgroup variation in relation to
population density and region of the country. ........................................................ 4
CHAPTER 2 BACKGROUND AND SIGNIFICANCE ......................................................... 5
2.1 HEROIN AND OTHER OPIOIDS ............................................................................... 5
2.1.1 Brief introduction of opioids .......................................................................... 5
2.1.2 Pharmacology of heroin and other opioids .................................................... 6
2.1.3 History of supply, manufacture, and use ....................................................... 9
2.2 CLASSIC STUDIES ON HEROIN EPIDEMIOLOGY............................................. 11
2.2.1 The Road to H (Chein et al., 1964) .............................................................. 11
2.2.2 The Epidemiology of Opiate Addiction in the United States (Ball
and Chambers, 1970) ............................................................................................ 13
2.2.3 The heroin epidemics: A study of heroin use in the US (Hunt and
Chambers, 1976) ................................................................................................... 15
2.3 HEROIN AND OPIOIDS EPIDEMIOLOGY IN THE 21ST CENTURY .................. 18
2.3.1 The Definition of Abnormal Use of Opioids ............................................... 18
2.3.2 Prevalence of Prescription Opioids and Heroin ........................................... 19
2.3.3 Burden of Disease Attributable to Opioid Use ............................................ 20
2.3.4 Relationship between prescription opioid (e.g., OxyContinÂŽ) and
heroin .................................................................................................................... 21
2.4 RESEARCH GAPS AND SIGNIFICANCE .............................................................. 23
CHAPTER 3 MATERIALS ...................................................................................................... 24
3.1 NATIONAL SURVEY ON DRUG USE AND HEALTH (NSDUH) ....................... 24
3.2 TREATMENT EPISODE DATA SET - ADMISSIONS (TEDS-A) ......................... 25

viii

CHAPTER 4 MANUSCRIPT 1 – EMPIRICAL EVIDENCE ON THE
ESTIMATION OF HEROIN INCIDENCE IN THE UNITED STATES: A BACKCALCULATION APPROACH ................................................................................................. 27
4.1 ABSTRACT ................................................................................................................ 27
4.2 INTRODUCTION ...................................................................................................... 28
4.3 METHODS ................................................................................................................. 31
4.3.1 Data source................................................................................................... 31
4.3.2 Sample selection .......................................................................................... 31
4.3.3 Organizing the case material ........................................................................ 32
4.3.4 Lag Correction Method ................................................................................ 36
4.4 RESULTS ................................................................................................................... 42
4.5 DISCUSSION ............................................................................................................. 52
CHAPTER 5 MANUSCRIPT 2 – A TRIGGER FOR NEWLY INCIDENT
HEROIN USE? EPIDEMIOLOGICAL EVIDENCE FROM THE UNITED
STATES: A CASE-CROSSOVER STUDY DESIGN ............................................................. 56
5.1 ABSTRACT ................................................................................................................ 56
5.2 INTRODUCTION ...................................................................................................... 57
5.3 METHODS ................................................................................................................. 59
5.3.1 Population Under Study and Sampling Approach ....................................... 60
5.3.2 Assessment Approach .................................................................................. 61
5.3.3 Post-Assessment Inclusions and Exclusions Approach ............................... 62
5.3.4 Hazard Interval and Control Interval Specifications ................................... 64
5.3.5 Analyze/Estimate Approach ........................................................................ 64
5.4 RESULTS ................................................................................................................... 66
5.5 DISCUSSION ............................................................................................................. 71
CHAPTER 6 MANUSCRIPT 3 – PATTERNS OF DURATION OF ELAPSED
TIME FROM EXTRA-MEDICAL PRESCRIPTION PAIN RELIEVER ONSET
TO FIRST HEROIN USE: A SURVIVAL ANALYSIS.......................................................... 74
6.1 ABSTRACT ................................................................................................................ 74
6.2 INTRODUCTION ...................................................................................................... 75
6.3 METHODS ................................................................................................................. 78
6.4 RESULTS ................................................................................................................... 81
6.5 DISCUSSION ............................................................................................................. 90
CHAPTER 7 FINAL DISCUSSION ........................................................................................ 92
7.1 SUMMARY OF FINDINGS ...................................................................................... 92
7.2 LIMITATIONS AND STRENGTHS ......................................................................... 92
7.3 FUTURE DIRECTIONS AND CONCLUSIONS...................................................... 94
CHAPTER 8 CONCLUSIONS................................................................................................. 95
APPENDICES ............................................................................................................................. 97
APPENDIX A THE LIST OF MUSICIANS OF THE ERA WHO DIED FROM
HEROIN OR OTHER OPIOID DRUGS OVERDOSE ............................................................... 98

ix

APPENDIX B THE LIST OF PRESCRIPTION PAIN RELIEVERS FROM
NATIONAL SURVEY ON DRUG USE AND HEALTH (NSDUH) ....................................... 100
APPENDIX C THE ESTIMATED NUMBERS OF NEWLY INCIDENT HEROIN
USERS BY HUNT’S METHOD DURING 1950-2012 FOR MALE AND FEMALE ............. 101
APPENDIX D THE ESTIMATED NUMBERS OF NEWLY INCIDENT HEROIN
USERS BY HUNT’S METHOD DURING 1950-2012 FOR FOUR REGIONS (i.e.,
NORTHEAST, MIDWEST, WEST, SOUTH)........................................................................... 102
BIBLIOGRAPHY ..................................................................................................................... 103

x

LIST OF TABLES

Table 4.2.1 A brief summary of the history of opium.................................................................. 30
Table 4.3.1 Literature review of age of first heroin use ............................................................... 34
Table 4.4.1 Number of people admitted to the treatment program by the type of primary
drug use problem for their first admission. Data from United States, Treatment
Episode Data Set –Admissions (TEDS-A), 1992-2012. ................................................... 42
Table 4.4.2 Persons age 12-54 years in the treatment programs by the type of drugs on
their primary admission, Data from United States, Treatment Episode Data Set –
Admissions (TEDS-A), 1992-2012 (n=11,327,152) ........................................................ 44
Table 5.4.1 Data used to estimate case-crossover relative risk in relation to the timing of
heroin use in relation to the pre-specified four-month hazard and control intervals
prior to the month of newly incident heroin use. Data from 447 newly incident
heroin users age 12-25 years of age, United States National Surveys on Drug Use
and Health (NSDUH), 2004-2014. ................................................................................... 68
Table 5.4.2 Estimates from post-estimation exploratory data analyses to shed light on
variations in the study estimates when alternative specifications for the 'd' interval
are made, and to illuminate estimates pertinent to more general trigger hypothesis
about EM use of prescription opioids and pain relievers. Data from 447 newly
incident heroin users age 12-25 years of age, United States, National Surveys on
Drug Use and Health (NSDUH), 2004-2014. ................................................................... 69
Table 5.4.3 Estimates from post-estimation exploratory data analyses to shed light on
variations in the study estimates before-and-after re-formulation of OxyContinÂŽ,
and to illuminate estimates pertinent to 2004-2011 and 2012-2014. Data from 335
and 112 newly incident heroin users age 12-25 years of age respectively for 20042011 and 2012-2014, United States, National Surveys on Drug Use and Health
(NSDUH). ......................................................................................................................... 70
Table 6.4.1 Discrete time data is analyzed in the survival analysis using logic function for
each of the time periods. The unadjusted Hazard Ratios (HR) and covariateadjusted Hazard Ratios (aHR) are estimated with 95% confidence intervals (95%
CI). Data from the National Surveys on Drug Use and Health (NSDUH), United
States, 2008-2014. ............................................................................................................. 89
Table A.1 The list of musicians of the era who died from heroin or other opioid drugs
overdose ............................................................................................................................ 98

xi

LIST OF FIGURES

Figure 2.1.1 Classification of opioid subtypes. (Adapted from Booth, 1999) ............................... 6
Figure 2.1.2 Chemical structures of examples of six opioids ........................................................ 8
Figure 2.2.1 An example of Macro-epidemic process (Hughes P. and Crawford G., Illinois Drug
Abuse Program, unpublished, November, 1972) .............................................................. 17
Figure 4.3.1 Flowchart of United States, Treatment Episode Data Set –Admissions (TEDS-A),
1992-2012. ........................................................................................................................ 32
Figure 4.3.2 Cross-table of number of individuals per cell by the age of first admission and age
of first use (i.e., age of onset)............................................................................................ 35
Figure 4.3.3 Uniform Distribution is applied to break down cells into multiple one-year-period
cells. .................................................................................................................................. 35
Figure 4.3.4 Conceptual framework of truncations with lag time, Data from United States,
Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012. ................................. 37
Figure 4.3.5 A distribution of lag-years between heroin onset and treatment admission. Data
from United States, Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012. . 39
Figure 4.3.6 Heroin lag distribution. Data from United States, Treatment Episode Data Set –
Admissions (TEDS-A), 1992-2012. ................................................................................. 40
Figure 4.3.7 Other opioids lag distribution. Data from United States, Treatment Episode Data
Set –Admissions (TEDS-A), 1992-2012. ......................................................................... 41
Figure 4.4.1 The proportion of first treatment admissions of heroin and other opioids changes
over time. Data from United States, Treatment Episode Data Set –Admissions (TEDS-A),
1992-2012. ........................................................................................................................ 43
Figure 4.4.2 The observed and estimated number of newly incident heroin users from 1950 to
2012. Data from United States, Treatment Episode Data Set –Admissions (TEDS-A),
1992-2012. ........................................................................................................................ 46
Figure 4.4.3 US population age 12-54 years by year, US census 1950-2010. ............................. 47
Figure 4.4.4 Heroin incidence rate (per 10,000). Data from United States, Treatment Episode
Data Set –Admissions (TEDS-A), 1992-2012. ................................................................. 48

xii

Figure 4.4.5 The observed and estimated number of newly incident users of other opioids from
1950 to 2012. Data from United States, Treatment Episode Data Set –Admissions
(TEDS-A), 1992-2012. ..................................................................................................... 49
Figure 4.4.6 Incidence rate of opioids other than heroin (per 10,000). Data from United States,
Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012. ................................. 50
Figure 4.4.7 Incidence rate of all opioid compounds (per 10,000). Data from United States,
Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012. ................................. 51
Figure 4.5.1 The Federal Drug Control Budget 1970-2015 ......................................................... 54
Figure 5.3.1 Flow chart illustrates the sample selection of case-crossover study design. Data
from 393,473 individuals age 12-25 years of age, United States National Surveys on
Drug Use and Health, 2004-2014. .................................................................................... 63
Figure 5.3.2 Case-Crossover study design ................................................................................... 66
Figure 6.2.1 Hunt’s Macro-Diffusion theory ............................................................................... 76
Figure 6.4.1 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first
EMPPR use. Data from the National Surveys on Drug Use and Health (NSDUH), United
States, 2008-2014. ............................................................................................................. 83
Figure 6.4.2 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first
EMPPR use. Stratified by MET-SUBRUL. Data from the National Surveys on Drug Use
and Health (NSDUH), United States, 2008-2014. ............................................................ 84
Figure 6.4.3 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first
EMPPR use. Stratified by Male-Female. Data from the National Surveys on Drug Use
and Health (NSDUH), United States, 2008-2014. ............................................................ 85
Figure 6.4.4 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first
EMPPR use. Stratified by Age of EMPPR Onset. Data from the National Surveys on
Drug Use and Health (NSDUH), United States, 2008-2014. ........................................... 86
Figure 6.4.5 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first
EMPPR use. Stratified by Race/Ethnicity. Data from the National Surveys on Drug Use
and Health (NSDUH), United States, 2008-2014. ............................................................ 87

xiii

Figure 6.4.6 Estimated hazard functions depict the risk of starting heroin use, plotted in relation
to elapsed time (in years) since the year of the first EMPPR use, with contrasts of
subgroups (i.e., MET-SUBRUL, sex, age of EMPPR onset, and race/ethnicity). Data
from the National Surveys on Drug Use and Health (NSDUH), United States, 2008-2014.
........................................................................................................................................... 88
Figure B.1 The list of prescription pain relievers from NSDUH ............................................... 100
Figure C.1 The estimated numbers of newly incident heroin users by Hunt’s method during
1950-2012 for male and female. Data from United States, Treatment Episode Data Set Admissions, 1992-2012. ................................................................................................. 101
Figure D.1 The estimated numbers of newly incident heroin users by Hunt’s method during
1950-2012 for four regions (i.e., Northeast, Midwest, West, South). Data from United
States, Treatment Episode Data Set -Admissions, 1992-2012. ...................................... 102

xiv

KEY TO ABBREVIATIONS

CBSA: Core Based Statistical Area
CI: Confidence interval
CSA: Controlled Substances Act
DTSA: Discrete-time survival analysis
DSM: Diagnostic and Statistical Manual
EM: Extra-medical
EMOXY: Extra-medical OxyContin
EMPPR: Extra-medical prescription pain reliever
FDA: Food and Drug Administration
HR: Hazard ratio
IRD: Internationally regulated drugs
LCM: Lag Correction Method
MET: Urban/Metropolis
MSA: Metropolitan Statistical Area
NESARC: National Epidemiologic Survey on Alcohol and Related Conditions
NSDUH: National Survey on Drug Use and Health
RR: Relative risk
SUBRUL: Suburban/Rural
TEDS-A: Treatment Episode Data Set – Admissions

xv

CHAPTER 1
OVERVIEW AND SPECIFIC AIMS
1.1 OVERVIEW
This dissertation focuses on the epidemiology of opioid compounds. As described in
Chapter 2.0, the history of human use of opioids dates back to prehistoric times; traces of opium
have been found in Paleolithic human settlements. Misuse of opium compounds has been
documented since the early 19th century. Morphine and heroin, the two early opium derivatives,
were introduced in the mid-to-late 19th century. Later, in the 20th century, semi-synthetic opioids
(e.g., hydrocodone, oxycodone) were derived.
Opioid compounds can be conceptualized of as ‘an agent’ in the general agent-hostenvironment model of human disease. The range of human disorders resulting from human
contact with this ‘agent’ include opioid overdose (generally with complications such as
respiratory depression and cardiac arrest), opioid dependence (as defined by the World Health
Organization or the American Psychiatric Association), and infections or other complications
associated with injection of opioid compounds (e.g., human immunodeficiency virus (HIV)
infection and related opportunistic infections; hepatitis C virus infection; sepsis). There are many
unanswered research questions about opioid epidemiology in human populations. This
dissertation attempts to answer three specific questions in this domain of opioid epidemiology.
The first question falls under the rubric of ‘quantity’ within the array of five main rubrics
outlined for epidemiology (Anthony & Van Etten, 1998). Namely, ‘How many people are
becoming heroin users, year by year?’ Death statistics for the United States (US) clearly indicate
occurrence of a new epidemic of heroin overdose and EMPPR use overdose deaths (Han,
Compton, Jones, & Cai, 2015; Paulozzi, Budnitz, & Xi, 2006). Heroin overdoses do not occur

1

until heroin use occurs. There are several ways to derive estimates of incidence of heroin use;
one of the most common approaches is using treatment records on new admissions of people
requiring heroin intervention.
The second of the unanswered research questions falls under the rubric of ‘causes’ within
the array of five main rubrics of epidemiology (Anthony & Van Etten, 1998). Of interest is
whether the use of sustained release oxycodone (in the formulation known as OxyContinÂŽ)
triggers or precipitates the onset of heroin use. It is widely believed that the epidemic of
OxyContinÂŽ use has fueled a current epidemic of heroin use, with newly incident OxyContinÂŽ
use triggering onset of heroin use that otherwise would not have happened. This ‘triggering’
hypothesis will be tested using the epidemiological case-crossover approach, and this dissertation
includes estimates of the OxyContinÂŽ-heroin triggering relationship.
The third research question I am going to address in this dissertation is “in estimating
time from extra-medical use of prescription pain reliever (EMPPR) into heroin use, is there
hypothesized variation by area characteristics such as population density?” This question falls
under the rubric of ‘location’ within the array of five main rubrics of epidemiology (Anthony &
Van Etten, 1998).

2

1.2 SPECIFIC AIMS
1.2.1 SPECIFIC AIM 1: To estimate the incidence of heroin use considering lag-time between
onset and admission to a treatment facility, and to compare treatment-based estimates with field
survey estimates.
For this specific aim, the treatment admissions data based on Treatment Episode Data Set
- Admissions (TEDS-A) is used to estimate the incidence of heroin use with using ‘indirect’
approaches used to provide projections, forecasts, and steady state descriptions of heroin use
incidence when the only available data are from treatment admissions. That is, the comparison of
heroin incidence estimates based on treatment admissions data versus estimates based on the
field survey data will be informative, irrespective of whether the various methodological
approaches yield the same or different estimates of incidence. The expectation is that the
treatment admission incidence estimates will not differ from those of the field survey data (i.e., a
null hypothesis).
1.2.2 SPECIFIC AIM 2: To estimate the triggering effect from OxyContinÂŽ use specifically
and from prescription opioids generally to heroin use via the case-crossover approach.
This specific aim is to learn whether prior use of OxyContinÂŽ might trigger or precipitate
onsets of heroin use (e.g., when OxyContinÂŽ becomes too expensive or cost-prohibitive and
heroin use becomes less cost-prohibitive). The data come from the month and year of onset
obtained from recent national surveys. The design involves a case-crossover approach such that
there are two estimated quantities of crucial significance: (1) the number of individuals whose
onset of heroin use is preceded by OxyContin® onset during a pre-specified ‘hazard interval,’
and (2) the number of individuals whose onset of heroin use is preceded by OxyContinÂŽ onset
during a pre-specified ‘control interval.’ The ratio of these two numbers can be used to derive an
evaluation of the degree to which OxyContinÂŽ use might be triggering onset of heroin use.

3

1.2.3 SPECIFIC AIM 3: To estimate the probability of transition from extra-medical EMPPR
onset to heroin onset with using survival analysis models, with attention to hypothesized
subgroup variation in relation to population density and region of the country.
In the second specific aim, the hazard interval is pre-specified from EMPPR onset to
heroin onset. To evaluate my hypotheses of the length of time from EMPPR onset to heroin
onset, I turn to using survival analysis approaches to study the duration of elapsed time and the
factors influencing this duration, with attention to hypothesized subgroup variation in relation to
population density and region of the country. The null hypothesis is that the estimated median
time from EMPPR onset to heroin onset will not differ by population density.

4

CHAPTER 2
BACKGROUND AND SIGNIFICANCE

2.1 HEROIN AND OTHER OPIOIDS
2.1.1 Brief introduction of opioids
The focal point of this dissertation is the human use of opioid drug compounds, as
studied from an epidemiological perspective. Opioids are prominent as beneficial medicinal
compounds and especially respected for their pain relieving effect. However, they also have
some toxic consequences, including case fatality when large doses are taken (White & Irvine,
1999). The term ‘opioid’ refers to derivatives of the opium poppy (e.g., morphine, heroin) or are
synthesized (e.g., methadone) compounds such as methadone. As just noted, opioids usually are
prescribed for pain relief. In the opium form, opioids have been used as pain relievers for
thousands of years (Booth, 1999).
The distinction between opiates and opioids is fairly straightforward. Opiates are
essentially the derivatives of opium poppy. Opioids encompass the opiates as well as semisynthetic and synthetic compounds, some of which can be derived in a laboratory when an opium
supply is not present. Methadone is a prototypical opioid. It was developed during the World
War II era by chemists in Nazi Germany, when the country became concerned that its access to
opium poppy fields might be constrained by wartime blockades, and efficacious opioid pain
relief was a pressing need (Defalque & Wright, 2007).
It is possible to discriminate between three types of opioids: 1) Natural Opioids (‘opiates’),
which are directly created from alkaloids in the opium poppy plant, most often from material in
the seedpods; 2) Opioid Derivatives (also called semi-synthetic opiates), which are derived from

5

natural opioids; 3) Synthetic Opioids, which are manufactured in chemical laboratories to yield
similar chemical structures to the truly ‘opiate’ compounds derived from opium poppy plants.
For clarity, morphine and codeine are classified as natural opiates, whereas heroin is a
derivative of morphine, as shown in the accompanying figure.

Figure 2.1.1 Classification of opioid subtypes. (Adapted from Booth, 1999)

2.1.2 Pharmacology of heroin and other opioids
Opioids work in the body by interacting with proteins called opioid receptors, which can be
found in the central nervous system (CNS) and gastrointestinal tract (GI). Opioid receptors are
responsible for analgesic effects of opioids in the body. The three main types of opioid receptors
are mu (𝜇), kappa (𝜅), and delta (𝛿). For example, when opioid binds to the 𝜇-receptor, it
produces the effects of analgesia. Morphine is the first chemical found to bind to the 𝜇-receptor.
The 𝜅-receptor is also associated directly with analgesia but not many significant agonists acting
on the 𝜅-receptor are known. Morphine and other commonly used opioid analgesics strongly
bind to the 𝛿-receptor but the 𝛿-receptor is mostly found in larger cells and plays an important
role in spinal analgesia (Jordan, Cvejic, & Devi, 2000).

6

The chemical structure of a drug determines its affinity and activity for receptors and the
ability to elicit a response to the body. Similarity, the affinity and activity of opioid receptors can
be determined by its chemical structure (Feinberg, Creese, & Snyder, 1976). The chemical
structures of various opioid compounds are displayed in the Figure 2.1.2.

7

Figure 2.1.2 Chemical structures of examples of six opioids
NATURAL OPIATES
Morphine
Molecular Formula: C17H19NO3

Codeine
Molecular Formula: C18H21NO3

SEMI-SYNTHETIC OPIATES
Heroin
Molecular Formula: C21H23NO5

Oxycondone
Molecular Formula: C18H21NO4

SYNTHETIC OPIATES
Methadone
Molecular Formula: C21H27NO

Fentanyl
Molecular Formula: C22H28N2O

Source: PubChem Substance and PubChem Compound database (https://pubchem.ncbi.nlm.nih.gov/)

8

2.1.3 History of supply, manufacture, and use
Opium poppy crops are found in many countries. In 1803, morphine was extracted
from opium resin for medical treatments. In 1827, Merck pharmaceutical company (Germany)
began the manufacture of morphine, after which it was widely prescribed to relieve pain, reduce
coughing, or improve sleep. In 1898, heroin was introduced by the Bayer Company in Germany.
It was developed chiefly as an alternative to morphine for cough suppressants, and the Bayer
Company promoted that heroin did not have morphine's addictive side effects at that time. Later
time was found that heroin might be more addictive than morphine (BAYER Pharmaceutical
Products, 1901).
Before 1914, laws concerning opiates were strictly imposed only on a local city or
state basis in the US. For example, one of the first laws was in San Francisco in 1875; smoking
opium became illegal in opium dens. Harrison Narcotics Tax Act was a US federal law approved
in 1914. It stated that
It is an act to provide for the registration of, with collectors of internal revenue, and to
impose a special tax upon all persons who produce, import, manufacture, compound, deal
in, dispense, sell, distribute, or give away opium or coca leaves, their salts, derivatives, or
preparations, and for other purposes (Harrison Narcotics Tax Act, 1914).
This law aimed to require professionals who prescribed narcotics (e.g., heroin) to
register and pay taxes. Therefore, it helped to manage distribution and sale of narcotics.
Harrison's legislation did not prohibit drugs specifically but the prescription of narcotics is only
permitted under the course of clinicians’ professional practices.
In 1924, US Congress banned the manufacture, importation, and sale of heroin. In
1970, it enacted a Controlled Substance Act (CSA), which involved placing selected drugs into

9

one of five schedules under existing federal law. Heroin was listed as a Schedule I drug, making
it illegal to make or possess. Drugs in this schedule have no currently accepted medical use in the
US. The United States government recognized that the epidemiological 'reservoir' for heroin,
opioids, and other drugs subject to the Controlled Substances Act of 1970 (CSA1970) often was
located outside the US. These 'agents' in the agent-host-environment model might originate in
the opium-growing regions of Asia, and for this reason, international regulation was proposed.
An international treaty analogy of CSA1970 regulations has been adopted by most countries
participating in the United Nations and World Health Organization under the name
"Psychotropic Drug Convention," which sets forth its own set of "schedules," many of which are
based on CSA1970 specifications. For this reason, and because within-country jurisdictions
sometimes depart from the CSA1970 and also from the Psychotropic Drug Conventions, we no
longer can declare drug use as 'illegal' or refer to 'illegal drug use' because it might not be illegal
in some jurisdictions. (For instance, heroin is used in the practice of medicine in Switzerland;
cannabis now is used in the practice of medicine in many of the states of the US, and can be used
recreationally in a growing number of states and in many countries.) For this reason, we refer to
the drugs regulated under CSA1970 as 'internationally regulated drugs' (IRD) and this
designation segregates them from widely used drug compounds that are not internationally
regulated at this point (e.g., alcoholic beverages, nicotine and tobacco products; caffeinecontaining compounds such as coffee; betel nut) (Blatman, 1971).
Increasing numbers of heroin users in the US reached the level of an epidemic in
1970s; at that time, more than 70% users were African Americans (Boyle & Brunswick, 1980;
Clayton & Voss, 1981; DuPont & Greene, 1973; Hunt & Chambers, 1976; O'Donnell, Voss,
Clayton, Slatin, & Room, 1976). Injection was the most common method of using heroin.

10

Celebrities with drug overdose deaths involving heroin or other opioids have raised
the visibility of heroin to the public (Just, Bleckwenn, Schnakenberg, Skatulla, & Weckbecker,
2016). Some famous musicians of the era who died at the young age because of drug overdose
such as heroin or cocaine. More recently, River Phoenix, a famous American actor and singer,
died at the age of 23 in 1993 from heroin and cocaine overdose. Kurt Cobain, the lead singer of
the band Nirvana, said “heroin is the only thing that’s saving me from blowing my head off right
now. I’ve been to 10 doctors and nothing they can do about it. I’ve got to do something to stop
this pain.” Cobain committed suicide while high on heroin and died at the age of 27 in 1994.
Prince, a famous American musician, died from an opioid overdose of prescription fentanyl in
2016. A list of musicians of the era who died from heroin or other opioid drugs overdose is
shown in Appendix A.
2.2 CLASSIC STUDIES ON HEROIN EPIDEMIOLOGY
This section provides an overview of background information on heroin epidemiology
based on historical classics, including the history of heroin epidemics and how it changed over
time.
2.2.1 The Road to H (Chein et al., 1964)
The “H” stands for heroin. In the early 1950s, juvenile heroin users often were found in
deprived areas of the inner city with a large proportion of disrupted or impoverished families.
However, the inter-relationships were not clear. To investigate drugs, delinquency, and
impoverishment, Chein and colleagues studied thousands of male adolescents aged 16-20 in New
York City (i.e., Manhattan, Bronx, and Brooklyn) and conducted an extensive set of analyses

11

examining neighborhood influences, psychosocial correlates, family influences, and individual
characteristics of young heroin users (Chein, Gerard, Lee, Rosenfeld, & Wilner, 1964).
The authors noted significant neighborhood effects, influencing the probability of
becoming a heroin user or addict and found an association with deprived social and economic
conditions. In describing the process of becoming addicted to heroin, youth frequently had
delinquent attitudes, which influenced their probability of using heroin (Chein et al., 1964).
Most young heroin users got their first heroin from peers. Among heroin addicts, 90%
became habituated users within 1-2 years after first use. The epidemic areas were the places
where many residents were of low-income status or low educational attainment. The relationship
between drug use and crime was found to be complicated; the authors provided an example of
financial resource to explain this relationship among young heroin users. The high price of
heroin in the 1950s made it difficult to have a regular basis for young habituated users. In order
to maintain supplies of heroin, some of them turned to illegal markets to sell narcotics for cash or
drugs. In the areas with a high rate of drug use, the moneymaking juvenile crime was increased
proportionally by the increasing number of drug use. However, drug use is not the cause of
delinquency. The authors mentioned that, the areas of highest incidence of drug use were the
areas of high delinquency. However, an area of high delinquency was not necessary to the spread
of drug use for this young population. Some areas of high delinquency have low incidence of
drug use. In these areas, the delinquency might have come from rape, assault, automobile theft,
or disorderly conduct. The authors concluded that the total amount of delinquency is independent
of the number of drug users.
The drug use was not evenly distributed in an area. Chein and colleagues identified
epidemic areas and non-epidemic areas by the number of drug users. Based on the study areas of

12

New York City, epidemic areas of Manhattan contributed 84% of the cases in Manhattan;
epidemic areas of Bronx contributed 83% of the cases in Bronx; and epidemic areas of Brooklyn
contributed 77% of the cases in Brooklyn. On average, epidemic areas had relatively higher
percentage of low-income families, low education attainment, crowded housing, and
disadvantaged minority groups (e.g., African American, Puerto Rican). As for cultural context,
the participants were asked about their point of view on their neighborhoods. In the areas with
the highest incidence of drug use and delinquency, the high incidence of juvenile drug use was
linked to negativistic aspects and the sense of futility of the delinquent neighborhoods.
The authors then discussed how these young people became addicted to narcotics. There
were four stages in the process becoming involved with narcotics use: experimentation, casual
use, habitual use, and efforts to break the habits. Most youths who lived in high-drug-use areas
heard about heroin or saw people use it at the time they were 15 years old or so, but there was no
evidence of pressure to experiment with heroin. Youths at age of 16 or 17 years were more
susceptible to using drugs. Their first heroin use was with peers in social or casual occasions. Of
those who became habitual users, most of them had taken heroin for occasional use for one or
two years already. Once youths became habitual users, they were more likely to be involved in
criminal activities to obtain money or drugs.
2.2.2 The Epidemiology of Opiate Addiction in the United States (Ball and Chambers, 1970)
Ball & Chambers collected articles to provide an overview of opiate problems in the
United States, emphasizing opiate addicts (1970). According to reports of opiate addicts admitted
to two federal hospitals, more than 70% were using heroin (Lexington and Fort Worth hospitals).
The data collected from active files of the Federal Bureau of Narcotics in 1963 showed that 80%
of people who were addicted to opiates were younger than 40 years old, and the highest

13

prevalence of opiate addicts was among people ages 20-29. Across all ages, approximately 1621% of opiate addicts were female. More non-White and Hispanic groups were found to be
opiate addicts after World War II, mostly African Americans, Mexicans, and Puerto Ricans. In
the 1960s, 48% of opiate addicts were white, and 36% of them were African Americans. Among
opioid addicts, the peak ages of opiate use onset were 16-17 years old. More than 60% of addicts
started using opiates before age 21.
Comparing drugs used by addicts between the 1930s and 1970s, 51% of addicts used
morphine and 43% used heroin in the 1930s. However, by the 1970s, more than 70% of addicts
used heroin; only 7% were addicted to morphine. Heroin had become the top drug used by opiate
addicts since 1960s. With respect to the variety of drugs being used, only 1% of addicts used
drugs other than heroin, morphine, or opium in the 1930s.The use of other synthetic opiates
increased over the next 30 years from 1% to 21% by the 1970s. In a comparison of geographic
distribution over time, the highest rates of admission for opiate addiction treatment were from
southern states such as Louisiana and Texas in the 1930s. However, by the 1970s the patterns
had shifted, and the states with the highest admission rates were in Washington DC and northern
states such as New York and Illinois. The number of opiate addicts had remarkably increased in
New York City and Chicago over these 30 years. A shift in age and race also occurred in the
addict population. The median admission age for male addicts decreased by eight years from the
1930s to 1960s (the shift for female was not reported).
Among opiate addicts admitted to hospitals in the 1960s, two different patterns were
found. In New York, Illinois, Arizona, New Mexico, and New Jersey, along with Washington
DC and Puerto Rico, the principal cause for diagnosis of opiate addiction was the use of heroin.
More than 70% of the persons from New York, Washington DC, and Puerto Rico were African

14

Americans or Puerto Ricans; and the age at admission from these three locations was younger
than other places (the range of median age at admission was 24-30 years). Another pattern was
found in Alabama, Georgia, and Kentucky. Addicts from these states mostly used opiates other
than heroin such as morphine, paregoric, Dilaudid (hydromorphone), or codeine; less than 5% of
them had an addiction because of using heroin. Of these persons, more than 90% were white and
admitted to hospitals at older ages (i.e., older than 40 years). The length of opiate use since onset
of opiate use before admission for addicts was 9.5 years on average. Mexican-Americans had a
shorter average time (8.8 years) compared to Whites (9.4 years) and African-Americans (9.9
years). According to data from the 1960s, 85% of addicts voluntarily sought treatment. Of them,
Mexican-Americans had the lowest rate (71%) and Puerto Ricans had the highest rate (94%) of
seeking treatment voluntarily (Ball & Chambers, 1970).
2.2.3 The heroin epidemics: A study of heroin use in the US (Hunt and Chambers, 1976)
Based on Ball and Chambers (1970), the use of heroin increased remarkably in the US
between the 1930s and the 1970s, and it became a serious problem for public health and caused
both social and political issues. Drug use, including heroin use, spread. It could have spread from
one population to another and/or from one place to another place. Three sections described the
US heroin epidemic. The first section described the spread of heroin use; the second one depicted
the characteristics of heroin users; and the third section contained comments on treatment and
prevention policy for the use of heroin.
Two models were presented that explained how heroin use spreads: Micro-diffusion and
Macro-diffusion. Micro-diffusion, also called Branching process shows how heroin use is spread
at a person-to-person level. It suggests that the initial heroin users entered a community and
spread the use of heroin to their susceptible friends or peers, and their susceptible friends spread

15

heroin use to secondary friends or peers, and so on. Based on this Micro-diffusion model, Hunt
& Chambers observed that each heroin user spread heroin use to peers only in his/her first year
of use, and a community can be completely exposed to heroin in about seven years after the
introduction of the first heroin user. This Micro-diffusion model also explained the epidemic of
heroin use in Chicago (Hughes, Barker, Crawford, & Jaffe, 1972). Based on this model, the
individual of each generation produces a random number of offspring (i.e., newly heroin users).
These newly incident heroin users become the successors of the next generation. Figure 2.2.1
shows the process of branching, produced by the probability of distribution, which describes the
number of successors of each individual. There are some limitations of this Micro-diffusion
model. For example, all successors within the susceptible group are subsumed by the successor
in the group. This model assumes that branching process is independent from one group to
another group. Therefore, this model might not be a good mathematical model for studying the
structure of heroin spreading due to the dependency between groups.
The second model, Macro-diffusion, depicts the use of heroin spread from urban areas
with high population densities to suburban and rural areas. For example, the use of heroin began
to develop in the 1930s, a time when the majority of illegal heroin was smuggled into the US
from other countries (e.g., China). After World War II, the widespread use of heroin was found
in large cities (e.g., NYC & Chicago). During the 1960s, most cities and their suburban
surroundings were exposed to heroin, and heroin users were increasingly found in these areas.
Results showed that new cases of heroin use increased more than ten times during a five-year
period (1963-1968). Despite the difficulties of identifying all cases precisely, the rapid increase
of heroin users could be described as reaching an epidemic level. In the 1970s, two resources
were used for estimating heroin incidence: 1) police arrest data and 2) treatment program data.

16

However, these two resources were neither comprehensive nor representative of a total
population.
Figure 2.2.1 An example of Macro-epidemic process (Hughes P. and Crawford G., Illinois Drug
Abuse Program, unpublished, November, 1972)

In the first section of this book, Hunt & Chambers portrayed how fast and how far heroin
use spreads (1976). Incidence of heroin use was a measure of spread. The prevalence of active
heroin users in a population was surprisingly not well known in 1970s. In their second section,
data from different sources produced varying values of heroin use prevalence, and characteristics
of heroin users. As aforementioned, treatment programs provided a data source on heroin users,
but it includes mostly heavy heroin users. Different characteristics were shown between different
treatment programs. For example, methadone treatment programs served principally older, longaddicted, persistently criminal, and non-White heroin users as compared to other treatment
programs. A majority of the heroin users arrested for drug law violations or those involved in

17

other crimes were not the same persons who entered treatment programs. One study showed that
only 18% of drug users enrolled in treatment programs had criminal records. Unfortunately, data
from a survey of the general population under-estimated the prevalence of heroin use. Another
study in New York City showed that a survey of the general population may underestimate the
prevalence of heroin use by 50%-400% as compared to other data sources. On the other hand,
heroin users in psychiatric hospitals were older and had different patterns of incidence of first
use than users from other groups, and of course, all of them had problems of drug addiction and
mental illness.
In summary, the challenge in estimating the prevalence of heroin users was counting
hidden heroin users regardless of the data source. Hunt projected that 12% of addicts would enter
treatment within 1 year, 22% within 2 years, 26% within 2-3 years, 26% within 3-4 years, 6%
within 4-5 years, 3% within 5-6 years and 3% within 6-7 years. More than 90% of addicts
entered treatment programs within 5-6 years since they first used heroin. Methods from this book
(Hunt & Chambers, 1976) will be used in this dissertation for estimating the incidence of heroin
from the treatment dataset (TEDS-A). More details will be provided in Chapter 4.
2.3 HEROIN AND OPIOIDS EPIDEMIOLOGY IN THE 21ST CENTURY
2.3.1 The Definition of Abnormal Use of Opioids
Opioids are prescribed for a range of legitimate purposes within medicine. However,
there are three types of what might be called abnormal drug use: 1) illegal use, 2) extra-medical
(EM) use, and 3) non-medical use, and these categories are not mutually exclusive. In the US,
heroin use is illegal. While legal, use of prescription opioids is different and more complicated
than the illegal use of heroin. Concerns about the inappropriate use or ‘misuse’ of prescription
opioids have been raised in recent years. The definition of what contributes inappropriate use or

18

prescription misuse has included the following: 1) non-prescribed use, 2) use for recreational
purposes, or 3) meeting criteria of the Diagnostic and Statistical Manual (DSM) for opioid
dependence (Barrett, Meisner, & Stewart, 2008). In this context, the two types of legal misuse of
prescription are included: non-medical use and extra-medical use. Both of them are defined as
using prescription drug without a physician’s approval. Non-medical prescription opioids use can
be described as the act of getting high, however uncertainty about “boundary users.” EM
prescription opioid use involves use to get high, getting it from someone other than a physician,
or using outside the bounds of the prescribed purpose (Anthony, Warner, & Kessler, 1994).
2.3.2 Prevalence of Prescription Opioids and Heroin
Global prevalence of opioid use is about 7 per 1000 people aged 15-64 years, and
includes EM prescription opioids and heroin, and roughly half of opioid users are heroin users.
In other words, globally, about 31 million people aged 15-64 years had used opioids including
heroin and prescription pain relievers. Of that population, approximately 13~21 million people
used heroin. Oceania (especially Australia and New Zealand) had the highest prevalence of
opioid use (3%), followed by the Americas (2.1%), Europe (0.7%), and Asia (0.4%) and Africa
(0.4%). North America has the highest prevalence of opioid use (4%); prevalence of EM
prescription opioid use in the United States was 4.3%. Overall, opioids now rank as the world's
second most widely used drug after cannabis (3.8%) (United Nations Office on Drugs and Crime
2012).
In the US (2011), an estimated 4.2 million Americans aged 12 or older (1.6 %) had used
heroin at least once in their lives, according to a 2012 National Institute on Drug Abuse (NIDA)
report (United States. Substance Abuse and Mental Health Services Administration, 2012).
About 4-5% of individuals aged 12 years or older used EM prescription opioids in the past year,

19

and 0.3% used heroin in the past year. Approximately 3.6% of recent extra-medical prescription
pain reliever (EMPPR) users had initiated heroin use within 5 years following the first EMPPR
use. An estimated 23% of heroin users become dependent (Anthony et al., 1994; Muhuri,
Gfroerer, & Davies, 2013).
Heroin use and non-medical opioid use is a significant public health problem in the US.
The findings by Grau and colleagues stated that poly-opioid use contributes to quicker transition
to heroin and other injection drugs use as compared to single opioid use (Grau, Dasgupta et al.
2007). A 2012 annual report from a national survey presented a finding that approximately 3.6%
of recent non-medical opioid users had initiated heroin use within 5 years following first EMPPR
use (Muhuri et al., 2013). OxyContinÂŽ is one of the most commonly used opioid prescription
drugs. In the past, people who used OxyContinÂŽ as a means of getting high by crushing the pill
so that they could snort the drug. The US Food and Drug Administration (FDA) approved the reformulated version of OxyContinÂŽ in 2010 to prevent users from misusing the drug by snorting
or injecting it. Consequently, heroin sometimes became a substitute for OxyContinÂŽ. In a
qualitative study, an interviewee said: “Most people that I know don't use OxyContin® to get
high anymore. They have moved on to heroin [because] it is easier to use, much cheaper, and
easily available” (Cicero, Ellis, & Surratt, 2012).
2.3.3 Burden of Disease Attributable to Opioid Use
Approximately 70,000-100,000 people die from opioid overdose each year worldwide,
many with drug dependence (UNODC, 2012). In 2010, 20 million DALYs, roughly 0.8% of
global all-cause DALYs, have been attributed to drug dependence. Furthermore, among
‘scheduled’ drugs, it is heroin or other opioid use that often has been followed by appearance of
a drug dependence syndrome (e.g., with withdrawal symptoms). This means opioids account for

20

much of the drug-related DALYs burden, estimated as 9.2 million DALYs in 2010 (Degenhardt
et al., 2013).
In the US, the heroin overdose death rate had increased by about 100% over 4 years, from
1 per 100,000 in 2008 to 2.1 per 100,000 in 2012 (Rudd et al., 2014). Data from the National
Vital Statistics System showed that the heroin-related overdose deaths nearly tripled over the last
decade, from 0.7 per 100,000 in 2000 to 2.7 per 100,000 in 2013. The greatest increase was in
the Midwest region (National Vital Statistics System 2014). The growth of overdose deaths due
to prescription opioid pain relievers increased faster than overdose deaths of heroin, from 1.5 per
100,000 in 2000 to 5.0 per 100,000 in 2013 (Hedegaard, Chen, & Warner, 2015).
2.3.4 Relationship between prescription opioid (e.g., OxyContinÂŽ) and heroin
Both theory and recent empirical evidence from the US support the idea that newly
incident heroin use may have been triggered when younger population started to use the original
formulated OxyContinÂŽ products extra-medically. The origins of theory and conceptual
foundations for asking research questions about the suspected EM OxyContinÂŽ-heroin triggering
hypothesis date back to early general 'steppingstone' and 'gateway' ideas about how use of one
drug early in a chain of temporally sequenced events might account for later drug use (Anthony,
2002, 2012; Morral, McCaffrey, & Paddock, 2002). Evidence pertinent to the EM OxyContinÂŽheroin triggering hypothesis has come largely from clinic-based studies of persons in treatment
programs, as well as a mix of field studies with local area recruitment approaches (Carise et al.,
2007; Cerda, Santaella, Marshall, Kim, & Martins, 2015; Cicero et al., 2012; Jones, 2013; Mars,
Bourgois, Karandinos, Montero, & Ciccarone, 2014; Young & Havens, 2012). Heroin possibly
becomes an alternative substitute for OxyContinÂŽ because it often has been cheaper and more
readily available. In an illustration of evidence from the qualitative research tradition, Cicero and

21

colleagues provide the following quotation from an EM user, “Most people that I know don't use
OxyContinÂŽ to get high anymore. They have moved on to heroin [because] it is easier to use,
much cleaner, and easily available” (Cicero et al., 2012).
Some clinicians and researchers in this field are skeptical about the suspected EM
OxyContinÂŽ-heroin triggering hypothesis, noting a basic principle that when Y follows X, it
does not mean that Y is caused by X. Indeed, ten years ago, drawing upon this basic principle,
estimates published challenged the causal significance of the more general 'gateway' sequence,
and advanced an alternative hypothesis that otherwise uncontrolled individual-level underlying
predispositions might account for a temporal sequence of passage from EM use of one drug to
EM use of another drug. This idea about individual-level underlying susceptibility traits has an
analogy in pharmacoepidemiological research in the form of 'confounding by indication', such
that it might not be the exposure to a medicine that triggers some unwanted later effect. Rather,
the apparent causal linkage from earlier medicine use to a later adversity sometimes might be
wholly explained by the underlying medical condition that prompted a clinician to prescribe the
medicine in the first place (i.e., the 'indication'). To illustrate, for reasons such as these, one
might be hesitant to blame an antidepressant medicine for subsequent suicide outcomes because
it truly might be the underlying depression that has prompted both the exposure to the
antidepressant medicine and also the later suicide attempt or successful suicide. With respect to
EM use of OxyContinÂŽ products, it might not be the EM OxyContinÂŽ use per se that has been
the trigger for new onsets of heroin use. Rather, there might some other shared source or sources
of variation accounting for both of these linked behaviors, which might be found anywhere
across the range from micro-level domains (e.g., genetic influences) toward macro-level domains
(e.g., drug policies).

22

2.4 RESEARCH GAPS AND SIGNIFICANCE
In the world, and in the US specifically, heroin and extra-medical prescription opioids
use has become a significant public health problem. The number of overdose deaths due to
heroin or other opioid compounds increased 4~5 times in the past two decades (Cicero, Ellis,
Surratt, & Kurtz, 2014; Muhuri et al., 2013). The epidemiology of heroin use and EMPPR use
now intersect considerably (Cicero, Ellis, & Harney, 2015; Compton, Jones, & Baldwin, 2016;
Jones, 2013; Mars et al., 2014; Young & Havens, 2012). However, there are gaps in
epidemiological research, with several important unanswered research questions. For example,
concerns about completeness of epidemiological estimates for incidence of heroin use exist. For
instance, if heroin users don't participate in field surveys, incidence rates based on treatment
admissions might be as good or better estimates. The idea that OxyContinÂŽ or other prescription
opioids is triggering onset of heroin use has not been examined completely. The current
estimates are subject to confounding and other biases that might create spurious associations.
Genetic susceptibility traits have not been held constant. Furthermore, more than 80% of heroin
users had previous EM prescription opioid use (Muhuri et al., 2013). However, research about
environmental factors and individual characteristics that influence probability of transitioning
from extra-medical prescription opioid use to heroin use are still limited and unclear.

23

CHAPTER 3
MATERIALS
3.1 NATIONAL SURVEY ON DRUG USE AND HEALTH (NSDUH)
This dissertation focuses on two U.S. study populations: 1) the general population from
an annual nationally representative survey, and 2) persons who entered treatment programs from
a national data system of annual admissions to substance abuse treatment facilities. The NSDUH
study population was designated to include civilian, non-institutionalized US residents age 12
and older, including residents of households as well as group quarters and homeless shelters. For
NSDUH, the national sampling frame and approach is based on lists of residential dwelling units
within primary sampling units in all geographic areas annually. Its nationally representative
samples are drawn using multi-stage area probability sampling for each of the 50 states in the US
and the District of Columbia. The surveys exclude relatively small segments of the current US
non-institutionalized population, such as homeless persons who do not reside in shelters, military
personnel on active duty; residents of institutional group quarters such as prisons and psychiatric
hospitals also are excluded. With the exceptions noted, all community-dwelling civilian residents
are eligible to be sampled and to participate as survey respondents. NSDUH seeks large
nationally representative community samples of non-institutionalized civilians age 12+ years,
generally achieving participation levels at 65%-70%. Repeated annually from 2004 through
2013, the NSDUH research approach involves multi-stage area probability sampling and
recruitment according to protocols approved by a cognizant Institutional Review Board. Each
year, the NSDUH yields an independent replication in the form of a nationally representative
sample of the study population. The “Survey Documentation and Analysis” (SDA) for NSDUH

24

is a public-use dataset and is available for download on the website
(http://www.icpsr.umich.edu/icpsrweb/SAMHDA/sda).
3.2 TREATMENT EPISODE DATA SET - ADMISSIONS (TEDS-A)
In epidemiological concepts, it is seldom possible to study incidence of heroin use
directly because many cases are unreported. Besides, the relationship between reported cases and
total cases may be unclear. Therefore, the inference of total cases is difficult to make. Nationally
representative survey data (e.g., NSDUH) helps in understanding the number of heroin cases in
the community by multi-stage sampling approach. However, this number might be underreported
from self-reported survey due to social undesirability or stigmatization. Therefore, this
dissertation includes an alternative national surveillance dataset collecting cases (i.e., TEDS).
Using treatment program dataset to estimate the incidence of heroin use, the lag between heroin
onset and the time of entering treatment program must be considered. For heroin users in
treatment programs, there are very few users entering treatment program in their first year of use
(~12%). Earlier in the U.S., it seemed that more than 90% of users entering treatment program
within 5-6 years since they first used heroin (Hunt and Chambers 1976). However, a recent study
showed that the lag time now is much longer, and that women have shorter length of time from
first use of heroin to treatment admission than men, approximately 17 years for men and 14 years
for women (Substance Abuse and Mental Health Services Administration 2011). On the other
hand, a large number of heroin users do not become dependent (Anthony, Warner et al. 1994).
TEDS is a national census data system in the US focused on persons aged 12 or older
who admitted to or discharged from substance abuse treatment facilities. TEDS-A is a national
census system of collecting annual admissions reported by public or private drug use treatment
facilities. Unfortunately, not all treatment facilities report their admission or discharge cases to

25

TEDS system. Only some treatment programs are required to report all of their admissions to
their state based on the types of funds they receive (i.e., the facilities that received public funds
must report). TEDS-A was collected since 1992, year by year. The most recently available
public-use TEDS-A datasets can be downloaded through a SAMHSA website
(http://www.icpsr.umich.edu/icpsrweb/SAMHDA/sda).

26

CHAPTER 4
MANUSCRIPT 1 – EMPIRICAL EVIDENCE ON THE ESTIMATION OF HEROIN
INCIDENCE IN THE UNITED STATES: A BACK-CALCULATION APPROACH

4.1 ABSTRACT
Heroin incidence trends in the United States study population are studied using the
TEDS-A from 1992 to 2012. Hunt’s Lag Correction Method is used to estimate the number of
incident heroin cases by adjusting reported numbers of heroin users visiting drug treatment
programs for the time lag between first use of heroin and first treatment admission (lag
distribution). Initial estimates from the Hunt model 1992–2012 TEDS-A analyses show more
than 11 million admissions age 12-to-54-years-old with heroin as primary drug, and clear
evidence of two heroin epidemic peaks, in 1969 (5.2 per 10,000) and in 2012 (4.2 per 10,000).
Moreover, for the incidence rate of opioids other than heroin, the first peak is observed in 1970
(1.5 per 10,000) and the second peak was in 2010, about 3 times higher than the first peak (3.8
per 10,000). However, the result from treatment data need to be interpreted cautiously, especially
in relation to the wider context of underlying trends in the population. Potential biases might
derive from underreporting heroin users and from the changes in the proportion of users in
treatment programs. This study discloses temporal patterns and relative heroin incidence rates.
Whereas the results may not serve as absolute numbers of newly incident cases, they suggest
accelerated heroin incidence rates in recent epidemic years. The evidence should improve our
understanding of heroin epidemiology, complementing field survey estimates.

27

4.2 INTRODUCTION
In this study, the main aim is to estimate the annual incidence of heroin use in the US
from World War II until 21st century. The first heroin epidemic in the US occurred after World
War II. The number of newly incident heroin cases rapidly increased between 1950s and 1970s
(DuPont & Greene, 1973; Greene, 1974; Hughes et al., 1972; Hunt & Chambers, 1976). About
30% of the veterans had experience of heroin use while serving in Asian conflicts (e.g., Vietnam
War, 1954-1975) and some continued heroin use after return to the U.S. About 20% of Vietnam
veterans became dependent (Robins, Davis, & Goodwin, 1974). During the 21st century, a
resurgent heroin epidemic in the U.S. has been discussed in the literature (Jones, Logan,
Gladden, & Bohm, 2015; Lipari & Hughes, 2015). Heroin overdose deaths have increased five
times since 2000 (0.7 to 3.5 per 1000). Heroin users in the 21st century were more likely to start
EM prescription opioids use before their first use of heroin (Cicero et al., 2015; Compton et al.,
2016; Jones, 2013; Young & Havens, 2012).
During the interval from 1950 through the 2010s, we have no systematic study of heroin
incidence estimates (i.e., from World War II until recently), and no way to compare the two US
heroin epidemics using the same data source and method. This study fills that gap, and provides
national-level estimates of heroin incidence from the 1950s to 2010s.
Identifying the new cases of heroin users from community surveys is difficult because the
prevalence of heroin use is relatively low: an estimated 0.1% of people aged 12 or older are
current heroin users (United States. Substance Abuse and Mental Health Services
Administration, 2014a). Voluntary reporting of illegal drug use may not be as one might hope. In
contrast to small number from community surveys, more heroin users can be obtained from
treatment program dataset. For example, from the dataset of Treatment Episode Data Set -

28

Admissions (TEDS-A), heroin as a primary drug accounted for about 13~16% of all treatment
admissions (United States. Substance Abuse and Mental Health Services Administration, 2014b).
TEDS-A data provide estimates required for Hunt’s Lag Correction Method (LCM) in order to
estimate heroin incidence, including age of first heroin use. The resulting ‘Lag length’ is defined
as the delayed years between heroin onset and subsequent entry into treatment (Hunt &
Chambers, 1976). The retrospective information about lag-time, provided in the treatment data
(TEDS-A 1992-2012), allows heroin incidence estimation, 1950 through 2012.
In this research, the most important topic is estimating the annual number of newly
incident heroin users and seeing ebb and flow of this public health problem in the U.S. This
study provides empirical evidence on the trend of heroin incidence over the span of 60 years in
the US. One clear strength and advantage of using treatment data relative to alternative
observational national surveys, is that more heroin users can be found in the treatment programs
than community or household surveys. The result is more stable estimates for heroin incidence.
Another strength of this study's research approach is that it is not restricted to samples of persons
in any specific location. Instead, this research takes into account all heroin users irrespective of
where they are located in the U.S. as long as their first treatment admission occurred between
1992 and 2012.
However, this study cannot lay claim to provide nationally representative estimates of
heroin incidence for the general US population. This study only seeks to estimate the patterns of
“relative heroin incidence” over the past 60 years, from one epidemic to the next, occurring more
than 100 years after Bayer marketed heroin in 1895, as shown in Table 4.2.1.

29

Table 4.2.1 A brief summary of the history of opium
Year
1803

1827

Events
Morphine is extracted from opium resin and used for medical
treatment of pain and chronic diseases.
Merck pharmaceutical company (Germany) begins the manufacture
of morphine.
Heroin was introduced by the Bayer Company in Germany. It was

1895

developed chiefly as a morphine substitute for cough suppressants
and Bayer Company claimed that it did not have morphine's sideeffects of dependence.
Harrison Narcotics Tax Act in the USA: To manage the distribution

1914

and sale of heroin, along with other opioids, permitted heroin to be
prescribed as well as sold for medical reasons.

Post World War II

Widespread use in metropolitan such as Chicago or New York City.

1950s – Early 1960s

New heroin users found in other big cities such as Detroit.

1960s

New heroin users peak in the late 1960s.

1970s

1980s-1990s

The speed of spread over geographically rises with increased size of
the cities (Hunt & Chambers, 1976).
Public health concern about emrging drug use including heroin
(Courtwright, 1983; Hughes & Rieche, 1995).
EM use of prescription opioid has emerged as one of main public

2000s

health problems for the United States (Bohnert et al., 2011; Compton
& Volkow, 2006; Paulozzi et al., 2006).

30

4.3 METHODS
4.3.1 Data source
TEDS is a national census data system in the US focused on persons aged 12 or older
who admitted to or discharged from drug treatment facilities. As described in Chapter 3, TEDSA covers a selection of both public and private drug use treatment facilities. Unfortunately, only
treatment programs with certain types of funding report their admission and discharge cases to
TEDS. Facilities that received public funds must report, but the privately funded treatment
facilities report voluntarily. These facilities include early intervention, outpatient treatment,
intensive outpatient, hospitalization, detoxification, general hospital, psychiatric hospital, mental
health program, and medication assisted therapies. TEDS-A data has been collected since 1992,
with results through 2012. The TEDS-A public use datasets are available for download at the
following ICPSR URL: http://www.icpsr.umich.edu/icpsrweb/ICPSR/series/56, last accessed on
4th November 2016.
4.3.2 Sample selection
A total number of 37,749,456 episodes of drug admission between 1992 and 2012.
Among total episodes, approximately14 million episodes were first admission episodes and the
first admission episode from each individual was treated as individual record in this study. For
those who were first admitted to the treatment programs, information on the primary drugs for
first admission, age of first use of primary drug, age at first admission, were collected from
treatment facilities. The flowchart shows the selection of cases of heroin and other opioids with
implementing exclusion rules (Figure 4.3.1). A total of 889,031 individuals were identified as
heroin users (primary drug of admission: heroin). In the same manner, 484,407 individuals were

31

identified as users of opioids other than heroin such as non-prescription methadone and other
semi-synthetics or synthetics (e.g., codeine, oxycodone, tramadol).

Figure 4.3.1 Flowchart of United States, Treatment Episode Data Set –Admissions (TEDS-A),
1992-2012.

4.3.3 Organizing the case material
Age of first admission and age of first use of heroin were categorized as 12-14, 15-17,
18-20, 21-24, 25-29, 30-34, 35-39, 40-45, 46-49, and 50-54. The tabulated table of these two
variables is shown as the Figure 5.3.2. Based on the data structure, the number of individuals in
these 55 cells will be calculated separately. Moreover, the data structure of TEDS-A can be used
extensively to have one-year period for each cell. The assumption of Uniform Distribution has
been made here to break down each cell into multiple one-year period by one-year period cells.
That is, the number of individuals in each cell is equally distributed. For example, the
assumption is made that the probability of 13-years-old youths having their first heroin use at age

32

12 is equal to the probability of their first heroin use at age 13. The equation is shown below.
One example is presented in the Figure 4.3.3.

Notation and formula
To calculate the number of cases by age of onset and age of first admission

The assumption of Uniform Distribution made here to break down each cell into multiple
one-year period by one-year period cells might not be satisfied and it can be improved through
mathematical simulation in the future. However, it is challenging to apply any specific
probability distribution on this issue because the mean age of heroin onset was different from
time to time or from population to population (Table 4.3.1) (Anglin, Brecht, Woodward, &
Bonett, 1986; Bailey, Hser, Hsieh, & Anglin, 1994; Boeri, Sterk, & Elifson, 2008; Carpenter,
Chutuape, & Stitzer, 1998; Concool, Smith, & Stimmel, 1979; Hser, Huang, Chou, & Anglin,
2007; Inciardi & Harrison, 1998; Kandel & Yamaguchi, 2002; Kelly, Cornelius, & Clark, 2004;
Khantzian & Treece, 1985; Lipari & Hughes, 2015; Lofwall, Brooner, Bigelow, Kindbom, &
Strain, 2005; McGlothlin & Anglin, 1981; Neaigus et al., 2006; Neaigus et al., 2001; Nurco,
Cisin, & Balter, 1981; O'Driscoll et al., 2001; Ochoa, Hahn, Seal, & Moss, 2001; Pugatch et al.,
2001; Rounsaville, Kosten, & Kleber, 1985; Rutherford, Cacciola, & Alterman, 1994; Seal et al.,
2001; Vaillant, 1973). Cicero and colleague reported that the age of heroin onset increased about

33

6 years across 40 years, from 16.5 year in 1960s to 22.9 years in 2000s (Cicero et al., 2014).
Advantages of using the uniform distribution includes minimizing the misleading weights in the
analysis with the feature of uniformly distributed random elements in each cell.
Table 4.3.1 Literature review of age of first heroin use
Study

Year of publication

Mean age of first heroin use

Vaillant

1973

23.0

Cooncool et al.

1979

18.2

McGlothlin & Anglin

1981

19.6

Nurco et al.

1981

18.2

Khantzian & Treece

1985

20.0

Rounsaville et al.

1985

20.3

Anglin et al.

1986

18.5

Rutherford et al.

1994

23.2

Bailey et al.

1994

18.1

Inciardi & Harrison

1998

21.0

Capenter et al.

1998

21.1

Neaigus et al.

2001

22.4

O’Driscoll et al.

2001

20.7

Ochoa et al.

2001

18.0

Pugatch et al.

2001

18.2

Seal et al.

2001

20.0

Kandel & Yamaguchi

2002

20.1

Kelly et al.

2004

23.9

Surratt et al.

2004

23.4

Lofwall et al.

2005

19.3

Neaigus et al.

2006

22.3

Hser et al.

2007

17.9

Boeri et al.

2008

19.0

Lipari & Hughes

2015

24.5

34

Figure 4.3.2 Cross-table of number of individuals per cell by the age of first admission and age
of first use (i.e., age of onset)

Figure 4.3.3 Uniform Distribution is applied to break down cells into multiple one-year-period
cells.

35

4.3.4 Lag Correction Method
After determining twenty-one datasets separately (TEDS-A 1992 to TEDS-A 2012) with
their numbers in one-year-period cell shown in Figure 4.3.2, Hunt’s method has been applied to
estimate the incident heroin cases for each calendar year. The Hunt’s LCM, one type of back
calculation method, is used to calculate backward. This method provides an estimate of the
unknown number of heroin users who were in the treatment program but the data has been
truncated (left truncation). That is, Hunt’s LCM assumes that heroin users are not always in the
treatment data. A lag time distribution is used to work out the number of newly incident heroin
users for each calendar year. Hunt's recommendation for estimated ‘Lag time’ is based on an
onset cohort that takes all heroin users who initiated heroin use in a given year and classifies
them into subsets by when they entered treatment (i.e., within one year, one to two years, etc.).
Hunt’s model is based on four assumptions: 1) the probability of admission is the same for
persons of specific duration of dependence regardless of year of onset; 2) there is no change on
clinical diagnosis of drug dependence; 3) the rate of mortality is constant; 4) and there is a fixed
amount of treatment facilities for admitting new persons (Hunt and Chambers 1976). Based on
the data TEDS-A 1992-2012, I have been able to carry out the ‘Lag time’ from 0 to 42 years,
specifying year of onset from 1950 to 2012 with left truncation (see Figure 4.3.4).

36

Figure 4.3.4 Conceptual framework of truncations with lag time, Data from United States,
Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012.

Therefore, the lag distribution is estimated from observed admissions for individuals
admitted to treatment programs during 1992-2012. The shifts of the lag distributions over
calendar time are shown in the Figure 4.3.5.
When admission cohort lag curves are stable for each calendar year, the percentages of
each year’s admission can be averaged to yield a mean cumulative entry curve, yielding a more
robust estimate than any single year provides. The patterns of lag distributions of each admission
years are illustrated for heroin and for other opioids in Figure 4.3.6 & Figure 4.3.7, respectively.
Lag distribution curves move toward the left from 1992 to 2012, which means that the lag time
37

from the first use to first admission was getting shorter. For example, the median lag time for
heroin is 10 years in 1992 and shifted to shorter lag time of 5 years in 2012. Therefore, the
averaged probability of entering treatment by lag time was applied to correct the number of
newly incident heroin users. ‘Lag’ effect can be calculated from the cumulative entry rate by
using the number of observed heroin users entering treatment program at onset year. For
example, if twenty percent of heroin users were admitted to treatment program in the first year of
use, it is assumed that the other eighty percent users of the same onset cohort did not enter
program but eventually will. In this manner, an inflation factor can be calculated as 5.
Accordingly, the number of newly incident heroin users is calculated from year 1950 to 2012.

Notation and formula
To calculate the number of incidence users for each calendar year based on Hunt’s LCM

38

Figure 4.3.5 A distribution of lag-years between heroin onset and treatment admission. Data
from United States, Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012.

39

Figure 4.3.6 Heroin lag distribution. Data from United States, Treatment Episode Data Set –
Admissions (TEDS-A), 1992-2012.

40

Figure 4.3.7 Other opioids lag distribution. Data from United States, Treatment Episode Data
Set –Admissions (TEDS-A), 1992-2012.

41

4.4 RESULTS
This study's estimates are from 11,327,152 persons aged 12-54 years -- i.e., those
admitted between 1992 and 2012, with an opioid as their primary drug problem as reported by
public or private treatment facilities where they were admitted. Among identified persons,
889,031 people (8% of treatment program admissions) had heroin as the primary drug problem
prompting treatment. These are the observations used to apply Hunt’s back- calculation approach
in order to estimate newly incident heroin cases by estimating numbers of being truncated, year
by year, based on those actually admitted for treatment, 1950-2012 (Table 4.4.1).

Table 4.4.1 Number of people admitted to the treatment program by the type of primary drug use
problem for their first admission. Data from United States, Treatment Episode Data Set –
Admissions (TEDS-A), 1992-2012.
Year of
Admission
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
Total

Heroin

Other opiates

Other drugs

Total

35,692
38,703
38,311
43,357
41,214
40,147
40,528
42,355
35,084
37,234
39,329
37,276
36,506
43,079
45,940
48,244
51,470
54,576
44,882
47,263
47,841
889,031

3,399
3,689
3,771
3,884
4,035
4,106
5,082
5,988
7,814
10,701
13,531
15,474
17,378
23,312
28,951
36,133
45,308
53,644
59,552
74,661
63,994
484,407

447,924
439,641
459,142
471,724
464,457
433,425
427,603
432,148
417,633
425,641
455,168
436,764
422,797
473,250
532,511
579,658
597,757
572,065
512,286
505,104
447,016
9,953,714

487,015
482,033
501,224
518,965
509,706
477,678
473,213
480,491
460,531
473,576
508,028
489,514
476,681
539,641
607,402
664,035
694,535
680,285
616,720
627,028
558,851
11,327,152

42

Figure 4.4.1 The proportion of first treatment admissions of heroin and other opioids changes
over time. Data from United States, Treatment Episode Data Set –Admissions (TEDS-A), 19922012.

In 1992, there were 35,692 persons (7.3% of the total admissions in 1992) admitted to
treatment programs with heroin as a primary drug problem; this number increased to 47,841 in
2012. Among them, those with 'primary heroin' increased from 7.3% in 1992 to 8.6% in 2012.
Additionally, the number of persons admitted with 'primarily some other opioid' increased from
3,399 in 1992 to 63,994 in 2012. As a proportion of admissions, the number shifted remarkably
from 0.7% of total admissions in 1992 to 11.5% in 2012. People admitted to treatment programs
because of their other opioids use increased approximately 16 times over two decades (Table
4.4.1 & Figure 4.4.1).

43

Table 4.4.2 presents an overview of selected socio-demographic characteristics of 12- to54-years-old patients by types of drugs (i.e., heroin, other opioids, and other drugs). As shown,
the majority of people first admitted to treatment programs are between their early 20s and 40s
(i.e., age of 21-40 years). Heroin and other opioid users are older than other drug users. The sex
ratio of males to females is 2:1 for heroin users and nearly 1:1 for other opioids. According to
region: more than one-third of heroin users are from the treatment facilities in the Northeast
(35%), followed by West (25%), South (22%), and Midwest (18%).
Table 4.4.2 Persons age 12-54 years in the treatment programs by the type of drugs on their
primary admission, Data from United States, Treatment Episode Data Set –Admissions (TEDSA), 1992-2012 (n=11,327,152)
Heroin
n
889,031

%
100

Other Opioids
n
%
484,407
100

Other Drugs
n
%
9,953,714
100

Age
12-14
15-17
18-20
21-24
25-29
30-34
35-39
40-44
45-49
49-54

494
9,469
57,424
122,444
153,724
145,854
140,117
123,306
86,809
49,390

0.06
1.07
6.46
13.77
17.29
16.41
15.76
13.87
9.76
5.56

1,242
9,640
38,263
91,981
113,011
79,980
55,526
42,746
32,264
19,754

0.26
1.99
7.9
18.99
23.33
16.51
11.46
8.82
6.66
4.08

216,606
1,121,941
919,419
1,308,734
1,515,933
1,413,297
1,265,207
1,028,402
731,288
432,887

2.18
11.27
9.24
13.15
15.23
14.2
12.71
10.33
7.35
4.35

Male
Female

581,249
307,477

65.38
34.59

247,562
236,718

51.11
48.87

6,898,352
3,050,314

69.3
30.64

Northeast
312,149
35.11
Midwest
158,757
17.86
South
194,886
21.92
West
220,713
24.83
Observations were omitted due to missing values

99,781
86,289
224,489
73,834

20.6
17.81
46.34
15.24

1,905,242
2,730,888
2,824,013
2,490,632

19.14
27.44
28.37
25.02

Sex
Region

Based on the TEDS-A 1992-2012, the number of observed newly incident opioids cases
increased from 1950 to 2012. Figure 4.4.2 shows observed and corrected number of newly
incident heroin cases each year. Differences between observed and corrected numbers are from
correction by lag distribution for the truncated information (see Figure 4.3.4 and Figure 4.3.6).

44

After factoring in population size for each calendar year based on US census (see Figure 4.4.3),
the peak of the first heroin epidemic was observed in 1969 (5.2 per 10,000). The second heroin
epidemic was first observed in 2000 (2.5 per 10,000), which increased to 4.2 per 10,000 in 2012,
but apparently has not yet reached the heroin incidence peak value seen in 1969 (Figure 4.4.4).
As for opioids compounds other than heroin, the observed and corrected number of newly
incident cases are shown in the Figure 4.4.5. The incident opioid cases peak in the 1970s
(~17,000 incident cases). In the 2000s, the prescription opioids become one of the most popular
drugs being used extra-medically such as “to get high.” Figure 4.4.5 shows that the incident
cases increased from 7,200 in 1990 to 63,500 cases in 2010, dramatically increase in the past two
decades. After factoring population size from U.S. census by calendar years, the first peak of
incidence rate was observed in 1969 (1.5 per 10,000). After 1969, the incidence rate was
gradually decreased until 1990 (0.5 per 10,000). There was a steep rise, starting in 1990 and
increasing nearly eight times to 3.8 per 10,000 in 2010 (Figure 4.4.6). The trends of incidence

rate over time differ between heroin and other opioids. Heroin has two clear epidemic peaks
in1969 and 2010, and these two peaks were nearly reach the same incidence rates (5.2 versus
4.8 per 10,000). Incidence rates for other opioids compounds increased markedly from the
late 1990s until 2012. Taking all opioids as a whole (see Figure 4.4.7), the peak incidence is
seen in the late 1960s, with slow increases to a large peak in 2010. The estimated opioids
incidence rate in 2012 is larger than the peak rate seen in 1969 (6.5 per 10,000 in 1969 versus
7.8 per 10,000 in 2012).

45

Figure 4.4.2 The observed and estimated number of newly incident heroin users from 1950 to
2012. Data from United States, Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012.

46

Figure 4.4.3 US population age 12-54 years by year, US census 1950-2010.

47

Figure 4.4.4 Heroin incidence rate (per 10,000). Data from United States, Treatment Episode
Data Set –Admissions (TEDS-A), 1992-2012.

48

Figure 4.4.5 The observed and estimated number of newly incident users of other opioids from
1950 to 2012. Data from United States, Treatment Episode Data Set –Admissions (TEDS-A),
1992-2012.

49

Figure 4.4.6 Incidence rate of opioids other than heroin (per 10,000). Data from United States,
Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012.

50

Figure 4.4.7 Incidence rate of all opioid compounds (per 10,000). Data from United States,
Treatment Episode Data Set –Admissions (TEDS-A), 1992-2012.

51

4.5 DISCUSSION
The main findings of this study may be summarized succinctly. First, the number of
opioids admissions to treatment programs has increased for heroin and other opioids. Second, the
lag time between first opioid use and their first admission for opioid use dropped from 1992 to
2012. Third, according to Hunt’s lag distribution approach, two heroin epidemics can be seen to
peak in the late 1960s and early 2010s (if no more increase is seen). As for incidence rates for
opioids other than heroin, incidence rates increased eightfold from 1995 to 2010.
Before detailed discussion of these results, several of the more important study
limitations merit attention. Of central concern is how lag distributions changed over time from
1950 to 2012. The lag distribution used in this study was from the averaged values of lag time.
The median lag time in this study is 9~10 years for heroin and 7~8 years for other opioid
compounds. Similar results have been reported in previous studies. For example, the median lag
year was 8 years and the mean was 9~10 years for heroin based on the California treatment
program dataset in the 1970s (Judson, Ortiz, Crouse, Carney, & Goldstein, 1980). Krebs et al.
reported that the median lag time was 10 years between the first heroin use and first treatment
admission based on treatment for opioid use disorder in California from 2006 to 2010 (Krebs et
al., 2016). In another study with use of National Epidemiologic Survey on Alcohol and Related
Conditions (NESARC), among those who sought treatment, the midpoints in the cumulative
probability distributions were 12 years for any drug from the time of first use to first treatment
contact (Blanco et al., 2015). According to the previous studies, the lag between first heroin use
and first treatment admission was approximately 10 years across the period from 1970s to 2010s.
The lag distribution used in this study to correct the number of heroin incident cases might be
over- or under-estimated; this is a concern to be investigated in future research.

52

In addition, there are some limitations with respect to the data source used in this study
for the estimates of heroin incidence. TEDS-A is a national census data system of annual
admissions to drug treatment facilities in the United States. However, many drug users do not
receive treatment (Blanco et al., 2013; Compton, Thomas, Stinson, & Grant, 2007; Edlund,
Booth, & Han, 2012; Hasin, Stinson, Ogburn, & Grant, 2007; Kessler et al., 1999; Olfson,
Kessler, Berglund, & Lin, 1998; Regier et al., 1993; United States. Substance Abuse and Mental
Health Services Administration, 2014a; Wang et al., 2005). The most recent report indicates that
only 4.5% of drug users felt they needed the treatment. Among the users who felt they needed,
35% of them did receive the treatment. Overall, only 1.6% of drug users felt they needed and did
receive treatment. The main reasons that drug users did not receive treatment: (1) no health
coverage and could not afford cost, (2) not ready to quit, (3) did not have knowledge of treatment
programs, (4) might have negative effect on job, (5) could deal with the problem by themselves
without treatment, and (6) did not feel the need for treatment at the time (United States.
Substance Abuse and Mental Health Services Administration, 2014a).
Even with existence of barriers to treatment, the increased number of patients in the
treatment programs might not only due to the increase of heroin users but also the increase of
capacity of treatment and more options of pharmacological treatments. Methadone is commonly
used to treat opioid dependence in the treatment maintenance program in the United States.
Methadone was introduced in late 1960s as a maintenance treatment for treating opioid
dependence (Dole & Nyswander, 1965). In 2002, buprenorphine was approved by FDA as
another medical treatment of heroin dependence other than methadone
(https://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatientsandProvid
ers/ucm191521.htm). Buprenorphine and methadone are the most effective treatment for opioid

53

dependence according to the World Health Organization (World Health Organization, 2009).
Therefore, patients with opioid dependence have more options of treatments since 2002 and this
might increase the number of heroin users admitted to treatment programs. Moreover, the federal
budget of drug treatment increased from <1 billion US dollars in 1970 to 10 billions in 2015
(Figure 4.5.1). The increased budget might result in the increase of the capacity of drug
treatment. These changes in the past decades might influence the proportion of heroin incident
cases extracted from TEDS-A database, and then influence the estimates of the heroin incident
cases in each calendar year. Moreover, with a very limited literature investigating the difference
between opioid users admitted to private and those who admitted to public facilities, the opioid
users who admitted to private treatment facilities were more likely to be non-Hispanic White,
younger, and both prescription opioid and heroin users. For those who admitted to public
treatment facilities have higher likelihood of trading or selling sex (O'Grady, Surratt, Kurtz, &
Levi-Minzi, 2014).

Figure 4.5.1 The Federal Drug Control Budget 1970-2015

Link: https://www.drugpolicy.org/sites/default/files/DPA_Fact_sheet_Drug_War_Budget_Feb2015.pdf
Supply Reduction: interdiction, eradication & law enforcement
Demand Reduction: education, prevention & treatment.
Source: White House Office of National Drug Control Policy (ONDCP); Sourcebook of Criminal Justice Statistics.

54

Notwithstanding limitations such as these, the study findings are of interest because this
study has produced heroin incidence rate estimates across 60 years, using Hunt's lag distribution
method as applied to TEDS-A admissions data, given us a new view of the ebb and flow of US
heroin epidemics. Similar patterns have been seen in prior estimation attempts with more
restricted data (i.e., not for 1992-2012), but the sixty-year view has value, and raises questions
about whether the current heroin epidemic already has reached its peak.
This study discloses temporal patterns and ‘relative’ heroin incidence. Whereas I do not
claim an absolute level of accuracy about numbers of newly incident heroin users, these
estimates should prompt new thinking about primary prevention, outreach, and treatment
resources for heroin users.

55

CHAPTER 5
MANUSCRIPT 2 – A TRIGGER FOR NEWLY INCIDENT HEROIN USE?
EPIDEMIOLOGICAL EVIDENCE FROM THE UNITED STATES: A CASECROSSOVER STUDY DESIGN

5.1 ABSTRACT
The aim of this study is to investigate whether extra-medical (EM) OxyContinÂŽ (EMO) use
might have triggered heroin onset among 12-25 year olds during 2004-2014, with probes for risk
variations across EMO level. A case-crossover study design is used. Setting: United States (US),
with community residents 12+ years sampled and recruited for National Surveys on Drug Use
and Health (NSDUH), 2004-2014, then assessed using standardized methods. There are 447
newly incident heroin users among participants assessed 2004-2014. Fine-grained self-report
month-of-onset data for heroin and EMO use meet temporal sequencing guidelines and help
constrain individual-level susceptibilities. The main finding is that excess risk for newly incident
heroin use is seen in a four-month interval immediately after onset of EMO use [case-crossover
relative risk (RR) = 1.9; 95% confidence interval (CI) = 1.13, 3.10; p = 0.01]. Post-estimation
exploratory analyses suggest no excess risk for EM users of other prescription pain relievers, and
indicate no excess risk correlated with new formulations of OxyContinÂŽ, per se. Based on
community-dwelling newly incident heroin users ascertained without prejudice, the estimates
shed light on previously published but more uncertain evidence on suspected EMO-heroin
associations. In conclusion, at least in the United States, the onset of EMO has statistically robust
excess risk of triggering heroin use, even if EMO onset, per se, is not a cause of heroin use.

56

5.2 INTRODUCTION
Both theory and recent empirical evidence from the US support the idea that newly
incident heroin use might have been triggered when young people started to use other opioid
compounds or the originally formulated OxyContinÂŽ products extra-medically (Al-Tayyib,
Koester, & Riggs, 2017; Banerjee et al., 2016; Carlson, Nahhas, Martins, & Daniulaityte, 2016;
Cerda et al., 2015; Cicero et al., 2015; Cicero et al., 2012; Compton et al., 2016; Grau et al.,
2007; Jones, 2013; Lankenau et al., 2012; Mars et al., 2014; Martins, Storr, Zhu, & Chilcoat,
2009; Palamar, Shearston, Dawson, Mateu-Gelabert, & Ompad, 2016; Young & Havens, 2012).
“Extra-medically” (EM) refers to the use of compounds to get high or for related feelings, and
otherwise outside boundaries set by prescribing clinicians as defined by Anthony et al. (Anthony
et al., 1994). The origins of theory and conceptual foundations for asking research questions
about the suspected EM OxyContinÂŽ-heroin triggering hypothesis date back to early general
'steppingstone' and 'gateway' ideas about how the use of one drug early in a chain of temporally
sequenced events might account for later drug use (Anthony, 2002, 2012; Morral et al., 2002).
Evidence pertinent to the EM OxyContinÂŽ-heroin triggering hypothesis has come largely from
clinically-based studies of persons in treatment programs, as well as from a mix of field studies
with local area recruitment approaches (Cicero et al., 2012; Grau et al., 2007; Martins et al.,
2009). Heroin use becomes a possible alternative for OxyContinÂŽ because heroin often has been
cheaper and more readily available. In an illustration of evidence from the qualitative research
tradition, Cicero and colleagues provide the following quotation from an EM user:
“Most people that I know don't use OxyContin® to get high anymore. They have moved
on to heroin [because] it is easier to use, much cleaner, and easily available” (Cicero et
al., 2012).

57

With this conceptual framework in mind, we recognized an emergent research
opportunity to shed new light and provide more definitive evidence on the idea that extramedical use of the originally marketed OxyContinÂŽ products might have accounted for
triggering of onsets of heroin use. We cannot claim that this research opportunity represents the
'perfect' experiment, nor does the opportunity possess all of the advantages of what might be
observed in what most likely would be an unethical randomized controlled trial (RCT). Instead,
the research approach is completely observational and non-experimental. Randomization is not
used to try to bring into balance potentially confounding influences such as might be achieved in
an RCT.
Rather, in an important departure from all prior research approaches used to study the
hypothesis about EM OxyContinÂŽ-heroin triggering (hereinafter abbreviated as 'EMO-heroin'
triggering), we try to hold constant underlying individual-level predispositions and other
confounding influences by turning to a 'subject-as-own-control' design known as the
epidemiologic case-crossover approach. Consider, for example, the idea that an underlying
genetic predisposition or susceptibility trait for self-administration of opioid drugs might lead us
to make an erroneous conclusion that onset of EMO use has triggered a subsequent onset of
newly incident heroin use. That is, it might not be the OxyContinÂŽ product, per se, that is
accounting for any observed excess risk of newly incident heroin use. Instead, it might be an
otherwise uncontrolled individual-level predisposition or susceptibility trait.
One clear strength and advantage of the case-crossover 'subject-as-own-control' design,
relative to alternative observational research designs, is that it holds constant all of these
individual-level or subject-specific predispositions and susceptibility traits up to the time of the
suspected causal exposure. This feature of case-crossover research has been explained in recent

58

research on the cannabis-cocaine association, where interested readers also will find references
to the original methodological articles on the epidemiological case-crossover approach (Maclure,
1991; O'Brien, Comment, Liang, & Anthony, 2012).
Another strength of this study's research approach is that we have not restricted the
samples to persons attending clinics, nor to users in any specific local area. Instead, we are
studying all newly incident heroin users irrespective of where they have been found in recent
national surveys of civilians living in US communities, subject to three fairly simple and
understandable exclusion rules: (1) all must be residing in non-institutional dwelling units (DU)
at the time of the survey, defined to encompass non-household DU such as shelters for homeless
persons; (2) the initial heroin use of these individuals must have occurred within 24 months prior
to the assessment date and cannot be a thing of the more distant past, in order to minimize any
memory and recall errors; (3) among the newly incident heroin users, we must have information
based on standardized survey items about what they recall and report about (i) the month and
year of the first use of heroin, and (ii) the month and year of the first EMO experience, if and
when there has been an EMO onset.
The main value of using case-crossover study design is to deal with the issue of betweensubjects confounding, thus avoiding the problems of comparability of control groups with regard
to confounding variables such as age and sex. The epidemiological estimates of this association
can be seen as a way to prevent heroin initiation among young population.
5.3 METHODS
Before we turn to a more detailed description of methods, there are several important
points to be made. First, as noted above, since 2010, the manufacturer has been distributing a
new 'abuse deterrent' formulation of OxyContinÂŽ, concurrent with efforts to reduce supply of the

59

original formulations; this study's estimates largely pertain to the original formulations available
between 2004 and 2011, and the re-formulations of OxyContinÂŽ between 2012 and 2014. We
did not exclude 2010-11 from our initial case-crossover estimates for original formulations
because it seemed logical that the original OxyContinÂŽ formulation might be available for
diversion from medicine cabinets, and possibly also within gray or black drug markets as well.
Second, with a focus on the basic 'Y after X' relationship, we saw no reason to base this study on
any hypothesis or idea about how OxyContinÂŽ and heroin might be interacting in a
pharmacological or toxicological sense. Third, there is a substantial overlap in US empirical
estimates of the correlates of EMO use and the correlates of heroin use, with some noteworthy
differences (Cicero et al., 2012; Grau et al., 2007; Martins et al., 2009). Nonetheless, this casecrossover research directly addresses the similarity of all individual-level correlates via its
'subject-as-own-control' nature. Fourth, whereas the NSDUH is designed to yield nationally
representative estimates about EM use of heroin and other internationally regulated drugs in the
US, we do not claim that this study's estimates should be regarded as 'representative' of this
nation as a whole. We judge that the research approach derives some strength because the
informative study participants have been sampled, recruited, and assessed 'without prejudice' via
a large-scale survey protocol designed to produce national estimates, but we do not ask readers to
assume or treat this study's estimates as 'nationally representative' in any sense of that phrase.
5.3.1 Population Under Study and Sampling Approach
The United States National Surveys on Drug Use and Health (NSDUH) seeks large
nationally representative community samples of non-institutionalized civilians age 12+ years,
generally achieving participation levels at 65%-70%, and with sample coverage of homeless
shelters and other non-household dwelling units in the community, as well as household dwelling

60

units. Repeated annually from 2004 through 2014, the NSDUH research approach involves
multi-stage area probability sampling and recruitment according to protocols approved by a
cognizant Institutional Review Board (United States. Substance Abuse and Mental Health
Services Administration, 2012)
During these years, the number of young people in aggregate NSDUH public use dataset
analytic samples (i.e., 12-25 year olds) exceeds 350,000 individuals. By the time this study's
exclusion rules are applied with a focus on newly incident heroin users who provided
information about month and year of heroin and EMO use, the effective sample size for our casecrossover research is 447 newly incident heroin users. That is, the vast majority of heroin users
found in NSDUH surveys started to use heroin in the past, and not within a 24 month interval
pre-dated the survey assessment date. For reasons explained in the article's INTRODUCTION,
we exclude these past-onset heroin users and focus the estimation task on the 447 newly incident
heroin users for whom we have values for month and year of heroin use, and if EMO onset has
occurred, for whom we also have values for month and year of EMO onset.
Methodological details of the IRB-approved NSDUH protocol for field survey work are
provided in many readily available online reports (e.g. see
https://nsduhweb.rti.org/respweb/homepage.cfm, last accessed 19 October 2016). These online
reports describe sampling, recruitment, and assessment procedures in great detail, provide
codebook documentation of the study variables, and include links to the original interview
assessment modules and the standardized items on the topics under study.
5.3.2 Assessment Approach
In brief, with respect to the assessment approach, after sampling and recruitment, each
participant completes the NSDUH modularized assessment with coverage of multiple behavior

61

and health domains, including past-onset and newly incident EM drug use for a selection of
drugs, including OxyContinÂŽ and internationally regulated drugs such as heroin. Almost all
NSDUH participants complete confidential computerized self-interviews on the topics under
study, with color drug product displays and other aids to enhance response validity, after choice
of an English or Spanish language version of the assessment.
Month-of-first-use is assessed for recent-onset newly incident EM users of these drugs,
but not for past-onset users (due to anticipated memory errors), with 'newly incident use' made
operational as EM use having first occurred within 24 months prior to the date of survey
assessment. All users are asked about days of EM use in the 12 months prior to assessment. The
exact wording of all NSDUH interview items used in this study is presented in the online
methods reports for each year http://www.samhsa.gov/data/population-data-nsduh/reports.
5.3.3 Post-Assessment Inclusions and Exclusions Approach
As shown in Table 6.4.1, a total of 447 newly incident heroin users are included in this
study's analytic sample. We have been careful to devise inclusion rules that allow us to consider
information from as many newly incident heroin users as is possible, and to implement exclusion
rules only to the extent required to create a basic case-crossover analysis of the type shown in the
Figure 5.3.1.

62

Figure 5.3.1 Flow chart illustrates the sample selection of case-crossover study design. Data
from 393,473 individuals age 12-25 years of age, United States National Surveys on Drug Use
and Health, 2004-2014.

*

The month of EM OxyContinÂŽ onset before heroin onset

63

5.3.4 Hazard Interval and Control Interval Specifications
Each epidemiologic case-crossover study requires pre-specification of a 'hazard interval'
of a given duration ('d'), as well as a 'control interval' of the same duration. Pre-specification of
this 'd' interval during which ‘triggering’ might occur generally involves a synthesis about what
is known or judged to be plausible with respect to the hypothesized 'induction' or 'incubation'
period after exposure and before a clinically significant event occurs.
Well in advance of the analyses reported here, we studied the available published data,
and chose a four-month interval as the duration of the hazard interval. That is, we pre-specified
the 'hazard interval' to be the four months just prior to onset of heroin use. For the 'control
interval' we specified the four-month interval just prior to the start of the hazard interval (i.e., an
equal four-months duration for the hazard interval and for the control interval). As noted
previously, when heroin use and EMO use started in the same month, we set aside information
about the newly incident heroin users; these individuals contribute no information to the RR
estimates of this study because to include them would introduce uncertainty about the temporal
sequencing of the two events under study: (1) newly incident heroin use, and (2) newly incident
EMO use.
5.3.5 Analyze/Estimate Approach
From the prior sections of this article and prior publications (Maclure, 1991; O'Brien et
al., 2012), it can be seen that the relative risk estimate from a case-crossover study uses the
'subject-as-own-control' design to hold constant an array of confounding influences such as
individual-level genetic or other susceptibility traits, and produces an RR estimate that conveys
the degree to which the evidence favors the suspected triggering hypothesis relative to the
evidence favoring the null hypothesis -- i.e., in this study, (a) evidence in favor of EMO-

64

triggering of heroin onset, expressed as the count of newly incident heroin users who had started
EMO use in the four month ‘hazard interval’ just before heroin onset (but not in other months),
versus (b) evidence against the triggering hypothesis, expressed as the count of newly incident
heroin users who had started EMO use in the four month ‘control interval’.
Whereas we had pre-specified the four-month duration of the hazard and control intervals
before any data analyses, we decided it would be best to conduct post-estimation exploratory
data analyses with other specifications for the 'd' duration value. We describe the estimates from
these post-estimation exploratory data analyses in the DISCUSSION section and the first column
of Table 5.4.2.
There was another post-estimation exploratory data analysis that we had not planned in
advance; it was prompted by comments from anonymous reviewers of our manuscript. Namely,
we specified the same four-month duration of the hazard and control intervals, replacing our own
suspected causal exposure variable (onset of EMO use) with a different suspected causal
exposure variable [onset of any EM use prescription pain reliever (PPR) covered in the NSDUH
assessments (e.g., hydrocodone), not counting OxyContinÂŽ products]. That is, we shifted the
research question from our original hypothesis about OxyContinÂŽ products in the direction of a
more general hypothesis about prescription opioids and other PPR. For this post-hoc analysis
task, we focused on the stratum of newly incident heroin users who had never used OxyContinÂŽ
products, but we also show estimates based on analyses that include this stratum. The
DISCUSSION section describes the estimates of this ancillary analysis; the middle and rightmost columns of Table 5.4.2 present the RR estimates and 95% CI.
In the final post-estimation exploration step, we probed into the issues of OxyContinÂŽ
formulation. We completed another post-estimation exploratory data analysis of risk variations

65

between original and new formulations of OxyContinÂŽ based on the year of EMO onset, i.e.,
2004-2011 vs. 2012-2014. The result is shown in the middle and right-most columns of Table
5.4.3.

Figure 5.3.2 Case-Crossover study design

5.4 RESULTS
Among 12-25 year olds at risk for becoming newly incident heroin users during the
observation interval of up to 24 months prior to the date of NSDUH assessment, a total of 447
qualified as newly incident heroin users and had non-missing month-of-onset data. The upper
right hand cell of Table 5.4.1 shows evidence in favor of EM OxyContinÂŽ-triggering of heroin
onset -- namely, the 43 exposure-discordant newly incident heroin users who started EM
OxyContin® use in the pre-specified 4-month ‘hazard interval’ just before the month of heroin
onset but not in the control interval. Evidence against the triggering hypothesis is in the lower
left hand cell of Table 5.4.1 and includes 23 exposure-discordant newly incident heroin users
who started EM OxyContin® use in the pre-specified 4-month ‘control interval’ just prior to the

66

hazard interval and not in the hazard interval (estimated RR = 43/23 = 1.9; 95% CI = 1.13, 3.10,
p = 0.01).
We conducted post-hoc analyses to see what we might have found with different
specifications for 'd.' The first column of Table 5.4.2 displays results from this post hoc analysis.
If the unethical 'cherry-picking' approach had been used, our research team would have been
wise to select d=6, with its much stronger RR estimate of 4.3 and p < 0.01 (95% CI = 2.4, 7.7).
In truth, our d=4 pre-specification produced one of the smallest of the observed potential RR
estimates shown in the first column set of Table 5.4.2; even so, the empirical p = 0.01 for this
d=4 pre-specification. The other two sets of columns of Table 5.4.2 show RR estimates from the
post hoc analyses that were made to include other prescription opioids in the analyses, which
involved re-casting our study focus on OxyContinÂŽ products in the direction of a more general
triggering hypothesis about EM use of prescription opioids and pain relievers (i.e., not focused
on OxyContinÂŽ products).
Exclusion of observations from the years after the manufacturer started releasing
reformulated 'abuse deterrent' OxyContinÂŽ partway through 2011 induced no appreciable
attenuation of the triggering effect estimate. The estimated RR for 2004-2011 is 1.8 (95% CI =
1.05, 3.22). We also conducted the analyses to compare with different specifications for 'd' for
the periods of before-and-after re-formulations of OxyContinÂŽ (see Table 5.4.3).

67

Table 5.4.1 Data used to estimate case-crossover relative risk in relation to the timing of heroin
use in relation to the pre-specified four-month hazard and control intervals prior to the month of
newly incident heroin use. Data from 447 newly incident heroin users age 12-25 years of age,
United States National Surveys on Drug Use and Health (NSDUH), 2004-2014.
EM OxyContinÂŽ onset: Hazard Interval
(Months 1-4 before heroin onset)
EM OxyContinÂŽ onset
in Control Interval
(Months 5-8 before heroin onset)

No

No

n=381 a

Yes

n=23 c

Yes
Total

n=43 b

Undefined

(Estimated case-crossover RR = b/c = 43/23 = 1.87, 95% CI = 1.13, 3.10, p = 0.01).
(85.3% of total n =447); b (9.6% of total); c (5.1% of total).

a

68

447

Table 5.4.2 Estimates from post-estimation exploratory data analyses to shed light on variations in the
study estimates when alternative specifications for the 'd' interval are made, and to illuminate estimates
pertinent to more general trigger hypothesis about EM use of prescription opioids and pain relievers. Data
from 447 newly incident heroin users age 12-25 years of age, United States, National Surveys on Drug
Use and Health (NSDUH), 2004-2014.

The 'd'
duration
of the
hazard &
control
intervals
d=4

Column Set #1
("First")

Column Set #2
("Middle")

Column Set #3
("Right-most")

Hypothesized Triggering
Exposure:
EM OxyContinÂŽ Use.

Hypothesized Triggering
Exposure:
EM Use of Any Prescription
Opioid or Other Prescription
Pain Reliever, including EM
OxyContinÂŽ Users in these
estimates.

Hypothesized Triggering
Exposure:
EM Use of Any Prescription
Opioid or Other Prescription
Pain Reliever, excluding EM
OxyContinÂŽ Users from
these estimates.

Estimated
RR

Lower
95%
CI

Upper
95%
CI

Estimated
RR

Lower
95%
CI

Upper
95%
CI

Estimated
RR

Lower
95%
CI

Upper
95%
CI

1.87

1.13

3.10

1.46

0.91

2.37

1.43

0.72

2.83

d=1
d=2
d=3
d=4
d=5
d=6

0.71
0.34
1.48
1.15
0.55
2.42
2.07
1.09
3.92
2.15
1.12
4.16
1.5
0.89
2.51
1.57
0.93
2.64
See d=4 in row 1
See d=4 in row 1
2.65
1.58
4.43
2.36
1.44
3.89
4.29
2.40
7.67
2.57
1.58
4.15
Key: Emboldened text with shading indicates p-value < 0.05

69

2.50
0.78
7.97
2.33
0.90
6.07
1.90
0.88
4.09
See d=4 in row 1
2.50
1.20
5.21
2.64
1.32
5.28

Table 5.4.3 Estimates from post-estimation exploratory data analyses to shed light on variations
in the study estimates before-and-after re-formulation of OxyContinÂŽ, and to illuminate
estimates pertinent to 2004-2011 and 2012-2014. Data from 335 and 112 newly incident heroin
users age 12-25 years of age respectively for 2004-2011 and 2012-2014, United States, National
Surveys on Drug Use and Health (NSDUH).
Column Set #1
("First")

Column Set #2
Column Set #3
("Middle")
("Right-most")
Hypothesized Triggering Exposure: EM OxyContinÂŽ Use

2004-2014

The 'd'
duration
of the
hazard &
control
intervals

2004-2011

2012-2014

(Before re-formulation)

(After re-formulation)

Estimated
RR

Lower
95%
CI

Upper
95%
CI

Estimated
RR

Lower
95%
CI

Upper
95%
CI

Estimated
RR

Lower
95%
CI

Upper
95%
CI

1.87

1.13

3.10

1.84

1.05

3.22

2.00

0.60

6.64

0.71
0.34
1.48
0.73
0.34
1.6
2.07
1.09
3.92
2.89
1.35
6.16
1.50
0.89
2.51
1.58
0.89
2.81
See d=4 in row 1
See d=4 in row 1
2.65
1.58
4.43
2.33
1.34
4.05
4.29
2.40
7.67
4.08
2.17
7.68
Key: Emboldened text with shading indicates p-value < 0.05

0.50
0.60
1.20

d=4
d=1
d=2
d=3
d=4
d=5
d=6

70

0.05
5.51
0.14
2.51
0.37
3.93
See d=4 in row 1
5.50
1.22
24.81
5.50
1.22
24.81

5.5 DISCUSSION
Drawing strength from a sample based largely on experiences in US communities before
the manufacturer's substitution of an 'abuse deterrent' formulation of OxyContinÂŽ products, this
study’s primary discovery is modest evidence of what might have been a triggering effect of EM
OxyContinÂŽ use with respect to risk of starting heroin use, as has been hypothesized by prior
clinician observers and epidemiologists, estimated here as RR = 1.9 (p = 0.01).
This study's assumptions about independent observations and no appreciable survey
design effects are substantiated by the fact that the RR estimate is based on a total of 66
informative newly incident heroin users (summed across the exposure-discordant off-diagonal
cells of Table 5.4.1), who qualify as independently observed cases distributed across 11 years of
the NSDUH survey interval (2004-2014) and the multiple local sampling strata distributed across
the US. Some readers might ask whether the complex multi-stage sampling approach for the
NSDUH requires special variance estimation methods as might be needed to take the nonindependence of observed cases and survey design effects into account. We reasoned that
standard variance estimation for matched pairs RR estimates would suffice, given that the
informative cases of newly incident heroin use (i.e., those with exposure-discordant pairs of
hazard and control intervals) most likely would arise from completely independent local areas
and dwelling units, spread across the independent survey fieldwork conducted from 2004 to
2011. If that is the case, assumptions about independent observations become reasonable, and
any survey design effect will be of trivial size. Indeed, an unweighted analysis without attention
to multi-stage sampling and non-independence of sampled observations might lead to different
variance approximations and perhaps wider confidence intervals as compared to what can be
obtained using analysis weights and the calculus methods such as Taylor series linearization that

71

we often use to handle survey design effects in our epidemiological field studies. In this study,
our reasoning was that we could conduct a post-hoc investigation to evaluate whether there
might have been an unlikely and generally implausible departure from standard assumptions
about rarely occurring events in survey research on newly incident heroin use (e.g., as might be
true if there are multiple exposure-discordant newly incident cases from the same sampling
stratum in the same survey year). If and when the post hoc investigation might disclose multiple
cases arising in the same survey year and within the same local sampling stratum, then it would
be possible to turn to approaches that have been used in epidemiological studies of neonates
when there are twins or other multiple births. To illustrate, just as investigators often select one
of the twins or multiple births at random from the set of liveborn multiple birth neonates, it
would be possible to select one newly incident heroin user at random from within any specific
local area that gave rise to more than one newly incident heroin user in a given survey year. As
it happened, it was unnecessary in this study to take a random draw of one of the exposurediscordant newly incident heroin users from the survey sampling stratum and to re-calculate the
matched pairs RR estimator and its 95% confidence intervals (CI).
As for weaknesses of this study, it should be noted that the NSDUH are cross-sectional
surveys. In order to simulate a prospective study and to align temporal sequencing correctly, it
has been necessary to turn to NSDUH retrospective data about fine-grained autobiographical
details such as the month of drug use onset. Memory flaws, confusion, and chronological errors
may be present, as suggested elsewhere (Anthony, Neumark, & Van Etten, 2000). In addition,
some concurrent EMO-heroin users might fall outside the sampling frame, or decline to
participate, as noted in this Chapter’s INTRODUCTION, but we do not judge this possibility to
be especially compelling with respect to bias in this study’s relative risk estimates.

72

Notwithstanding limitations such as these, we judge that this new evidence on the EMOheroin triggering hypothesis represents an important contribution and step forward in our
understanding of the epidemiology of heroin use in specific and to the epidemiology of
prescription pain reliever use in general. As for implications with respect to clinical or public
health policy and actions, these findings highlight some potential unintended consequences of
sustained release opioid re-formulations in the pharmaceutical research and development process
(Jones, Muhuri, & Lurie, 2016), but perhaps it is of most importance that the case-crossover
approach rules out predispositions and susceptibility traits that have not been controlled in the
prior investigations of this topic, yielding more definitive evidence than has been available from
past inquiries. These results tend to confirm the initial EMO-heroin triggering hypothesis
articulated by observant clinicians and in anecdotes from heroin users, as well as observations
from prior field study investigations without benefit of the case-crossover approach, including
those conducted prior to the manufacturer's introduction of the 'abuse deterrent' formulation. If
the findings can be confirmed in future replications, the survival analysis estimates may have
value in clinical or public health practice. Once the future investigations of case-crossover
research on the suspected EMO-heroin triggering hypothesis during the interval 2011 to 2020 are
completed, it will be possible to be more confident that the intent of ‘abuse deterrence’ has been
achieved (Cicero & Ellis, 2015). Moreover, a better understanding for EMO-heroin triggering
would consider a cohort study design, but we are not aware that any cohort study has been
published. However, there is an ongoing study in Australia that is evaluating the potential impact
of re-formulated OxyContinÂŽ via gathering cohort data of regularly opioids misusers
(Degenhardt et al., 2015). This prospective cohort study might help future studies to understand
how re-formulated OxyContinÂŽ correlated with heroin or other opioid use.

73

CHAPTER 6
MANUSCRIPT 3 – PATTERNS OF DURATION OF ELAPSED TIME FROM EXTRAMEDICAL PRESCRIPTION PAIN RELIEVER ONSET TO FIRST HEROIN USE: A
SURVIVAL ANALYSIS
6.1 ABSTRACT
Population density was associated with heroin use in the 1970s. Heroin incidence peaked
earlier in the cities then shifted to smaller cities and rural areas. During the 21st century, many
studies have found that prior use of extra-medical prescription pain relievers (EMPPR) is
associated with increased risk of heroin use. This study aims to estimate the degree to which
living in a metropolitan area is associated with transitioning from 1st EMPPR use to 1st heroin
use. Its study population included non-institutionalized United States civilians age 12 years and
older. Survival analyses produced time-specific risk estimates of initiating use of heroin among
EMPPR users. Estimated peak risk for 1st heroin use was found three years after EMPPR onset.
With 10 years of observation, approximately 5% of EMPPR users had initiated heroin use. The
corresponding proportion 6% for males as compared to 3% for females. About 7% of EMPPR
users who started EMPPR during adolescence initiated heroin use within 10 years. Estimated
transition probabilities for these subgroups did not vary with population density. EMPPR users
who are male, White, and with early EMPPR onset have increased risk of initiating heroin use.
Although the outcome does not support the proposed hypothesis, the findings may also help
clinicians to understand a more vulnerable population for heroin use (e.g., early EMPPR onset)
and help prevention programs to identify target populations in the clinical setting.

74

6.2 INTRODUCTION
Heroin use is one of the leading drug use problems in the US. An estimated 4.2 million
Americans aged 12 or older (1.6 %) have used heroin at least once in their lives (United States.
Substance Abuse and Mental Health Services Administration, 2012), and the prevalence of
heroin use increased from 0.33 percent to 1.60 percent in the past decade (Martins et al., 2017).
Emerging evidence suggests that prior EM of prescription opioids is associated with heroin use,
and the heroin incidence is 19 times greater among those who had prior use of EMPPR. Nearly
80% of heroin users had experienced EMPPR use before initiating heroin use (Compton et al.,
2016; Muhuri et al., 2013; Young & Havens, 2012). The transition from prescription opioids to
heroin is a growing public health problem in the US though the fraction of prescription opioid
users who shifted to use heroin was estimated under 4% (Compton et al., 2016; Jones, 2013;
Mars et al., 2014; Muhuri et al., 2013).
Among those people who started using opioid drugs in the 1960s, heroin was the first
used. More than 80% were men, nearly equal white to nonwhite ratio, and most heroin users
lived in urban or metropolitan areas. Moreover, the emerging heroin users shifted over time from
metropolitan cities to smaller cities, and they were not homogeneously distributed in the regions
of US. During Post-World War II years, heroin outbreaks first emerged in metropolitan cities
especially in the inner cities, such as Chicago and New York City; as time passed the epidemic
diffused to smaller cities. The temporal manifestation of peak of heroin incidence in the cities
reflected population size in the cities, from densely populated areas to less populated areas
(Hughes et al., 1972; Hunt & Chambers, 1976). In Figure 6.2.1, the clear pattern of heroin
incidence indicated that heroin spread from higher densely populated states to lower populated
states in 1960s. The peak of heroin epidemic began in the East Coast (e.g., New York and

75

Massachusetts) and West Coast (e.g., California) and shifted to the Midwestern states (e.g.,
Michigan and Minnesota). In the more recent 21st century heroin epidemic, most people started
using heroin after EM use of prescription opioids, female heroin users now are as many as male
heroin users, more than 80% are white, and more heroin users are living in non-urban areas
(Cicero et al., 2014).

Figure 6.2.1 Hunt’s Macro-Diffusion theory

Source: Hunt, L. B., & Chambers, C. D. (1976). The heroin epidemics: a study of heroin use in the U.S., 1965–75.
Holliswood, NY: Spectrum, p.48.

It also has been hypothesized that several environmental factors are associated with the
transition from EMPPR to heroin, including the easy access to heroin and the cheaper price of
heroin, relative to prescription opioids, in the street market may be underlying explanations

76

(Cicero et al., 2014; Unick, Rosenblum, Mars, & Ciccarone, 2014). Drug use is no longer an
urban problem in the US. Some studies report that people who live in the rural areas might have
a higher risk of using opioids non-medically because the cost of prescription drugs might be less
expensive in smaller towns or rural areas than inner cities, and the increased availability in rural
than urban areas (Cicero, Surratt, Inciardi, & Munoz, 2007; Havens, Oser, & Leukefeld, 2007;
Keyes, Cerda, Brady, Havens, & Galea, 2014; Paulozzi & Xi, 2008). Young and colleagues also
indicated that people who live in rural areas had much higher drug problem severity of
nonmedical use of prescription opioid than those who live in urban areas due to the routes of
using prescription opioids such as snorting and injection rather than the oral route (Young,
Havens, & Leukefeld, 2010). On the other hand, as for the shift of the environment, heroin users
have shifted from low-income neighborhood in the cities with majority of African Americans in
1960s to middle-class neighborhood in the suburban and rural areas with primarily White
populations in the 21st century. Heroin apparently has become a cheaper and more accessible
alternative to prescription opioid in non-urban than in urban areas (Cicero et al., 2014; Cicero &
Kuehn, 2014).
The heroin epidemic in 2010s is linked to the shifting of demographic characteristics
(e.g., age, sex, race/ethnicity) of heroin users and the areas where people live (e.g., urban vs.
rural). Environmental factors (e.g., population density) might play an important role on spreading
of drug use (Cicero & Kuehn, 2014; Cooper, Bossak, Tempalski, Des Jarlais, & Friedman, 2009;
Dew, Elifson, & Dozier, 2007). However, little is known about the factors influencing heroin
spread in relation to environmental conditions and individual characteristics, and whether
transitioning from EMPPR use to heroin use is faster in rural areas. The motive of this study is

77

strongly encouraged by Hunt’s micro-diffusion theory (Hunt & Chambers, 1976). The aim of
this study is to estimate the probability of transitioning from first EMPPR use to heroin use.

6.3 METHODS
The study population is that of the United States civilians age 12 years and older, 20082014, as sampled for the National Surveys on Drug Use and Health (NSDUH), with details
provided here: https://www.icpsr.umich.edu/icpsrweb/ICPSR/series/64, and in Chapters 3 and 5,
including sampling, recruitment, participation, and assessment plans, as well as sample sizes.
The study protocol was reviewed and approved by the cognizant institutional review board for
protection of human subjects in research.
The key response of interest in this study is occurrence of heroin use among EMPPR
users who had experience of using EMPPR but they did not use heroin previously. The duration
of transition from first use of EMPPR to first heroin use was measured through the difference
between age of heroin onset and age of EMPPR onset. The suspected causal determinant or
covariate of central interest is living area (i.e., metropolis with densely populated areas vs.
suburban or rural with less population). The metropolitan area was identified by two variables:
population density and metropolis status. Population density was measured via the census on
Core Based Statistical Area (CBSA) classifications provided by the Office of Management and
Budget. The metropolis status was reported and classified into large metro, small metro, and
non-metro. The Urban-Metropolis area (MET) in this study is defined as the large metropolitan
area where containing a core urban area of one million or more, versus suburban-rural area
(SUBRUL) defined as the small metro or non-metro area with a core area of less than one
million populations.

78

In order to probe more completely into the temporal sequencing issues, the plan for data
analysis was organized in relation to standard "explore, analyze, explore" cycles, in which the
first exploratory steps involved cross-classifications and Kaplan-Meier curves. Thereafter, the
analyze/estimate step involved discrete-time survival analysis (DTSA) via STATA version 13.0
(College Station, TX: StataCorp LP; 2013) software. In this work, the newly incident EMPPR
users with 'tied' year of heroin onset are excluded from the analysis due to temporal sequencing
concerns. The formula of discrete-time survival analysis is shown below.

79

In the initial analysis step, the task was to estimate the probability of transition from first
use of EMPPR to time of first heroin use, for which the statistical approach was Kaplan-Meier
analysis which provides a visual display of the heroin onset risk experiences of these newly
incident EMPPR users, after pre-specified stratification into 'SUBRUL' versus 'MET' of EMPPR
users lived when survey was being conducted. In subsequent analysis steps, the statistical
approach involved DTSA with logistic model to estimate the risk of heroin initiation among
EMPPR users. The outcome of interest was time to onset of heroin use. Each individual's time
origin for survival analysis started at the year of first EMPPR use, after which year of heroin
onset was ascertained. Right-censoring occurred as heroin onset did not happen until the year of
assessment.
The final exploratory analysis steps compared the risks of heroin initiation for metropolis
versus non-metropolis (MET vs. SUBRUL) with the stratification of sex (i.e., male vs. female),
age of EMPPR onset (i.e., <12, 12-17, 18-25, >25 years), and race/ethnicity (i.e., non-Hispanic
White, non-Hispanic Black, Hispanic, and others). In specific, using DTSA models, I traced
year-by-year experiences of more than 60,000 aged 12 year or older NSDUH 2008-2014
participants who qualified as EMPPR users and who did not have experience of using heroin
before their EMPPR onset which allowed us to deal with temporal sequencing issues. For this
research, I stress precision of the study estimates with a focus on 95% confidence intervals; pvalues are presented as an aid to interpretation. Cox proposed an extension of the proportional
hazards model to discrete time and is used to fit DTSA by working with the conditional odds of
heroin use at each time period (Cox, 1972).

80

6.4 RESULTS
Each Kaplan-Meier survival curve conveys information about the timing of heroin use
onsets among the EMPPR users. There are 5% of EMPPR users who began using heroin within
10 years since EMPPR onset (see Figure 6.4.1). With stratification by population density [that is,
suburban-rural (SUBRUL) versus urban-metropolis (MET)], an estimated 4.9% of EMPPR users
in SUBRUL areas used heroin within 10 years since EMPPR onset, versus 4.6% of EMPPR
users in MET areas. There is no clear difference between MSA and non-MSA (p-value=0.15, the
log-rank test) (Figure 6.4.2). Male EMPPR users have faster transition from first EMPPR to first
heroin use than female EMPPR users (Figure 6.4.3). Age of EMPPR onset during adolescence or
young adulthood is associated with faster transition to heroin use, especially for adolescents who
age 12-17 years. Approximately 7% of EMPPR adolescents initiated heroin use within 10 years
since their first EMPPR use, compared to 3.5% for those started their first EMPPR use during
young adulthood (18-25 years old) and 1.5% for those started their first EMPPR use during
adulthood (26 years or older). The result is shown in the Figure 6.4.4. Moreover, the
corresponding probability of heroin onset is significantly higher for non-Hispanic Whites than
other subgroups such as non-Hispanic Black or Hispanic individuals. Approximately 6% of nonHispanic White EMPPR users had experienced using heroin as compared to 3% for Hispanic and
1% for non-Hispanic Black subgroups, and there is no variation by subgroup with regard to
metropolitan or non-metropolitan areas they lived (Figure 6.4.5).
The hazard function describes the instantaneous rate of starting heroin use at specific
time points over the observed 10 years and the Figure 6.4.6 is presented as a graphical display to
illustrate the risk for heroin onset over time for each subgroup. In other words, the hazard
function produced a conditional rate of heroin onset in a specified time interval since first

81

EMPPR use given that the subject had not used heroin ('at-risk') at each time interval's start. The
risk for heroin onset peaks in the third year since EMPPR onset for most EMPPR users. Men
have a hazard rate at peak of 0.8%, compared to the peak of 0.5% for women. Non-Hispanic
White EMPPR users have a highest hazard rate over time among these race/ethnicity subgroups.
They have a hazard rate of heroin onset at peak of 0.8% in the third year since EMPPR onset,
compared to the peak of 0.4% for Hispanic individuals and 0.1% for non-Hispanic Black. For
adolescents (12-17 years) and young adults (18-25 years), the peak of the risk is on the third year
since EMPPR onset (1% for adolescents and 0.5% for young adults). Additionally, for people
who started EMPPR younger than 12 years, their risk of heroin onset is increasing from the third
year to the sixth year since EMPPR onset. There in no remarkable peak of hazard on heroin onset
for those who are older than 25 years.
Under the DTSA with logit model for analysis-weighted data, with Taylor series
linearization for variance estimation, unadjusted hazard ratios (HR) and covariate-adjusted
hazard ratios (aHR) with their corresponding 95% confidence intervals (CIs) are estimated. The
results shown in Table 6.4.1. The risk of heroin onset shows no variation for EMPPR users who
lived in MET versus SUBRUL areas. Male EMPPR users are an estimated 1.5 times more likely
to become heroin users than female (HR=1.6, 95% CI= 1.4, 1.7). EMPPR users who started
EMPPR during adolescence are about five times more likely to initiate heroin use within 10
years since EMPPR onset than those who started EMPPR later in their lives (HR=4.5, 95% CI=
3.7, 5.4). Moreover, non-Hispanic Whites have the highest hazard of the transition from the first
EMPPR use to heroin onset among subgroups of race/ethnicity. For example, non-Hispanic
White EMPPR users are about six times more likely than non-Hispanic Blacks to start heroin use
within 10 years since EMPPR onset (HR=5.8, 95% CI= 4.5, 7.4).

82

Figure 6.4.1 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first EMPPR
use. Data from the National Surveys on Drug Use and Health (NSDUH), United States, 20082014.

83

Figure 6.4.2 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first EMPPR
use. Stratified by MET-SUBRUL. Data from the National Surveys on Drug Use and Health
(NSDUH), United States, 2008-2014.

84

Figure 6.4.3 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first EMPPR
use. Stratified by Male-Female. Data from the National Surveys on Drug Use and Health
(NSDUH), United States, 2008-2014.

85

Figure 6.4.4 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first EMPPR
use. Stratified by Age of EMPPR Onset. Data from the National Surveys on Drug Use and
Health (NSDUH), United States, 2008-2014.

86

Figure 6.4.5 Kaplan-Meier survival curves for newly incident EMPPR users 12 years or older
who started heroin use, plotted in relation to elapsed time (in years) since year of first EMPPR
use. Stratified by Race/Ethnicity. Data from the National Surveys on Drug Use and Health
(NSDUH), United States, 2008-2014.

87

Figure 6.4.6 Estimated hazard functions depict the risk of starting heroin use, plotted in relation
to elapsed time (in years) since the year of the first EMPPR use, with contrasts of subgroups (i.e.,
MET-SUBRUL, sex, age of EMPPR onset, and race/ethnicity). Data from the National Surveys
on Drug Use and Health (NSDUH), United States, 2008-2014.

88

Table 6.4.1 Discrete time data is analyzed in the survival analysis using logic function for each
of the time periods. The unadjusted Hazard Ratios (HR) and covariate-adjusted Hazard Ratios
(aHR) are estimated with 95% confidence intervals (95% CI). Data from the National Surveys on
Drug Use and Health (NSDUH), United States, 2008-2014.
Parameter

Estimates
HR (95% CI)

n (%)

aHR (95% CI)

a

Metropolitan areas
MET
SUBRUL
Unclassified
Sex
Female
Male
Age of EMPPR onset
<12
12-17
18-25
>26
Race/Ethnicity
Non-Hispanic White
Non-Hispanic Black
Hispanic
Others

24,095 (40.07)
26,440 (43.97)
9,592 (15.95)

Ref.
1.06 (0.98, 1.15)
1.03 (0.92, 1.14)

Ref.
0.94 (0.86, 1.02)
0.89 (0.79, 0.99)

28,918 (48.09)
31,209 (51.91)

Ref.
1.55 (1.43, 1.67)

Ref.
1.51 (1.40, 1.63)

2,698 (4.49)
27,136 (45.13)
23,661 (39.35)
6,632 (11.03)

1.74 (1.29, 2.34)
4.45 (3.65, 5.42)
2.26 (1.84, 2.77)
Ref.

2.08 (1.54, 2.80)
4.57 (3.75, 5.56)
2.26 (1.84, 2.77)
Ref.

40,922 (68.06)
5,811 (9.66)
8,347 (13.88)
5,047 (8.39)

5.76 (4.45, 7.44)
Ref.
3.15 (2.38, 4.17)
3.37 (2.51, 4.52)

5.67 (4.39, 7.33)
Ref.
2.93 (2.21, 3.87)
3.24 (2.42, 4.35)

a

MET: a metro area contains a core urban area of one million or more populations; SUBRUL: a suburban or rural area contains a
core urban area of less than one million populations.

89

6.5 DISCUSSION
The main findings of this study may be summarized succinctly. Over 10 years, among
EMPPR users of age 12 years or older, an estimated 5% initiated heroin use. The peak of the risk
of heroin onset is in the third year since first EMPPR use. The population density of the area
(MET vs. SUBRUL) is not associated with the transition from EMPPR to heroin use among
EMPPR users. However, some factors are associated with heroin onset among EMPPR users
such as age of EMPPR onset, sex, and race/ethnicity.
Before detailed discussion of these results, several of the more important study
limitations merit attention. This study discloses temporal patterns of heroin onset among EMPPR
users. Of central concern is a recall bias in the retrospective self-reported survey between the age
of first EMPPR use and age of first heroin use. This bias can be attributed to recall period, age of
interview, or other personal characteristics. Memory flaws, confusion, and chronological errors
may be present, as suggested elsewhere (Anthony et al., 2000). In this study, the MET-SUBRUL
distinction was defined by two variables: (1) Population density identified with CBSA which is a
US geographic area defined by the Office of Management and Budget (i.e., Segment in a CBSA
with 1 million or more persons, fewer than 1 million, and Segment not in a CBSA), and (2)
Urbanicity identified by the categories provided from NSDUH (i.e., large metro, small metro,
and non-metro). Therefore, the classification of this variable might not be clear enough to reflect
urban or rural areas in the smaller cities or town.
The results of this study from the survival analysis hazard ratio (HR) estimates and
survival curves could not help substantiate the idea that population density plays an important
role of the transition from first EMPPR use to heroin onset. Hunt’s macro-diffusion theory from
densely populated areas to less densely populated areas might not work for heroin users in 21st

90

century who had EMPPR use prior to heroin use (Hunt & Chambers, 1976). The non-significant
results of population density might be due to the use of the internet as a source of drugs. Online
drug markets and social media are fast-growing trends in contemporary drug consumption.
Online drug marketplaces such as ‘Silk Road’, ‘AlphaBay Market’ or ‘Dream Market’ may make
drugs more readily available in rural areas (Christin, 2013; Dasgupta et al., 2013; Rhumorbarbe,
Staehli, Broseus, Rossy, & Esseiva, 2016; Van Hout & Bingham, 2013; Van Hout & Hearne,
2017). As is the case for social media, internet sources can be used to spread out the information
of drug and sale of prescription drugs or illegal drugs as well, such as Craigslist (Tofighi, Perna,
Desai, Grov, & Lee, 2016).
However, even though there is no relationship between metropolis status and the shifting
from EMPPR to heroin, this study still provides some evidence that characteristics such as nonHispanic White, male, and early EMPPR onset are associated with the shifting from EMPPR use
to heroin onset. These results are convergent with what has been found in previous studies. For
example, heroin users have shifted to primarily non-Hispanic White populations (Cicero et al.,
2014; Muhuri et al., 2013).
The estimated cumulative probability curves and hazard rate for onset of first heroin use
after the year of first EMPPR use may be useful in future investigations of this type because they
show a peak of hazard rate of heroin onset at the third year after first EMPPR use. The results
may also help clinicians to understand the vulnerable population among EMPPR users, and then
helps prevention programs to identify target populations in their primary setting.

91

CHAPTER 7
FINAL DISCUSSION
7.1 SUMMARY OF FINDINGS
The study in Chapter 4 applies Hunt’s back calculation approach to estimate incident
cases for heroin and EMPPR use in US. Based on Hunt’s model via re-calibration approaches
using Treatment Episode Data Set - Admissions datasets (TEDS-A), heroin onset increased
160% from 2000 to its peak in 2010; the incidence rate in 2012 is similar to the first heroin
epidemic in 1969. As for opioids other than heroin (e.g., prescription opioids), the incidence
increased more than 250% from 2000 to 2010. The study in Chapter 5 investigates whether EM
OxyContinÂŽ use might have triggered heroin onset among 12-25 year olds between 2004-2014.
The excess risk of newly incident heroin use is seen in a four-month interval right after onset of
EM OxyContinÂŽ use (case-crossover risk ratio = 1.9). Post-estimation exploratory analyses
suggest no excess risk for EM users of other prescription pain relievers, and indicate no excess
risk correlated with new formulations of OxyContinÂŽ per se. In the Chapter 6, the peak risk for
transitioning from onset of EMPPR use to heroin onset within 10 years is found at the third year
since first EMPPR use. Approximately 5% of participants initiated heroin use in 10 years since
EMPPR onset. The estimates of these subgroups remain similar regardless of population density.
EM prescription pain reliever users who are male, White, and have early EM prescription pain
reliever onset have an increased risk of initiating heroin use.
7.2 LIMITATIONS AND STRENGTHS
In the Chapter 4, Hunt’s lag distribution method is used to estimate the incident cases for
heroin and EMPPR. However, the change of lag distribution over time from 1950 to 2012 might
be a central concern when the averaged value of lag distributions was used to estimate the cases

92

of truncation. In this study, the median lag year in this study is 9~10 years for heroin and 7~8
years for other opioid compounds and the estimates are similar to other previous studies (Blanco
et al., 2015; Judson et al., 1980; Krebs et al., 2016). The lag distribution used in this study to
correct the number of heroin incidence cases might be over- or under-estimated; more research is
needed on this aspect of the back-calculation. Moreover, there are some limitations with respect
to the data source used in this study for the estimates of heroin incidence. TEDS-A is a national
census data system of annual admissions to drug treatment facilities in the United States.
However, many drug users do not receive treatment (Blanco et al., 2013; Compton et al., 2007;
Edlund et al., 2012; Hasin et al., 2007; Kessler et al., 1999; Olfson et al., 1998; Regier et al.,
1993; United States. Substance Abuse and Mental Health Services Administration, 2014b; Wang
et al., 2005). Based on the dataset of TEDS-A, with only a small proportion of drug users in the
treatment program, the numbers of heroin incident cases produced in this study may be undercounts. Even so, the observed patterns of relative heroin incidence over six decades from the first
heroin epidemic to the recent heroin epidemic in the 2010s may prove to be valuable evidence
and it may help health policy makers planning responses for heroin use in the U.S.
As for weaknesses of using NSDUH datasets in the Chapter 5 and Chapter 6, it should be
noted that the NSDUH are cross-sectional surveys. In order to simulate a prospective study and
to align temporal sequencing correctly, it has been necessary to turn to NSDUH retrospective
data about fine-grained autobiographical details such as the month of drug use onset. Memory
flaws, confusion, and chronological errors may be present, as suggested elsewhere (Anthony et
al., 2000). Estimates of drug use through national household surveys may involve underreported
use because of social pressure or legal consequences of drug use. However, there is evidence on
validity and reliability of drug use onset age (United States. Substance Abuse and Mental Health

93

Services Administration, 2010). Other major advantages from NSDUH are large sample sizes
drawn based on a multistage area probability design, and the national representative sample for
the entire US civilian population age 12 years and older.
7.3 FUTURE DIRECTIONS AND CONCLUSIONS
The evidence produced in this dissertation research adds to our understanding of heroin
epidemiology. It discloses temporal patterns and ‘relative’ heroin and EMPPR incidence in the
past 60 years and it has helped clarify relationships linking EMPPR use with heroin use. The
estimates comparing past and recent epidemic years should prompt new thinking about primary
prevention, outreach, and treatment resources for heroin users. The results may also help
clinicians to understand risks encountered in populations of EMPPR users, which include risks
associated with starting heroin use. The results may also help leaders of prevention programs to
identify target populations in their primary settings.

94

CHAPTER 8
CONCLUSIONS
I conclude my doctoral dissertation with a few reflections on this project and what it
might mean to future researchers in this area. At the beginning, I concentrated on estimating the
incidence of heroin and other opioid compounds to provide empirical evidence on the incidence
trend over decades in US history. Factors associated with the transition from extra-medical
prescription pain reliever (EMPPR) to heroin use were examined in terms of environmental
conditions and individual characteristics thought to be of possible importance for prevention
policies or practices, and to future researchers or policy makers in this area.
By the end of the process, I had shifted my attention to the increase of opioid incidence
over recent decades for non-heroin opioid compounds such as EMPPR. Moreover, I found the
transition probability from EMPPR onset to heroin onset to be greater for some interesting
subgroups, especially male EMPPR users and those starting EMPPR use at an earlier age.
For this reason, I will offer no more than a few words about overall conclusions from my
research, and the public health challenges faced by anyone who is trying to reduce global health
burdens attributed to heroin and other opioids use. First, a recent surge of newly incident opioid
users has implications. Considering the lag-time between onset and treatment admission (heroin:
9-10 years; other opioids, 7-8 years), we may face increasing treatment needs. Second, my
recommendation is about age of EMPPR onset and risk of 1st heroin use. Here, primary
prevention programs are needed, and few prevention researchers now focus specifically on
EMPPR use. Finally, I should add that in future research it should become possible to study
differences in opioid incidence trends at the state-level or for supra-state geographical regions.

95

Future research of this type should provide a more a complete understanding of US opioid
epidemics of the past and present, setting the stage for better public health work in the future.

96

APPENDICES

97

APPENDIX A
THE LIST OF MUSICIANS OF THE ERA WHO DIED FROM HEROIN OR OTHER OPIOID
DRUGS OVERDOSE
Table A.1 The list of musicians of the era who died from heroin or other opioid drugs overdose
Musicians
1960s
Rudy Lewis
Frankie Lymon
1970s
Janis Joplin
Jim Morrison
Brian Cole
Gram Parsons
Zeke Zettner
Tim Buckley
Gary Thain
Tommy Bolin
Elvis Presley
Gregory Herbert
Sid Vicious
Lowell George
Jimmy McCulloch
1980s
Darby Crash
Tim Hardin
Mike Bloomfield
Pete Farndon
Paul Gardiner
Gary Holton
Phil Lynott
Paul Butterfield
Jesse Ed Davis

Age of death

Year of death

Overdose drugs

27
25

1964
1968

Heroin
Heroin

27
27
29
26
25
28
27
25
42
28
21
34
26

1970
1971
1972
1973
1973
1975
1975
1976
1977
1978
1979
1979
1979

Heroin
Suspected heroin
Heroin
Morphine and alcohol
Heroin
Heroin, morphine, and alcohol
Heroin
Heroin, cocaine and alcohol
Assorted drugs
Heroin
Heroin
Heroin
Morphine

22
39
37
30
25
33
36
44
43

1980
1980
1981
1983
1984
1985
1986
1987
1988

Heroin
Heroin
Heroin
Heroin
Heroin
Morphine and alcohol
Heroin
Heroin
Heroin

98

Table A.1 (cont’d)
Musicians

Age of death

Year of death

Overdose drugs

1990s
Andrew Wood
24
1990
Heroin
Iosu ExpĂłsito
31
1992
Heroin
River Phoenix
23
1993
Heroin and Cocaine
Dave Rubinstein
29
1993
Heroin
Rob Jones
29
1993
Heroin
Kurt Cobain
27
1994
Suicide while high on heroin
Kristen Pfaff
27
1994
Heroin
Dwayne Goettel
31
1995
Heroin
Bradley Nowell
28
1996
Heroin
John Kahn
48
1996
Heroin
Jonathan Melvoin
34
1996
Heroin
West Arkeen
36
1997
Opiate
Nick Traina
19
1997
Heroin
John Baker
44
1999
Heroin
Saunders
David McComb
37
1999
Heroin
2000s
Allen Woody
44
2000
Heroin
Joachim Nielsen
36
2000
Heroin
Dee Dee Ramone
50
2002
Heroin
Robbin Crosby
42
2002
Heroin
Howie Epstein
47
2003
Heroin
Jeremy Michael
27
2003
Heroin
Ward
Tim Hemensley
31
2003
Heroin
Robert Quine
61
2004
Heroin
Gidget Gein
39
2008
Heroin
Jay Bennett
45
2009
Fentanyl
2010s
Paul Gray
38
2010
Morphine and Fentanyl
Mikey Welsh
40
2011
Heroin
Dave Brockie
50
2014
Heroin
Prince
57
2016
Fentanyl
Source: Wikipedia: List of drug related deaths, accessed on 27 July 2017.
https://en.wikipedia.org/wiki/List_of_deaths_from_drug_overdose_and_intoxication

99

APPENDIX B
THE LIST OF PRESCRIPTION PAIN RELIEVERS FROM NATIONAL SURVEY ON DRUG
USE AND HEALTH (NSDUH)

Figure B.1 The list of prescription pain relievers from NSDUH

Adapted from Substance Abuse and Mental Health Services Administration (SAMHSA) (https://www.samhsa.gov

100

APPENDIX C
THE ESTIMATED NUMBERS OF NEWLY INCIDENT HEROIN USERS BY HUNT’S
METHOD DURING 1950-2012 FOR MALE AND FEMALE

Figure C.1 The estimated numbers of newly incident heroin users by Hunt’s method during
1950-2012 for male and female. Data from United States, Treatment Episode Data Set Admissions, 1992-2012.

101

APPENDIX D
THE ESTIMATED NUMBERS OF NEWLY INCIDENT HEROIN USERS BY HUNT’S
METHOD DURING 1950-2012 FOR FOUR REGIONS (i.e., NORTHEAST, MIDWEST,
WEST, SOUTH)

Figure D.1 The estimated numbers of newly incident heroin users by Hunt’s method during
1950-2012 for four regions (i.e., Northeast, Midwest, West, South). Data from United States,
Treatment Episode Data Set -Admissions, 1992-2012.

102

BIBLIOGRAPHY

103

BIBLIOGRAPHY

Al-Tayyib, A. A., Koester, S., & Riggs, P. (2017). Prescription opioids prior to injection drug
use: Comparisons and public health implications. Addict Behav, 65, 224-228.
doi:10.1016/j.addbeh.2016.08.016
Anglin, M. D., Brecht, M. L., Woodward, J. A., & Bonett, D. G. (1986). An empirical study of
maturing out: conditional factors. Int J Addict, 21(2), 233-246.
Anthony, J. C. (2002). Death of the 'stepping-stone' hypothesis and the 'gateway' model?
Comments on Morral et al. Addiction, 97(12), 1505-1507.
Anthony, J. C. (2012). Steppingstone and gateway ideas: a discussion of origins, research
challenges, and promising lines of research for the future. Drug Alcohol Depend, 123
Suppl 1, S99-S104. doi:10.1016/j.drugalcdep.2012.04.006
Anthony, J. C., Neumark, Y. D., & Van Etten, M. L. (2000). Do I do what I say? A perspective
on self-report methods in drug dependence epidemiology (A. Stone, J. S. Turkan, C. A.
Bachrach, J. B. Jobe, & H. S. Kurtzman Eds.). Mahwah, NJ: Lawrence Erlbaum
Associates.
Anthony, J. C., & Van Etten, M. L. (1998). Comprehensive Clinical Psychology (A. Bellack &
M. Hersen Eds.): Oxford: Elsevier Science Publications.
Anthony, J. C., Warner, L. A., & Kessler, R. C. (1994). Comparative epidemiology of
dependence on tobacco, alcohol, controlled substances, and inhalants: Basic findings
from the National Comorbidity Survey. Experimental and Clinical Psychopharmacology,
2(3), 244-268.
Bailey, R. C., Hser, Y. I., Hsieh, S. C., & Anglin, M. D. (1994). Influences Affecting
Maintenance and Cessation of Narcotics Addiction. The Journal of Drug Issues, 24(2),
249-272.
Ball, J. C., & Chambers, C. D. (1970). The Epidemiology of Opiate Addiction in the United
States. Springfield, IL: Charles C Thomas.
Banerjee, G., Edelman, E. J., Barry, D. T., Becker, W. C., Cerda, M., Crystal, S., . . . Marshall,
B. D. (2016). Non-medical use of prescription opioids is associated with heroin initiation
among US veterans: a prospective cohort study. Addiction, 111(11), 2021-2031.
doi:10.1111/add.13491

104

Barrett, S. P., Meisner, J. R., & Stewart, S. H. (2008). What constitutes prescription drug
misuse? Problems and pitfalls of current conceptualizations. Curr Drug Abuse Rev, 1(3),
255-262.
BAYER Pharmaceutical Products. (1901). HEROIN - HYDROCHLORIDE. American Journal
of Pharmacy, 73.
Blanco, C., Iza, M., Rodriguez-Fernandez, J. M., Baca-Garcia, E., Wang, S., & Olfson, M.
(2015). Probability and predictors of treatment-seeking for substance use disorders in the
U.S. Drug Alcohol Depend, 149, 136-144. doi:10.1016/j.drugalcdep.2015.01.031
Blanco, C., Iza, M., Schwartz, R. P., Rafful, C., Wang, S., & Olfson, M. (2013). Probability and
predictors of treatment-seeking for prescription opioid use disorders: a national study.
Drug Alcohol Depend, 131(1-2), 143-148. doi:10.1016/j.drugalcdep.2012.12.013
Blatman, M. E. (1971). The Controlled Substances Act of 1970. Pa Med, 74(5), 45-46.
Boeri, M. W., Sterk, C. E., & Elifson, K. W. (2008). Reconceptualizing early and late onset: a
life course analysis of older heroin users. Gerontologist, 48(5), 637-645.
Bohnert, A. S., Valenstein, M., Bair, M. J., Ganoczy, D., McCarthy, J. F., Ilgen, M. A., & Blow,
F. C. (2011). Association between opioid prescribing patterns and opioid overdoserelated deaths. JAMA, 305(13), 1315-1321. doi:10.1001/jama.2011.370
Booth, M. (1999). Opium: A History. New York City, NY: St. Martin's Griffin.
Boyle, J. M., & Brunswick, A. F. (1980). What happened in Harlem? Analysis of a decline in
heroin use among a generation unit of urban Black youth. J Drug Issues, 10, 109-130.
Carise, D., Dugosh, K. L., McLellan, A. T., Camilleri, A., Woody, G. E., & Lynch, K. G. (2007).
Prescription OxyContin abuse among patients entering addiction treatment. Am J
Psychiatry, 164(11), 1750-1756. doi:10.1176/appi.ajp.2007.07050252
Carlson, R. G., Nahhas, R. W., Martins, S. S., & Daniulaityte, R. (2016). Predictors of transition
to heroin use among initially non-opioid dependent illicit pharmaceutical opioid users: A
natural history study. Drug Alcohol Depend, 160, 127-134.
doi:10.1016/j.drugalcdep.2015.12.026
Carpenter, M. J., Chutuape, M. A., & Stitzer, M. L. (1998). Heroin snorters versus injectors:
comparison on drug use and treatment outcome in age-matched samples. Drug Alcohol
Depend, 53(1), 11-15.
Cerda, M., Santaella, J., Marshall, B. D., Kim, J. H., & Martins, S. S. (2015). Nonmedical
Prescription Opioid Use in Childhood and Early Adolescence Predicts Transitions to
Heroin Use in Young Adulthood: A National Study. J Pediatr, 167(3), 605-612 e602.
doi:10.1016/j.jpeds.2015.04.071

105

Chein, I., Gerard, D. L., Lee, R. S., Rosenfeld, E., & Wilner, D. M. (1964). The Road to H:
Narcotics, Delinquency, and Social Policy. New York, NY: Basic Books.
Christin, N. (2013). Traveling the Silk Road: A Measurement Analysis of a Large Anonymous
Online Marketplace. Paper presented at the In: Proceedings of the 22nd International
Conference on World Wide Web.
Cicero, T. J., & Ellis, M. S. (2015). Abuse-Deterrent Formulations and the Prescription Opioid
Abuse Epidemic in the United States: Lessons Learned From OxyContin. JAMA
Psychiatry, 72(5), 424-430. doi:10.1001/jamapsychiatry.2014.3043
Cicero, T. J., Ellis, M. S., & Harney, J. (2015). Shifting Patterns of Prescription Opioid and
Heroin Abuse in the United States. N Engl J Med, 373(18), 1789-1790.
doi:10.1056/NEJMc1505541
Cicero, T. J., Ellis, M. S., & Surratt, H. L. (2012). Effect of abuse-deterrent formulation of
OxyContin. N Engl J Med, 367(2), 187-189. doi:10.1056/NEJMc1204141
Cicero, T. J., Ellis, M. S., Surratt, H. L., & Kurtz, S. P. (2014). The changing face of heroin use
in the United States: a retrospective analysis of the past 50 years. JAMA Psychiatry,
71(7), 821-826. doi:10.1001/jamapsychiatry.2014.366
Cicero, T. J., & Kuehn, B. M. (2014). Driven by prescription drug abuse, heroin use increases
among suburban and rural whites. JAMA, 312(2), 118-119. doi:10.1001/jama.2014.7404
Cicero, T. J., Surratt, H., Inciardi, J. A., & Munoz, A. (2007). Relationship between therapeutic
use and abuse of opioid analgesics in rural, suburban, and urban locations in the United
States. Pharmacoepidemiol Drug Saf, 16(8), 827-840. doi:10.1002/pds.1452
Clayton, R. R., & Voss, H. L. (1981). Young men and drugs in Manhattan: a causal analysis.
NIDA Res Monogr, 39, 1-187.
Compton, W. M., Jones, C. M., & Baldwin, G. T. (2016). Relationship between Nonmedical
Prescription-Opioid Use and Heroin Use. N Engl J Med, 374(2), 154-163.
doi:10.1056/NEJMra1508490
Compton, W. M., Thomas, Y. F., Stinson, F. S., & Grant, B. F. (2007). Prevalence, correlates,
disability, and comorbidity of DSM-IV drug abuse and dependence in the United States:
results from the national epidemiologic survey on alcohol and related conditions. Arch
Gen Psychiatry, 64(5), 566-576. doi:10.1001/archpsyc.64.5.566
Compton, W. M., & Volkow, N. D. (2006). Major increases in opioid analgesic abuse in the
United States: concerns and strategies. Drug Alcohol Depend, 81(2), 103-107.
doi:10.1016/j.drugalcdep.2005.05.009

106

Concool, B., Smith, H., & Stimmel, B. (1979). Mortality rates of persons entering methadone
maintenance: a seven-year study. Am J Drug Alcohol Abuse, 6(3), 345-353.
doi:10.3109/00952997909001723
Cooper, H. L., Bossak, B., Tempalski, B., Des Jarlais, D. C., & Friedman, S. R. (2009).
Geographic approaches to quantifying the risk environment: drug-related law
enforcement and access to syringe exchange programmes. Int J Drug Policy, 20(3), 217226. doi:10.1016/j.drugpo.2008.08.008
Courtwright, D. T. (1983). The hidden epidemic: opiate addiction and cocaine use in the South,
1860-1920. J South Hist, 49(1), 57-72.
Cox, D. R. (1972). Regression models and life tables. Journal of the Royal Statistical Society,
34(2), 187-220.
Dasgupta, N., Freifeld, C., Brownstein, J. S., Menone, C. M., Surratt, H. L., Poppish, L., . . .
Dart, R. C. (2013). Crowdsourcing black market prices for prescription opioids. J Med
Internet Res, 15(8), e178. doi:10.2196/jmir.2810
Defalque, R. J., & Wright, A. J. (2007). The early history of methadone. Myths and facts. Bull
Anesth Hist, 25(3), 13-16.
Degenhardt, L., Larance, B., Bruno, R., Lintzeris, N., Ali, R., & Farrell, M. (2015). Evaluating
the potential impact of a reformulated version of oxycodone upon tampering, nonadherence and diversion of opioids: the National Opioid Medications Abuse Deterrence
(NOMAD) study protocol. Addiction, 110(2), 226-237. doi:10.1111/add.12746
Degenhardt, L., Whiteford, H. A., Ferrari, A. J., Baxter, A. J., Charlson, F. J., Hall, W. D., . . .
Vos, T. (2013). Global burden of disease attributable to illicit drug use and dependence:
findings from the Global Burden of Disease Study 2010. Lancet, 382(9904), 1564-1574.
doi:10.1016/S0140-6736(13)61530-5
Dew, B., Elifson, K., & Dozier, M. (2007). Social and environmental factors and their influence
on drug use vulnerability and resiliency in rural populations. J Rural Health, 23 Suppl,
16-21. doi:10.1111/j.1748-0361.2007.00119.x
Dole, V. P., & Nyswander, M. (1965). A Medical Treatment for Diacetylmorphine (Heroin)
Addiction. A Clinical Trial with Methadone Hydrochloride. JAMA, 193, 646-650.
DuPont, R. L., & Greene, M. H. (1973). The dynamics of a heroin addiction epidemic. Science,
181(4101), 716-722.
Edlund, M. J., Booth, B. M., & Han, X. (2012). Who seeks care where? Utilization of mental
health and substance use disorder treatment in two national samples of individuals with
alcohol use disorders. J Stud Alcohol Drugs, 73(4), 635-646.

107

Feinberg, A. P., Creese, I., & Snyder, S. H. (1976). The opiate receptor: a model explaining
structure-activity relationships of opiate agonists and antagonists. Proc Natl Acad Sci U S
A, 73(11), 4215-4219.
Grau, L. E., Dasgupta, N., Harvey, A. P., Irwin, K., Givens, A., Kinzly, M. L., & Heimer, R.
(2007). Illicit use of opioids: is OxyContin a "gateway drug"? Am J Addict, 16(3), 166173. doi:10.1080/10550490701375293
Greene, M. H. (1974). An epidemiologic assessment of heroin use. Am J Public Health, 64 Suppl
12, 1-10.
Han, B., Compton, W. M., Jones, C. M., & Cai, R. (2015). Nonmedical Prescription Opioid Use
and Use Disorders Among Adults Aged 18 Through 64 Years in the United States, 20032013. JAMA, 314(14), 1468-1478. doi:10.1001/jama.2015.11859
Hasin, D. S., Stinson, F. S., Ogburn, E., & Grant, B. F. (2007). Prevalence, correlates, disability,
and comorbidity of DSM-IV alcohol abuse and dependence in the United States: results
from the National Epidemiologic Survey on Alcohol and Related Conditions. Arch Gen
Psychiatry, 64(7), 830-842. doi:10.1001/archpsyc.64.7.830
Havens, J. R., Oser, C. B., & Leukefeld, C. G. (2007). Increasing prevalence of prescription
opiate misuse over time among rural probationers. J Opioid Manag, 3(2), 107-111.
Hedegaard, H., Chen, L. H., & Warner, M. (2015). Drug-poisoning deaths involving heroin:
United States, 2000-2013. NCHS Data Brief, 190, 1-8.
Hser, Y. I., Huang, D., Chou, C. P., & Anglin, M. D. (2007). Trajectories of heroin addiction:
growth mixture modeling results based on a 33-year follow-up study. Eval Rev, 31(6),
548-563. doi:10.1177/0193841X07307315
Hughes, P. H., Barker, N. W., Crawford, G. A., & Jaffe, J. H. (1972). The natural history of a
heroin epidemic. Am J Public Health, 62(7), 995-1001.
Hughes, P. H., & Rieche, O. (1995). Heroin epidemics revisited. Epidemiol Rev, 17(1), 66-73.
Hunt, L. B., & Chambers, C. D. (1976). The heroin epidemics: a study of heroin use in the U.S.,
1965–75. Holliswood, NY: Spectrum.
Inciardi, J. A., & Harrison, L. D. (1998). Heroin in the Age of Crack-Cocaine (Vol. 6): SAGE
Publications, Inc.
Jones, C. M. (2013). Heroin use and heroin use risk behaviors among nonmedical users of
prescription opioid pain relievers - United States, 2002-2004 and 2008-2010. Drug
Alcohol Depend, 132(1-2), 95-100. doi:10.1016/j.drugalcdep.2013.01.007

108

Jones, C. M., Logan, J., Gladden, R. M., & Bohm, M. K. (2015). Vital Signs: Demographic and
Substance Use Trends Among Heroin Users - United States, 2002-2013. MMWR Morb
Mortal Wkly Rep, 64(26), 719-725.
Jones, C. M., Muhuri, P. K., & Lurie, P. G. (2016). Trends in the Nonmedical Use of OxyContin,
United States, 2006-2013. Clin J Pain. doi:10.1097/AJP.0000000000000426
Jordan, B. A., Cvejic, S., & Devi, L. A. (2000). Opioids and their complicated receptor
complexes. Neuropsychopharmacology, 23(4 Suppl), S5-S18. doi:10.1016/S0893133X(00)00143-3
Judson, B. A., Ortiz, S., Crouse, L., Carney, T. M., & Goldstein, A. (1980). A follow-up study of
heroin addicts five years after first admission to a methadone treatment program. Drug
Alcohol Depend, 6(5), 295-313.
Just, J. M., Bleckwenn, M., Schnakenberg, R., Skatulla, P., & Weckbecker, K. (2016). Drugrelated celebrity deaths: A cross-sectional study. Subst Abuse Treat Prev Policy, 11(1),
40. doi:10.1186/s13011-016-0084-z
Kandel, D. B., & Yamaguchi, K. (2002). Stages of drug involvement in the U.S. population. In
D. B. Kandel (Ed.), Stages and pathways of drug involvement: Examining the gateway
hypothesis (pp. 65-89). Cambridge: University Press.
Kelly, T. M., Cornelius, J. R., & Clark, D. B. (2004). Psychiatric disorders and attempted suicide
among adolescents with substance use disorders. Drug Alcohol Depend, 73(1), 87-97.
Kessler, R. C., Zhao, S., Katz, S. J., Kouzis, A. C., Frank, R. G., Edlund, M., & Leaf, P. (1999).
Past-year use of outpatient services for psychiatric problems in the National Comorbidity
Survey. Am J Psychiatry, 156(1), 115-123. doi:10.1176/ajp.156.1.115
Keyes, K. M., Cerda, M., Brady, J. E., Havens, J. R., & Galea, S. (2014). Understanding the
rural-urban differences in nonmedical prescription opioid use and abuse in the United
States. Am J Public Health, 104(2), e52-59. doi:10.2105/AJPH.2013.301709
Khantzian, E. J., & Treece, C. (1985). DSM-III psychiatric diagnosis of narcotic addicts. Recent
findings. Arch Gen Psychiatry, 42(11), 1067-1071.
Krebs, E., Min, J. E., Evans, E., Li, L., Liu, L., Huang, D., . . . Nosyk, B. (2016). Estimating
State Transitions for Opioid Use Disorders. Med Decis Making, 272989X16683928.
doi:10.1177/0272989X16683928
Lankenau, S. E., Teti, M., Silva, K., Jackson Bloom, J., Harocopos, A., & Treese, M. (2012).
Initiation into prescription opioid misuse amongst young injection drug users. Int J Drug
Policy, 23(1), 37-44. doi:10.1016/j.drugpo.2011.05.014

109

Lipari, R. N., & Hughes, A. (2015). The NSDUH Report: Trends in Heroin Use in the United
States: 2002 to 2013. The CBHSQ Report: April 23, 2015. Substance Abuse and Mental
Health Services Administration, Center for Behavioral Health Statistics and Quality.
Retrieved from Rockville, MD.:
https://www.samhsa.gov/data/sites/default/files/report_1943/ShortReport-1943.pdf
Lofwall, M. R., Brooner, R. K., Bigelow, G. E., Kindbom, K., & Strain, E. C. (2005).
Characteristics of older opioid maintenance patients. J Subst Abuse Treat, 28(3), 265272. doi:10.1016/j.jsat.2005.01.007
Maclure, M. (1991). The case-crossover design: a method for studying transient effects on the
risk of acute events. Am J Epidemiol, 133(2), 144-153.
Mars, S. G., Bourgois, P., Karandinos, G., Montero, F., & Ciccarone, D. (2014). "Every 'never' I
ever said came true": transitions from opioid pills to heroin injecting. Int J Drug Policy,
25(2), 257-266. doi:10.1016/j.drugpo.2013.10.004
Martins, S. S., Sarvet, A., Santaella-Tenorio, J., Saha, T., Grant, B. F., & Hasin, D. S. (2017).
Changes in US Lifetime Heroin Use and Heroin Use Disorder: Prevalence From the
2001-2002 to 2012-2013 National Epidemiologic Survey on Alcohol and Related
Conditions. JAMA Psychiatry. doi:10.1001/jamapsychiatry.2017.0113
Martins, S. S., Storr, C. L., Zhu, H., & Chilcoat, H. D. (2009). Correlates of extramedical use of
OxyContin versus other analgesic opioids among the US general population. Drug
Alcohol Depend, 99(1-3), 58-67. doi:10.1016/j.drugalcdep.2008.06.013
McGlothlin, W. H., & Anglin, M. D. (1981). Long-term follow-up of clients of high- and lowdose methadone programs. Arch Gen Psychiatry, 38(9), 1055-1063.
Morral, A. R., McCaffrey, D. F., & Paddock, S. M. (2002). Reassessing the marijuana gateway
effect. Addiction, 97(12), 1493-1504.
Muhuri, P. K., Gfroerer, J. C., & Davies, M. C. (2013). Associations of nonmedical pain reliever
use and initiation of heroin use in the United States: Substance Abuse and Mental Health
Services Administration.
Neaigus, A., Gyarmathy, V. A., Miller, M., Frajzyngier, V. M., Friedman, S. R., & Des Jarlais,
D. C. (2006). Transitions to injecting drug use among noninjecting heroin users: social
network influence and individual susceptibility. J Acquir Immune Defic Syndr, 41(4),
493-503. doi:10.1097/01.qai.0000186391.49205.3b
Neaigus, A., Miller, M., Friedman, S. R., Hagen, D. L., Sifaneck, S. J., Ildefonso, G., & des
Jarlais, D. C. (2001). Potential risk factors for the transition to injecting among noninjecting heroin users: a comparison of former injectors and never injectors. Addiction,
96(6), 847-860. doi:10.1080/09652140020050960

110

Nurco, D. N., Cisin, I. H., & Balter, M. B. (1981). Addict careers. I. A new typology. Int J
Addict, 16(8), 1305-1325.
O'Brien, M. S., Comment, L. A., Liang, K. Y., & Anthony, J. C. (2012). Does cannabis onset
trigger cocaine onset? A case-crossover approach. Int J Methods Psychiatr Res, 21(1),
66-75. doi:10.1002/mpr.359
O'Donnell, J. A., Voss, H. L., Clayton, R. R., Slatin, G. T., & Room, R. G. (1976). Young men
and drugs-a nationwide survey. NIDA Res Monogr(5), i-xiv, 1-144.
O'Driscoll, P. T., McGough, J., Hagan, H., Thiede, H., Critchlow, C., & Alexander, E. R. (2001).
Predictors of accidental fatal drug overdose among a cohort of injection drug users. Am J
Public Health, 91(6), 984-987.
O'Grady, C. L., Surratt, H. L., Kurtz, S. P., & Levi-Minzi, M. A. (2014). Nonmedical
prescription drug users in private vs. public substance abuse treatment: a cross sectional
comparison of demographic and HIV risk behavior profiles. Subst Abuse Treat Prev
Policy, 9, 9. doi:10.1186/1747-597X-9-9
Ochoa, K. C., Hahn, J. A., Seal, K. H., & Moss, A. R. (2001). Overdosing among young
injection drug users in San Francisco. Addict Behav, 26(3), 453-460.
Olfson, M., Kessler, R. C., Berglund, P. A., & Lin, E. (1998). Psychiatric disorder onset and first
treatment contact in the United States and Ontario. Am J Psychiatry, 155(10), 1415-1422.
doi:10.1176/ajp.155.10.1415
Palamar, J. J., Shearston, J. A., Dawson, E. W., Mateu-Gelabert, P., & Ompad, D. C. (2016).
Nonmedical opioid use and heroin use in a nationally representative sample of us high
school seniors. Drug Alcohol Depend, 158, 132-138.
doi:10.1016/j.drugalcdep.2015.11.005
Paulozzi, L. J., Budnitz, D. S., & Xi, Y. (2006). Increasing deaths from opioid analgesics in the
United States. Pharmacoepidemiol Drug Saf, 15(9), 618-627. doi:10.1002/pds.1276
Paulozzi, L. J., & Xi, Y. (2008). Recent changes in drug poisoning mortality in the United States
by urban-rural status and by drug type. Pharmacoepidemiol Drug Saf, 17(10), 997-1005.
doi:10.1002/pds.1626
Pugatch, D., Strong, L. L., Has, P., Patterson, D., Combs, C., Reinert, S., . . . Brown, L. (2001).
Heroin use in adolescents and young adults admitted for drug detoxification. J Subst
Abuse, 13(3), 337-346.
Regier, D. A., Narrow, W. E., Rae, D. S., Manderscheid, R. W., Locke, B. Z., & Goodwin, F. K.
(1993). The de facto US mental and addictive disorders service system. Epidemiologic
catchment area prospective 1-year prevalence rates of disorders and services. Arch Gen
Psychiatry, 50(2), 85-94.

111

Rhumorbarbe, D., Staehli, L., Broseus, J., Rossy, Q., & Esseiva, P. (2016). Buying drugs on a
Darknet market: A better deal? Studying the online illicit drug market through the
analysis of digital, physical and chemical data. Forensic Sci Int, 267, 173-182.
doi:10.1016/j.forsciint.2016.08.032
Robins, L. N., Davis, D. H., & Goodwin, D. W. (1974). Drug use by U.S. Army enlisted men in
Vietnam: a follow-up on their return home. Am J Epidemiol, 99(4), 235-249.
Rounsaville, B. J., Kosten, T., & Kleber, H. (1985). Success and failure at outpatient opioid
detoxification. Evaluating the process of clonidine- and methadone-assisted withdrawal. J
Nerv Ment Dis, 173(2), 103-110.
Rudd, R. A., Paulozzi, L. J., Bauer, M. J., Burleson, R. W., Carlson, R. E., Dao, D., . . .
Fernandes, J. C. (2014). Increases in Heroin Overdose Deaths — 28 States, 2010 to 2012.
MMWR, 63, 849-876.
Rutherford, M. J., Cacciola, J. S., & Alterman, A. I. (1994). Relationships of personality
disorders with problem severity in methadone patients. Drug Alcohol Depend, 35(1), 6976.
Seal, K. H., Kral, A. H., Gee, L., Moore, L. D., Bluthenthal, R. N., Lorvick, J., & Edlin, B. R.
(2001). Predictors and prevention of nonfatal overdose among street-recruited injection
heroin users in the San Francisco Bay Area, 1998-1999. Am J Public Health, 91(11),
1842-1846.
Tofighi, B., Perna, M., Desai, A., Grov, C., & Lee, J. D. (2016). Craigslist as a source for heroin:
a report of two cases. Journal of Substance Use, 21(5), 543-546.
Unick, G., Rosenblum, D., Mars, S., & Ciccarone, D. (2014). The relationship between US
heroin market dynamics and heroin-related overdose, 1992-2008. Addiction, 109(11),
1889-1898. doi:10.1111/add.12664
United States. Substance Abuse and Mental Health Services Administration. (2010). Reliability
of Key Measures in the National Survey on Drug Use and Health. Office of Applied
Studies, Methodology Series M-8, HHS Publication No. SMA 09-4425.
United States. Substance Abuse and Mental Health Services Administration. (2012). Center for
Behavioral Health Statistics and Quality. Results from the 2011 National Survey on Drug
Use and Health: Summary of National Findings. NSDUH Series H-44, HHS Publication
No. (SMA) 12-4713. Rockville, MD: Substance Abuse and Mental Health Services
Administration. Retrieved from
http://www.samhsa.gov/data/nsduh/2k11results/nsduhresults2011.htm

112

United States. Substance Abuse and Mental Health Services Administration. (2014a). Results
from the 2013 National Survey on Drug Use and Health: Summary of National Findings,
NSDUH Series H-48, HHS Publication No. (SMA) 14-4863. Retrieved from
http://archive.samhsa.gov/data/NSDUH/2013SummNatFindDetTables/NationalFindings/
NSDUHresults2013.htm
United States. Substance Abuse and Mental Health Services Administration. (2014b). Treatment
Episode Data Set (TEDS): 2002-2012. National Admissions to Substance Abuse
Treatment Services. Retrieved from Rockville, MD:
http://www.samhsa.gov/data/sites/default/files/TEDS2012N_Web.pdf
UNODC. (2012). United Nations Office on Drugs and Crime, World Drug Report.
Vaillant, G. E. (1973). A 20-year follow-up of New York narcotic addicts. Arch Gen Psychiatry,
29(2), 237-241.
Van Hout, M. C., & Bingham, T. (2013). 'Silk Road', the virtual drug marketplace: a single case
study of user experiences. Int J Drug Policy, 24(5), 385-391.
doi:10.1016/j.drugpo.2013.01.005
Van Hout, M. C., & Hearne, E. (2017). New psychoactive substances (NPS) on cryptomarket
fora: An exploratory study of characteristics of forum activity between NPS buyers and
vendors. Int J Drug Policy, 40, 102-110. doi:10.1016/j.drugpo.2016.11.007
Wang, P. S., Berglund, P., Olfson, M., Pincus, H. A., Wells, K. B., & Kessler, R. C. (2005).
Failure and delay in initial treatment contact after first onset of mental disorders in the
National Comorbidity Survey Replication. Arch Gen Psychiatry, 62(6), 603-613.
doi:10.1001/archpsyc.62.6.603
White, J. M., & Irvine, R. J. (1999). Mechanisms of fatal opioid overdose. Addiction, 94(7), 961972.
World Health Organization. (2009). Guidelines for the psychosocially assisted pharmacological
treatment of opioid dependence. Retrieved from
http://www.who.int/substance_abuse/publications/opioid_dependence_guidelines.pdf
Young, A. M., & Havens, J. R. (2012). Transition from first illicit drug use to first injection drug
use among rural Appalachian drug users: a cross-sectional comparison and retrospective
survival analysis. Addiction, 107(3), 587-596. doi:10.1111/j.1360-0443.2011.03635.x
Young, A. M., Havens, J. R., & Leukefeld, C. G. (2010). Route of administration for illicit
prescription opioids: a comparison of rural and urban drug users. Harm Reduct J, 7, 24.
doi:10.1186/1477-7517-7-24

113