Received 11/1/96; revised 3/18/97; accepted 3/19/97.
The costs of publication of this article were defrayed in part by the payment of
page charges. This article must therefore be hereby marked advertisement in
accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
I This work was supported by Grants CA55769 (to M. R. S) and CA68437 (to
W. K. H.).2 To whom requests for reprints should be addressed, at Department of Epide-
miology, Box 189, The University ofTexas M. D. Anderson Cancer Center, 1515Holcombe Boulevard, Houston, TX 77030.
Vol. 6, 565-571, Augur: 1997 Cancer Epidemiology, Biomarkers & Prevention 565
Identifying and Recruiting Healthy Control Subjects from a Managed
Care Organization: A Methodology for Molecular Epidemiological
Case-Control Studies of Cancer1
Karen Suchanek Hudmon, Susan E. Honn, Hong Jiang,
Robert M. Chamberlain, Wenguang Xlang, George Ferry,
Wayne Gosbee, Waun Ki Hong, and Margaret R. Spi&
Departments of Epidemiology [K. S. H., S. E. H., H. J., R. M. C., W. X.,
M. R. S.], Behavioral Science [K. S. H.], and Thoracic Head and Neck Medical
Oncology [W. K. H.], The University of Texas M. D. Anderson Cancer Center,
and Kelsey Seybold Foundation [G. F., W. G.], Houston, Texas 77030
Abstract
Case-control studies with stringent matching criteria
require large pools of healthy subjects from which toselect matched controls. This paper describes a successful
method of identifying a large pool of potential control
subjects to participate in two molecular epidemiological
case-control studies of lung cancer, each enrolling 400case subjects and 400 control subjects. These studies are
not population based, and the study base is not well-
defined. Therefore, potential control subjects are beingidentified and recruited through 20 area clinic sites of a
large multispecialty health maintenance organization.
Because the research focus is driven by genetic
hypotheses and we are controlling for multiple smoking-
related variables, representativeness is of lesser concern.
To identify potential control subjects, a one-page
questionnaire is distributed to patients in the waitingroom to assess contact information as well as data
relevant to the case-control matching process. An average
of 2,228 questionnaIres are returned monthly toward a
target pool of 40,000; of these, 59% of the respondents
fuffill eligibifity criteria as a control subject for one of the
studies and are not averse to being contacted in thefuture for the purpose of research. When compared to
former smokers and never smokers, current smokers in
the control population were least likely to refuse further
contact. A collaborative arrangement with a managed
care organization offers a feasible mechanism through
which researchers can access a large, ethnically diverse
population of potential control subjects.
Introduction
Although there is a growing body ofliterature describing meth-
ods of recruitment and factors associated with participation in
clinical trials, less is known about these issues as they pertain
to epidemiological case-control studies. Molecular epidemio-logical studies require active consent from participants because
biological samples, in addition to personal interview data, must
be collected. The identification of case and control subjectstypically requires separate recruitment methodologies, andwhen matching criteria are stringent, it is necessary to establish
a large population of eligible subjects from which to select
controls. Logistically, the conduct of large case-control studies
can be difficult, costly, and time consuming.
The number and type of molecular epidemiological case-
control studies being conducted is growing rapidly. Although
the literature is replete with discussions of the pitfalls in controlselection, little methodological guidance is available for epide-
miologists who must identify and recruit healthy persons to
serve as control subjects for large case-control studies, espe-
cially when the cases are hospital-based and the study base is
not well-defined. This paper describes a method for developinga large pool of approximately 40,000 control subjects from
which to select participants for two hospital-based molecularepidemiological case-control studies of lung cancer: Genetic
Markers of Cancer Susceptibility in Former Smokers (the “For-
mer Smoker Study”) and the Ecogenetics of Lung CancerStudy. Control subjects are being identified and recruitedthrough a large, multispecialty health maintenance organiza-
tion. Additionally, we describe our methods of identifying case
subjects, sociodemographic factors that affect potential control
subjects’ willingness to participate, and agreement betweenwritten questionnaire and personal interview data among the
currently enrolled control subjects.The Former Smoker Study examines interindividual dif-
ferences in genetically determined susceptibility to tobacco
carcinogenesis in former smokers (i.e., individuals who stopped
smoking at least 1 year before cancer diagnosis), a group thatstill exhibits substantially increased risks for cancer when com-
pared with never smokers. The Ecogenetics of Lung CancerStudy extends the aims ofthe Former Smoker Study to examinesimilar susceptibility markers among current and never smok-
ers. Both studies are designed to assess lung cancer suscepti-
bility at several stages in tobacco carcinogenesis, includinggenetic modulation of carcinogen activation and detoxification
(metabolic polymorphisms), in vitro chromosome sensitivity totobacco mutagens, DNA repair capacity, and levels of tobacco-
related DNA adducts in current smokers.
Each study will accrue 400 lung cancer case subjects;
control subjects will be matched to case subjects in a 1 : 1 ratiowith respect to sex, age (±5 years), ethnicity, smoking status
(ever, never, and former), duration of smoking cessation in
566 RecruitIng Healthy Control Subjects
3 The abbreviations used are: UTMDACC, University of Texas M. D. Anderson
Cancer Center; KS, Kelsey Seybold.
former smokers (±5 years), and pack-year smoking history informer smokers (± 10 pack-years). To be included as a casesubject, individuals must be a minimum of 18 years old with a
histologically confirmed lung cancer who have not had chem-otherapy or radiotherapy within the preceding year. Case sub-jects are enrolled without regard to sex, ethnicity, cancer stage,or cancer histology. Control subjects must have no previoushistory of cancer, with the exception of nonmelanoma skincancer. The study has been described in detail previously (1).
Materials and Methods
Mechanisms for Identifying Case Subjects. Because cytoge-netic studies are being performed, recent prior chemotherapy orradiotherapy are exclusion criteria. Therefore, the challenge incase enrollment is early ascertainment, before initiation of
therapy.Although the majority of cases are being identified
through UTMDACC3 in Houston, procedures are in place foradditional case ascertainment through the KS Foundation andseveral hospitals in San Antonio (to enhance the Mexican-American proportion of cases) as well as the Ben Taub County
Hospital and the Veterans Affairs Hospital in Houston. Here,we describe the methods of case ascertainment at UTMDACC.
Staff interviewers conduct a daily review of computerizedappointment schedules for the hospital outpatient clinics thatserve lung cancer patients (thoracic medical and surgery din-
ics). Each new patient is asked to complete a brief questionnairethat assesses smoking status and prior therapy as well as the
individual’s interest in participating in an epidemiologicalstudy. Interviewers match the questionnaires with their com-
puterized listings to identify potentially eligible subjects. After
obtaining permission from the patient’s physician, interestedpatients are given an informational brochure, and the inter-viewer explains the study to the potential participant. Afterinformed consent is obtained, an in-depth interview is con-
ducted, and a 30-mI blood sample is procured and hand-deiv-ered to the epidemiology research laboratories for processing
and storage. Each patient who provides a blood sample andcompletes the interview receives a $20 voucher for use at a
local supermarket.
Selection of a Comparison Population. The principles un-derlying appropriate control selection have been well de-scribed (2-4). However, in practice, adherence to theseprinciples is often problematic and constrained by resources,time, and feasibility. Furthermore, these difficulties are mag-nified when the cases derive from an urban cancer center thatserves as a tertiary referral center. Wacholder et a!. (3) have
outlined the benefits and drawbacks of choosing medicalpractice controls. Because population-based control selec-tion was not considered an option for our research, we
decided to use the diverse catchment area of the largesthealth maintenance organization in the Houston metropoli-
tan area. This comparison population is not necessarily arepresentative sample of the population base of the cases and
may differ by socioeconomic status and other unmeasureddeterminants. However, our study design mandated closematching on smoking status; therefore, we are indirectly
matching on correlates of smoking behavior (e.g., education,
health behaviors, and others). Moreover, because our re-search focus was entirely driven by genetic hypotheses (i.e.,
cancer risk is associated with a susceptible, exposed sub-
population), representativeness was less of a concern than if
we were considering only environmental determinants of
lung cancer.
Mechanisms for Identifying Control Subjects. Because weare matching a comparison group on restrictive smoking criteria
as well as by age, sex, and ethnicity, our study requires a large
pool from which to select control subjects. Thus, healthy sub-
jects to serve as controls are being recruited through the KS
organization, with the assistance of the KS Foundation research
staff. KS is a large, private, multispecialty health-maintenance
organization that services capitated, Medicare, Medicaid, and
fee-for-service patients. The organization has provided clinical
services for over 50 years and has developed a large and
heterogeneous client base. Approximately 800,000 patient con-
tacts are recorded annually from approximately 250,000 mdi-
viduals enrolled in its Houston area clinics. The KS Foundation
is a not-for-profit organization that is interested in outcomes
research, clinical trials, and cancer prevention. We have estab-
lished a collaborative agreement with the KS clinics and Foun-
dation to identify, enroll, and interview a healthy comparison
population for our studies. It is our goal to identify 40,000 KS
enrollees to comprise the healthy subject pool from which we
will select matched control subjects.
We developed and pilot tested a one-page questionnaire
that elicits all relevant criteria for control selection and evalu-
ates participant interest. This form is distributed by KS person-
nel during registration at a scheduled appointment. Because the
questionnaire is attached to KS forms routinely distributed
during registration, we designed our form to be distinctly dif-
ferent from other KS forms (such as paper color and format).Patients are instructed to complete the questionnaire and de-
posit it in a designated collection box that has been placed
conspicuously in the clinic waiting area. The collection boxesare designed to coordinate with the questionnaire color and
style. A Plexiglas holder is attached to the box to hold pencils
and additional questionnaires that may be completed by persons
accompanying patients. Leaflets summarizing the study and
listing the telephone number ofthe project director are provided
with the collection boxes. Both the questionnaires and the
leaflets are printed in English on one side and Spanish on the
reverse side.
Prior to full-scale implementation across the 20 clinic
sites, the distribution methods were pilot tested in the largest
KS clinic site. The pilot study, conducted over a period of 4
weeks, provided feedback to refine the questionnaire and
method of distribution. The final questionnaire assesses: (a)
variables pertinent to the matching criteria for the case-control
study, including sex, date of birth, ethnicity, smoking status,
age at smoking initiation, age at smoking cessation (former
smokers), number of years smoked, number of cigarettessmoked per day, and history of previous cancer; (b) contact
information, including address and telephone number; and (c)willingness to be contacted for further questioning, posed by
the question,”Would you be willing to answer more questions
if contacted?” Subjects who respond either “yes” or “maybe”
are classified as potential participants. For the purpose of this
paper, willingness to be contacted will be considered as will-ingness to participate in one of the case-control studies. An-
other item, added for the benefit of the KS organization, in-quires if respondents would like to receive smoking cessation
materials from the KS health promotion department. The over-
all study and the final questionnaire have been reviewed and
1995 1996
Fig. I. Monthly accrual of questionnaires from KS clinic sites. The pilot phase
was conducted at KS’s largest clinic site. Initiation of questionnaire administra-
lion at the remaining 19 clinic sites occurred over the span of months 1-4. No
questionnaires were distributed between the pilot phase and month 1.
Cancer Epidemiology, Biomarkers & Prevention 567
approved by the Institutional Review Boards for the Protection
of Human Subjects at UTMDACC and KS.
Over a period of 4 months, questionnaire distributionprocedures were phased in at the remaining 19 clinic sites.
Each Thursday, a designated person at each clinic site for-wards all completed questionnaires to the KS Foundation.
Questionnaires are tallied by the KS staff and retrieved by amember of our study personnel on the following Monday. At
this time, the questionnaire data are routed through data
cleaning and consistency checks, followed by coding and
data entry. Questionnaires are entered into the project data
base within 7 days of their receipt, and variables that are
imperative to the case-control matching process are double-
keyed to ensure accuracy.Questionnaire accrual rates are monitored through
weekly reports that enumerate the number of questionnaires
obtained at each clinic site. Each Monday, KS is provided
with a report describing the previous week’s questionnaire
accrual number; on the second Monday of each month, themedical director of the KS Foundation receives the same
report as well as an aggregate report characterizing the study
population. These reports are generated weekly so that prob-lems in accrual may be identified and addressed rapidly. The
majority of feedback provided to the clinic sites is admin-
istered by the KS Foundation; however, our project coordi-
nator and a KS representative visit each site approximately
every 3 months to re-establish contacts and provide incen-
tives, thereby enhancing the study’s visibility.
For each case subject accrued, there is a computer-
generated list of up to five potential matched control sub-
jects. The selected control subjects are contacted by tele-
phone to confirm willingness to participate, and an
appointment is scheduled at a KS clinic site convenient to
the participant. At the designated meeting time, the control
subject is greeted by a KS or UTMDACC interviewer and
escorted to a KS phlebotomist who draws the necessary
blood sample. The interviews, which take approximately 1 h,
are conducted by a trained interviewer from the UTMDACC
study staff. Each participating control subject is compen-
sated with a $20 voucher for use at a local grocery store and
parking validation. The blood samples are delivered to the
epidemiology laboratory by the staff interviewer.
Analytic Procedures. Characteristics of the respondents are
summarized using descriptive statistics. For the analysis of
factors associated with willingness to participate among
eligible control subjects, a dependent variable (“willingness
to participate”) was created by combining subjects who
responded “yes” with subjects who responded “maybe” to
the questionnaire item, “Would you be willing to answer
more questions if contacted?” This resulted in a dichotomous
yes/no variable. Because smoking status was associated with
willingness to participate as well as multiple predictor van-
ables, analyses were stratified by smoking status (never,former, and current). Relationships with the dependent van-
able were assessed using x2 tests of independence for cate-
gonical variables and t tests for continuous variables. Among
former and current smokers, multivariate tests of associa-
tions between smoking-related variables and willingness to
participate were conducted using logistic regression, con-
trolling for sex, race, and age. For the 121 control subjects
who had been enrolled into the study at the time of manu-
script submission, we assessed concordance between the KSwaiting room questionnaire data and personal interview data.
2,000�0a)
i� 1,500C,)
CC.2 1,000
a)a-
.� 500a)
.0
E
z
Pilot 01 02 03 04 05 06 07 08 09 10 11
Month
All statistical procedures were conducted using SPSS forWindows, version 6. 1 (5).
Results
Accrual Rates for Control Population Questionnaires. As
described above, accrual to the pool of control subjects is
monitored over time. The KS Foundation is unable to pro-vide denominator data by clinic; therefore, actual responserates cannot be computed. Questionnaire accrual rates fluc-
tuate with seasonal variation in acute illnesses and holidaysand will decrease over time as the proportion of KS patientswho have already completed a questionnaire increases (be-
cause of repeat visits during the accrual period). Fig. Idepicts accrual of questionnaires over time, including ac-
crual during the pilot phase. Since initiation of questionnairedistribution at the final KS site (as of month 5), we havereceived an average of 2228 questionnaires each month. Acomparison of the cumulative statistics on a weekly basis
now shows little change in the summary statistics data as
new respondents are added to the pool of controls.
Characteristics of Questionnaire Respondents and Willing-ness to Participate. After 12 months of accrual (including I
month of accrual during the pilot phase), questionnaires havebeen received from 14,624 KS patients. Of these, 11,763(80.4%) provided sufficient data for subject contact andmatching purposes. There were 454 persons deemed ineli-gible to serve as a control subject because they had a per-sonal history of cancer other than nonmelanoma skin cancer.
As shown in Table 1, of 14,624 questionnaires received,8,691 (59.4%) of respondents: (a) have provided sufficientquestionnaire data for contact and matching purposes; (b)have no personal history of cancer other than nonmelanoma
skin cancer; and (c) are not opposed to participation in thestudy. The majority of eligible respondents were female(66.2%) and white (59.0%), with a mean age (SD) of 42.1(14.0). Over one-half never smoked, and 22.7% designated
that they formerly smoked but had quit.Current smokers were least likely to refuse participation
when compared with former and never smokers (Table I;18.7% refusal for current smokers versus 25.5% and 20.8%for never and former smokers, respectively; x2’ 2 df = 53.0,P < 0.0001). Table 2 shows the relationships ofsex, age, andrace with willingness to participate, stratified by smoking
568 RecruIting Healthy Control Subjects
Table I Characteristic
Variable
s of study-eligible KS questionnaire respondents who provided adequate data for contact and matching purposes, b
Smoking status
y smoking status
n 11,309Category Never smoker Former smoker Current smoker
n 6,599 n 2,572 n 2,138
Sex Male 1,813 (27.5r 1,180 (45.9) 834 (39.0)
Female 4,786 (72.5) 1,392 (54.1) 1,304 (61.0)
3,827 (33.8)
7,482 (66.2)
Race White 3,433 (52.0) 1,860 (72.3) 1,379 (64.5)
African-American 1,824 (27.6) 416 (16.2) 474 (22.2)
Hispanic 1,018 (15.4) 248 (9.6) 228 (10.7)
Other 324 (4.9) 48 (1.9) 57 (2.7)
6,672 (59.0)
2,714(24.0)
1,494 (13.2)
429 (3.8)
Age in years 18-39 3,631 (55.0) 708 (27.5) 1,067 (49.9)
40-49 1,605 (24.3) 672 (26.1) 598 (28.0)
50-59 727 (1 1.0) 515 (20.0) 296 (13.8)
60+ 636(9.6) 677 (26.3) 177 (8.3)
5,406(47.8)
2,875 (25.4)
1,538 (13.6)
1,490 (13.2)
History of cancer Cancer free 6,510 (98.7) 2,495 (97.0) 2,122 (99.3)
Nonmelanoma skin cancer 89 (1.3) 77 (3.0) 16 (0.7)
1 1,127 (98.4)
182 (1.6)
Willingness to Yes 2,870 (43.5) 1,359 (52.8) 1,142 (53.4) 5,371 (47.5)
participate Maybe 2,045 (31.0) 679 (26.4) 596 (27.9)
No 1,684 (25.5) 534 (20.8) 400 (18.7)
3,320(29.4)
2,618 (23.1)
a Number (percentage).
Table 2 Factors associated with willingness to participate: age, sex, and race, by smoking status
Smoking status Variable Category Willing Not willing P”
Never smokers Sex Male 1,328 (73.2)” 485 (26.8)
Female 3,587 (74.9) 1,199 (25.1)
Age 18-39 2,615 (72.0) 1,016(28.0)
40-49 1,237 (77.1) 368 (22.9)
50-59 571 (78.5) 156 (21.5)
60+ 492 (77.4) 144(22.6)
Race White 2,537 (73.9) 896 (26.1)
African-American 1,372 (75.2) 452 (24.8)
Hispanic 765 (75.1) 253 (24.9)
Other 241 (74.4) 83 (25.6)
0.16
0.00
0.71
Former smokers Sex Male 965 (81.8) 215 (1 8.2)
Female 1,073 (77.1) 319(22.9)
Age 18-39 552 (78.0) 156(22.0)
40-49 521 (77.5) 151 (22.5)
50-59 416 (80.8) 99(19.2)
60+ 549(81.1) 128(18.9)
Race White 1,492 (80.2) 368 (19.8)
African-American 319 (76.7) 97 (23.3)
Hispanic 190 (76.6) 58 (23.4)
Other 37 (77.1) 1 1 (22.9)
0.00
0.26
0.27
Current smokers Sex Male 684 (82.0) 150 (18.0)
Female 1,054 (80.8) 250(19.2)
Age 18-39 873 (81.8) 194 (18.2)
40-49 473 (79.1) 125 (20.9)
50-59 260 (87.8) 36(12.2)
60+ 132 (74.6) 45 (25.4)
Race White 1,084 (78.6) 295 (21.4)
African-American 421 (88.8) 53 (11.2)
Hispanic 182 (79.8) 46(20.2)
Other 51(89.5) 6(10.5)
0.49
0.00
0.00
a p associated with g� tests of independence.
b Number (percentage).
status. Among never smokers and current smokers, the pro-portions of men and women interested in participation were
similar. For former smokers, however, men were signifi-cantly more willing to participate than women (81.8% versus77.1%). Age was a significant predictor of willingness toparticipate among never smokers and current smokers, with
willingness being highest in individuals between 50 and 59years old. Race predicted willingness to participate only in
current smokers. Whites and Hispanics had similar refusalrates, which were nearly double the refusal rates for mdi-viduals of African-American or “other” origin.
Table 3 shows relationships of smoking-related variableswith willingness to participate. Controlling for sex, age, and
race, former smokers who did not refuse participation were
heavier smokers who had smoked for a longer duration. Incurrent smokers, persons who were not opposed to participation
0
Age stopped smoking (Na68)
-13-8 .6 -5 -4 -3 -2 -1 0 1 2 3 4 8 10141523
Discrepancy in years for reported age at smoking cessation
Number of cigarettes smoked per day (N63)
IDiscrepancy in reported number of cigarettes smoked per day
Cancer Epidemiology, Bioinarkers & Prevention 569
Table 3 Factors associated with willingnes s to participate: smokin g variab les, by smoking status
.Smoking status
.Variable
n
Willing
Mean (SD) n
Not willing
Mean (SD)P’�
Multivariateb
P
Former smokers
Current smokers
Age (years) at smoking initiation
Number of cigarettes/day
Years smoked
Pack-years smoked
Age (years) at smoking initiation
Number of cigarettes/day
Years smoked
Pack-years smoked
2,031
2,030
2,038
2,033
1,718
1,691
1,719
1,681
18. 1 (4.4)
19.9 (14.5)
15.5 (1 1.7)
18.7 (22.7)
18.4 (4.9)
17.1 (10.9)
19.1 (12.4)
18.9(20.8)
532
533
534
533
391
385
394
383
1 8.4 (5. 1)
16.8 (13.6)
13.7 (1 1.7)
14.4 (21.7)
18.7 (5.3)
17.2 (12.3)
18.5 (12.9)
18.7 (22.7)
0. 12
0.00
0.00
0.00
0.19
0.87
0.42
0.88
0.20
0.00
0.02
0.00
0.02
0.31
0.02
0.15
a p associated with Student’s t tests.b Derived using logistic regression, controlling for sex, age, and race (control variables categorical, as shown in Table 1).
Age started smoking (N=108)
Discrepancy in years for reported age at smoking initiation
Years smoked (Na106)
�Ji�.23-13-10-7 .6 -5 .4 .3 -2 .1 0 1 2 3 4 5 6 9 101624
Discrepancy in years for reported number of years smoked
Fig. 2. Concordance between KS waiting room questionnaire and personal interview data. Difference scores were calculated as the value reported during personal
interview minus the value reported on the waiting room questionnaire. For “years smoked” data, the values for current smokers were adjusted for time elapsed between
KS questionnaire completion and personal interview data collection.
began smoking at an earlier age and smoked a greaten numberof years than did their counterparts.
Concordance between KS Questionnaire Data and PersonalInterview Data. To date, 121 matched control subjects havebeen enrolled in the study. Of these, we had complete agree-ment between KS questionnaire data and personal interviewdata for 99% of subjects on the variable sex, 100% of subjectsfor race, 99% of subjects for date of birth, and 100% of subjectsfor smoking history (ever smoker versus never smoker). Fig. 2shows the extent of agreement for four smoking variables: agein years at smoking initiation, age in years when stoppedsmoking (former smokers), number of cigarettes smoked per
day, and the number of years smoked. For the number ofcigarettes smoked per day, the comparison was limited to
former smokers because this construct would not be consideredstatic among current smokers. Discrepancy was calculated as
the value reported during personal interview minus the valuereported on the waiting room questionnaire, and the zero values
on the graphs indicate agreement between the two methods of
data collection.
For age at smoking initiation (current and former smokers)
and age at smoking cessation (former smokers), 84 and 74%had agreement within 2 years, respectively. Among the former
smokers enrolled as a control subject, 64% had acceptablereliability in reported number of cigarettes smoked per day,with values differing by no more than five cigarettes per day.
For the reported number of years smoked among current and
former smokers, 84% had agreement within 5 years.
570 Recruiting Healthy Control Subjects
Discussion
This paper describes a successful methodology applicable to the
managed care environment that may be useful in studies re-quiring the identification of a large pool of healthy subjects.
Additionally, we contrasted characteristics of the pool of po-
tential control subjects who reported an interest in participating
in the medical research with those who requested that they not
be contacted for the purpose of research. This comparison helpscharacterize our “eligible, willing control pool” with respect to
the population of otherwise eligible subjects who presented at
a KS clinic and completed a waiting room questionnaire during
the accrual period.To attain a sufficiently large pool of eligible and willing
control subjects to enable stringent matching of case and con-
trol subjects, questionnaires are being administered to all adultspresenting at Houston-area KS clinics. To keep KS adminis-
trative personnel closely informed about the project’s progress,
we provide feedback that includes a weekly report of question-name accrual at each clinic site as well as a monthly report of
prevalence data describing the smoking patterns of their clinic
patient population. Also, a list of the names of questionnaire
respondents who indicate that they are interested in receiving
smoking cessation materials (for themselves or for others) isforwarded to the KS health promotion department; approxi-mately 20% of respondents make this request. Continuous
feedback to the KS organization is important in facilitating
successful long-term collaboration.Despite close monitoring of questionnaire accrual rates,
we experienced a temporary reduction in questionnaire accrual
rates after the initial kick-off at the KS clinics. It becameapparent that regular site visits to clinics, phone calls, and small
incentives were not enough to maintain the KS clinic person-
nd’ s interest in our project. The purpose and benefit of theproject, as well as the critical role of the KS personnel in
distributing questionnaires to patients, had not been communi-cated sufficiently at all clinics.
In month 10, we addressed this problem by hosting aluncheon at UTMDACC. Key personnel from each clinic were
invited, and the invitations were distributed by the KS director
of clinic operations. All but one KS clinic was represented atthe meeting. In preparation for the meeting, we videotaped
personnel at three clinics that had been consistently successfulin questionnaire distribution. Each of the three clinics demon-
strated the mechanism they used to distribute questionnaires at
their facilities. We anticipated that the videotapes would helpthe personnel at other clinics to generate ideas for improving
distribution at their own locations. We also videotaped theprocess used at UTMDACC to recruit lung cancer case subjectsinto the study. The results of our efforts appeared to be positive,
with response rates increasing from 791 questionnaires col-
lected in month 10 to 1495 questionnaires in month 11, an 89%
increase.
To assess possible bias in willingness to participate among
the control pool (those who completed a questionnaire), wecompared study-eligible subjects who were not interested inparticipation with those who were either willing or were unsure.Several differences were observed, but without having con-
ducted more in-depth assessments, we are not able to describereasons underlying the differences. For example, it is not clear
why female former smokers were more likely to refuse partic-ipation than male former smokers, yet there was no difference
by sex among never smokers and current smokers. Because ofthe practical limitations associated with a waiting-room ques-
tionnaire, we could not assess the multiple sociobehavioral
factors that previously have been shown to be associated with
cancer prevention behavior and participation in trials in general
(6-26). It was surprising, however, that current smokers were
least likely to refuse participation because, in general, individ-
uals interested in participating in cancer prevention research
tend to be more highly educated, have higher annual family
incomes, be regular vitamin users, and be more concerned
about getting cancer (22), characteristics not typically associ-
ated with smokers.
We were able to gain some limited information character-
izing the pool of potential control subjects who were not averse
to participation with respect to the overall sample of KS en-
rollees who completed a questionnaire. The extent to which the
characteristics of the control pool reflect the general population
of the KS patient base is not known, and our methods do not
afford the opportunity to assess the proportion of waiting room
patients who completed a questionnaire. Furthermore, it is
possible that the proportion of respondents who indicated that
they may be willing to participate and those who are matched
as controls and actually do participate, once contacted, may
differ. To date, 121 of 139 subjects (87%) contacted to serve as
a control subject have been enrolled as controls. The number of
subjects who chose not to participate when contacted (n = 18)
currently is too small to reliably assess for possible participa-
tion bias occurring at this phase of the recruitment process. In
the future, as our estimate becomes more stable, data regarding
the participation rate among contacted persons will be factored
into the overall “willingness to participate” rates. In comparing
the waiting room questionnaire data with personal interview
data collected from the enrolled control subjects, we observed
high levels of agreement for sociodemographic factors and
reports of ever smoking. In contrast, there was some variation
in reports of age of smoking initiation and smoking cessation,
number of years smoked, and number of cigarettes smoked per
day. When the interview data are unacceptably discrepant from
the KS questionnaire data, hence voiding the “match,” we
reselect another matched control subject. The original control is
kept for future matching to a case.
It is important to reinforce that because the cases do not
derive from defined populations, the case-control studies de-
scribed here are not population based. Consequently, compa-
rability of the two base populations cannot be assumed, and
there are inherent limitations associated with our methods of
subject selection (2, 3). This potential bias is less problematic,
however, when the study hypotheses are focused on genetic
susceptibility factors as opposed to lifestyle- or exposure-de-
pendent factors.
In summary, it is our experience that although the appro-
priateness of the population may be controversial for some
studies, developing a collaborative agreement with a large
managed care organization offers a feasible and workable
mechanism through which investigators can identify and access
large numbers of healthy persons of diverse ethnicities and
socioeconomic backgrounds who are interested in participating
in medical research. Once established, the pool becomes a
valuable resource for the selection of healthy subjects for both
case-control and cohort studies. Because the number of molec-
ular epidemiological case-control studies is growing rapidly,
studies of perceived benefits and barriers of participation, as
well as factors associated with willingness to participate in
these types of studies, would be worthy contributions to the
literature.
Cancer Epidemiokaj�, Biomarkers & Prevention 571
Acknowledgments
We thank Joanne Sider for questionnaire formatting and development, Betty
Henry and Pat Pillow for data inspection and cleaning, Mary Harris and SecnobiaEaster for data entry, and Pain Porter and Mary McCabe for conducting panic-
ipant interviews. We also thank all KS Foundation personnel who have made this
project possible, including Terry Litchfield, Fran Greene, Cynthia Sustaita, andthe clinic personnel.
References
1. Spitz, M. R., Hsu, T. C., Wu, X. F., Fueger, J. J., Amos, C. I., and Roth, J. A.
Mutagen sensitivity as a biologic marker of lung cancer risk in African Amen-
cans. Cancer Epidemiol., Biomarkers & Prey., 4: 99-103, 1995.
2. Wacholder, S., McLaughlin, J. K., Silverman, D. T., and Mandel, J. S.
Selection ofcontrols in case-control studies. I. Principles. Am. J. Epidemiol., 135:
1019-1028, 1992.
3. Wacholder, S., Silverman, D. T., McLaughlin, J. K., and Mandel, J. S.
Selection of controls in case-control studies. II. Types of controls. Am. J.
Epidemiol., 135: 1029-1041, 1992.
4. Wacholder, S., Silverman, D. T., McLaughlin, J. K., and Mandel, J. S.
Selection of controls in case-control studies. HI. Design options. Am. J. Epide-
miol., 135: 1042-1050, 1992.
5. SPSS, Inc. SPSSR for Windows�”. Advanced Statistics, Release 6.1, Chicago,
IL, 1994.
6. Johnson, R. A., and Murata, P. J. Demographic, clinical, and financial factorsrelating to the completion rate of screening mammography. Cancer Detect. Prey.,
11: 259-266, 1988.
7. Evans, A. M., Love, R. R., Meyerowitz, B. E., Leyenthal, H., and Nerenz,
D. R. Factors associated with active participation in a cancer prevention clinic.Prey. Med., 14: 358-371, 1985.
8. Rutledge, D. N., Hartmann, W. H., Kinman, P. 0., and Winfield, A. C.
Exploration of factors affecting mammography behaviors. Prey. Med., 17: 412-
422, 1988.
9. SandIer, R. S., DeVellis, B. M., Blalock, S., and Holland, K. L. Participation
of high-risk subjects in colon cancer screening. Cancer (Phila.), 63: 221 1-2215,
1989.
10. Yeomans-Kinney, A., Vernon, S. W., Frankowski, R. F., Weber, D. M.,
Bitsura, J. M., and Vogel, V. 0. Factors related to enrollment in the breast cancerprevention trial at a comprehensive cancer center during the first year of recruit-
ment. Cancer (Phila.), 76: 46-56, 1995.
1 1. Halabi, S., Vogel, V. G., Bondy, M. L., and Vernon, S. W. Recruiting older
women for screening mammography. Cancer Detect. Prey., 17: 359-365, 1993.
12. Roetzheim, R. G., Van Durme, D. J., Brownlee, H. J., Herold, A. H.,
Woodard, L. J., and Blair, C. Barriers to screening among participants of a
medi-promoted breast cancer screening project. Cancer Detect. Prey., 1 7: 367-
377, 1993.
13. Lerman, C., Rimer, B. K., Trock, B., Baishem, A., and Engstrom, P. F.
Factors associated with repeated adherence to breast cancer screening. Prey.
Med., 19: 279-290. 1990.
14. Fletcher, S. W., Morgan, T. M., O’Malley, M. S., Earp, J. L., and Degnan, D.
Is breast self-examination predicted by knowledge, attitudes, beliefs, or sociode-mographic characteristics? Am. J. Prey. Med., 5: 207-215, 1989.
15. Zapka, J. G., Stoddard, A. M., Costanza, M. E., and Greene, H. L. Breast
cancer screening by mammography: utilization and associated factors. Am. J.
Public Health, 79: 1499-1502, 1989.
16. Fajardo, L L, Saint-Germain, M., Meakem, T. 3., Rose, C., and HilIman,
B. J. Factors influencing women to undergo screening mammography. Radiology,
184: 59-63, 1992.
17. Champion, V. L. Compliance with guidelines for mammography screening.
Cancer Detect. Prey., 16: 253-258, 1992.
18. Calnan, M. The health belief model and participation in programmes for the
early detection of breast cancer� a comparative analysis. Soc. Sci. Med., 19:
823-830, 1984.
19. Rimer, B. K., Trock, B., Engstrom, P. F., Lerman, C., and King, E. Why do
some women get regular mammograms? Am. J. Prey. Med., 7: 69-74, 1991.
20. Rimer, B. K., Davis, S. W., Engstrom, P. F., Myers, R. E., and Rosan, J. R.
Some reasons for compliance and noncompliance in a health maintenance orga-nization breast cancer screening program. J. Compl. Health Care, 3: 103-114,
1988.
21. Sutherland, H. J., Carlin, K., Harper, W., Martin, L J., Greenberg, C. V., Till,3. E., and Boyd, N. F. A study ofdiet and breast cancer prevention in Canada: why
healthy women participate in controlled trials. Cancer Causes & Control, 4:
521-528, 1993.
22. Mettlin, C., Cummings, K. M., and Walsh, D. Risk factor and behavioral
correlates of willingness to participate in cancer prevention trials. Nutr. Cancer,
7: 189-198, 1985.
23. Miller, H. H., Bauman, L. J., Friedman, D. R., and DeCosse, J. J. Psycho-
social adjustment of familial polyposis patients and participation in a chemopre-
vention trial. hit. J. Psychiatry Med., 16: 21 1-230, 1986.
24. Lerman, C., Rimer, B. K., Daly, M., Lustbader, E., Sands, C., Balshem, A.,
Masny, A., and Engstrom, P. F. Recruiting high risk women into a breast cancerhealth promotion trial. Cancer Epidemiol., Biomarkers & Prey., 3: 27 1-276,
1994.
25. Hudmon, K. S., and Kinney, A. Y. Issues of enrollment bias in cancer
chemoprevention trials. Cancer Bull. (Houst.), 47: 339-342, 1995.
26. Hudmon, K. S., Chamberlain, R. M., and Frankowski, R. F. Outcomes of aplacebo mn-in period in a head and neck cancer chemoprevention trial. Controlled
Clin. Trials, 18: 228-240, 1997.