Tobacco Documents Online

such as rao:, ethaicity, lifestyle, sodoeconomr"c ttatus, diet, and typc of tobacco smoked need to be addressed. Add•iti'odal confounders may be related to age, gender, family history of lung cancer or tttbertulocis, medications, aewpatioaal or household exposure to products from combusi',o4 and eaqosure to tsdoo. The epidemial4c studies Yary.rtth re:pm to potential eonfounderr taken into aooo+mt in the study desip oa the analysis, e{, faMura to adjust even for ap or marital itaws are oot anaommoa Tbe Report abouid talte sources of confounding into aecoRmt for each imdiwidual study and attempt to evaluate the impact ik 6ad oo the retahs. Comparisoos acrost studies should then be cooduaed to ideatify potential confounding variabtet that may play a signi .fiwtt role. Age appears to, be the only confounding variable reported to have ouc5 inDuepee, however, from what is atrrentliy known. Adjustmeot for confounding factors may be of particular value in countries where there is a high background rate of lung cancer in nonsmokers. Differences in susceptibility to lung carcinogens between raees is a poWbility. S3. BIAS An upward bias is expected from married female ever-smokers, i.e, current and former smokers, .vbo misreport tbemsetves as never-smokers, coupled with a tendency for smokers to be married to smokers. Numcricat adjustment for tbis misdassification effect is done incorrectly in the Report. Tbe figure for bias, which should be hie6er, is sufficient to explain the association seen in men and can explain a very 4rae proportion of the association seen in US. and Western European women. A much higher association, Lowever, would be required to esplaia the association seen in Asian women. The nutliodoloV Jor adjurernenr of tnioker mitclassijicario~t ii a oonteiuious topic in du Arcnm+nc. One woiewrr nrportr thot the .dusment /or wusclour%icariai biat in du Rspon is ewdi too larA if oow maxes o+r ad3usbneM aPP^opiate to each brdivi&al mdy priow to conzb=M rardy auca»es bvtsad of oftowwd (ue Sectiorr 63). Upward bias may result also from considering only studies reported io the open literature. An attempt should be made to include unpublished atudies, which may not Dave been published because no 11

aign`tFiant results were found. The omission of three or more studie:.vith RR - 1 would account for the significanoe of the combined study results. An upward bias results in the overall estimate of relative risk because some sttadiu lack comparability between cases and oootrols in the circumstances wrder which data .vere collected and those atudie: tend to bave aigas5antly higher observed relative risks. Aspects of a atudj+s desigq or aaulysi: of the data..vtth respect to the treatment of potential confounding variables is a source of bias, with the dvoetioo of the bias tnu]ear. For esample, some stndia did oot adjust for age in the design or in the analysis. Matching by age and other factors in attd'tes that.Nere intended for multiple purposes may not apply to the vibset of the data on ETS. In particutar, a number of studies included active smokers or former smokers in the total study.vitbout matching on smoking status for analysis of data on ETS. Some studies included former smokers or single persons, treating the single persons equivalent to som-etcposed marrieds. Additional sources of bias are present to varying degrees in most of the studies. Lung cancer pitients may tend to overstate their exposure to spousal smoking as an explanation for their illness. Bias may result from depending on memory recall of a subject's exposure to spousal smoking. Estimates of relative risk may differ markedly between data collected from the subjects and data obtained from a surrogate, such as their children. The potential bias may not be consistently in the upward or downward direction, e.g., a weaker association is observed from the data for surrogates in the Varela study (also, the Janericb study),.vhile the reverse is observed in the case-control study of Garfinkel. Histologic verification of lung cancer was not conducted in all studies and the error rate may be substantial, e.g, 13% of the lung cancer cases in the case-control study of Garfinkel et. al. were found to be incorrectly diagnosed when the histotogy was reviewed by ooe of the authors. In geseral, the Report oeeds to address potential sources of bias and confounding (we Section 52) for each study iadiwiduaHy. 5.4. kIISMLOGY Studies suggesting an elevated lung cancer prevalence associated with spousal smoking appear ioconsistent with regard to the distribution of tumor types that may be associated with ETS exposure. More . 12

oonsistent evidence regarding the type of tumor assoaated with ETS exposure would be expected if there were a causal relationship. Also, the tumor evidence in epidemiologic audics on ETS is bot consistent with v" has been observed for active smoking (a much :trooger assocaatio® with squamous and amall a1l cancer than with adeaocarcinoma and Iarge oell anotr), which might be expected if it is claimed that both active amd passive smoking are a lung cancer barsrd, and maiastrcam and aidestream smoking are qualitatively aimilar. Ia the case eootrol' atudies* some inm:t4atoR restricted the selection of lung cancer case: to a apeciGc histological type(s) and some did not. For attsdie: in which data are avaUbk by histological type, the comparison acoss studies of relative risks calculated by histological type pve sa indication of a 6istologicsl type(s) more Wcely to be associated with ETS exposure. Account needs to be made, bowwer, of potential errors in diagnosis of primary carcinoma of the lung (see Section S3). 5.5. DRAWING CONCLUSIONS FROM EPIDEMIOLOGIC STUDIES Epidemiologic studies are notoriously unreliable in outcome. An observed relative risk of less than 1S-2.C? (some would up to 3.0): is inadequate to reject the hypothesis of no etlect. T6e overaIl relative risk calculated across studies is well below a minimal value for seriously attributing it to the presence of a real effect, ie., it is within the raage easily due to the `Doise in epidemiologic data resulting from the limitations and vagaries iatriasic to the metbodology and its applicatioa. This same conclusion also applies to ncarly all of the studies on an individual basis. Anot6er rtasoo for conservative interpretation of the ETS atttdies is that several studies are of poor quality (good textbook examples of bow aot to do an epidemiologic study) and some were originally designed for adiHerent, or broader, purpose than a««ing bealth risks Gom ETS 'Iie EPA':1989 Workshop Report on the ttse of human evidencc lays out seven criteria for judging the adequacy of epidemioio®c studies and seven additional criteria for judging the strlengtb of support for a O causal iafereace. These criteria provide a useful framework for summarizing the weight of the epidemiologic W evidence (even if they ue not official EPA guidelines at this time). Studies not satisfying a criteria for ~ adequacy should not be considered further. The staticticil outcome has little meaning in a study judged to ~ 13

be inadequate, :iaoe statistial signifinnce magweld be the resutt of an artifus. The Report ateds to be based on guiding pcineipks, tucb as the two sets of aevea aiteria reculting Eom the EPA workshop oo that topic 7bar me .lro adtQia bi the titcrature and taftob applicable 10 epiderniafogic Nudes 1k jeno14 as woted by sevrrol reWewcrs. 6. TOPICS TREATED IIdAC.'CURATELY OR INAPPROPRIATELY (WNG GNCER) 7htr isction perraLw to cor++rneatr ditgaeirRtfindmr+enraly witJi d,e Repor[7r Asacuion of mnie topics snd the opproprietencss of some methods of mwysis auployrd h does nor oddrea ano+r dlarc wilf be e+eeed in an erraaua (or fn a nevisiori) acepc when they mial,t subuanrivey effecx coachrsians. Xoa+nrc., t+vo csarnpks are presented hen. ?7u violuer of the ttaastic S in Table }Q applicobk to crne{ono,ol studies with .n odlusred aruaysis, conraiirs du vdLes fro.rc du ori jinal cola~fotions by BROW dnd SVEN Jiorn whidi the S violuer of 1.78 mid 1.89 arr daiwed Ahanauve methods of daivin8 this statistic yield nan fiJnfJicarrt nrsults for these tA.n saidies oRly. If tJuu oiternaarr merhods arr use4 the aurnba of sWftcant awca•net (oae-arFfed p- s!iobie kss dran 0.05) !s 3(instead of 5) out of ll. 7Iu ovenoAF concGexion, hawrvrer, is eua:daly unahend-the chance of tJkrre or aiorr hudrs njeairea the hypod,e,rfs of no arsodatiore whar It is bue (Type I erma in statix8rel farpn) Lt sdll tnaall (QOIS). SeNnaW nevievert feei' that the p-valua used should be doubk4 wluch would impy t/mt the alternadw hypotliais jnd4des the possibiliry thar an ETS e,dect could influence the rrJarrve ritlt in either an upward or downwmd dinctioa, i.e, to <I or to >l. Two neviewrr: pobsted out duu some foamulac in Appendix B had eypogrophical errors, but that the calculations appeQned to be doru eorrrcety. The fbnreulQ for Populaxion .lmibutabk Risk (WA eq. B-3 is incor+rct but use of the eorrcu fonmulo Jieldt nemy ldenicard rrsults. 66.2. COHORT STUDIES The Japanese (Hirayama) study is iaaoasrately portrayed as a model by which to campare the American (Garfinkel) study. Tbere is no consensus on the Hirayama study. Although it has undergone oaasiderable scutiny, there are mumerous criticisms of it that remain unanswered. TDere are deficiencies 14

evidcat in the dcsigu, conduct, and analysis of the study, but Hirayama has refused to make study data available and to: adequately describe other aspects of the study that would permit an objective assessment. Some daims of the study that have aot been adequatefjr.eriSed ooee+ern the 'esasW of the study population, oompkteaess of the foII'ow-up, iotera:T iaoooitsteade: in the reported cohort mortality aqerieaee, and drnparate ruults when stratification of age at entry to the study reters to the wife or the husband. A setious defea iavoives the aeaoracy of the d'tapos'ts of Imng caaoe:. Data were obtained from death onti6ates which are aotoriously tm'reliable and have alarp built-in e:ror in 6vor of lung esaoer. H'irayama has never conducted swvival adjusted analyses, which would be aseful io assessiog study data. The unresolved issues regarding the H'irayama study ritse doubts about its quality and the credi'bility of the resuht reported. 'Fbe possibt'Lt y of upward bias must be considered siso, considering the nature of the issues left unanswered. The Report erroneously describes the study/s analysis as based on data grouped by time imtervals, instead of grouped by subject's entry age. Q also states iooorreuly that the relative rislt for .romea married to amoken ieeeases with age (declines aith age is correct). Tbe Report appears to have relied oo, and over-iaterpreted, summary remarks in the NRC and SG reports and elsewhere suggesting that the Hitayama had withstood thorough sautiay, witbout considering recent critical reviews (they are not tated in the report). T6e discussioa in the Report comparing the Hirayama and Garfinkel studies is incorrect and misleading. Tbe Garfinkel study has many merits not eonsiderad. The comparison of the Hirayama and Garfinkel studies is an unfair and misleading attempt to acptaia the observed respoose-itrversion in the Garfinkel study, or to support EPA's preconceived 'mteatioo to demonstrate an ETS lung caoeer risk. 62. USE OF NMA-ANALYSIS Combining study results stateitticagr (meta-aaalyus) is inappropriate in view of the marked d'issimilarities in study quaTity, dcsiga. location, and other characteristics (see Section S.i). Methods of ineta- ~ aaalysis are intended ody for highly sims7ar studies, whether they are experimeots, surveys, clinical trials, etc., 0 a condition severely violated by the epidemiologic studics. Furtbcrmore, overall relative risk from a meta- N ~ ~ u

analysis is meaningkss, because base-line risk of lung cancer from non-ETS sources differs between studies, i.e,, hmg ancer ratcs aot assodated with e3pocure to tobacco smoke are much higher in some countries where studies were conducted than io others. Coo:eqoendy, tbere a only a bypotbetical population to.vhich an overall relative risk from meta-anilytis would apply. SeNvot nwicWO% WtJyoW oUAo.riv dK use of n,aa.anaysi; oa"cerdeC dwr thar it no M!"+d basis Jor iwAudn, f non-U.1 smd+ct !n a combinsd miaTysis )ivr inoriaics on Jie U.S. If study results are combined for me US, for the Asean ooantries, and for the European countries indvidually, tbere is no ioc-tioa of an effect in the US, while the results are itatistically significant for both Europe and Asia. Furtbermore, there is a statistically significant difference between the US. and Europe, between the U. S. and A* and a non- eigaiGcant difference between Asia and Europe. The Varela study should be included in any meta-analy:is oonducted, using statistical methods appropriate for study results from adjusted methods of analysis as from raw data. Thc VancJa mdy is not ,atuaeafty sirijrcan; sacspt ar mi am+e,ruy Jupl, oevrc of espo,strnr to spasd tnoft and t/uur wnura,K,t *lter the ncrulu by conanent dtscrr3cd above. If the Vardh study is included in a meta-aaa}ysis of al1 the Sud'ie:, the ouen!a relative risk drops from the Report'r 1.41(4S% C.I. L26, L57) to 1.23 (95% C1.1A9, 138) (or, alternatively, to (113 (L13,. L35), with the Varela study, alI data for males and females; L2$ (1.16, L42), with the studies of Varela, Shimizu, and the recent study by Sobue et al., femak: only). Tlkc values showee waic roken /Fom Affe,iait ftnicws. When adjusted for upward bias due to smoker m'esrlissificstion (see Section 63), the resultant value is further reduced and does not support the conclusion of an asiodation. Two recent studies, Kabat a aL (their second study in the US.) and Sobue et al (Japan) do not aTport an a:sodatioo between E7`S and lung eaneer risk. They should be added to the Report, along with the Varels study. Siom neither study reports significant fmdimgs, their inclusion in a meta-aaalyrsic would reduce an extimue of overaD rr.lative risk. 16

63. ADJUST'MF.Nt' FOR SMOKER M1SeUSSIFICATION nu Rrporr's ntodrl to odjusr dwt o•+wou rrlamw ,itk dowrcwwd to accowr )roe dIt crpeatd cD'eul of rmoaka rnisclars~~casion ft a mirror cwtsion of she IVRC tPpraocl+. 7lit tcc)uzical iuua nJatcd to nialdg on .djusonent mtd the dcturnFnation of par,arneter viaArm so use how been a saurce of corttention in the liraatun as work on dus lopwr has been e++oUM or+v scmmJhems. Tbs m+e+aer's comne,its rsnd to be sonwM+hat tedYniral mid too latjtlry m duat3s wrU br this bricf rrNew. Tbe Repoet'i adjustment for snssrhte;ficat;,s sboWd; (1) implement concordance ratio and frequency of exposure in a.ray not aSected by mi:cEauiScation, (2) treat persoas incorrectly dassified as .ever-smoker: as a percentage of ever smokers ('mstesd of never-smokers), (3) dearly distinguish between risks relativd to never-smokers, in general, and risks relative to never-smokers not exposed to EFS, and (4) make the adjustment for misdassifcation to each study individual'ly instead of to an overall risk estimate. The report uses inappropriate estimates of the number of lung cancer deaths occurring annually among never smokers in the US.. Suitably corrected (excluding item (4)), an observed relative risk of 1.19 is expected when the true value is 1, i.e., if there is not eRect of ETS. ?he aorsespondina "e pvrn in the Report ir bh The Report should observe fwe basic principles, that include (2) and (4) above. Tbe remaining three recommendations refer to using only female misclassification data for application to study data on females, using parameters for adjustment in each individual study that reflect the time period and locale of the study, and basing the proportion of mis¢lassibed smokers on self-reportcd never-smoken only (instead of a+on-ltsers). Implementation of the pritadpks, Muditg adjustment to individual studies (item (4) above), indicates that only about 3% of the observed relative risk is due to smoker mischssifiation, on average.. 7Iu torrrspouEM NaTus for bioa in ths acport is IOWo, obtoined ftrn I.II/1.2$whe,e 1.28 is the or+croll nlooMr risx odjusud for ntiscJatsi jricotion . ~ Wubout taking estreme combinations from the plausible rsnge of parameter values given in the Report, observed relative risks of L1O -L1S could easily be observed in cohort studies when the true relative risk is 1. For the csse-cantrol studies, values in the range of 225-130 would not be unlikdy, due to smoker 17

ausdassibcatioa alone. These results, taken togetbu,lead to the overaU' conclusion that an overall estimated reLtive risk of 123 (obtained by induding the Varela study) should be adjusted to LOS-L10 to aacount for the kvels of bia: due to -;-4•«:ficxd= .SUlfBlQfcd dOtO !dT POrfl Co1nDU1adOliT of fwwf1lJCl mhiu in dK =ieC pMpl I11 fJlc RCpOrf jQbCQlt AgOi obflAKd oM00gIClOQ1K rLTk6r War 1JgCOlildo=ft, dY! l0 pnoAvm1lC~rS~~CQfJOrt t•10s OAplt, W6Ot lbc Vsjc VaJtu tr Z &<. ADJUSTMENT FOR BACKGROUND EXPOSURE The adjustment to risk for background esposure is predicated on ETS being causally related to increased occurrence of lung cancer. Since that condnsion is not supported by the weight of evidence, there is no basis for adjusting risk. Z3e nse of catiniae data is an unacceptable surrogate for use in adjusting for background exposure to ETS. It is not known to correlate with uptake of lung carcinogens in ETS. 'Ibere is some evidence that nicotiae, and hence its metabolite, is not specific to tobacco smoke as normally assumed. AdditiomaDy, the adjustment for background risk is based on an assumed linear dose-response relationship that is uveritud. 6.5. F.X'IRAPOIATION OF POPLJUI7ON ATTRIBU'TABLE RISK Sorne nWewers liavr fed tlw dota for inaJu =d janalu shoufd be combined in dhe oveeall analysir. Odias JiaNc wped thar data far nealu mrd Jaaoles tJForrtd be kept upmWe, p'our3lY devrloprig afsk uwnmu in P"llcl bi du Rtpnrt The Report obtaims an e:timate of risk applicable to the population of never-smoking.romen anarried to a smoker. ZLere is inadequate justification for eaQrapoTNion of risk to the larger population of esposed females, and then further earapolatioo to include males and former tmoke'rs. The data oa males .rere aot analyzed in the study. Ttiere is little known about the risk to formu-smokers. Ile total estimate of population risk may be very misleading. 7be conclusion that the number of lung cancer deatbs per year N O N CJ N ~ N ~ ~ 19

attrLbiuabk to passive smoking is between 1900 and 6100 is insupportable. Given the amy of sources of .neertainty it is irresponsible and misleading to make such an eact claim. 7. RESPIRATORY DISORDERS IN CHILDREN Cbnsidnobly /ew irview+ers .ddrrsted Mpttr S of die Report dm the dapeas at JM emua. A few rctpondattr, IiowrMCr, toFdy oddnsstd the anateriaJ on nspirotay dfsorderr and did so tn emsfdtrobJe detm2 Tbe Report's conclusion that parental smoking is associated with incrased prevalence of respiratory sYmptoms and dtseases, and with roduced pulmonary itmetiod, is not justified. Tbe Report is incomplete and ieaecttrate in reporting and interpreting the relative literature. SuppatinS amMpJss wer Latluded in the ieuiewAko, the treatment of data and concepts is oversimpliCed. Studies do not appear to have been .iewed critically, potential confounding variables are inadequately addressed, and statistical testing is omitted. Overall, the Report appean superGcial, evea using direct quotations from the NRC and the SG reports .ritbout complete attrtbution. A comprehensive review of all the pertinent literature suggests an association between parental (primarily maternal) smoking and respiratory symptoms and certain diseases in pre-school age children. The observed associations, however, could be due to a number of factors otber than passive smoking, e.g., ioadequate consideration of socioeeonomic status and other bctors, greater sensitivity of younger children to ETS or other agents, reported effects of maternal smoking on lactation or on the child's development in at . No consistent association edus, 6o.reru, for older children or other endpoints in chi]drea of any age. Wadiag asthmae:. s Fsposure assessment is inadequate in the epidemiologie uudia available. It is targeiy' based on measures of parental (or matetnal) smoking that vary between studies. Adequacy of questionnaire data regarding ED'S exposure of children has not beta supported by studies o[ cotinine levels in children. Very few of the fifty or more potential confounding variables that may affed the analysis of tbese studies have been controlled in the reported studies. 'ILis is a very important concern that cannot be ignored in the 19

F Report. The most important aonfoundiag vuiabks include Stnetic aisceptability to childhood reapiratory Moesset cocs-infectioo among family mcm6cr: arithio the bome, aoaiocconom.ic wtcu, number of bonsebold memben, demogtaphir cbaracteriuia of the stndy poQulatiaio, birth wtot, aursiag practiacc infant switioaal aattes, growtb rates, pryr~ [scuxs, age prevakox of pareatal rupintory qmptoms, damp bou4mg ontdoa air pollatioq fatbw'a occupation, iafoctioai acquired in day we aatem aatri6oo, bmily health habit:, parcatal Gtatyler and ot>ura The Report eaodudes avsst body of impoctaat 6toratura in tlsis a:ea which introduces mbstaatial Dias in the conclusions and aaifioes the Report's arodibOty.. ZO