Tobacco Documents Online

1Cli~x Epidemtot Vol. 50, No. 3, pp. 253-~.63, 1997 Copyright © 1997 Elsevier Science Inc. ELSEVIER 0895-4356/97/$ PII S0895-4356(96)00378-2 Ro National Incidence of Smoking and Misclassification Among the U.S. Married Female Population Michael W. Ogden,* Walter T. Morgan, David L. Heavner, Riley A. Davis, and Thomas J. Steiche.n J. REYNOLDS TO~^CCO COMPANY, RESEARCH (~ DEVELOPMENT, WINSTON-SALEM, NORTH CAROLINA, 27102-1236 ABSTRACT. Because of a lack of representative data on smoking status misclassification among U.S. married females, a two-part study was conducted. Part I w-as conducted to obtain nationally representative estimates of the percentage of U.S. women who report themselves to be current, former, and never smokers, to determine the concordance ot~ smoking habits among spouse pairs, and to establish field quotas and probability weightings for Part II. Part II was conducted to determine smoker misclassification rates using salivary cotinine as an indica- tion of active smoking. Part I, conducted in January 25-29, 1992, utilized random-digit dialing telephone inter- ,clewing throughout the 48 contiguous United States. Part II, conducted from February 19, 1992 to March 7, 1992, was a mall-intercept study in nine geographically disperse U.S. cities and it involved interviewing and saliva collection. Among married U.S. women, 25% reported they were current smokers, 22% reported they were former smokers, and 53% reported they were never smokers. Using a cotinine concentration of either >35 ng/ml or > 106 ng/mt to indicate regular smoking, 3.61% and 2.55% of regular smokers, respectively, reported themselves to be never smokers. The concordance ratio, an important parameter in correcting for non-differential misctassificadon bias, was found to be 5.52. In addition, an indication of substantial differential misclassification was found between exposed and unexposed populations. This type of misctassification bias has previously not been accounted for in the adiustment of epidemiology-based risk assessments of tobacco smoke exposure and lung cancer. Taken together, these data suggest that misclassification bias alone is likely to explain any lung cancer risk elevation observed in the U.S. epidemiolog3" of environmental tobacco smoke exposure among non- smoking women. ~ CLaN Eems.mo~. 50;3:253-263, 1997. © 1997 Elsevier Science Inc. KEY WORDS. Bias, concordance, cotinine, misclassilication, smoking, tobacco INTRODUCTION Misrepresentation of smoking status is potentially a major bias in the epidemiology-based risk assessment of environ- mentaI tobacco smoke (ETS) exposure [1,21. Proper adjust- ment of individual epidemiology studies or of groups of recta-analyzed studies requires that misclassification rates be known precisely for the target population under study. Al- most exclusively, the epidemiologic investigations that have attempted to study ETS risk have used "spousal smok- ing" as the surrogate for ETS exposure of self-reported never smoking women. As a result, a major target population for ETS-related epidemiology is married females, and the most important type of misclassification is the degree to which regular smoking, married females report themselves to be r~ever smokers. Several studies that are occasionally cited f'or smoking • status.misclassification in the U.S. are those of Haddow et 'Address for correspondence: Dr. Michael Ogden, R. J. Reynolds Tobacco ~0., R&D, P.O. Box 1236, Winston-Salem, NC 27102-1236. Ac-cegted for publicatiori on 18 October 1996. a/. [3,4], Coultas et a/. [5], and Cummings eta/. [6]. Although these studies warrant consideration, each is severely limited in representativeness for the target U.S. population. Appli- cation of misclassification rates derived from these studies to epidemiologic risk estimates may seriously bias the out- come. For example, one of these studies did not differentiate never smokers from former smokers and employed only Caucasian, pregnant subjects [4]. Another study used only Hispanic subjects in New Mexico [5] and anotb.er used only patients in a cancer-screening clinic who knowingly volun- teered for a study on tobacco smoke exposure [6]. These studies comprise the bulk of the U.S. data on misclassifica- lion employed in a recent risk assessment of ETS and respi- ratory heakh effects, including lung cancer [7]. To date, there has been no effort to study the prevalence of smoking status misclassification in a random cross-section of the targeted (i.e., married female) U.S. population. Be- cause there is a substantial lack of understanding about this important parameter, the current study was implemented. To this end, the current study defined exposure in accor- dance with the vast majority of the U.S. female spousab

254 M.W. Ogden er al. smoking epidemiology wherein the definition of "ETS-ex- posed" is strictly based upon the smoking status of the sub- ject's husband. MATERIALS AND METHODS Part I--National Incidence Survey A two-part study was designed and executed. Part I, the National Incidence Survey (NIS) was conducted to esti- mate the percentage of U.S. women (age 18+, married, and living with their spouse) who smoke. Other objectives were to estimate smoking status concordance among married couples and to establish cell quotas for the field determina- tion of current smoker misclassification rares (Parr II). Cells were defined by all combinations of the subject's smoking status (current smoker, former smoker, never smoker) and her husband's current smoking status (smoker, nonsmoker). Criteria used to assign smoking status were as follows. A current smoker was one who reported current use of ciga- rettes. A former smoker was one who, at any time in her life, had ever smoked more than five cigarettes per week for a period of at least six months but reported complete abstention for at least six months prior to interview. A never smoker was one who had never smoked more than 20 cigarettes in her life. Former smokers were also asked if they were currently using any type of nicotine-containing product as an aid in smoking cessation (gum, patch, inhaler, etc.) and, if so, they were excluded. Subjects not meeting these criteria for cell assignment were also excluded. All telephone interviews were conducted by a nationa[ market research agency from a central location on Januat3" 25-29, 1992. An ending sample size of 1200 was chosen to minimize sampling error (---3%). The beginning sample size needed to complete the intetn'iewing process was estimated to be approximately 11,000 based on assumptions regarding the percentage of non-working telephone numbers, respon- dent rethsal rates, and qualif3"ing households. However, dur- ing the interviewing process, an additioiaal sample popula- tion (3750) was needed due to higher than expected refusal rates (especially in the Pacific time zone), non-qualified households (percentage of households without a female 18+ years who was currently married and living with her spouse), and respondent non-availability (percentage of an- swering machines, no answers, and callbacks). The sample was divided into 13 replicates of 1100 and one replicate of 450 (Pacific time zone only). The sample frame for the study was all telephone households in the con- tiguous 48 states. A modified random digit dialing (RDD) sample was selected through an IN-HOUSE GENESYS SAMPLING SYSTEM® (lVlarketing Systems Group, Phi[a- delphia, PA). The sample was drawn by county, within state, in proportion to the population ( 1.990 Census). Selec- tions within state and county were proportionate to the size of the telephone exchanges within the political boundaries. QUESTIONNAIRE DEVELOPMENT. All. questionnaires (Parrs I and II) were developed by the national market re- search agency staff in consultation with client technical staff (the authors). The technical staff's responsibility was to ensure the scientific validity of the study design. The marker research agency staff's responsibility was to ensure that the questionnaire wording was as clear and unambigu- ous as possible and to ensure that the objectives of the study and the identity of the study sponsor were not revealed to either the subjecr.s in Part I and II or to the local field service staff/interviewers in Part The standard etiologic definition of "ETS exposed" was used where the female was "exposed" only if her spouse smoked. This is the definition most commonly employed in ETS epidemiology and risk assessment [7]. INTERVIEWING PROCEDURE. All supervisors were briefed with regard to the interviewing procedure/protocol, sample monitoring, and survey confidentiality. Supervisors and in. terviewers participated in hands-on computer briefing (questionnaire wording, skip patterns, dispositions, etc.). During interviews, supervisors ensured that interviewers strictly followed the protocol and questionnaire. Screening and interviewing were conducted via the Athena-CATI ("Computer Assisted Telephone Interviewing") system (CRC Inc.. New York, NY). Interviewing (via WATS from a central location) began on Saturday, Januar3," 25 and ended Wednesday, January 29, 1992. Each inter,'iewer was monitored randomly for correct administration of the ques- tionnaire and accuracy of data entry.. ~XTERVtEWINO PROTOCOl.. Throughout the five days of interviewing, interviewer hours were scheduled and moni- tored" to ensure that each listed household was given an equal opportunity of being contacted. Interviewing hours were spread across the four time zones in proportion to the quantity of numbers in each time :one. Furthermo're, inter. viewer hours were allocated equally across each day (9 am- 1 pm, I pm-5 pm, and 5 pro-closing) within each time :one. (Closing refers to the hours of operation of the marke:" research agency telephone center, which was always at 9 pm EST. Later hours were used for calls to the Pacific time :one.) As a replicate was exhausted, i.e., dispositior~ of all remaining telephone numbers was "Busy," "No An. swer," "Answering Machine," or "Callback," the next rep{i. care was queued. Call histories (disposition printouts) were analyzed twice daily to control/apportion interviewin~ hours and monitor sample utilization. Part II~Srnoker Misclassification Study The primary objective of Part II was to determine the rat~ at which current, regular smokers report themselves to 1:, never smokers (regular smoker misclassification rate). Suk ject qualifications (age 18+, married and living wid

InCidence of Smoking and Misclassiflcation 255 TABLE 1. Census region population percentages and number of sampling sites needed per region for Part II (smoker misclassi. fication study) Census Population No. of test region States within region (% of total) sites needed I CT, MA, ME, NH, NJ, NY, PA, RI, VT 19 2 2 IA, IL, IN, KS, MI, MN, MO, NE, ND, OH, SD, WI 24 2 3 AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, OK, SC, TN, TX, VA, WV 35 3 4 AZ, CA, CO, ID, MT, NM, NV, OR, UT, WA, WY 22 2 spouse) and criteria for assigning smoking status (current, former, never) were the same as used in Part I. Question- naire data on various lifestyle factors were also obtained, but are not discussed here. Site selection. An ending sample of 900 was chosen for this study, selected in nine geographically disperse regions of the 48 contiguous United States. According to the !.990 Census population in each of the four census regions, the number of locations in each regign was allocated as shown in Table I. Nine states were chosen with regard to geographic dis- bursement and relative demographic make-up compared with that of the 48 contiguous states. In general, this meant selecting larger "metropolitan statistical areas" MSA as used in the census. Within each of the nine states, the national market research company selected local field services on the basis of reliability, efficiency, and availability within the study time frame. Local field service supervisors/interview- ers were trained by the national agency supervisors to ensure correct administration of the protocol/questionnaires. Nei- ther the study purpose nor the identity of the sponsor were ever revealed to the local field services. Supervisors and in- terviewers of the local field services were informed only that respondents were being invited to their facility to "partici- pate in a diet/health/lifestyle study." Quotas for each city were calculated on the basis of the individual state's contribution to the region's population and the region's contribution to the population of the 48 contiguous United States. Selected locations, test dates, and the final number of subjects completing Part II are listed in Table 2. FIELD QUOTAS. Actual field quotas were established based on a combined consideration of the results from the NIS (Part I) and the minimum number of participants needed in each cell for statistical power. Table 3 shows the actual quotas selected for the field operations. The numbers in parentheses indicate the exact quotas predicted from Part I. Cell designations are based on the standard etiologic definition of "exposed," wherein the female was "ETS ex- posed" only if her spouse smoked. Quotas for each cell in each city were obtained by applying the cell proportion in Table 3 to the total quota per city. The final number of completes per city (Table 2) was divided into cells by virtu- ally the proportions shown in Table 3. RESPONDENT RECRUITMENT. Each respondent was inter- cepted in a mall environment and was administered a 5- minute screening questionnaire by a professional inter- viewer. Each respondent was qualified on the basis of age ( 18 + ) and marital status (married and currently living with spouse). Each participant was also assigned to a cell ac- cording to smoking status and history of ETS exposure. The final number of subjects in each cell differed slightly from the quota: Cell 1,199; Cell 2, 101; Cell 3,200; Cell 4, I98; and Cell 5, 200 (total n = 898). For cell definitions, see Table 3. CENTRAL LOCATION SURVEY. Each qualified respondent was invited to a private office or office suite within the mall where personal interviewing occurred. Respondents .pro- vided a saliva sample and completed a 20-minute question- naire. This questionnaire addressed consumer lifestyle (diet, health, products used in the home, work status, etc.) and perceived exposure to ETS in the home and workplace en- vironments. Each participant who completed the study re- ceived a gratuity of $10. SALIVA SAMPLING AND COTININE DETERMINATION. Sa- liva samples were collected with. Salivettes® (Sarsredt Inc., Newton, NC). The devices consisted of clear polystyrene conical centrifuge tubes with a snap-on stopper and an un- treated, sterile cylindrical cotton swab. Participants re- moved the snap-on stopper, placed the cotton swab in their mouth, and chewed vigorously for one minute. Then they returned the cotton swab to the conical centrifuge tube. After saliva sample collection, the Salivettes were placed in individual zip-lock plastic bags. The individual bags, in turn, were placed in larger zip-lock bags and stored in either a freezer or an insulated container (Polyfoam Packers Corp., Wheeling, IL) containing approximately .50 pounds of dry ice. In addition, nine field blanks were prepared in each city by adding 2 ml of distilled, deionized water to each cotton swab. Field blanks were stored in the same manner as the samples. Samples and blanks were transported frozen to the laboratory. Upon arrival at the laboratory, samples and blanks were stored at approximately -I0°C in a freezer

256 M.W. Ogden et 02. TABLE 2. Selected test sites and test dates for Part II (smoker misclassification study) Census No. of region Test site completes Feb. 19-21 Test dates (1992) Feb. 26-28 Mar. 4-6 Freehold, NJ 83 Buffalo, NY 85 Muncie, IN 107 Des Moines, IA 111 Dallas, TX 100 Ft. Myers, FL 106 Atlanta, GA 102 Los Angeles, CA 102 Denver, CO 102 x × × X × X prior to cotinine determination by radioimmunoassay (RIA). The RIA method used is as described in detail else- where [8] except that 100/A of saliva (instead of 10/A) was used in each assay to increase sensitivity. Limit of detection, defined as three times the standard deviation of the blanks [9 p. llS], averaged 0.52 ng/ml. QUESTIONNAIRE CODING AND DATA ANALYSIS. All tele- phone survey data from Parr I (NIS) were tabulated and output directly from the Athena-CATI system and provided to client personnel. All questionnaire data from Part II were rid&edited by a representative of the national market re- search agency and the supervisor of the local agency at each test site. Final editing and data coding were performed by the national agency staff. Data were provided to the client on magnetic tape. Summary statistics and data comparisons were computed using SAS® software (SAS Institute, Cary, NC). RESULTS Part I-NIS CONCORDANCE RATIO. Table 4 contains the results of Part I, the NIS. Note that the standard etiologic definition of"exposed" is used; i.e., the participant is "ETS exposed" only if her spouse smoked. From these data, the extent of smoking habit concordance between subjects and spouses can be calculated. Calculated as the cross-product ratio of the 2 × 2 exposure matrix of current smoking status, the concordance ratio is expressed (for the data in Table 4) as: [(Cell l)(Celt 4 + Cell 5)/(~etl X)(Cell 2 + Cell 3)] or [(158)(749)/(142)(151)] = 5.52 In other words, the odds that a woman's spouse is a current smoker is 5.52 times greater if she is a current smoker than if she is not a current smoker. The 1986 NRC report on ETS ([10], p. 239) uses the term "aggregation" to describe this non-random marriage pattern and computes a smoker aggregation factor by a similar method to that used here for the concordance ratio. Part II-Smoker Misclassification Study DENIOGRAPHICS. T~e age distribution of subjects is out- lined in Table 5 and racial distribution ~s outlined in Table 6. The distributions of respondent's educational level and total family income are depicted in Figures i and 2, respec- TABLE 3. Actual field quotas and percentages (and predicted quotas from the National Incidence Survey, Part I) for the smoker misclassification study (Part II) Subject code ETS exposed ETS unexposed Totals Current smoker* Cell i 200 200 (225) 22.22% 22.22% (25.00%) Former smoker Cell 2 Cell 4 300 I00 (41) 200 (155) 33.33% 11.11% (4.58%) 22.22% (17.17%) Never smoker Cell 3 Cell 5 400 200 (72) 200 (407) 44.44% 22.22% (8.00%) 22.22% (45.25%) Totals 500 (232) 400 (668) 900 55.56% (25.75%) 44.44% (74.25%) 100.0% (225) (25.00%) (196) (21.75%) (479) (53.25%) (900) (100.0%) *Current smokers are not categorized according to ETS exposure (i.e., husband's smoking) status.

Ir~cidence of Smoking and Misclassification 257 TABLE 4. National incidence survey (Part I) results representing current self-reported smoking status and ETS exposuxe status" of the U.S. married, female population subject code ETS exposed ETS unexposed Totals Current smoker Cell 1 Cell X 158 (13.17%) 142 (11.83%) 300 (25.00%) Former smoker Cell 2 Cell 4 55 (4.58%) 206 (17.17%) 261 (21.75%) Never smoker Cell 3 Cell 5 96 (8.00%) 543 (45.25%) 639 (53.25%) Totals 309 (25.75%) 891 (74.25%) 1200 (100.00%) =ET~ exposure status corresponds to the current smoking status of the subject's husband; e.g., ETS exposed indicates husband is a smoker. For additional discussion, see text. lively. A slight shift is seen toward lower educational level and lower family income for self-reported current smokers. COTININE RESULTS. The mean salivary cotinine concen- tration for all self-reported current smokers was 352.9 ng/ ml; city averages ranged from 287.5 ng/ml in Los Angeles, CA to 400.9 ng/ml in Atlanta, GA. Salivary cotinine con- centrations are summarized in Table 7 according to the sub- ject's self-reported smoking status. Breakpoints were chosen as follows: 10 ng/ml is a generally-recognized concentration level for maximum sensitivity and specificity in delineating current smokers from nonsmokers Jill; 35 ng/ml is 10% of the mean level determined in all self-reported current smokers; and 106 ng/ml is 30% of the mean level deter- mined in all self-reported current smokers. The I0% and 30% levels were used in a recent risk assessment [7] to delin- eate occasional smokers (i.e., those with 10-30% of smok- er's mean level) from regular smokers (i.e., those with >30% of smoker's mean level). Accordingly, under this sce- nario [7], people with less than i0% of the mean level deter- mined in self-reported current smokers were re-classified as nonsmokers, even if they reported themselves to be smokers. The cotinine levels determined for all self-reported cur- rent smokers are depicred in Figure 3. As seen, there is a wide range of salivary cotinine levels for women who report themselves to be current smokers. According to EPA's definitions of occasional and regular smokers [7], 83.4% of self-reported current smokers would be characterized as reg- ular smokers (EPA assumes 90%) and 7.5% would be char- acterized as occasional smokers. Under these same criteria, an additional 9.0% of self-reported current smokers would be re-classified as nonsmokers. SMOKING STATUS MI$CLASSIFICATION. The numbers of self-reported nonsmokers (both never and former smokers) with saliva cotinine concentrations inconsistent with being a nonsmoker are summarized in Table 8. For simplicity, we designate these subjects as "deceivers." Note also that this table categorizes the deceivers by ETS-exposure category and three different cotinine cut-point concentrations. The rates of misclassification are necessarily reported in different ways depending on the nature of the misclassifica- lion adjustment model used to correct epidemiology-based observed risk estimates. The data as reported in Table 7 will enable calculation in a variety of ways. For example, Lee [I,2] typically expresses never smoker misclassification rates as the percentage of reported never smokers who were deter- mined to be current regular smokers. From Table 7, the ob- served misclassification rates are 3.5% (14/400) and 2.5% (10/400) corresponding to cotinine cut-point concentra- tions of >35 ng/ml and > 106 ng/ml, respectively, to indi- cate a regular smoker. The misclassification model employed by EPA [7], how- ever, requires a regular smoker misctassification rate that is expressed as the ratio of deceivers to the reported number of current smokers who are also regular smokers. EPA's definition of regular smokers includes only those with coil- nine values exceeding 30% of the mean level found in all self-reported smokers. TABLE 5. Age distribution by self.reported smoking status for Part II (smoker misclassification study) Self.reported smoking status Age (years) Mean SD Median Min. Max. Current 37.3 14.8 34 18 76 199 Former 43.2 15.5 41 18 81 299 Never 40.6 14.7 39 18 77 400 A.I1 40.7 15. I 38 18 81 898

258 M.W. Ogden et al. TABLE 6. Racial distribution by self-reported smoking status for Part II (smoker misclassification study) Serf-reported Native smoking White/ Black/ American status Caucasian African Hispanic Indian Asian Other Totals Cu~ent 158 21 16 I 3 0 199 Former 23l 42 22 i 2 1 299 Never 293 76 28 1 2 0 400 All 682 139 66 3 7 1 898 4O 35 3O o 25 O ~0-~ Current Smokers Former Smokers Never Smokers Respondent's Educational Level HGURE 1. Distribution of respondent's educational level by self-reported smoking status for Part II (smoker misclassifi- cation study). Abbreviations: Sch. -- School; H.S. = High School; Compl. = Completed: Gr. Sch. -- Graduate School. 3O 25 ~ 20 5 0 ~ Current Smokers ~ Former Smokers ~ Never Smokers . Respondent's Total Family Income HGURE 2. Distribution ot: respondent's total family income by self-reported smoking status (in thousands of dollars) for Part II (smoker misclassification study). Recall, however, that the ceil populations in Part II were fixed to ensure adequate numbers of participants in each cell. Thus, nationally representative misclassification rates for the U.S. married, female populatioh are obtained by weighting the observed rates according to the cell propor- tions determined ir~ Part I (Table 4). A seemingly more logical way to express this important misclassification rate would be as the ratio of deceivers to true regular smokers. Thus, this ratio necessitates adding the deceivers to the denominator resulting in an expression that is truly a probability (since the numerator is now a subset of the denominator). Observed and weighted mischssification rates calculated by two methods with two different cotinine cut-point concentrations used to indicated regular smoking status are listed in Table 9. TABLE 7. Salivary cotinine concentrations by self.reported smoking status Salivary Self.reported smoking status cotinine Never Former Current (ng/ml) smoker smoker smoker <I0 379 256 14 • 10-35~ 7 12 4 35-106s 4 8 15 >106 10 23 166 ~35 ng/ml is 10% of the mean concentration determined in all self-reported current smokers. ~106 ng/ml is 30% of the mean concentration deterrmned in all self- reported current smokers.

incidence of Smoking and Misclassification 259 16 14 l 12 • ~ 8 Salivary Cotinine (ng/mL) HGURE 3. Distribution of salivary cotinlne concentrations among self-reported current smoking, married U.S. females, age 18+. DISCUSSION The age range of subjects in this study was 18 to 81 years, the central age was approximately 40 years, and the distribu- tion was slightly skewed with the longer tail toward older age. In comparison, although epidemiologic studies of lung cancer in the U.S. have a similar age range (e.g., 20-79 years [12]), they have a central age of approximately 60- 65 and a distribution that is fairly skewed with the longer tail toward younger age. Nevertheless, it is not clear what effect, if any, the differences in age distribution may have on estimates of smoking status concordance or misclassifi- cation. Our study was designed using two-stage sampling that ob- tained most of its subjects from the under-65 age group. Calculation of an age-adjusted result based on an age distri- bution similar to cancer subjects would place most of the weight on only a small proportion of the observed subject pool and almost no weight on the remaining subjects. Pre- sentation of such a weighted result was deemed inappropri- ate because the reported result would have so little precision as to be essentially meaningless. Alternatively, age trunca- tion of the observed subject distribution might be consid- ered. This alternative does nothing to resolve the differ- ences in distribution, assigns zero weight to the excluded subjects, and still requires weighting of the remaining sub- jects. Therefore, the reported results are based only on the TABLE 8. ETS exposure status of self.reported nonsmokers with saliva cotinine levels inconsistent with being a nonsmoker No. Of self-reported nonsmokers wth saliva cot/nine above cut-point concentration >I0 ng/ml >35 ng/ml >100 ng/mP ETS Forme~ Never Former Never Former Never ' exposure smokers smokers smokers , smokers smokers smokers Exposed 20 17 13 12 11 9 Unexposed 23 4 18 2 12 I "These results remain unchanged with a cut-point concentration of 106 ng/mL TABLE 9. Observed and weighted regnlar smoker misclassificadon rates calculated by two methods at two cotinine concentra- tions used to identify regular smokers Regular smoker misclassLRcation rates Observed %" Weighted %b Percent of Percent'of true According true According Cotinine reg~|ar to EPA reg-lar to EPA concentration smokers model" smokers model* >35 ng/mi 6.19 7.73 3.61 4.10 >106 ng/m{ 5.03 6.02 2.55 2.81 ~)bserved rotes in Part II (smoker misclassification study). SObserved misclassification rates in Part II, weighted for population percentages determined in Part I (national incidence survey). 'Rate as required for use in EPA misclassification adiustment model ([7L Appendix B).

260 M. W. Ogden et age distribution obtained from the Part II sample and are reweighted to match the distribution of exposed and unex- posed current, former, and never smoking subjects in the age 18+ U.S. married, female population. The definition of never smoker used in this study was selected to be more conservative than equivalent definitions used in many of the epidemiologic investigations of lung cancer in nonsmoking women. For example, in the studies of Wu et al. [i3] and Brownson eta/. [14], no clear defini- tions of lifetime nonsmokers are given. In another study [15t, a nonsmoker was defined as "one who reported he or she had never smoked or smoked only occasionally but had never smoked regularly"; however, "occasionally" and "reg- ularly" were not defined. In more recent studies where clearer definitions are given, never smokers were defined as those having never smoked more than i00 cigarettes [12,16]. We chose a more conservative definition of never smokers, those having smoked fewer than 20 cigarettes. However, we used a more traditional definition of former smokers, those with complete abstinence for at least 6 months and who previously had smoked at least five ciga- rettes a week for 6 months (ca. 125 cigarettes total). These definitions exclude potential subjects who reportedly had smoked a number of cigarettes between these limits. Re- ported former smokers (however defined) tend to misrepre- sent their current smoking status at a rote more than twice that of reported never smokers (see p. 157 [2] and Table 7). Therefore, restricting the never-smoker category to exclude borderline former smokers is expected to yield a conserva- tive estimate, if not an underestimate of the regular smoker misclassification rate. Since the misclassification rate of cur- rent smokers as former smokers is of lesser overall impor- tance as a bias, a more traditional definition of former smok- ers was used. Random digit dialing was chosen as the preferred means of selecting a representative, population-based sample for Part I of this study [17]. Obtaining an unbiased estimate of cell proportions (Table 4) was of prime importance due to the necessary weighting of the observed misc[assification rates in Part II and due to the heavy dependence of EPA's misclassification model [7] on the number of current smok- ers in the population. We noted a larger than expected re- fusal rate in the telephone screening survey (Part I), particu- larly in the Pacific time zone. It is unknown what effect, if any, this may have had on the cell distributions determined. Part II of this study utilized mall-intercept sampling as a means of recruiting subiects for direct interviewing and sa- liva sample collection. One potential disadvantage of this type of sampling is that it is a form of convenience sampling rather than probability sampling. Other forms of conve- nience sampling [18] include, for example, sampling from church groups, subscriber lists, hospitals, clinics, respon- dents to advertisements, etc. Virtually all of the studies cited by EPA in the context of smoking status misclassifica- lion [7] also used forms of convenience sampling. In general, mall shoppers tend to be disproportionately white/Caucasian, female (an advantage for this study), middle/upper middle class, and urban/suburban [181. Sam- pling sites and times for this study were selected in an at- tempt to minimize the impact of these factors on the study population. For example, some malls would be considered upscale, while others were not, and all malls tended to re- flect the racial distribution of the surrounding community. Specifically, mall shoppers in Freehold and Muncie tended to be predominantly white/Caucasian; shoppers in Atlanta tended to be predominantly black/African (as was a fairly high proportion in Denver); and the mall populations in Dallas and Los Angeles were well balanced with Hispanics. The resulting subject base was not disproportionately white/ Caucasian (Table 6). Likewise, the distribution of mall types and their geographic locations did not produce an overall subject base that appeared to be disproportionately upper middle class (Fig. 2). Over 45~% of reported current smokers and 30% of reported never smokers had total family incomes less than $25,000 per year. Within each mall, sampling "was conducted Wednesday through Friday. Saturday was reserved as a back-up day in each location, but was not used. Weekend sampling was avoided in an attempt to maximize contact with the adult shopping population. Also, women are more likely to be found in malls on weekdays, rather than on weekends [18]. Moreover, sampling was conducted in all mails from approx- imately i 1 am to 8 pm, essentially throughout the mall's business hours. By emploqing these measures, in addition to quota sam- pl~ng and strategic stationing of interviewers at or near main mall arteries, we have incorporated most of the controls sug- gested by Sudman [19] for approximating probability sam- pling with mall-intercept data. The hypothesis that imple- menting certain selection procedures in mall interviewing results in a probability sample has been tested. Murry et ad. [20] found significant demographic and lifestyle differences between comparable mall-intercept data and random tele- phone surveys without Sudman's controls. On the other hand, Bush and Hair [21] found no such differences when employing the controls suggested by Sudman. Overall, they observed a trend toward better data quality using the mall- intercept method rather than the telephone method. In the context of convenience sampling, mall inter- viewing has an additional advantage over other subject pools. Sudman [19] cites studies indicating that 90-95% of all adults shop in a mall at least once a year, and two-thirds of all households shop in a mall in any given two-week pe- riod. The same cannot be said of other convenience sam- pling surveys reported in the context of smoking status mis- classification: being pregnant [4], living in an all-Hispanic neighborhood [5], or attending a cancer screening clinic [6], for example. Whether or not mail-intercept data sampling can be made to approximate a true probability sample seems open 0 £0

Incidence of SmokinB .~nd lvI.Lscl.~ssi.fic~cion 361 to debate. However, with the incorporation of the numer- ous factors outlined above in our Part II mall.intercept study (many of the controls suggested by Sudman, such as the use of nine geographically and demographically diverse sam- pling centers with the resulting distributions of age, income, and ethnicity; quota sampling and weighting to the nation- ally representative probability sample of Part I; and the use of a conservative definition of never smoker), the resuking data are more likely to be representative than most of the data relied on previously [7] to estimate the bias of smoking status misclassification. The concordance ratio is important in the spousal-smok- ing epidemiology of ETS [2,7,10] in that the higher the con- cordance of smoking habits, the larger the bias in observed relative risk (RR) due to misclassification. In other words, with larger concordance ratios, smaller misclassification rates explain larger RR biases. Concordance ratios for fe- males in the U.S. have been shown to range from about 2.6 to 6.0 ([2], p. 160]. The concordance ratio of smoking habits among spouse pairs in this work was found to be 5.52. Misclassification of smoking status is important as a bias in the epidemiology of ETS if the proportion of deceivers differs between exposed and unexposed (or control) groups. Further, these proportions can differ between the exposed and unexposed groups in two ways. First, the proportion of deceivers may simply be different among exposed and unex- posed subjects. That is, the deception rates are "differen- tial." Second, the rates may be the same (i.e., "non-differen- tial") between exposed and unexposed subjects. But, because of spousal concordance and smoking distributions, deceivers are more likely to be in the exposed-case category. When misclassification corrections are made, only this sec- ond way is typically considered. Thus, the concordance ratio is an attempt to show, in general, the magnitude of non-differential deception to- wards the exposed group. Underestimating spousal concor- dance (of smoking habits) in risk assessment and epidemiol- ogy undercorrects non-differential misclassification bias. In Part I, we determined this ratio to be 5.52; however, as the data in Table 8 show, there also appears to be substantial differential deception as calculated from cotinine determi- nation in Part II. Reported never smokers who are truly regular smokers (>106 ng/ml) are shown to be nine times more likely to be ETS exposed (married to smokers). Although these two data sets (Parts I and II) are from two different samples of the same population and the infer- ence of significant differential misclassification is drown from limited data (Table 8), there is an indication that con- cordance ratios alone, as typically used in misclassification adjustment models, may underestimate the bias attributable to smoker misclassification. As an example, the NRC esti- mated that a true RR of 1.0 could be observed to be 1.3 due to non-differential misclassification bias ([10], p. 236). Additional consideration of differential bias, as calculated in this study, would increase the observed RR from 1.3 to 1.4. If the level of differential misclassification suggested by the data in Table 8 is real, differential misclassification is a major bias in the epidemiology of reported ETS exposure and lung cancer. Further research is required to better deter- mine the impact of differential misclassification bias. There are several types of misclassification; however, only two can be detected with any degree of reliability by coti- nine measurement. These are the rates at which current smokers report themselves to be either former smokers or never smokers. True former smokers (with necessary smok- ing abstinence of at least I-3 days) who report themselves to be never smokers cannot be detected by cotinine assay. The rate at which current smokers misreport themselves as never smokers is, generally, the misclassification bias of greatest interest, since it contributes the largest bias to ob- served risk estimates. However, an exact cotinine concentration does not exist for unequivocally delineating "smokers from nonsmokers. The arbitrary cut-points of 10% (occasional ~moker) and 30% (regular smoker) of mean cotinine levels in all reported current smokers [7] appear to be too high. From the distribu- tion of cotinine levels presented in Figure 3, more reason- able salivary cotinine concentration cut-points for discrimi- nating current smoking status may be: <10 ng/ml (nonsmoker); 10-35 ng/ml (occasional smoker); and >35 ng/ml (regular smoker). This definition of nonsmoker would also be in better agreement with the recent review of salivary cotinine levels by Etzel [11]. A change from the arbitrary cotinine level used by EPA for delineating regular from occasional smokers to a more reasonable level would increase the misclassification rate, see Table 9, and have the net effect of further reducing risk estimates observed in epidemiology studies. The 83.4% of self-reported current smokers who can be considered regular smokers under the EPA-recommended guidelines can be another important variable in misclassifi- cation adjustment. We are not aware of any published data for the U.S. that give the ratio of regular to reported current smokers. The value determined here is less than the value used recently (90%) in one risk assessment [7] to calculate misclassification rates from some studies. A shift from 90% to 83.4% would increase the misclassification rates calcu- lated from several studies [7]. As noted above, saliva sampling in Part II was conducted Wednesday through Friday from approximately 11 am to 8 pm in each city. While cotinine concentrations have been shown to depend partially (---20-30%) on the time of day when sampling is performed [22,23], we do not expect this to have any substantive impact on the results reported here. Sampling times were naturally randomized throughout the time period of interviewing. There is the possibility that some regular smokers interviewed early in the day may not have reached steady-state cotinine concentrations. If so, and if such a subiect also happened to self-report herself to be a never smoker, it is possible this misclassified subject o~

262 M. W. Ogden et o/. would not be correctly reclassified with the high cotinine cut-point used here. If this occurred, the regular smoker mis- classification rate reported here would be biased low. Reclassification of smoking status based on cotinine is potentially subject to other errors as well. For example, due to genetic factors, some certified active smokers may have cotinine concentrations below generally accepted values for smokers. Cholerton eta/. [24] estimate that as much as 10% of the population poorly metabolize nicotine due to poly- morphism in the CYP2D6 isozyme. Additionally, Benowitz eta/. [25] report a metabolic deficiency to convert nicotine to cotinine via an anomaly in the C-oxidation pathway in up to 4% of the population. Among self-reported current smokers in our study, 9% are necessarily reclassified as non- smokers under the EPA-recommended guidelines. How- ever, a significant number of these subjects may be regular smokers with genetically determined low cotinine. If true, then we must assume that some percentage of true regular smokers who report themselves to be never smokers must also be undetectable by cotinine determination. However, the misclassificadon rates presented in Table 9 would not appear to be biased since both the numerator and denomi- nator would appear to be biased by the same percentage. This assumes, of course, that there is no interaction be~een cotinine deficiency and propensity to misreport current smoking status. To illustrate the importance of the regular smoker mis- classification rate (MRslss), we show the effect of changing this single parameter on the meta-analyzed relative risk esti- mate for lung cancer reported recently by EPA [7]. In a manner analogous to EPA's calculations, we performed the misclassification calculations individually for each epidemi- ology study using the EPA-documented input parameters (for all but the one misclassification rate in question, MRslss), then we meta-analyzed the corrected relative risks. Figure 4 shows the results of such calculations as the M~sr~s rate is changed. A change in misclassification rate from the EPA-cited value of 1.09% [7] to the rates found here (2.81-4.10%) is a seemingly modest increase in regular smoker misclassifi- cation; however, this change drastically alters the conclu- sions regarding the recent meta-analyzed relative risk esti- mate for lung cancer among exposed nonsmokers [7]. As seen in Figure 4, application of a regular smoker misclassifi- cation rate of about 3% in the EPA meta-analysis results in a non-significant elevated risk (even with 90% confidence intervals as used by EPA). As also depicted in Figure 4, a misclassification rate of 2.81% (using >106 ng/ml as the cotinine cut-point) is nearly sufficient to drive the recta- analyzed result to statistical non-significance. Applying the rate of 4.10% (using >35 ng/ml as the cotinine cut-point) clearly reduces this recent risk estimate to non-significance. Even higher misclassification rates than those reported here have recently been observed [26]. In three clinical set- tings dispersed across the U.S., 15.96% of self-reported non- 1.4 ~.2 "~ 1.0 UJ 0.9 0.8 0% 2% 4% 6% Misclassification Rate HGURE 4. EPA's meta-analyzed relative risk estimate for lung cancer as a function of regular smoker misclassiflcation rate (M~s[Ns) ([7] Appendix B). RR estimate is the center line; upper and lower lines are the 90% confidence levels. smokers were observed to have urinary cotinine levels in excess of 500 ng/ml (a level approximately equal to 100 ng/ ml salivary cotinine [27]). For comparison, our data must be put in a different format (since the data in [26] are for all nonsmokers and are not categorized as never and former smokers). The comparable rate obtained from our study (ob- served rate from Tables 5 and 7 = [(10 + 23)/(400 + 299)] = 4.72%) and weighted for ceil percentages determined in Part I is 4.01%. Thus, the recent result from clinical settings is nearly four times higher than the result we report here. Misclassification of current smoking status among self- reported never smoking married females in the U.S. is com- mon. The results from this representative study of subiects categorized in accordance with the majority of ETS epide- miology studies indicate that many individual studies and one recent meta-analyzed risk estimate from pooled studies may have underestimated this important bias. A regular smoker misclassification rate of 3.05% (or larger) is suffi- cient to explain all the observed risk in one meta-analysis of U.S. epidemiology studies performed recently [7]. The current study reveals this critically-important parameter to be 2.81% or 4.10% (depending on cotinine concentration used to define regular smoker) for married, never smoking U.S. females. Further, none of these studies, including the recent EPA mera-analysis [7], considered the effect of differential mis- classification bias. Although limited data are available at present, the effect of differential misclassification appears to be significant. Failure to account for this bias results in fur- ther inflated risk estimates. c c C

Incidence of Smoking and Misclassification 263 The. authors gratefully acknowledge the contributions of M. F. Stiles, R. B. Hege and S. B. Sears. References I. Lee PN. Misclassification of Smoking Habits and Passive Smoking: A Review of the Evidence. Berlin: Springer-Ver- lag; 1988. 2. Lee PN. Environmental Tobacco Smoke and Mortality. Ba- sel: Karger; 1992. 3. Haddow JE, Palomaki GE, Knight GJ. Use of serum cotinine to assess the accuracy of self reported non-smoking. Br Med J 1986; 293: 1306. (Letter) 4. Haddow JE, Knight GJ, Palomaki GE, et al. Second-trimester serum cotinine levels in nonsmokers in relation to birrhweight. Am J Obstet Gynecol 1988; 159: 481-484. 5. Coultas DB, Howard CA, Peake GT, et aI. Discrepancies be- tween self-reported and validated cigarette smoking in a com- munity survey of New Mexico Hispanics. Am Rev Respir Dis 1988; 137: 810-814. 6. Cummings KM, Markelto SJ, Mahoney M, et al. Measurement of current exposure to tobacco smoke. Arch Environ Health 1990; 45: 74-79. 7. United States Environmental Protection Agency. Respira- tory Health Effects of Passive Smoking: Lung Cancer and Other Disorders. Washington, D.C.: Office of Research and Development, U.S. Environmental Protection Agency; 1992. (EPA publication no. EPA/600/6-90/006F) 8. Langone JJ, Van Vunakis H. Radioimmunoassay of nicotine, cotinine, and ~(3-pyridyt)-~oxo-N-methylbutyramide. In: Langone J J, Van Vunakis H., Eds. Immunochemical Tech- niques. Methods in Enzymology. New York: Academic Press; 1982; 84: 628-640. 9. Miller JC, Miller JN. Statistics for Analytical Chemistry. 2nd ed. Chichester: Ellis Horwood Limited; 1988. 10. NRC Committee on Passive Smoking. Environmental To- bacco Smoke Measuring Exposures and Assessing Health Effects. National Research Council Committee on Passive Smoking. Washington, D.C.: National Academy Press; 1986. 11. Etzel RA. A review of the use of saliva cotinine as a marker of tobacco smoke exposure. Prey Med 1990; 19: 190-197. 12. Fontham ETH, Correa P, Reynolds P, et aI. Environmental tobacco smoke and tung cancer in nonsmoking women. JAMA 1994; 271: 1752-1759. 13. Wu AH, Henderson BE, Pike MC, Yu MC. Smoking and other risk factors for lung cancer in women. J Natl Cancer Inst 1985; 74: 747-751. 14. Brownson RC, Alavanja MCR, Hock ET, Loy TS. Passive smoking and lung cancer in nonsmoking women. Am J Pub Health 1992; 82: 1525-1530. 15. Garfinkel L. Time trends in tung cancer mortality among non- smokers and a note on passive smoking. J Nail Cancer Inst 1981; 66: 1061-1066. 16. Janerich DT, Thompson WD, Vareta LR, eta/. Lung cancer and exposure to tobacco smoke in the household. N Engl J Med 1990; 323: 632-636. 17. Orden SS, Dyer AR, Liu K, et al. Random digit dialing in Chicago CARDIA: Comparison of individuals with unlisted and listed telephone numbers. Am J Epidemiol 1992; 135: 697-7O9. 18. Dupont TD. Do frequent mall shoppers distort mall-intercept survey results? J Advertising Res 1987; 27: 45-51. 19. Sudman S. Improving the quality of shopping center sam- pling. J Marketing Res 1980; 4:423-43 I. 20. Murry JP, Lastovicka JL, Bhalla G. Demographic and lifestyle selection error in mall-intercept data. J Advertising Res 1989; 29: 46-52. 21. Bush AJ, Hair JF Jr. An assessment of the mall intercept as a data collection method. J Marketing Res 1985; 22: 158- 167. 22. Benowitz NL, Kuyt F, Jacob P, et al. Cotinine disposition and effects. Clin Pharmacol Ther 1983; 34:604-61 I. 23. Benowitz NL, Jacob P. Metabolism of nicotine to cotinine studied by dual stable isotope method. Clin Pharmacol Ther 1994; 56: 483-493. 24. Cholerton S, Arpanahl A, McCracken N, et al. Poor metabo- lizers of nicotine and CYP2D6 polymorphism. Lancet 1994; 343: 62-63. 25. Benowitz NL, Jacob P, Sachs DPL. Deficient C-oxidation of nicotine. Clin Pharmacol Ther 1995; 57: 590-594. 26. Apseloff G, Ashton HM, Friedman H, et al. The importance of measuring cotinine levels to identify smokers in clinical trials. Clin Pharmacol Ther 1994; 56: 460-462. 27. Wall MA, Johnson J, Jacob P, Benowitz NL. Cotinine in the serum, saliva, and urine of nonsmokers, passive smokers, and active smokers. Am J Public Health 1988; 78: 699-701. o 0~ o.~ 0