Tobacco Documents Online

CORRESPONDENCE Re: Environmentall Tobacco Smoke and Lung Cancer Risk in Nonsmoking Women The report by Stockwell et' al. ( l'), adds little to the data on environmen- tal tobacco smoke and lung cancer. Various biases could have contributed to the asaociations noted between lung cancer and some indices of environ• mentall robacco, smoke exposure. One concern relates to the use of healthyy control subjects who were obtained byy random-digit dialing, resulting in pos- sible recall and nonresponse bias. Another concern, perhaps more so than in other lung cancer-znviron- mentai tobacco smoke studies, is information bias. Thus, all control subjects provided data directly. but surrogates providedl data for 6717c off case patients, many of whom were dead. There were also notable case- control differences in the proportions of inter. iews conducted face to face, by telephone, or by mail. MUch, at- tention has been, given to, bias from, the misclassification of smoking hubit.~ (2), but, although Stockwell et al. (l') refer to this misclassification. they attempt no statistical adjustment and do not present a comparison of smoking status as recorded at various staees of their study. They also fail "to consider confounding by dizt. This failure is remarkable, since in another paper (3), apparently based on the same study, they report a strong pro- tzctive effect against lung, cancer among nonsmokers that is associated witK toral, vegetable consumption and withiintake of carotene, and, as I' have reviewed elsewhere (2), a reduced consumption of veg etables is associ- ated with marriage to a smoker. Ad- justment for this source of bias alone could welli render the reportzd' asso- ciation between lung cancer and' ex- posure to environmental tobacco smoke not statistically significant. There are aiso' severe problems regarding presentation of results in • the reporn by Stockwell et al. How can, one update meta-analyses for diftlerin, indices of environmental tobacco~ smoke exposure when risk ' estimates are presented only for those indii;es for which an association is reported? What were the odds ratios and confidence intervals for environ- mental tobacco smoke exposure at work or during' social activities'? hteta-analysis would also be assisted by presenting„ as other investigators do (4,S), numbers of cases and controls by exposure. Another diffi- culty is the unusual method of' anal- ysis for spousal environmental to- bacco smoke exposure, with~ the referent group not, as is customary (2), marrie& women whose husbands di& not smoke, but instead women. (married' and unmarried) ~ unexposed to household environmental tobacco smoke from any source. The relative risk estimate is thus not comparable with that for other studies (2)', Fur- thermore, because of the strong asso- ciation of the indices of environ- mentall tobacco smoke used with both marital status and household size (the larger the family, the more likely is e:rposure), there are additional pos, sibilities of confoundins. The referent exposure group is identical for all analvses in Tables 2 an& 3 of the report by Stockwell et al. (11): there- fore, the cited risks for different indexes of environmental tobacco smoke exposure are not independent and could all be affected' by an unusually low proportion of unex- .posed cases, perhaps resulting from recall bias. In any event, the results of the study by Stockwell et' al. (1) have no real effect on the overall d'atat Based on alla data available before this study, meta-analysis (Table I ) gives no overalll statistically significant asso- ciation of lung cancer to workplace or childhood environmental tobacco smoke exposure and! only a small positive association with a husband's smoking, which, as I have previously shown, can be explained in terms of misclassification of smoking habits, confounding, publication bias, and specific study weaknesses (2)'. PETER N. LEE. M!.A. P: N: Lee Stutisties and Compuring Lrd: I Z Cednr Rd: Strltnn, Strrre~, StN2 5DA. England References tt) STOCKWELL HG,. GOLOnIU, AL. LYSIAN GHI ET AL: Envirnnmental tobacco smoke and lung cancer risk' in nonsmokine women. 1 Nal Cancer Inst'3.t"11J7-113^; 199? (2) LEE PN: Environmental Tobacco Smoke andl Mortallry. Basel: S. Karger. 199-"' (.f,) CA\DELORA EC. ST,OCKwELL HG. AR+urRONG AW, Er Au, Dittary intake and rick' of lung cancer in women who never smoked. Nutr Cancer 17:263-270; 199? (-).):WALD NJ, NANC'H:AH.AL. K. THO\IP50.e' SG. ET AL: Does breathine other people'c tobacco vmuke cause lung cancer' Br \ted J(Clin Res Ed) 293:1217-1'-32: 1936 (5) NarIO.sL R8sE.+RCH COL.aIL: En%tron- mental tobacco smuke. Measuring ex- po.ures and a.~sescine health effects. Wash- tneton:, Natl' Acad Press„ 1968'. (t5) FONTH.a.V ET, CORREA P„ WL'-WILLLIN'IS. A. Er AL: Lune' cancer in nonsmokine women: A multiirnter case-control studv. Cancer Epidemioll Biomarktrs Prev ' 1:3 5- 43. 1991' 171 Ltu ZY. HE XZ. CHHrutA.N RS: Smukineand other risk factors for lune cancer Ini Xuanwei.,Cfiina, Int JiEpidemiol 201'-6-3d., 1991 2nd Letter Stockwelll et al. (1) reported : relative risk of 1.6 for women whr never smoked' and who were marriet to smokers. This relative risk wa adjusted for aae, racz, and educatio but' nut for ~ other potential cor Table 1. Mtw-analysis' of studies of environmental tobacco smoke and lung cancer amor reported lifelong nonsmokers Source ofl environmental No. of studies tobacco smoke expo.eure providing data Workplace 1 1 Childhood t I Husband 30 Relative risk 95% conficem estimates interval 1.02 0 9„1.!'-' 0.96 0:85-1.09' 1.19 1.09-1.31 a 'Based omdata presented in ('2) inTabtes 3.14F: 3.21, and'3.'_3, with the addition of data t in two studies (6;7). 7-48' CORRESPONDENCE Jburnal of the Nationall Cancer Ihstitute, Vol. 85. No. 9, Ntati• .` 1993 ~'

; frnd'ers.such as dfet. occupation. an l prior Ilune disease. In fact, nuo ntiun was made of data huvi4 + c.n collected on thosefactor. How- i e-er. in another report. Candelora et a (2) discussed an analysis of diet II J lunc cancer in a subset of sub- ; J.,ts from the same study. Among i, her results. they reported strong i \eo,z associations between luns ncer and both, total veeetable con- ~ mtptiun and! total carotene intake. I or example. the relative risk for the I rghetit consumption quartile versus. iiz lowest quartile was 0.2 for total egetable consumption and 0.3 for i i otal carotene intake. Although -alendLlrt et al. stated~ that informa- ;ion was collected on occupation. exposure to known lung carcinooens. j personal medicall history. and! f'arnily. ; histor~ of cancer and respiratory di.zases, the diet anaiysis was not adjusted for any of those variables or for en\ironmtntal tobacco smoke exposure. Other studies (13',-1) have noted ,Verse associations between dietary tors andilune cancer among people w-ho never smoked, and' three studies (5-7) have reported inverse associa- tions between environmental tobacco smoke exposure and ~-carotene intake amone women in the United States who never smoked. Clearly, diet is an important potential confounder in re- ported associations between envi'ron- mentall tobacco smoke exposure and luno cancer. Since associations were, reported, between lung cancer and both environmental tobacco smoke exposure and diet in the reporo by Stockwell et al. (1), it would be interesting to know if there were associations between dietary factors and environmental tobacco smoke exposure that would give rise to confounding of the llung cancer asso- ciations. A multiple logistic regres- sion analysis that considerediall of the potential risk factors for which data were collected would be useful in elucidating these relationships. A historyy of nonmalignant lung _ sease is another potential con- founder ini reported associations be- tween environmental tobacco smoke exposure and''lung cancer. In a large case-control study of women, who never srnoked, Brownson et al. (R) 1I reported a spousal smoking-lung can- cer relative risk of 11.0. This rellrtive risk wu, adjusted for prior lung disease, but the extent of the adju.t~ ment was not stated. In another report on the same study. Alavanja et al. (9) estimated that 1i617c of lung cancer cases among, women who never smoked' were attributable to prior lung disease. The importance of adjusting esti~- mates of' associations between en vironmental tobacco smoke exposure and lung cancer for potential con- found'ers is emphasized by the weak- ness of' the overall epidemiologic data. From a meta-analysis of the 13 currently available studies of U.S. womem including the studies by. Brownson et al. (8), and Stockwell et al. (1). 1 have calculated a summary spousal smoking relative risk estimate of 1.07 (9517r confidence interval = 0~95~1?'1i). This estimate was adjusted for smokine status misclassification. using the assumptions and methods of the U.S. Environmental Protection Agency (10). Only two of those studies (8:1J) adjusted for prior lung disease, and none adjusted for dietary factors. BothiSidney et all (5) and Le Marchand et al. (6) estimated that confounding by R-carotene intake could inflate environmental tobacco smoke-lung cancer relative risk esti- mates by about IO%; therefore, the very weak overall U.S. association could conceivably be explained' by and lune cancer ri.k' in nnn.mok,ne wumen. JNail Cancer In.t 94 l-i 17-14?'-; 199_' (])~CA NUra.uRA EC. STUc>,HeLI. HG. AKn,.tiTk n.u AWFT AL: Dietarv mial•c and riek of lune cancer in wnmen,wh'n never smnked. Nutr Cancer 17:_6:i-'-70: 199? (3) Kiw LC: Dietary hahii.5 and lune cancerr risk among Chinese females in Hone Kon¢ wh'o, never smoked. Nuir Cancer 1!11:1:55-172. 1988 ('fJ) FRASF.R GE. BFFSON WL. PHILLIPSRL:. Diet and lune cancer in California Seventh-day Adventists, Am J Epideminl 133:683-693, 1991 (5) SIDNEY S, CAAN B1. FRtEDntAn: GD: Dietary intake nf earotenc in noncmokern with and without passive smoking ai home. Am J Epidemibt 139: 1305-1 .109, 1989 ((S) LE MARCHAND~ L, WILKENSLR,. HAtiKIN JH. ETI AL: Dietary patterns of female nonsmokerS with and without, exposure to environmental tobacco smoke. Caneer. Causes Controll ?:l 1-16, 1991' (7) SHIBATA A, PacANl.a-HtLL A. Ross R'K'. ET AL: DietarM1•p-carocene. ciearerte smok- ine, and lung cancer in men. Cancer Causes Control 3:207-? l-il 199? (R) BROWNSON: RC, ALAVANIAMC. Hl1CK. ET. ET AL: Passiive smokin¢ and lune cancer in nonsmoking women. Am J Public Health 82:1i5?5-1-530• 1992 (9) ALAVANJA MC• BROwNso,. RC. Buice JD JR. ET AL: Preexisting lung disease and lune cancer amon¢ non.cmoking women. Am J Epidemiol I36:6?_+-63?. 199?' (10) ENVIRONMENTAL PROTECTION A6ENCY: Respiratory Health Effects of: Pa.csi%e Smokine: Lung Cancer and Other Disor- ders. EPA/600/6-901006B. Washinednn, DC. 1992 (I1)GARFINKEL'.L:.Tlme trendsIn lun£Cancer mortality ' among nonsmokers and' a note on passive smoking. JNCI 66:1061-1066. 1981 (12) BURNS DM: Environmental tobacco smoke: The price of scientific certaintc: 1 Natl Cancer lhst 8-t:1 387-1388, 1992 t that single factor. Note In a recent Journali editorial. Burns (12) asserted that a causall relationship between environmental tobacco smoke exposure and lung cancer has been established with what he characterized as "srientific certainty." The above considerations, and many other uncer- tainties in the environmental tobacco smoke-lune cancer epidemiology, lead me to believe that Burns' conclusion is unjustified. MAXWELL W.LAYARDt Lavard Associates 2242 Sart Anronin Ave. Alantedu. CA 94507. References UJ'STOCKwELL HG. GoLDtitAN. AL, LYSUN GH. ET AL: Environmental tobacco smoke Joumall of' the 1`ationall Cancer Institute, Vol. 85, No. 9: May 5. 1993 'Author's note: The author is a partner in Layard Acxociates, a statistical consulting fcrm, He consults for the Tobacco Ins,tituie. 3rd Letter While another investigator (1) sees the report by Stockwell et al. (2) as an affirmation of a discernible link between environmental tobacco smoke exposure and lung cancer,the evident inconsistencies pointed out in the Stockwell report should give one pause. For example, the adenocar- cinoma data show no statisticall)• significant relationship to environ- mental tobacco smoke exposure and no pattern of dose response, in shurp CORRESPONDENCE --+y N' 0 N~ W 0 ~ +~

distinctitnn to the large U.S. study by Fontham et al. 04 that rzporttdl an riz~ated ri.k only Ior adznocar- ,:int,nta.. As a second example. the ~I apparent relationship between the ex- I' po,ure information source and the reported risk is opposite to thatt reported in the large study by Gar- tinkel et al. (4). The point is that hunting expeditionn through the datal uf an epidzmiologic study cani easily produce inconsistent artifacts where the exposure effects. if any. are likely to be very ,ntall. M y second point concerns the multiplicity of risk estimates in Table 2 of the StocrkwtJl rzport, which were reported to be statistically significant. The impression some may have is that ot' repzuttd! affirntations that ,nrznvhen the claim of a consistent zltzct. However, the reported ri,kk zstiniutes are o~erlappino. It appears that there may be only a single >tatisticallv si2niificant resuln, ile.. the reported risk associated with non- udenocarcinomas with total exposure grzatzr than 40 smoke-years. This sin2le result can account also for rzported! statistical sionificance for all lun2 cancers, for adulthood exposure. tuo~childhood txposure, und for all ' the related siznificant P values for trendl Furthermore, we were given no information regarding any ~possible a„ociation between childhood r:x- po~ure and adulthood exposure. lnevitably, choices were made in both the condtJct of the study and the reporting of the datat It appears that some or all of the stated conclusions could be affected by inclusiont exclu- ;iont or red6stributioni of a small I number of cases. We should be told to what extent' the investigators' choices could have affected such conclusions. Examples are choices related to geographic and temporal cutoffs for the selection of cases, the dzfinition of exposure classes, the choice of adjustmenn variables and adjustment procedures, and' grouping or .plitting of cell types and respond- ent categories.. Idzally, science would be better ,zrvzd if the study protocols and reporting procedures were published in advance of data collectiont The final report of the study could then distdngui,hi between planned and un- planned materiall Perhaps this Journal' could help to promote ,uch prestudy publication. The po..,ible impact' of other poten- tial biases in aJditinnton the potentiall selection biases described abuve, de- serve discussion. For example. smoker-nonsmoker misclassification errors were mentioned, but their im- pact was not assessed. Possibly of reater importance. there was no di.cu>.ion of preferentiully lower zn- vironntzntal tobacco smoke exposure antong nonrespondent ca.e putient, or aniong case patients excluded because of inadequate information oni active .mokine. It would have been helpful if Table 2 of the Stockwell report had included an additionall column indicating how many cases fell into each of the exposure categories as well as the number unexposed. Such direct re- porting of observed frequencies. while they are not demographically adjusted~ relative to the controls. provides a fuller appreciation of the underlyina data. Finally, the Stockwe{I report notes in a single sentence that the study also lookzd! at environmental tobacco smoke exposure at work and during social activities and found no stntas- tically significant estimated increase in lung cancer risk. This failure to report in detail is a fine example of' a publication bius where practically no mention is made of a necyative result, and it is therefore unlikely that this study would ever be included in a meta-analysis of workplace exposure studies. PAUt. SwtT7FR, PH.D.t Drpcu•tutent of Statistics Stttnfnrd University Srculford. CA 94305-4065 References (/) Bcrt,%a D,14: Em'irunmental tnhacco .muke: The price uf' .cientitic cerwinty:: i Natl. Cancer tn>t ti.i:l 3R7-1 3;iX, 199? (?) Stoe'r;wta.t_ HG. G<rLu+.... AL. l1ti'nt.au, GH, t•:T AL: Environmental tohaccu .muke and lune cancer risk im non.mukine :-^_: women. 1' Natl Cancer tnst Y1:1117-14"-^_ 1992 (Jt For:rnA,t ET. CORkEA P. wO-wn.uA..rs A., Er AL: Lune cancer in nun..mukine women: A muttii:enter case-cuntrnl •tuw. Cancer Eptdrmiul Biuinarker. Prev I:}5- 11. 1y41 (4) G,\reFi.,;i-i. L., rYi ~uH.\en O. Jui q~Hr L: In.uluntarv muking and lung cancer: A ca.e-nniriil •tudv JNCI 74,1h:1_i6y 19N5 Note 'Authur'• note: 'Phe*e cutnments have been prepared at the reque.t ut the Tt haucn ln.utute and repre>ent the views uf the author. not neee..arily th se ut the Tnhacco Ih.uitute or ot'' Stanfurd Universitv. Response In response to comments regarding the potential for dietary factors to confound the relationship between environmental' tobacco smoke ex- posure and the subsequent risk of lung cancer (!), it is important to reco2niae that the possible role of dietary factors is important for both smokers and nonsmokers. The ques- tion beins addre,sed' in our report. however. was whether nonsmokin2 women who~ wzre exposed to environ- mental tobacco smoke ha& an in- creased risk of developing lung cancer compared with women who were not exposed. Our results indicated! thart exposure to environmental tobacco smoke can increase the risk of dz- vzloping lung cancer in nonsmokers. A question that should be considered separately is whether dietary factors can exert a protective effect. i.e.. reducine the risk of lung cancer among those exposed to tobacco smoke. Analysis of our data on this question is not yet complete. but the results should be available shortly. It was suggested, in the correspond- tnce by Layard. that prior lung disease may have contributed to the lung cancer risk in these women. As tobacco smoke is known to contribute to the development of both malignant and nonmali2nant respiratory diseases in smokers, al shared common ex- posure to environmental tobacco smoke would appear a more likely explanation. In the correspondence by Lee. the use of data from surrogate respond- ents was questioned. Because lung cancer, is a rapidly fatal disease. the use of data from sutroeate respond- ents was necessary in some cases. These data were presented in Table 3 ot' our report I/L showina the results Vul. 56, No; 9. Iv1ay 5. t9`.` 750 CORR'ESPOti'DENCE Journal of the Natiunal Cancer Institute.

otl anal)~ses performed Neparately~y for se1ff report, and for surrngate re.ptmd- nts. The odds ratios associated with m ironmentai tobacco smoke ex- posure were actuall% greatza whzn the analysis was linrited to living sub- jects. This finding suggested that, had it been possible to inten•iew all case patients directly, the data might ha\,e indicated an e%en stroneer aysociation between Ilane cancer ~risk an& ex- posure to zn~•irunmental tobacco smoke than they did when surrogate reipondents were included. Lee also indica(ed that he considered it to be unusual to use wonien who had' not been exposed to environmental to- bacco smoke a> the referent group because mot;t, previous ,tudies~ had used only marTied women whose husbands had not, smoked. Con:ider- ine spousal exposure as the only source of household tobacco smoke,. however, ignores the possibility of exposure from other household mem- bers. Janerich et al. h?) reported that exposures to high leNels of household smoke during childhood and adoles- cence doubled the risk of lung cancer amon2 non,mokerN. To consider only women married to a smoker as exposed. regardless of other reportedl exposures to smokerti in the house- hold, could result in the mi.,elbs.ifica- tion of exposed women as untxposed, possibly causing an artifici'aI reduc- tion in, the odd>~ ratio: Alat. consid- eration of differences in household size, which could ha%•t an impact on the number of ptotential smokers in the home, did not \-ary by case or control status. Lee also stated' that all associations between emironmental tobacco smoke exposure and lung cancer from all available data can be explained by issues in study design. Howe%er. it must be noted that this study (11) increases the total number of studies with positive findings be- tween environmental tobacco smoke exposure and luna cancer. and as these studies continue to be reported (3). dir;missall of all such findings become, inereasinalyy difficult. 1"IF.A7HF.R G. STOC'Kw'F.LL. AILt.AN L. ~'.lQi.f)k1AN CHARI.ES 1. Noss E1.1ZARF.TH' C. C.1NdDE1.ORA AI?A?.1 VV, ARMSTROr:C; Depurtrnenr of Ener,gr Office of' Health Gernlcattown, Md. References I/'I Siur r;w'cl.i, HG. Gin.l>MAN AL. Lvai \. GH., F;i- ,\ir. Envirunmentall ttrhaccn innhee and lunc cancer risk in nun.mnkin_ women, 1 Nail Cancer In,.+ iJ:1317-I l?~,. 1992 (') 1.vNr:Rtcit DT. Tr+or.iPsr» WD. VAk~-.i.A LR. Er AL: Lung cancer and' expoi.urc uo tnhaccn smoke in the hnusehold. N Encl J IW ed 3'_ ±:6?2-b:z5. 1990 (3 1 Fcu.ntaat ET• CUkHF.A P. Wl-WIIJ.Ia l% A. FT .\t.: Luneca.ncer in nnnsmnl.inL unmen: A multicenter cu.e-.ontrol Stud\ . Cancer Epid'eminl Blumarkerr Pre\ I:±S- 1;• 1991 Note Cnrreapundrnre m: Heather G, , Stnckarll., Sc.D.. Department of Enerey. OPtice nf Health, EH•;_: 199(1'1 Germaninu•m Rd.. Germanmwn. M11D1 2oK7J. i t'YvrXVAIALVUOF GOVERNMENT BOOKS The U.S. Government Printing Office has a free catalog of nev~, and popular books sold by the all about. Send for youz Jree ca talog: Government. Books about Free Catalog agFiculture, energy, children, space, health, history, business, vacations, and much more. Find out what Government books are P.O. Box 37000 Washington. DC 2001 3-7000 Journal of'the National Cancer Institute. Vol. 85. No. 9. May 5. 1993 CORRESPONDENCE 7_el