WWWWWII

”WNW

Wi

III‘

137
533
THS

 

 

 

This is to certify that the

thesis entitled

ESTIMATION OF SMALL AREA SMOKING PREVALENCE IN PRIMARY
CARE MEDICAL PRACTICES: A LOGISTIC MODEL

presented by

David P. Weismantel, M.D.

has been accepted towards fulfillment
of the requirements for

M. S . degree in Epidemiology

 

/
CPJC @Uttdtp

Major professor

v7

9 M3 ’6)

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

LIBRARY
Michigan State
University

 

 

 

PLACE IN RETURN Box to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJClRC/DateDuo.p65-p.15

IESTTNLAIICIbICfl?ShdAll.fiUKEl\Sthfl§IhK3IHKE\DALEHVC]3IPJIWUHAFUKYWCAJKE
MEDICAL PRACTICES: A LOGISTIC MODEL

By

David Paul Weismantel, MD.

A THESIS

Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of

MASTER OF SCIENCE
Department of Epidemiology

2003

ABSTRACT

ESTIMATION OF SMALL AREA SMOKING PREVALENCE IN PRIMARY CARE
MEDICAL PRACTICES: A LOGISTIC MODEL

By

David Paul Weismantel, M.D.

Data from a controlled trial within 87 primary care medical practices, the US.
Bureau of the Census, and the Michigan Department of Community Health was used to
estimate a small area smoking prevalence for each participating site.

The smoking prevalence within this medical practice cohort as estimated from a one-

day exit survey was 18.6 i 1.8% (95% confidence interval, (I) with a range of 0% to 60%,

significantly lower than the reported statewide smoking prevalence in Michigan of 26.1 i
1.8% (95% CI). With a binomial proportion of smokers obtained from the patient exit
survey within each of the practices serving as the dependent variable, a multivariate logistic
regression model was constructed. Significant predictors of practice-level smoking
prevalence within the final model were the estimated regional smoking prevalence from the
community health department, proportion uninsured and Medicaid coverage within each of
the practices, and the average patient age as determined from the exit survey. After
application of the resultant regression coefficients to individual practice site data, the overall
estimated smoking prevalence remained 18.6%, but with decreased range of 8.8% to 38.4%.
These results should be viewed and interpreted with caution as this model has not been

validated against more precise determinations of smoking prevalence at the practice level.

I Ii. Ii t“.'fi"

Copyright by
David Paul Weismantel, MD.
2003

ACKNOWLEDGEMENTS

I wish to thank Pramod K. Pathak, PhD., William C Wadland, M.D., MS, and
Nigel Paneth, M.D., MPH. who have provided invaluable support of my education and
development. Dr. Pathak has guided my work, demonstrated a quiet confidence in my
abilities, and helped me view problem as potential opportunities for growth. Dr. Wadland
initially identified my potential and interest in research; he has remained a steadfast advocate
by allowing me to serve as both epidemiologist and statistician. My TRECIDS fellowship
under the direction of Dr. Paneth and the faculty of the Department of Epidemiology at
Michigan State University allowed me to develop skills and confidence to further pursue a
research career.

I must also thank my wife, Arlene, and daughter, Miranda, for their love, patience,

and understanding of missed dinners and long nights.

iv

TABLE OF CONTENTS

Objectives

Data Management and Statistical Analysis
Protection of Human Subjects

Study Design

Model Selection

Variable Selection

Multiple Imputation of Missing Values
Calculation of Practice-Level Smoking Prevalence

15

32

35

37

46

LIST OF TABLES

Table 1

Relative Efficiencyof Multiple Imputation...

Table 2

US. Census 2000 Demographic Variables for Practice Site... ..

Table 3

Michigan Department of Community Health Variables for Practice Location... .

Table 4

Practice SurveyDemographic Variables... .. . ....

Table 5

Patient Exit SurveyDemographic Variables...

Table 6
Univariate Analysis of Factors within a Binomial Logistic Regression Model
of Practice Smoking Prevalence... .

 

Table 7

Initial Multivariate Regression Model of Practice Smoking Prevalence _

with Parameter Estimates...

Table 8

Multiple Imputation Variance Information...

Table 9
Final Multivariate Regression Model of Practice Smoking Prevalence

with Multiple Imputation Parameter Estimates and Odds Ratios... ..

vi

14

18

18

19

.. 20

.. 21

.22

22

23

LIST OF FIGURES

Figure 1

Geographic Distribution of StudyPractice Sites within Michigan... 7
Figure 2

Distribution of Exit SurveySample Size... 15
Figure 3

Distribution of Survey Smoking Prevalence within Primary Care Medical Practices . 16
Figure 4

Log Odds (Logit) Transformation of Exit Survey Smoking Prevalence ...................... 17
Figure 5 ‘

Age of Practice Population as a Function of Practice Specialty..... 24
Figure 6

Practice Smoking Prevalence - Logistic Model Estimate... 25
Figure 7

Logistic Model Deviance Residual Values vs. Predicted Smoking Prevalence ............... 26
Figure 8

Distribution of Exit Survey and Model Estimate Smoking Prevalence

bySurveySample Size... 27
Figure 9

Smoking Prevalence by Practice Site: Exit Survey and Model Estimates ..................... 28
Figure 10

Model Estimate Smoking Prevalence as a Function of

MDCH Regional Smoking Prevalence Estimate... 29
Figure 11

Model Estimate Smoking Prevalence as a Function of

Uninsured 8CMedicaid Proportion Coverage... .. 30
Figure 12

Model Estimate Smoking Prevalence as a Function of

Average Age of Adult Patients Within Practice... .. 30
Figure 13

Baseline Provider Referral Rates by Practice Site: Exit Survey and Model Estimates ...... 31

vii

INTRODUCTION

Tobacco smoking is an important risk factor for many diseases including coronary
artery disease, peripheral vascular disease, chronic obstructive pulmonary disease, and
transitional cell carcinoma of the bladder. The direct cost in human suffering caused by
tobacco related diseases is quite evident to patients, families, and physicians. The World
Health Organization (WHO) estimates that 90% of lung cancer, up to 20% of other cancers,
75% of chronic bronchitis and emphysema, and 25% of deaths from cardiovascular disease
at ages 35 to 69 years are attributable to tobacco (1). The financial burden of these illnesses
is shared by all of society. An analysis by the Centers for Disease Control and Prevention
and the University of California estimated $50 billion were spent on direct medical costs
attributable to tobacco use. This does not include medical care for diseases caused by
second-hand smoke, fetal complications of smoking in pregnancy, or burn care for smoking-
related fires (2). Indirect costs to society arise from lost economic productivity due to illness
and premature death (3).

Smoking cessation interventions have been shown to be both efficacious and cost-
effective in primary care settings according to a systematic review of controlled studies
undertaken by the task force for the Agency for Health Care Research and Quality (AHRQ)
guidelines on smoking cessation (4). These guidelines substantiate the value of identifying all
smokers, advising them to quit, assisting those who are ready to quit, and eventually
arranging follow-up care. Brief smoking cessation advice by the physician alone results in
long-term smoking quit rates of less than ten percent. Smoking risk identification and

cessation rates are enhanced by more systematic approaches in primary medical care (5).

Cognitive behavioral therapies, delivered primarily in groups and supplemented with
nicotine replacement, report some of the highest long-term cessation rates in practice (6,7).
These services typically offer coping strategies to prevent relapse and attend to follow-up
care; however; less than 5% of smokers will ultimately accept a referral and attend group
sessions (8).

Surveys of primary care physicians demonstrate that they understand the importance
of smoking cessation and espouse its value, yet the actual implementation of the key
elements of practice-based methods for smoking cessation remains quite limited in primary
care medical settings (9,10). Prior to the AHRQ guidelines, the National Cancer Institute
advocated the value of brief counseling for smoking cessation (11). In 1989, Cummings et.
al. reported that many physicians in both private and managed care practices never use these
recommended strategies, with 33-44% never advising on quit dates, 27-48% never assisting
with self-help materials, and 68-75% never arranging follow-up care (9). A subsequent
national survey of office-based physicians in 1998 showed little interim change in physician
practice patterns regarding smoking cessation (10). The physicians identified patient
smoking status at only 67% of visits and showed no change between 1991 and 1996.
Specialist physicians generally performed worse than primary care physicians on the
recommended components. Although follow-up care clearly enhances long-term success
rates, physicians rarely schedule smokers for follow-up smoking cessation visits. 'Ihese
clinical performance surveys clearly demonstrate that physicians fall short of the AHRQ
guidelines and the national goals for a more healthy Armrican population.

A new model for enhanced primary care (12) may be needed for physicians to meet
advocated standards in smoking cessation and other preventive health measures. Systematic

reviews of interventions to improve physician behavior in health screening and preventive

services do not suggest a single or simple solution. Traditional continuing medical education
(CME) programs improve short-term knowledge and performance, yet long-term behavior
remains unaffected (13). Provision of office-based systems, team support, and specific
feedback information on actual performance in preventive care have been shown to enhance
the long-term performance of physicians (14-16).

Several community and health system-based studies have demonstrated that
relatively high long-term smoking cessation rates of 20-36% may be achieved by combining
physician identification, advice, and referral for follow-up care with a telephone-support
counseling service (17-20). These services not only offer proactive follow-up care, but also
integrate some elements of relapse prevention counseling known to be helpful in cognitive
group therapy. By combining the systematic supports of an organized health care system
with accessible counseling services and specific feedback on referral rates and outcomes,
physicians will be more able to meet the AHRQ guidelines on smoking cessation.
Promoting maintenance of behavior change in large populations is best delivered by an
integrated and comprehensive approach across a system of care.

Specific feedback to providers regarding referral rates to smoking cessation
counseling services requires rather specific practice- and provider-level information. Work
intensity and a practice-level smoking prevalence are essential variables needed to construct a
valid measure of smoker contact opportunities for intervention or referral. Although
providers are able to accurately estimate a simple number of patients seen in a day or
number of days worked in a typical week, they have a limited sense of the smoking
prevalence within their own practices. The literature is limited regarding the estimation of

any health behavior at the medical practice level.

Previously described risk factors or predictors of smoking are race, gender, age, and
socioeconomic status. Tobacco use and smoking also vary within and among racial and
ethnic groups: American Indians have the highest prevalence; African American and
Southeast Asian men also have a relatively high prevalence of smoking. Asian American and
Hispanic women have the lowest prevalence. In most racial/ ethnic groups, men have a
higher prevalence of cigarette smoking (21-23). Smoking prevalence is noted to decrease
with increasing age. Although lower socioeconomic status has been found to be one of the
best predictors of smoking prevalence, there exists a complex interaction of race with
socioeconomic status (23-28). Any attempt to predict or focus effort upon populations at
increased of smoking should first account for those members with lower educational and

socioeconomic status.

METHODS

Objectives

The principle objectives of this study are to identify the significant risk factors for
smoking within a cohort of primary care medical practices and then model these associations
within a binomial logistic model to provide a refined small area or practice-level smoking
prevalence estimate for each practice. Furthermore these model estimates of smoking
prevalence will be compared to the initial exit survey estimates in order to describe the

potential effect of this procedure.

Data Management 8!. Statistical Ana_ly_sis

All data management, transformations, and analyses were conducted using SAS
version 8.2 (SAS Institute Inc, Cary, NC).

The LOGISTIC procedure was used to build, refine, and apply the binomial logistic
regression model of practice smoking prevalence. The M1 and MIANALYZE procedures
were used to multiply irnpute missing data and appropriately combine the model parameter
estimates from each of the imputed data sets.

The frequencies of responses for categorical variables and the means, standard
deviations, and distributions for continuous variables were initially determined. The
frequency of unknown or missing responses was counted for each variable, and steps were
taken to assess patterns of missing data to determine if methods should be used to exclude

cases or fields as they were, or irnpute additional information using standard procedures.

Protection of Human Subjects
The study protocol (# 01-789) was reviewed and approved by the Michigan State

University Committee on Research Involving Human Subjects (UCRIHS). (Appendix 1)

Study Desrg' n
This exercise was completed within the Evaluation of Organizational Oranges to

Promote Smoking Cessation within Managed Care Study (Robert Wood Johnson
Foundation: Grant # 43968), a collaborative effort of Michigan State University and Blue
Cross Blue Shield of Michigan. The study is an l8-month, communitybased, randomized,
and controlled trial designed to evaluate the effect of a targeted and comparative provider
feedback intervention upon smoking cessation counseling and referral behaviors within
primary care medical practices. Figure 1 illustrates the geographic distribution of

participating practices within the state of Michigan.

Emmi.
Geographic Distribution of Study Practice Sites within Michigan

0 Study Practice Sites

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

A total of 87 primary care practices including 308 providers in 39 counties were ultimately
enrolled in the study.

Practice, provider, and one-day patient exit surveys (Appendices 2 - 4) were
completed for each participating practice during a 6-month period prior to the start of the
randomized intervention in January 2003. The practice surveys were completed by the
business or practice managers while the provider surveys were completed by the
participating physicians, nurse practitioners, and physician assistants. The one-day patient
exit survey attempted to query all patients presenting to a participating clinic after their clinic
visits.

The primary study measure is the number of referrals to a designated telephone
smoking cessation counseling service with the medical practice as the primary unit of
analysis. Although the referrals are pooled for analysis between groups, a quarterly feedback
report was designed to reflect individual provider behavior and practice pattems. To this
end, an estimated referral rate was designed to report the number of referrals completed for
every 100 smoker visits. The numerator, or number of referrals, is obtained from BCBSM
on a quarterly basis for each provider within the study. The denominator, or estimated
smoker visits per quarter, is a synthesis of information obtained through survey of each
provider and an estimate of a practice-level smoking prevalence. The following formulas
describe the calculation of this denominator:

Quartedy Smoker Visits =-

Adjusted Adult Visits/ Day * Days Worked/ Quarter * Smoking Prevalence
1212}:
Adjusted Adult Visits - Total Visits - Pediatric Visits - (0.25 * Obstetric Visits)
The adjustment for obstetrics is designed to correct for the relatively increased frequency of

visits by obstetrics patients as compared to the more general primary care medical

population.

Model Selection

Although the exit survey queried patients regarding their smoking status, the
relatively small sample sizes in many of the practice sites limited its use as a practice-level
estimate of smoking prevalence. The provider survey queried physicians regarding their
estimate of smoking prevalence within the practice, yet this correlated poorly with regional
and exit survey results. In order to calculate more specific and accurate referral rates, a snnll
area smoking prevalence was estimated for each practice site through the application of a
binomial logistic regression model.

Smoking prevalence, most commonly viewed as a binary variable, lends itself to
analysis through logistic regression, investigating the relationship between discrete responses
and a set of explanatory variables. For binary response models, the response, Y, of an
individual maytake one of two possible discrete values. Suppose x is a vector of explanatory
variables and p - Pr(Y - 1 | x) is the response probability to be modeled. The linear
logistic model has the general form:

logit(p) E log (TI-’71?) = a + War

where a is the intercept parameter and fl is the vector of slope parameters. Logistic models
enable the specification of both continuous and categorical explanatory variables. It should
be noted the response probability p within a logistic model is bounded by the values of 0 and
1. For these reasons and since the unit of analysis for the smoking prevalence is the medical
practice, a logistic regression model was used to calculate the parameter estimates and odds
ratios for a variety of potential explanatory variables and their association with a practice-

level smoking prevalence.

Variable Selection

Potential explanatory variables were assembled from several levels with regard to the
primary care medical practice:

1. 2000 US. Census Data (29)

a. Per Capita Income (5)
i. County
ii. City
b. Prevalence of Poverty (%)
i. County
ii. City
c. Population Density (Persons per Square Mile)

2. Michigan Department of Community Health (MDCH)
Behavioral Risk Factor Surveillance System (30)

a. Designated Health Region
(Categorical: 1-12)
b. Regional Smoking Prevalence 1995-99 (%)

3. Practice Survey

a. Practice Type
(Categorical: Solo vs. Group)
b. Practice Specialty
(Categorical: Family Practice, Internal Medicine, Obstetrics/ Gynecology)
c. Estimation of Population Age
(Proportion Adults 265 Years of Age)
d. Estimated Proportions of Race/ Ethnicity within the Practice
(White, African American, Hispanic)
e. Estimated Proportions of Insurance Coverage within the Practice
(Private Prepaid, Private Fee-For-Service, Medicare, Medicaid, Uninsured)

4. Patient Exit Survey

a. Current Smoking Status
(Categorical: Yes / No)
b. Age (Years)
c. Gender
(Categorical: Male/ Female)
d. Race/ Ethnicity
(Categorical: White, African American, Hispanic, Asian, Native American)

10

After logarithmic or square root transformation of some positivelyskewed variables
to better approximate a normal distribution, all potential continuous and categorical
explanatory variables were evaluated within a univariate logistic model to screen for potential
inclusion within a subsequent multivariate model. All variables with a threshold value of
P 9.25 were initially entered into the model, with variables of highest variance removed one
at a time in a backward fashion and the model recalculated until all remaining variables had
P $.05. The null hypothesis that the odds ratio of the variable was equal to 1 was rejected if
the probability P of the association was found to be equal or less than 0.05 in this analysis
(or-0.05). Confidence intervals for the regression parameters and odds ratios were calculated

based on the profile likelihood function (31,32).

Multiple Imputation of Missmg' Values

Since 75 of the 87 practice site units were missing data for at least one of the
significant variables and were excluded from the initial analysis as incomplete cases in a list-
wise fashion, a method to complete the data set and further refine the smoking prevalence
estirrntes was needed. A potential strategy for handling missing data is simple imputation, in
which one substitutes a value for each missing value. For example, each missing value can
be imputed with the variable mean of the completed cases, or it can be imputed with the
variable mean conditional on observed values of other variables. This approach treats
missing values as if they were known in the complete-data analysis. Single imputation does
not convey the uncertainty about the predictions of the unknown missing values, and the
resulting estimated variances of the parameter estimates will be biased toward zero (33-39).

Instead of filling in a single value for each missing value, multiple imputation

replaces each missing value with a set of plausible values that represent the uncertainty about

11

the correct value to irnpute. The multiply imputed data sets are then analyzed by using
standard procedures for complete data and combining the results of these analyses. Multiple
imputation does not attempt to estimate each missing value through simulated values but
rather to represent a random sample of the missing values. This process results in valid
statistical inferences that appropriately reflect the uncertainty due to missing values; for
example, confidence intervals with the correct probability coverage. It should be noted that
a base assumption for multiple imputation is that the data is missing at random (MAR). By
definition, the missing data for variable Y are rrrissing at random if the probability of missing
data on Y is unrelated to the value of Y, after controlling for other variables in the analysis.
Multiple imputation inference involves three distinct phases:

1. The missing data are filled in mtimes to generate mcomplete data sets.

2. The mcomplete data sets are analyzed by using standard statistical analyses.

3. The results from the 772 complete data sets are combined to produce inferential

results.

With 772 irnputations, mdifferent sets of the point and variance estimates for a parameter Q
can be computed. Let Q and U, be the point and variance estimates from the i-th imputed
data set, i=1, 2, , 771. Then the combined point estimate for erom multiple imputation is

the average of the mcomplete-data estimates:
1 m .
Q = ‘7; Z; Qz'
1:

Let U be the within-imputation variance which is the average of the m complete-data

estimates:

1 m
e=_ze.

12

and B be the between imputation variance

Then the variance estimate associated with Qis the total variance
—— 1
T = U + (1 + —)B
m

The degrees of freedom 1;, depends on mand the ratio

(1 +m"1)B
U

 

7‘:

The ratio r is called the relative increase in variance due to nonresponse. When there is no
missing information about Q, the values of r and B are both zero. With a large value of mor
a small value of r , the degrees of freedom vwill be large. Another useful statistic is the
fraction of missing information about Q.

r+2/(v+3)

3,:
r'+1

 

The relative efficiency (RE) of using the finite m imputation estimator, rather than using an
infinite number for the fully efficient imputation, in units of variance, is approximately a
function of mand A

RE = (1+ —’\—)-I
m

The following table shows the relative efficiencies with different values of m and A For

cases with little missing information, only a small number of irnputations are necessary.

13

Table 1.
Relative Efficiency of Multiple Imputation

 

A

 

10%

20%

30%

50%

70%

 

gswwa

0.9677
0.9804
0.9901
0.9950

0.9375
0.9615
0.9804
0.9901

0.9091
0.9434
0.9709
0.9852

0.8571
0.9091
0.9524
0.9756

0.8108
0.8772
0.9346
0.9662

 

 

 

 

With approximately 10% (N =11) of the variable for uninsured 8?. Medicaid status
missing and the values of exit survey variables from a single practice, multiple imputation
was performed with the MI procedure with the number of imputations Ira-=10. All other
variables with significant association to the dependent variable of smoking prevalence with
P $.25 were included in the imputation procedure. The MIANALYZE procedure was then
used to combine the results of individual binomial logistic regression analyses for the 10

multiply imputed data sets.

Calculation of Practice-Level Smolgrg Prevalence

The resultant parameter estimates were then applied to each individual practice site
in order to calculate a model estimate smoking prevalence. Subsequently, this model
estimate was directly compared to the initial exit survey estimate, and the baseline provider

referral counts were used to calculate referral rates based upon a uniform or model estimate

smoking prevalence.

14

RESULTS

The exit survey in 86 of the 87 participating primary care practices yielded the
proportion of smokers, the dependent variable upon which potentially predictive variables
were modeled. Of the 3,619 eligible adults present during the 1-day survey, a total of 3,180
(87.9%) were approached and 1,966 (54.3%) agreed to complete the exit survey. Figure 1
illustrates the wide range and distribution of the exit survey sample size within these primary

care practices.

Em.
Distribution of Exit Survey Sample Size

 

35

 

 

 

 

10-

 

 

 

 

 

 

 

 

 

 

 

 

 

I l I l l l

25 35 45 55 65
Sample Size N

Ul—l
y—b
0|

 

Sample (N) Mean Sample Size Standard Deviation Minimum Maximum

 

 

86 22.9 13.6 4 65

 

15

Evaluation and analysis of the practice smoking prevalence revealed a distribution
approximating lognorrml with a weighted mean of 0.186 and an estimate of 0 for 2 of the 86

surveyed practice sites (Figure 2).

figure—3.
Distribution of Survey Smoking Prevalence within Primary Care Medical Practices

 

40—.

35-

 

 

 

10~ , \

s— \

 

 

 

 

 

 

 

 

 

 

 

 

 

 

T I l

l I
0 0.1 0.2 0.3 0.4 0.5 0.6
Smoking Prevalence

 

Sanuale (N) Weighted Mean Standard Deviation Minimum Maximum
86 0.186 0.047 0 0.6

 

16

A natural log transformation of the odds (logit) of the surveyed smoking status of
patients within each of the practice sites demonstrated a nearly symmetric distribution
approximating normality (Figure 3). It is to this distribution that a binomial logistic

regression model of potential risk factors may be fit.

Emmi.
Log Odds (Logit) Transformation of Exit Survey Smoking Prevalence

 

40
— Normal

35-4

 

/"\

 

 

 

 

 

 

 

 

 

 

 

 

 

l T T l I l l

-3.3 -2.7 -2.1 4.5 -0.9 -0.3 0.3
Log Odds Smoking Prevalence

17

Table 2 summarizes the information obtained from the US. Bureau of the Census

and includes income and poverty measured at both county and city levels.

Table 2.
US. Census 2000 Demographic Variables for Practice Site (N -87)

 

N Mean Standard Deviation Minimum Maximum

 

Per Capita Income, County {5} 87 20790 3684 15078 32534
Per Capita Income, City {5} 87 19354 4567 12691 32622
Poverty % - County 87 10.7 3.5 3.4 20.4
Poverty % - City 87 14.1 8.5 3.2 37.2
Population Density {Pr Sq Mile} 87 5322 741.0 243 3356.1

 

The Behavioral Risk Factor Survey (BRFS) administered annually by the Michigan
Department of Community Health provided estimates of smoking prevalence for each of 12

designated health regions as demonstrated in Table 3.

Table 3.
Michigan Department of Community Health Variables for Practice Location (N -87)

 

N Mean Standard Deviation Minimum Maximum
Smokr_ng' Prevalence % 87 25.5 2.9 18.1 30.1

 

 

N (%) SmokingPrevalcnce (%)

 

Health Region
1 16 (18.4) 26.6
2 8 (92) 30.1
3 1 (1.1) 22.2
4 14 (16.1) 23.4
5 5 (5.8) 24.3
6 2 (2.3) 25.9
7 10 (11.5) 24.6
8 7 (8.0) 28.1
9 5 (5.8) 23.4
10 3 (3.4) 26.5
11 4 (4.6) 23.7
12 12 (13.8) 27.5

 

18

Table 4 summarizes the demographic information as obtained from the practice
surveys. The specialty of well over half of the participating offices is family practice. Each
of the five medical insurance categories is well represented within this cohort of primary care

rrredical practices.

Table 4.
Practice Survey Demographic Variables

 

N Mean Std Dev Minimum Maximum

 

Participating Providers 87 3.5 2.6 1 10
Race (Estimated Percent)
White 82 80.0 21.4 0 100
African American 82 10.4 13.8 0 65
Hispanic 82 4.9 6.7 0 40
Insurance
Private Prepaid 76 20.3 22.1 0 96
Private Fee For Service 76 30.5 21.2 0 85
Medicare 77 23 .2 17.9 0 70
Medicaid 76 16.0 15.9 0 85
Uninsured 76 5.3 8.8 0 70

 

 

N Percent %

 

Specialty
Family Practice 51 58.6
Internal Medicine 20 23.0
Obstetrics/ Gynecology 16 18.4
Practice Type
Solo 13 14.9
Group 74 85.1

 

19

Table 4 summarizes the patient demographic information from the exit surveys. The

study population is noted to be predominately white with a majority of women participants.

 

 

Table 5.
Patient Exit Survey Demographic Variables
N (%) Mean Std Dev Minimum Maximum
Age, Individual {Years} 1937 47.6 18.4 18 98
Age, Practice Mean {Years} 86 47.2 9.3 22.4 66.6
Gender {0/0}
Male 597 (31.0) 30.5 20.8 0 85.2
Female 1330 (69.0) 69.5 20.8 14.8 100
Race (%)
White 1712 (88.8) 86.6 20.2 o 100
African American 111 (5.8) 7.5 15.7 0 85.7
Hispanic 35 (1.8) 2.2 5.4 o 37.5
Asian 22 (1.1) 1.3 4.5 o 36.4
Native American 19 (1.0) 1.1 3.2 0 22.2
Other 28 (1.5) 1.3 2.5 o 12.5

 

Each of the identified continuous and categorical variables was then individually
assessed within a univariate logistic regression model with the binomial proportion of
smokers as the dependent variable. For the practice-level estimate of proportion insurance
coverage, linear combinations of the categories were constructed and also tested for
potential significance; the combined category of uninsured and Medicaid was found to be
more predictive than either category alone. A composite listing of each factor and its

associated regression coefficient, standard error, and significance levels P is reported in

Table 5.

20

Table 6.

Univariate Analysis of Factors within a Binomial Logistic Regression Model of Practice

 

 

 

 

 

 

 

 

Smoking Prevalence
U.S. Census 2000 N )3 Standard Error P
Income (County) 86 -0.00006 0.00002 0.0022
Log Transformation” 86 - 1.23710 0.39278 0.0016
Income (City) 86
Log Transformation” 86 0.0395
Poverty % (County) 86 0.01110 0.01750 0.0524
Poverty % (City) 86 -0.00205 0.00670 0.7595
Population Density 86 0.00002 0.00010 0.8447
Log Transformation” 86 -0.08680 0.04760 0.0680
Race (Percent)
White” 86 0.00967 0.00628 0.1240
African American 86 -0.07500 0.00740 0.31 1 1
Hispanic” 86 -0.05220 0.02560 0.0413
Asian” 86 -0.10750 0.03890 0.0057
Native American 86 0.04130 0.04150 0.3199
MDCH N 3 Standard Error P
Smoking Prevalence (MDCI-l)” 86 0.04660 0.02040 0.0223
MDCH Region 86 0.2837
Practice Survey N B Standard Error P
Specialty” 86 0.0006
Practice Type (5010 vs. Group) 86 0.8855
Age (Proportion Adults 265 Years) 72 0.31920 0.34160 0.3501
Race (Estimated °/o)
White 81 0.00068 0.00349 0.8449
African American 81 -0.00240 0.00520 0.6439
Hispanic 81 -0.00861 0.01180 0.4667
Insurance (Estimated o/o)
Private Prepaid 75 -0.00094 0.00280 0.7376
Private Fee For Service” 75 -0.00392 0.00292 0.1797
Medicare” 76 -0.01050 0.00372 0.0047
Medicaid” 75 0.01890 0.00395 0.0001
Uninsured” 75 0.03440 0.01410 0.0150
Uninsured 8: Medicaid” 75 0.01740 0.00358 0.0001
Square Root Transformation 75 0.14190 0.03150 0.0001
Exit Survey N 3 Standard Error P
Age (Average Age)” 86 -0.02310 0.00601 0.3472
Gender (Percent Male) 86 -0.00322 0.00293 02716
Race (Percent)
White 86 -0.00267 0.00390 0.4931
African American 86 0.00475 0.00523 0.3639
Hispanic 86 -0.00152 0.01520 0.9204
Asian 86 -0.01040 0.01890 0.5811
Native American 86 0.01680 0.01920 0.3808
Other 86 0.00083 0.02530 0.9738

 

” indicates those variables significant at the p S 0.25 level.

21

Analysis within a multivariate binomial logistic regression model yielded the
parameter estimates as shown in Table 7. The remaining variables include the estimated
regional smoking prevalence, proportion uninsured and Medicaid, and the average patient

age as determined from the exit survey.

Table 7.
Initial Multivariate Regression Model of Practice Smoking Prevalence
with Parameter Estimates (N =75)

 

 

fl 95% CI P
Intercept -2.399 -3.648 - 1.176 0.0001
Smoking Prevalence (MDCH) {°/o} 0.056 0.014 0.010 0.0100
Uninsured 8L Medicaid (Practice) {°/o}
Square Root Transformation 0.106 0.040 0.173 0.0018
Avegge AgeiExit) {Years} -0.019 -0.034 -0.005 0.0096

 

Since 11 practice sites were missing the estimates of insurance coverage and another
the average age of patient per exit survey, multiple imputation was undertaken to replace this
missing data and allow the model estimation of smoking prevalence in all 87 participating

practices. A summary of multiple imputation variance information is provided in Table 8.

 

 

Table 8.
Multiple Imputation Variance Information
Variance Relative Fraction
Between Within Total DF I???“ rn Miss“);
anance Information
Intercept 0.01106 0.34067 0.35284 7562.2 0.0357 0.0348
Smoking Prevalence (MDCH) 0.00001 0.00044 0.00045 8251.4 0.0342 0.0333
Uninsured 8C Medicaid (Practice)
Square Root Transformation 0.00010 0.00098 0.00110 829.9 0.1162 0.1063
Avefle Age (Exit) 0.00000 0.00005 0.00005 4383.1 0.0475 0.0457

 

22

The final multiple imputation model parameter estimates, odds ratios, and
significance levels P are reported in Table 9. There is only minimal change noted in these

estimates after multiple imputation.

1&2.
Final Multivariate Regression Model of Practice Smoking Prevalence
with Multiple Imputation Parameter Estimates and Odds Ratios (N =87)

 

 

 

 

 

B 95% CI P
Intercept -2.410 -3.57 4 - 1.245 0.0001
Smoking Prevalence (MDCH) {°/o} 0.056 0.014 0.098 0.0087
Uninsured 8: Medicaid (Practice) {°/o}
quareRoot Transformation 0.103 0.038 0.168 0.0019
Aveer_ag WAS early) {Years} -0.018 -0.032 -0005 0.0077
Odds R360 95% CI
Smoking Prevalence (MDCH) {°/o} 1.058 1.014 1.105
Uninsured 8C Medicaid (Practice) {°/o}
Square Root Transformation 1.112 1.041 1.189
Avegge 554121111) {Years} 0.981 0.966 0.995

 

23

The association of age with practice specialty is illustrated in Figure 5 with the

internal medicine patient population significantly older than the family practice and

 

 

obstetrics/ gynecology populations.

Elm;

Age of Practice Population as a Function of Practice Specialty
70
60 ‘

 

4|-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

E]

 

 

20-

 

] T

I
Family Practice Internal Medicine Obstetrics / Gynecology

 

N Mean Age Standard Deviation Minimum Maximum

 

Specialty
Family Practice 51 47.1 6.7 22.4 58.3
Internal Medicine 20 57.0 6.1 44.7 66.6
Obstetrics/ Gynecology 16 35.1 5.2 28.7 45.6

ANOVA : P <0.0001 with a significh difference between each of the groups.

24

The distribution of the logistic model estimate smoking prevalence shown in Figure
6 is again approximates a lognormal distribution but with a significantly decreased range and

variance.

ems.
Practice Smoking Prevalence - Logistic Model Estimate

 

80

70-

 

 

 

104 \

0 1 l I
0 0.1 0.2 0.3 0.4 0.5 0.6
Estimated Smoking Prevalence

 

 

 

 

 

 

 

—
‘

 

Sarryle (N) Weighted Mean Standard Deviation Minimum Maximum
87 0.186 0.025 0.088 0.384

 

25

Figure 7 illustrates the model deviance residuals versus the predicted smoking

prevalence for each practice site.

 

 

 

 

 

 

Fm 7.
logistic Model Deviance Residual Values vs. Predicted Smoking Prevalence
3 .1
I
I
I
I
I
1.5 4 ._ I.
a I 'I I
I I I I
1?. ' ' II. . '
8 I I . ~ 5
f 0 I . ' ... ! I F
o I l I I
a ' H ' '
'5 I ' II
‘9 I .I I II ' I
Q I .1 I -
I ~ I
-1.5- I ' I
I
I I
.3 ‘ I I I I I
0.080 0.158 0.235 0.313 0.390
Predicted Smoking Prevalence

26

Figure 8 is a graphical representation of the decreased range and variance of the

model estimate as compared to the exit survey determination of smoking prevalence.

 

 

 

 

 

 

 

 

Figure 8.
Distribution of Exit Survey and Model Estimate Smoking Prevalence by Survey Sample Size
0.6~ 0 .
Exrt Survey
0.54
. I
I
8 04
o ‘ I
>
£5 03‘ o o. o o .
bl) o I .
:5 o o o. I. 0 ° .
o ' 0 °
3 0.2 .0 . . .' ° ° . . ° 0 o
I. I .. I. .. . o. I . : I
0.1< I. o . .
I I
I . . . .
0 o o
6 io 2'o 3'0 4'0 s'o 6'0 70
SurveyN
0.64 .
Model Estimate
0.5.
Q.)
U 0.44
B
H
D. 0.3 o
no I
i ':.-'I ' ' .r- a '
a 02 ' I. I I. I. ... .. I2. . . I
(I) . o .0 .. 0 I e g ' . °
. I
0.1~ . ... . o z . .. ' . O
o.
6 i0 20 50 40 50 60 7'o
SurveyN

Figure 9 offers a direct comparison of the model and exit survey estimates of
smoking prevalence; the Pearson correlation coefficient r is 0.52. The greatest variance is
observed at the extremes of exit survey prevalence values, as these were much more likely to

have smaller sample sizes and less precise point estimates of smoking prevalence.

Biggie}
Smoking Prevalence by Practice Site: Exit Survey and Model Estimates

 

06‘ H Exit Survey
H Model Estimate

0.54
8
0.44
g
>
8
0.0.3 l
l. ‘ ‘ l
(1302‘ “11434)! I i I '1 i (I f I
'1 *1 ...)
0.14

 

 

 

Practice Sites (N =86)

28

The individual contributions of each of the independent risk factors to the model
estimate smoking prevalence are illustrated and summarized in Figures 10-12. It should be
noted that each of the factors covers an estimated smoking prevalence change of 010-015
for a plausible set of risk factor values. The minimum model estimate smoking prevalence
would be approximately 0.078 with an MDCH regional smoking prevalence estimate of
20%, no uninsured or Medicaid coverage, and an average adult patient age of 65 years.
Likewise, the maximum model estimate smoking prevalence would be approximately 0.463
with an MDCH regional smoking prevalence estimate of 30%, total proportion uninsured or

Medicaid coverage of 100%, and an average adult patient age of 25 years.

PM 10.
Model Estimate Smoking Prevalence as a Function of MDCH Regional Smoking Prevalence
Estimate

 

 

 

 

0.24
Assuming no uninsured or Medicaid coverage .
and average adult patient age of 25 years. "
0 22‘ Model prevalence estimate increases by ~ 0.0086 for each '/
g ' percentage point change in the MDCH regional I,
a smoking prevalence estimate. "
'5
I
m 0.204 ’3”
0 v"
u
8
'3
>
g 0.18. a
i
2.1
0.16‘
19,.
0.14 <
20 21 22 23 24 25 26 27 23 29 30

MDCH Smoking Prevalence %

29

Emil.
Model Estimate Smoking Prevalence as a Function of Uninsured 8C Medicaid Proportion

Coverage

 

 

 

 

0.35
Assuming average age of adult patients within practice of
0.325 25 years and MDCH regional smoking prevalence of 20%. l
1 .—”'
Model prevalence estimate increases by ~0.018 for each ..x' .
0 0,3. 10% change in the uninsured and Medicaid proportion ."
a coverage. ”,"Ir
a 0.275 ,
m ',. "
8 0.254 I ,
g . ,
§O.225l . ’
a. '
1 -1 1’77 v
.8 0.2 x‘,
2 0.17s..»—""'
0.15- I
0.125"
0 10 20 30 40 50 60 70 80 90 100
Uninsured &Medicaid in Practice °/o
F1gu_r§' ‘ 12

Model Estimate Smoking Prevalence as a Function of Average Age of Adult Patients Within
Practice

 

 

 

 

0' ' Assuming no ininsured or Medicaid coverage
~ and MDCH regional smoking prevalence of 20%.
0.144
\‘\ Model prevalence estimate decreases by ~ 0.175 for each
10 year change in the average age of adult patients
g 0.13« i within practice.
I ‘3 ~
Bi 0.12.
o
g .
,fi 0 11 8‘ .
9' 0.10‘
'3 ~.
‘8 ‘r.
2 0.09« ‘
0.084 I
0.074
25 35 45 55 65

Average Age of Adult Patients Within Practice (Years)

30

Figure 13 demonstrates the initial effect upon the baseline referral rates of the 58
providers recording at least one referral; the comparison rates are calculated using the mean
(0.186) or the model estimate smoking prevalence. The overall Pearson correlation
coefficient r is 0.92, although there does appear to be some decreasing correlation as the

referral rate increases.

 

 

 

Fm. 13.
Baseline Provider Referral Rates by Practice Site: Exit Survey and Model Estimates
H Calculated with Mean Smoking Prevalence
X—X Calculated with Model Smoking Prevalence
20-
.3
a
'3
.310.
82
XX W. e x _
0- X x X" 1);!“ ”5 xx X'F i i“.
I l l

 

 

 

31

DISCUSSION

This exercise has resulted in a model to estimate a small area health behavior,
primary care practice smoking prevalence, given a small survey sample at each site and
modeling these binomial proportions with possible predictive variables to further refine and
calibrate the initial survey estimate.

The smoking prevalence determined by exit survey is significantly lower than that
reported by MDCH for the year 2001. The average smoking prevalence among adults
within the participating practices was 18.6 i 1.8% (95% confidence interval, CI) as
compared to the reported statewide smoking prevalence of 26.1 i 1.6% (95% CI). (47) This
discrepancy could in fact be due to a relatively decreased absolute prevalence of smoking in
primary care medical practices as a result of inherent age and socioeconomic differences.
Yet one must consider the possibility of a differential sampling bias with only a 54%
effective survey response rate.

The measure of uninsured and Medicaid status within the medical practice is a strong
predictor of smoking prevalence. This is almost certainly due to its ability to act as a proxy
for income or socioeconomic status. Interestingly, Medicare status within the practice was
replaced within the multivariate model by average age of the adult patient population; so in
this case, a defined insurance status is serving as a proxy for age. Furthermore, the
uninsured and Medicaid status replaced the US. Census estimates of income at both the
county and city levels; this would seem to be appropriate as a more refined measure of the
small area versus surrounding community socioeconomic status.

Contrary to the general principle of measures taken in more geographic proximity to

the population of interest being more accurate, the income as measured at the county level

32

was a stronger univariate predictor of smoking prevalence than that measured at the city
level. A possible explanation is that a medical practice often draws from a larger geographic
region than its immediate vicinity; therefore, any measure within the practice is likely to
reflect a composite of its catchment area rather than a more focused description of its more
immediate neighborhood or municipality. This may in fact be the reason that the MDCH
smoking prevalence regional estimate remains significant despite adjustment for
socioeconomic status and age within the multivariate model.

Comparison of the exit survey and model smoking prevalence estimates reveals only
a moderate correlation, with the greatest variance noted at the extremes of the initial exit
survey estimates (Figure 9). This is in all likelihood a result of the decreased precision
associated with the relatively small sample sizes for these estimates. There is very little effect
of the model estimate upon the individually calculated baseline provider referral rates (Figure
13). However, there does appear to be decreased correlation between the rates based upon
mean and model prevalence estimates as referral rates increase. This would suggest that if
this exercise was of limited initial utility, it may in fact offer a better description of provider
referral behavior if the primary study intervention is effective.

Each of the three risk factors included in the final model may individually alter the
smoking prevalence estimate by 0.010 - 0.015 (Figures 1012). There does not appear to be a
dominant factor, but a web of demographic and socioeconomic influence upon the primary
care practice smoking prevalence.

Several other strategies have been employed to estimate a health behavior at the
small area level. The most trusted and reliable approach is to perform a large survey to
establish more accurate point estimates with narrow confidence intervals. This would

certainly be best for establishing an estimate for a limited number of sites, but the time and

33

expense become prohibitive when applied to multiple sites. Another approach is to apply
careful weighting of pre-existing demographic information from larger surveys in order to
draw valid inferences about a nested small area population (40-44). This would almost
certainly be ineffective within our current study population as it has been demonstrated that
the most proximal measures may not in fact be the most predictive for primary care patient
populations. In fact, Twigg 8C Moon have developed a multilevel predictive model based
upon weighting of survey data; the model performed quite well in predicting smoking
prevalence, yet correlated less well with increased alcohol consumption (44-45). As there is a
growing interest in attempting to alter potentially harmful lifestyle behaviors, it is becoming
increasingly important to be able to measure or estimate these at a small area or clinic level.

The model developed within this study is actually a composite of both approaches.
The small survey served to calibrate and refine the estimates of smoking prevalence
otherwise determined through the modeling of known risk factors and regional prevalence
estimates. If needed, this model could be adapted for use in other settings and with other
variables. As demonstrated in the modeling process, many variables are associated with
smoking prevalence at the primary care practice level and, if missing, would likely be
replaced with another established risk factor or significant proxy.

These results should be viewed with caution as this model has yet to be validated
against established surveys of smoking prevalence at the primary care practice level. The
multivariate model was also unable to establish any racial or ethnic influence, although this
may be a result of the limited power to detect these differences in this practice cohort with
only 13.4% minority patient representation. Despite these limitations, this model offers a
potential alternative to the estimation of a small area health behavior without extensive local

survey data.

34

CONCLUSIONS

Several conclusions may be drawn from this study of smoking prevalence within
primary care medical practices:
1. The average smoking prevalence among adults within the participating practices
was 18.6 i 1.8% (95% CI). This is significantly lower than the reported
statewide smoking prevalence of 26.1 i 1.6% (95% Cl); this may be a
consequence of inherent population differences in age or socioeconomic status
or sampling bias from differential survey participation.
2. Significant univariate predictors of the small area smoking prevalence included:
a. Per capita income at both county and city levels (US. Census)
b. Race (US. Census)

Regional estimate of smoking prevalence (MDCH -IBRFSS)

P

d. Practice specialty (practice survey)

Proportion of patients with Medicaid or uninsured (practice survey)

9

{'05

Age of patients within practice (exit survey)
3. Significant and independent predictors of the small area smoking prevalence
within a multivariate binomial logistic regression model included:
a. Regional estimate of smoking prevalence (MDCH - BRFSS)
b. Proportion of patients with Medicaid or uninsured (practice survey)
c. Age of patients within practice (exit survey)
4. The association between practice specialty and small area smoking prevalence
was dependent upon the age of the patient population within family practice,

internal medicine, and obstetrics/ gynecology practices.

35

. The regional estimates of race and income were replaced by a practice-level
measurement of insurance distribution as a significant predictor of small area
smoking prevalence. The proportion of patients who are uninsured or with
Medicaid serves as a small area proxy for socioeconomic status.

. The final logistic regression model estimate smoking prevalence retains the
original distribution approximating lognormal, but with a decreased range and
variability.

. Although the weighted mean smoking prevalence is identical between the model
and exit survey smoking estimates, there is only a moderate correlation between
the individual practice-level estimates.

. A very strong correlation exists between provider referral rates utilizing a
uniform population mean smoking prevalence and those calculated with the
more variable regression model estimate smoking prevalence. Despite this
strong correlation, there is some evidence of decreasing correlation as provider
referral rates increase.

. It would be ideal to validate the results of this study through a more
comprehensive survey of smoking prevalence within this or other primary care

practice cohorts with even more diverse patient populations.

36

APPENDICES

Appendix 1 University Committee on Research Involving Human Subjects
(UCRIHS) Review

Appendix 2 Practice Survey
Appendix 3 Provider Survey

Appendix 4 Patient Exit Survey

37

Apmndix 1

University Committee on Research Involving Human Subjects (UCRIHS) Review

 

MICHIGAN STATE
U N I V E R SIT Y
January7.2003

T0: William WADLAND
8101 Clinical Center
MSU

 

RE: IRB I 01-780 CATEGORY: FULL REVIEW
RENEWAL APPROVAL DATE: January 0. 2003
EXPIRATION DATE: December S. 2003

TITLE‘ EVALUATION OF ORGANIZATIONAL CHANGES TO PROMOTE SMOKING
' CESSATION WITHIN MANAGED CARE

TheUniversIty CanmmeeonReeeardilnvdvhgflmensmiecb'(UCRIHS)nviewofWeproieci
bWWlmemmmmmmdhmmmwb
beedequddywdectedmdmeflwdswobtehhfamedmmwmefihudamm
UCRIHS APPROVED TRIS PROJECTS RENEWAL ,

RENEWALS:UCRIHSWDV8HWGMWMWW.WW
beyondihledalemueiberenewedmmereneweiform. Ameximmnoffoureuchexpediied
Mmpoesbie. lmeeugaiorewlshmioconmueeproiectbeyondmetheneedlomrlie
5-yearenewalepplicetionior compietereview.

REVISIONS: UCRIHSmuetreviewanydwigeshpmceduvahghumanwbiecu,pflorb
hilielionolihechenge.Nhieiedmeatflreiineoirenewalmieaeehdudeerevieionfonnwhh
renewal.Torevieeenapprevedproiocoletmyoiherllmedunngmeyear.eendyourvwitlenrequeet
wimmedadiedrevbioncovereheeltoiheUCRIHSCheir. requeelirrorevisedapprovdend
reierenchgihepmject’isBiiandtiiie. inciudehyourrequeetedwbiionotmechmgeendmy
revised heiruments, consent formsoredvenisements ihetere applicable.

PROBLEMSICHANGES: Shouldeiiheroiirrefoliowhgar'lsedunngmecoureeolmeworknow
UCRIHS promptly: 1)probleme (unexpected eideeffecte. complaints. eic.)hvolvhghunm

abject
__ ' aflmhmmmWanewwonnefionhdicetthMbmm

 

mmmerdetedwhenmmeproiocdweeprevbudymiewedmdeppmved

iiwecmbeoiluihereeeletmce. pleaeecontectueet517 355-21800rvie email:
UCRIHSGmeuedu.

more“ Ashlr Kumar M. D.

film“:

sums-mo

UCRIHS Chair

38

Apgndix 2

Practice Survey

Health Care Practice Survey _ Site ID #

 

Thank you for participating in the Enhancing Smoking Cessation in Michigan Medical Practices Study
(ESCMMP)! The purpose of this survey is to collect information about your practice and its smoking
cessation services. This study is funded by a grant from the Robert Wood Johnson Foundation to Michigan
State University and Blue Cross Blue Shield of Michigan. You indicate your voluntary agreement to
participate in this study by completing and returning this survey.

Please complete all 3 pages and return it in the enclosed envelope within 2 weeks.

If you have questions about this project you may contact Dr. Jodi Holtrop at (517) 353-3544 (ext 432)
(jodi.holtrop(a)ht.msu.edu1. If you have questions about being a human subject of research you may
contact the MSU University Committee on Research Involving Human Subjects (UCRIHS) at

(517) 355-2180 or ucrihs@m_su.edu. Thank you.

I. Contact Person

 

 

Name
Title
11. Practice Characteristics

1. With what hospital or medical school are you affiliated, if any?

 

2. For 2001, what is the approximate total annual patient visits (total number of visits made by all
patients) for all providers (entire practice)?

3. For 2001, what is the approximate total number of active patients in the practice (total number of
patients who are “signed up” to be a patient at this practice and have had at least one visit in the past
three years)?

4. What is the practice type? (V m only)
__ 8010 (one clinician such as physician, nurse practitioner or PA)
_ Two person (two clinicians of same specialty)
__ Group practice (three or more clinicians of same specialty)
__ Multi-specialty group practice (two or more clinicians of different specialization)

5. Who owns the practice? (I % only)
_ Physicians
_ Hospital or health system
__ Managed care organization
__ Federal, state or local government
_ Other, please specify:

 

6. Please write in the number in each category for your practice. Place a zero if there are none.
_ Physicians — How many are there in each specialty?

_Internal medicine _Farnily practice _Med/Peds _Ob/Gyn _Other
_Physicians assistants _Nurse practitioners
_RN Nursing staff _LPN Nursing staff _Medical assistants

_Other health staff (lab, x-ray, psychologists, dietitians, etc.) please specify:

Page 1 of 3

39

III.

10.

11.

12.

l3.

l4.

Wisdxsized‘tbearmmizybrubfiyflxpmmkrislamai? [/ganljj
__ <5000

_ 5000— 10,000

_10,000— 25,000

_ 25,000-100,000

__ >100,000

How close is this practice to a major city?
_ n_ot within 25 miles of major city
__ in a major city or within 25 miles ofa major city

Patient Characteristics

What is the approximate percent of patients that fall within the following age categories for this
practice? Please write in the percent for each age grouping. This should add up to 100%.

_ <12 years

__ 12-17

_ 18-64

_ 65 and over

100% TOTAL

What is the approximate breakdown of race/ethnicity in this practice? This should add up to 100%.
__ White, non-Hispanic

_ Black, non-Hispanic

__ Hispanic

_ Other

100% TOTAL

Approximately, what are the payment methods for patients in this practice? Please write in the percent
for each insurance type. This should add up to 100%.

__ Private health insurance (prepaid)

_ Private health insurance (fee for service)

_ Medicare

__ Medicaid/other government assistance

__ Other

_ Uninsured

100% TOTAL

For 2001, approximater what percent of the patients in this practice are covered by some
type of Blue Cross Blue Shield insurance plan ? %

Under what Blue Cross Blue Shield plan(s) are the patients covered? (V all that apply)
_ Traditional Blue Cross Blue Shield Plan
__ Community Blue PPO
__ Blue Choice
_ Blue Care Network
_Other:

 

Approximater what percent of the patients are smokers? Check here if unknown D

Page 2 of 3

4O

IV. Practice Handling of Smoking Patients

15. Which of the following does your practice have in place for encouraging smoking cessation?
Please check one box for each line.
fig
Clinic is designated as smoke-free ..................................
No smoking signs/postings are in the clinic ...........................
Practice has an official policy restricting smoking on site ................
Posted signs in the reception area offering help in smoking cessation ......
Written patient educational materials on smoking cessation are available . . .
m, are the materials directly available to patients? ................
Other types of patient education materials on smoking cessation are available
Please specify:
Staff member designated to maintain smoking cessation materials .........
System to identify smoking status at every visit ........................
Documentation of smoking status in medical record (such as on problem list)
Presence of follow-up system for patients involved in quitting .............
Other: (please specify)

 

0000 0000000
0000 0000000%

 

V. Approval for Participation in the Study
14. Has the medical director of the office reviewed the project proposal and agreed to conducting the
research at this practice? (/ pn_e only)
Yes No

Signature of Site Medical Director Date

T hanlr you for completing this questionnaire!
Please return in the self-addressed, stamped envelope provided.

Page 3 of 3

41

_Apmndix 3
Provider Survey

Smoking Cessation Survey for Health Care Providers Study ID #

Please answer questions based on this practice location only.
Return the completed survey within the next two weeks in the envelope provided. Thank you.

1. Please check the appropriate box to describe yourself.

CI Physician Cl Nurse Practitioner D Physician Assistant

2. What is your medical specialty?
D Family practice Cl Internal Medicine Cl Obstetrics/Gynecology
C] Other:

 

3. In your personal outpatient practice, how many patients do you see in an average day?
Total Patients

4. Of the patients seen in an average day, how many patients are:
Pediatric (<18 years of age)
Obstetric

5. On an average day of outpatient practice, how many of your patients (age 12 and over) are smokers?
Patients are smokers

6. On average, how many days do you work in your outpatient practice per month?
Days/month

7. How often do w ask your patients if they smoke? (Circle fl number)

Never Always
1 2 3 4 5
8. How ofien does your staff ask your patients if they smoke? (Circle Qt; number)
Never Always
1 2 3 4 5

9. Of all patients you know who are smokers, about what percent do you ask at eveg visit about
willingness to make a quit attempt?

%

10. Of all patients you know who are smokers, about what percent do you advise at every visit to quit?
%

Counseling for smoking cessation is defined for this survey as the time spent discussing the possibility of or
methods of quitting smoking or staying smoke-free.

11. Of those smoking patients interested in making a quit attempt, about what percent do you provide
counseling regarding quitting? %

12. What is the average number of minutes you spend counseling. . ..

A new patient on smoking cessation? Minutes Check here if you do not counsel D
The same patient on follow-up visits? Minutes Check here if you do not counsel CI

42

13. For patients who wish to grit. how often do you provide... (Circle pg; number for each)

 

 

 

 

 

Never Alwa s
Prescription for buproprion or nicotine replacement ..... 1 2 3 4 5
Brochure, educational material or website information. . . 1 2 3 4 5
Referral to smoking cessation telephone quitline ....... 1 2 3 4 5
Follow-up visit to discuss progress with quitting ....... 1 2 3 4 5
Other, please specify: 1 2 3 4 5
14. For patients not willing to make a quit attempt, how often do you... (Circle one number for each)
. Never Always
Ask patient to let you know when he/she is ready to quit .. 1 2 3 4 5
Ask patient to identify his/her reasons to consider quitting 1 2 3 4 5
Ask patient to identify his/her barriers to quitting ....... l 2 3 4 5
Discuss health risks of his/her smoking ............... 1 2 3 4 5
Encourage him/her to consider quitting ............... l 2 3 4 5
Give advice on how to quit ........................ l 2 3 4 5
Give brochure on quitting smoking .................. l 2 3 4 5
Other: 1 2 3 4 5
15. Overall, how would you rate your knowledge about helping people stop smoking?
CI Poor CI Fair Cl Good Cl Very Good C] Excellent
16. Have you ever received any formal training in smoking cessation intervention strategies?
CI Yes CI No
If YES, please answer - From what source(s)? (V a_ll that apply)
0 Continuing education course/program
CI Organized study club
CI Professional course or curriculum in medical school or residency
C] Other, specify:
17. How willing are you to receive such training? (I pp; only)
Cl Very willing D Somewhat willing Cl Somewhat unwilling CI Not interested at all
18. How confident are you in your ability to help someone stop smoking? (Circle fl number)
Not confident Very confident
I 2 3 4 5
19. How successful have you been in helping patients stop smoking? (Circle o_np number)
Not successful Very successful
1 2 3 4 5
20. In your opinion, how important is smoking cessation as a component of overall health care provided in

medical practices? (Circle o_ne_ number)
Not important Very important
1 2 3 4 5

Please complete PAGE 3. Thank you.

43

21. To what extent are the following a barrier to incorporating smoking cessation activities into your
practice? (Circle ppg number for each)

Never

Always

Patient resistance/complaints ...................... 1 . 2 3 4 5
Amount of time required ......................... 1 2 3 4 5
Lack of reimbursement mechanisms ................. 1 2 3 4 5
Resistance by staff ............................... 1 2 3 4 5
Concerns about effectiveness ...................... 1 2 3 4 5
Availability of patient education materials ............ 1 2 3 4 5
Availability of adequate referral resources ............ 1 2 3 4 5
Your lack of knowledge ........................... 1 2 3 4 5
Other: 1 2 3 4 5

 

22. Which of the following smoking cessation educational opportunities or practice helps would assist you
in enhancing smoking cessation care to your patients? (Circle p15 number for each)

 

Not at all Very
Helpful Helpful
Attendinga course .............................. 1 2 3 4 5
Reviewing audiotapes ................................ 1 2 3 4 5
Training for your staff ................................ 1 2 3 4 5
Patient access to a quitline program .................... l 2 3 4 5
Feedback on your referrals to a quitline ................. 1 2 3 4 5
Educational materials to give to patients ................. l 2 3 4 5
Reimbursement for providing brief advice ............... 1 2 3 4 5
Other: 1 2 3 4 5

 

23. Approximately what percent of your patients are covered by any type of Blue Cross Blue Shield
insurance including Blue Care Network HMO, Blue Preferred PPO . . p %

24. Are you aware of the Blue Cross Blue Shield “Quit the Nic” quitline smoking cessation program?
CI Yes C] No

If YES, have you ever referred patients to this program? Cl Yes D No
25. What year did you graduate from your residency or clinical training program?
26. Your gender: D Male Cl Female
27. What is your smoking status? C] Current smoker Cl Former smoker Cl Never smoker

28. You may be receiving feedback on your rates of smoking cessation referral. Would you like to receive
this feedback electronically? Cl Yes Cl No

If YES, please provide your email address:
(this information will be kept confidential)

 

Thank you for completing this survey. Please return in the self-addressed, stamped envelope provided.

44

Apmndix 4

Patient Exit Survey
Smoking Cessation Counseling by Your Health Care Provider Study ID #
1. What is the name of the health care provider you saw today? Name:
2. What was the reason for your visit today?
3. During today ’5 visit, did your health care provider ask you if you smoke? D Yes Cl No
4. During today ’3 visit, did anyone else in the practice ask you if you smoke? CI Yes Cl No
5. Has any provider in this office asked you in the past year if you smoke? D Yes D No
6. Have you smoked at least 100 cigarettes in your entire life? D Yes D No
7. Have you smoked a cigarette, even a puff, in the last 7 days? Cl Yes C] No
If NO - Please SKIP TO QUESTION 12 If YES — Please answer the remaining questions.
8. On the average, how many cigarettes do you now smoke a day? Cigarettes
9. Which of the following BEST describes your plans regarding smoking? Check m box only please.
Cl Seriously considering quitting in the next month
Cl Seriously considering quitting smoking in the next 6 months
Cl Not seriously considering quitting smoking in the next 6 months
10. During t’oday 5 visit with your health care provider, did anyone do any of the following?
Advise you to stop smoking ...................................... D Yes C] No
Ask you about your interest in quitting smoking ...................... D Yes D No
Ask if you were willing to set a quit date ............................. Cl Yes D No
Give you a telephone number to call for help quitting ................... D Yes Cl No
Give you information about counseling classes or programs to help you quit D Yes D No
Refer you to someone in the office for more information about quitting ..... Cl Yes Cl No
Suggest a follow-up visit or phone call about quitting smoking ............ Cl Yes D No
Recommend using a nicotine patch or gum to stop smoking .............. Cl Ye Cl No
Recommend using a nicotine inhaler or nasal spray to stop smoking ........ Cl Yes D No
Give you a prescription for Zyban (buproprion, Wellbutrin) to stop smoking Cl Yes D No
Provide you with reading materials on quitting smoking ................. D Yes D No
11. During today ’s visit, did you agree to make an attempt to quit smoking? 0 Yes Cl No
12a. How old are you? Years 12b. Are you male or female? Cl Female Cl Male
13. Which one or more of the following describes your race and ethnicity?
CI Black — non-Hispanic D White - non-Hispanic Cl Other:
Cl Asian CI Native Hawaiian/Pacific Islander
Cl American Indian/Alaskan Native Cl Hispanic
14. What is your health insurance? Please make a check in any box that applies to you
D No insurance 0 Medicare CI Medicaid
Cl Blue Cross Blue Shield - Please check one if you know what type of plan:
B Community Blue Cl Blue Choice [3 Blue Care Network Cl Traditional

 

 

 

C! Other private insurance not listed above:

 

Thank you! Please write your comments on the back.

45

10.

11.

12.

13.

14.

REFERENCES

World Health Organization. World Health Report: 1999. Curbing the Tainan E prdemc
Geneva, Switzerland. World Health Organization; 1999.

Bartlett JC, et a1. Medical care expenditures attributable to cigarette smoking - United
States, 1993. MM IVR. 1994;43:469-472.

Kumra V, Markoff BA. Who’s smoking now? The epidemiology of tobacco use in the
United States and abroad. am in Overt Mariam 2000;21(1):1-9.

Fiore MC, Bailey WC, Cohen SJ, et al. Tradirg Tduao Use and qunlene Clerical Radar:
Graddire. Rockville, MD: US. Department of Health and Human Services, Public
Health Service; June 2000.

Ockene JK, Kristeller J, Golberg R, Amick TL, Pekow PS, Hosmer D, Quirk M, Kalan
K. Increasing the efficacy of physician-delivered interventions: a randomized clinical
trial] Gmlm‘emMed. 1991;6:1-8.

Stevens VJ, Hollis JF. Preventing smoking relapse, using an individually tailored skills-
training technique. ] Caedtirgand Crawl Pad). 1989;57:420-424.

Hill RD, Rigdon M, Johnson S. Behavioral smoking cessation treatment for older
chronic srrroker. Bdrau'or'flempy 1993;24: 321- 329.

Hughes JR, Giovino GA, Klevens RM, Fiore MC. Assessing the generalizability of
smoking studies. Addicn'm 1997;92(4):469-472.

Cummings SR, Stein M], Hansen B, Richard R], Gerbert B, Coates T]. Smoking

cessation and preventive medicine: A survey of mtemists in private practices and a health
maintenance organization. ArdrIntemMed 1989; 149 .-345 349.

Thorndike AN, Rigotti NA, Stafford RS, Singer DE. National patterns in treatment of
smokers by physicians. ]AMA. 1998;279(8):604-608.

Glynn T], Manley MW. Howto help jao'patrbas sap smkirg A Natz'mal W [retrace
MamalfiJerjsr'aktm US. Department oh Health, NIH 89-3064; March, 1989.

O’Connor PJ, Solberg LI, Baird M The future of primary care: the enhanced primary
care model. ]FP 1998;47(1):62-67.

Oxman A, Thomson MA, Davis DA, Haynes RB. No magic bullets; a systematic
review of 102 trials of intervention to improve professional practice. Gm Mai Assoc].
1995; 153:1423-1431.

Carney PA, Dietrich AJ, Keller A, Landgraf J, O’Connor GT. Tools, teamwork, and
tenacity: an office system for cancer prevention. ] Paumtt 1992, 35: 388- 394.

46

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

Thompson,RS, Taplin SH, McAfee TA, Mandelson MT, Smith AE. Primary and
secondary prevention services in clinical practice. Twenty years’ experience in
development, implementation, and evaluation. ]AMA .1995;273(14): 1130-1135.

Solberg LI, Kottke TE, Brekke ML, Calomeni CA, Conn SA, Davidson G. Using (121
to increase preventive services in clinical practice — going beyond the guidelines. Prev
Med 1996;252:59-67.

Wagner EH, Curry SJ, Grothhaus L, Saunders KW, McBride CM. The impact iof
smoking and quitting on health care use. A rdrIrrtemMai. 1995;155:1789-1795.

Zhu SH, Stretch V, Balabanis M, Rosbrook B, Sadler G, Pierce JP. Telephone
counseling for smoking cessation. Effects of single session and multiple session
interventions. I thonsult and anal Pad).1996;64(1):202-211.

Legorretta AP, Kashian C, Franklin C. Results of a smoldng cessation program in a
managed care setting. A m ] Mamgad Care 1996;2(7):831-836.

Wadland WC, Stoffelrnayr B, Ives K. Enhancing smoking cessation of low-incorne
smokers in managed care. J Fam Pract 2001;50(2):138-144.

US Department of Health and Human Services. Tainan UseA meg US. Racial/E dmic
Mbu'ity (310475. A Report tfde Surgeon Gaeml Washington (DC): US Department of
Health and Human Services, Centers for Disease Control and Prevention, 1998.

Oren VW. Smoking and the health gap in minorities. A malt q’ E prdrmdogy
l993;3(2):159-164.

Freeman HP. Poverty, race, racism, and survival. Annals cf E pzdemdcgy 1993;3(2): 145-
149.

Adler NE, Boyce T, OresneyMA, Cohen S, Folkrnan S, Kahn RL, et a1. Socioeconomic
status and health: the challenge of the gradient. Anen'amstdrdogrst. l994;49(1):15-24.

Kroeber Al, Kluckhohn C. A Critical Reu'ewcanepts arqu‘im'tims. New York: Vintage
Books, 1963.

Iiberatos P, Link BG KelseyJL. The measurement of social class in epidemiology.
EpidemdogicReu'ew. 1988;10:87-121.

Pappas G. Elucidating the relationships between race, socioeconomic status, and health
[editorial]. Arren'canjaoml fPuHicHazlth. 1994;84(6):892-893.

Williams DR, Collins C. US socioeconomic and racial differences in health: patterns and
explanations. A mm! RenewrfSaadogy 1995;21:349-386.

47

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

US Bureau of the Census. 2000 Chasm cfdePrprJatim: Soaalzdemn'c Wilts,
Um'tad States. Washington (DC): US Department of Commerce, Economics and Statistics
tration.

Michigan Department of Community Health. Region! Estimates by Cannorz'tyHaaldr
Assssmmt Regan 1995-1999. Lansing: Michigan Department of Community Health,
Bureau of Epidemiology.

Agresti A. Categm'cal Data A mljsis. John Wiley, New York, 1990.

Hosmer DW, Iemeshow S. Applied Logistic Regmsion John Wiley 8c Sons, 1989.

Zhou XH, Eckert GJ, TiemeyWM. Multiple imputation in public health research.
StatistMai. 2001;20:1541-1549.

SAS Institute Inc. SA S/S TA TSqimre Orarga amiEnlammrents, Release 8.2. Cary, NC:
SAS Institute Inc, 2001.

SAS Institute Inc. SA S/S TA T User’s Guide, Version 8. Cary, NC: SAS Institute Inc, 1999.

SAS Institute Inc. SA S/S TA TProcaimer Grade, Version 8. Cary, NC: SAS Institute Inc,
1999.

Schafer JL. A rubsis flrmrplete Mtdtiran'ate Data. London: Chapman and Hall, 1997.

Rubin DB. M ultzpk: Inputatimjbr Nomespmse in Surrey. New York: Wiley, 1987.

Schemper M, Heinze G. Probability imputation revisited for prognostic factor studies.
Stalist Mal 1997;16:73-80.

Ray NF, Thamer M, Fadillioglu B, Gergen PJ. Race, 1ncome,urbanicity, and asthma
hospitalization in California. a small area analysis. Crest 1998;113:1277-1284.

Malec D, Davis WW, Cao X. Model-based small area estimates of overweight
prevalence using sample selection adjustment. Statist Mai 1999;18:3189-3200.

Singh SP, Srivastava AK, Sisodia BVS. Evaluation of a synthetic method of estimation
for small areas. ]amral cprplied Statistic. 2002;29(8):1147-1151.

Ghosh M, Rao JNK. Small area estimation: an appraisal. Statistiad Sa'am1994;9:55-93.

Twigg L, Moon G. Predicting small area health-related behaviour: a comparison of
smoking and drinking indicators. Social Sa'eme ardMah'cire.2000;50:1109-1120.

Twigg L, Moon G. Predicting small area health-related behaviour: a comparison of
multilevel synthetic estimation and local survey data. S odd Salem? are! M edidm
2002;54:931-937

48

46. Black MA, Craig MA. Estimating disease prevalence in the absence of a gold standard.
Smart Med 2002.

47. Michigan Department of Community Health. Health Risk deru'ors in #29 State cf M rdrzgan

2001 Bdamr'oml Risk Factor Surzallarxe Sjstem Lansing: Michigan Department of
Community Health, Bureau of Epidemiology.

49

   
 
  

      

Tfiii‘fljijijjljlj]"fill

61