Tobacco Documents Online

DO11' 96 16 :47 wICSBR ' Failure ; ~ Analysis ' Assodates, FIU+A-SF R-90-09-10 P.3 ~~sow%w 6woft I~w Cam~onwsam Dnve. 04. bR 3016 M~nb Rsk, C~+YkrNO ML~l6 Uf5)J~6P/OD 7G4210 Fa(UiS)Lb40?? Review d Heelth BYi+ects or Paalve Bmoklam Astesaaent ot Laq Canar In Adults and Rapiretory Disorders !n C6lidren (Review Draft; May,1990) dc Office of Retearch and Development Office of Air aod Radiation U.S. Enviroamcatel Protectioo Agenoy by Willian! Y. autler, Ph.D. Fariure Analytij Al3odatet,lac.a 149 Commonwealth Drwa Menlo Park CA 94025 September, 1990 wmm c.NAoM •~.•,. n...M «rren w~+4vur ~rer. Asr.. rr.i~aa l~rwr wo.i~OC tf'r.wR vwrr WbMj ratift eC@ 1t j*0

. OCT 11 '90 16s 47 41CS8R Review of: Health Effects of Passive SmokinS: Assessment of Lung Cancer in Adults and Respiratory Disorders la Children (Review Draft; May,1940) 1. Intrbductlon This review addresses only the U.S. Environmental Protection ASency's (USEPA) consideration of lung cancer and exposure to environmental tobacco smoke (M). It does not uddreas the USEPA's consideration of E'TS and respiratory disorders tn children. hurther, this review focuses on the first step in the risk assessment proceu for E'I'S and lung cancer, that Is, hazard identifIcation.t The determinatior tlwt an agent posas a hazard to human health is necessary before proceeding to the subsequent steps in the process. In the absence of such a determination, the wbsequent steps are not rekvant. 'ihe USEPA's guidelines for mrcinoSen risk assmmnt consist of scven elements that should be Included In a hmrd Identification when tnformation from eacb is available: phyYicu1-chemicasl properties and routes and patterns of expowre; structuro-activlty relationships; metabolic and pharmscokinettc properties; toxicologic effecu; short-term tests; lonS-tarm animal studies; and human (epickmioloSic) studies. (USEPA,1486, pg 339941 '1,tu USEPA's review of ET'S and lung cancer considers only one of theso elemenu; that 116 epidemiolaSic studies. Because of this incomplete hazard asses:ment, the USEPA's report does not provide a risk assessment of BTS and lung cancar witicb is In compliance with USEPA YutdeUnes, Consequently, the conclusions in the USEPA's report canaot be described as being based on these Suideliret. iUSEPA,1990, pS 1-33. In faa, the USEPM hazard identification addraesed a aM lisnitod questioa TLe USEPA writes: This rgises the quession of whether any dired evidenee exists for the relationship between ETS exposure aAd lung cancer in the Seneral population and what i aLmplication~ may be for publicaM " i ns th document addresses t question by reviewin and ~ cumulative evidence from epIdemioloSlc studies. tI~SF.PA,199~pS 1•3) 1rM foW .t00 tn tbc r6tc u.eMMst prooaa Wt t1nrd t6oattttatton; dors rapont. waMmeat; axpoturo unsrne.nt; W rhk dur.cirrirstioa. (Natioo.l Aeie:e& Comd' 19" P.4 F.Ihr~e ~ -•- -.. ......,-.WJ M^s+4 BzeEt 06. 1 t 100

4C1' 11 '90 15:48 jacsw P.S It is not clkar why the USEPA chose to rcitriet its attention anly to epidemtoloSte studies sinee, lu Its own guidelines advise, they "are tnharaatty capable of datecting oWy compmtively tWV increases In ttu relative rtsk of comr. tUSEPA,19K pS 33995, emphuls added] Sinae prevIoiu raviewj of ETS and lung caacer clted by tM USEPA have reported relatively =ala magnitudes of association from epidemiologic studies (National Research Coundl, 1986; U.S. Surgeon GenetaI, 198b), the need for evidenoe from the other elemenu of a baurd jdentMatioA should have been recognized from the beginnittS. 'I'lu potential distortion from confoundto,g and misclauiflcaation is a major reason why cettala epidemiologic study deslBns are recogniaed as providing limited Information for coududinS causalitSr when the magnitude of the estimated relative risk Is smalL Acknowle4iq these sources of bias, the USEPA has set three neceisary (thouSh not eufficiant) conditions for determining whether an epidemioloStc association between an &Sent and cattcet is, In fact, causal: 1,'Y'bere Is no Identified bias that could explafa the atfociation. Z The possibility of eonfouAft has been considered an6 ruled out u Zaintq the Assodation a astociation is unlilcely to be due to cbance. ~IJSEPA,1986, pS 339991 The epidecnioloSic information on •BTS and lung coacer 1a not atlficient to meet ttase three eondidoAs. In fact, a combination of miscles:IQcatioa biet, confounding and chance could oplain masnitudet of association even greater than thoee whicb have been obseived for ipousal smoldq and lung caaoer. Further, tbe abseaea of Information on other epidemioloSic aiteria relevant for assessing causality (that ir, ooherenoe, temporality, specldcity, .tc) also preeludes concluding that RI'S CxpOfllre is a ptWC of lung cancer a=n$ non-ulfokerf. +jude that ST'S racndt~r,e.h , t cau,_ ~ of ttne c3naer in n~mnlrert ~ -Z- - GG1~1M1JN ~an~td~ WOa~ 13191 •C. Ii 100

OCT 11 '90 16:49 WCS8d2 2. Cafculatlon of a Summary Relative R1sk I I The USEPA Identified twenty-one wse-oontrol and three cohort studies which addressed the ussociation between spousal smoking and lung cancer. They combined the information from these studies itt a meta-analysit and calculated a summary relative risk. However, their statistical analysis i: flawed in the following ways: 1) The USEPA used unadjoterd estimates of relative risk from the case-control studies instead of the preferred adjusted measures. 'I7teir choice is clearly wrong. Adjusted estimates are calculated usinS methods designed to remove part or all of the potential distortions In the estimated raEative risk which are introduced by the variables for which adjustmentc were performed. For this reason, the adjusted estimates are preferred. Of course, the concept of adjusted versus unadjusted estimates does not reilert a true dichotomy since the degree of adjustment depends on the number utd types of confounding variables which are included In the adjustment process. ASo, race and source of the case sro the wriables for which 4uttment was typically pcrforrtud In the epidemiologic studies of spousal smoking and lung cancer. (USEPA, ]990, Table 3-1j Thc potential Impuat of additional confounding variables which were not usually included In the adjustment process in thesa studies Is addressed In Section 4 of this review. 2) The USEPA did not include in their meta-analysis the results from two case- amtrol etudies for which they had not yet been abie to obtain raw data; namely, VARS and SHIM? This IL not a valid criteria for excluding relevant epidemioiogic datA In a meta-analyiis. Imporutatly, both of these studies roported adjusted estimutee of rolative risk which were tower titan the summary measure reported by the USFPA. Zbtu, indusion of these t;tudies Is expecced to lower the summary estimate of relative risk reported by the USBPA. 3) The Mantel-Haenszel method was tuad by the USEPA to combine the euradfiWed estimates of relative rWc. Tbis method was developed for use with raw data (hence, the USEPA'c restrlctlon on studies) and Is not aa:tly applied to adjusted estimates of relative ri:k. I'ha preciuon-based approach used by t!b National Research Councll(1986, Appendix B1 to calculate a summary relative risk Is ea:ily applied to such analyses and, In the absenes of an approprlate alternative, is the tYSethod that should be used by the USBPA.3 2 The aamyms md by ths USBPA w idsAtify Mdtes an akd in tb4 reYtow a %iq. 'i'he ehrtioat of ~e ~nd'iridwt aod'i.c are pro..ated tn Tat~it 3-1 of tbe USBPAti raparc. The MuW1-Haeawt aad prWWon-bsud eiatsoe4 produos idaelto.l wMUouy .ulmatN &aW aoarwoaoe Ieqarvrk uetV raw dsa u+nw ft eroup of stuXw coeuid.rod by tbc USEPA. . -3- • .~ ~ ~uuu 7 v_1/ Y J 1.1 P1 ]I J P.6 raaun. N++iyw 12101 06. 11 1D0

OCT 11 190 16 :50 wS&R 4) A case^oontrol study of spousal smoking and lung auloer by Sobue et al, (1990J was not identified by the USEPA. This study was published only recently and, thus, may not have been available to the USEPA when they performed their literature search.'t The tummary estimates of relative risk which are presented here correct the flaw: listed above. They are based on the twenty-five available *pidecniologic atudjes: nlneteet: Case-oontrol and three cohort :tudies used by the USEPA in their meta- analysis; the two case-oontrol studies identified but not used by the USEPA; and the additional cxse-control study by Sobue et a1.ne data reported by the USEPA for the twenty-four studies they identifled are used In this uWysiss AD swnwAary point e:timates and confidence intervAls are calculated ustng the precision-based method. Location Number of Studies: case-control/ Cohort Relative MOM 9596 Cod tnter. United statn $ 1 S ' Asia 1~ l 1 4 as;1. a i Europe 4 1.75 1.14,270 All /3 1.?8 L16,1.42 Comparison between locations: >~~ U.S.vsAda U.s. vs 0.04 U.S. vi AsiaJEurope Asia vs Europe The sumrnary estimate of relative risk calculated from &1H twenty-Hve studies it 12& This mewre indlcYtee sn exeeas risk only sliShtly:nore than two-thirds of the value of 1.41 reported by the USEPA (199Q, pg 4-24j ( 0.68 - .28 j.41j. 4 rroae iba the study by wWt.on et at. liM) m.y b..nodW wur+o. ot.pteea~toa~c iatormnto" as this copk. Howdver, it is aet dear that their saa t~rould $uotv aa .o~aeaiiutioa ot the auoeiadoa betweea sPa+rN anokft aod !wW aaar. Sam t1eir .tady 4 4rP re>wtlwe to Oaay o[ tbo otben jacluded by the USFPA, u is rooommended tbat omsmnnlatt(oe aft Dr. AnLoa be a"mPted to dowmine whather theu data aei be WndW in 0e USBPAti npat. Cortotpoodoaoa wlah Df. ~toeLai was initialed by d3~ wtbar, but i rerpome $on ~m bu aot yet been rooelMed. Tba d~ta for the oohott uu~et arc t~koa trom ZYbie 4~1 ai t6e US)7TAti report aod ue tba ~.me .. tha. wed by t!4 Nattoaal RaearsA CoMe>! W 'P6a data for tW aw~ Aadka an e7QlaCtOd rfOfltWAOi 3.5 (IlaadjUtted tiliellat0l) a" 3-6 (4Uitid MLiA1itN) of fbe USUA's repOfL Tbe data for uaWjuited astimata from aae-oowrol audiet differ djk1y betwm 'lWAw 3,S aod 4.1 of the UMAti reMc. Dir.ct oommudesison wSrh Dr. Sbimiee provided abe tx.t-.vatt.bb iotormatba on the oonfiderue interval for ttit oatimale ot ibs r4ative rhk that h, RR - 3.1 with a 05% ooeAdeoae ldtetvat o(o.6S,1.aS. P.? - 4 • w,.Fa:~ ~aa~nVJ S1S11FINN aWnnttiA uoaA aZtEt 06. it 130

OCT 11 190 16 - 51 WCS8d2 I Ichwevor, substantial heterogeneity In the maSnitude of the relative risk Is observed -across the three locations of the studies. The summary estimate for the YJ.S. is tcututantially (and significantly) lower than that observed for studiee oonducted on Asian and European populations. Further, the summary ettimate for the U.S. Is not xlgnlficantly different from unity. This heteroSeneity In summary relative risks by location conflicts with, the USEPA's statement regarding the consistency of response aerots the available studies. {USEPA, 1990, pS 1-41 As pointed out by the USEPA, homogeneity in response reinforces the conclusion of a causal relation'hip between the exposure and response. However, this is not the ca:e here, suSgestin,S that factors In additioa to ipowsal smoking may Aeed to be considered in explaining the observed snociation. The USEPA indir+ecxly adutowlcdged the ditferenoe in the magnitude of association betwttn U.S. and Asian populations in their discussion of the findings of the two major cohort studies. [USEPA, 1490, Section 3.7] The USEPA conjecxured that a number of lifestyle variables which differ between the populations miSht alter the observed essociation of apousal smoking with lung cancer. 'rbese variables iacluded house alzes, the number of smokers per volume of afr, divorce rates and percentage of office workers amon,S females. Though there might be such differences between Asian and US. populations„ these variables would not be eacpected to explain the greater mSnitude of difference in relatfve risk between US. and European populations. Further, these arguments conflict with the fact that the Woest summary relative risk Is observed among European (and not Asian) populations. 'rhe statistical combination of epidamioloSic data In a atata-analyats as perfoimed here attd by the USEPA is appropriate when the variation eGrou mtdies is entlrely (or at 1east mostly) attributabie to random vuiadon. It it not appropriate to calculate summary values when heterogeneity among itstdlas i: attrtbutable to difrerenoes either in the populattons'sus cepdbilities to the exposure or In the types and magrtitudes of the effects of us,coiurolled confounden. [Kleinbaum, Kupper aad MorSenstern, 1982; Breslow and Day, 19801. Citrron the dupftude of the statistically significant ditCerences in assoAatfon between the U.S. and other populadoers, 1t Is not appropriate to iater that the rnqnitude of tha assoeiatioA calculated aaosc all populations ts valid for amry of the populations. Spedficauy, the tummary estimate of 1.09 (and not 1A1 as reported by the USUA t149a pg 4-2u]} should be used to atse:t the association between spousal tmofdnS aad lM caaoer In U.S. populatiau. Stncx this estimate Is not stati:ticdl+y signifkant, the USEPA's condltion that chance be ut unlikely explanation for the observed epidemtotogtc associatjon Is not met for U.S. populations. (USEPA,1986, pg 33949j P.8 r.uUFO r. _ s . - ArakA1s nIni1NWM aanirdd woai EaseT 06o it 100

OCT 11 '96 16:52 WCSSR 3. MixlasNffcation ot Smokm si Non-smokert The USEPA acknowledged that mi:cl.:sification of smokua as non4mokers could retult in s biased estimate of the magnitude of the association between tpoMl smoking and lung cancer. Itey provided a function of the observed reluive risk which produces an estimated relative risjc corrected for misclauiitcation. (USEPA, 1990, Appendix Bj Corrected estimates of tbe summary relative rLtke presented in the previous section were calculated using the USEPA'r method and are listed below: Adjusted for Hver-Smoicers Misclassitied as Non-Smokers? No Yes "tion RR (95% Cor,f Iater) RR (95% Conf Inter) United States 1.09 4.94,1 0.94 0.78,1.14 Asht 1.44 1.25,1.66 132 1.12,1.56 Europe 1.75 1.14, 2.70 1.66 1d1Q, 2.69 All 1.2$ 1.16,1,42 1.15 1.02,1.30 The corrected estimate of the relative risk for U.S. populations b less than unity. Tfiis shows that, using the USEPA't method of 4ustment, tntaclassif[catfon alone could explain the observed aaoeiaiion between ETS exposure and lung cancer in U.S. populations. For the other locations, the suiarnary estimates are reduced but sdll above unity. The function used by the USEPA to obtain the corrected estimate of relative risk requires the specification of a number of factors. Zhese factors are listed below along with the values for each whlch were used by the USEPA and the range of vtlues for each wbicb the USEPA reported as betng reaaonabla. v" U..d PUooe by dK ttWA wnaa ?0" or Clazedt emohn aaUelwi6ed u uwrawkm LS96-Z3l6 Foawr smahm mLd.mtfud „naW AootcM 4% x.Mi-"9i Newr amokm w5o m merW to auer.at mok.n 60% 4SlF7S% Formar aaakers who ve marrkd to uvreat =ok.n 82% 71%A2li Current Ona'sm a6o are m&rried to axnet aaokect 12% ?1964i2l6 1teladNS rLk otluae eancer 6or aMM smok 5 VAO me Mbdudw U WW am*= 1070 Former sawkm who rre w{khwified as nmr smokors ~ Z.70-3.t6 P. 9 • 6- f ~r~inr. eee • 7l.LJ ~ t~i~1NNM aHf11 I li d wOaA f= lC 1 86. t t 120

OCT 11 ' 90 16 :S3 tiJCS8R The values used by the USEPA are close to the midpoint of the ranges for each factor. Using the USEPA's values for these factors, a true relative tisk of 1.00 would he expected to generrte an observed relative risk of 1.14. Thua, if the values for each lactor as apeciflect by the USEPA are correct for the population being conridered, then an observed relative risk for apoual smoking and lung atnoer should be compared to 1.14 (and not unity) to assess the presence of an association. Of course, the values used by the USEPA are only one posaible combination for those fnctorx. Other combinations of values would be expected to generate observed relative risks greater or lea than 1.14 when the true relative risk wos 1.00. To examine the range of relative risks generated by different combinations of values for each factar, we performed a simulation ia which values for each factor were chasen uniformly from the ranges specifted by the USEPA. The resulting distribution of obderved relative risks Is shown In Ffgur+e l. it Is seen that the distribution Is bounded on the left by 1.00, Is slightly skewed to the right and hu a mode at 1.14. More than half (62.6%) of the combinations of values resulted In an observed relative risk of 1.14 or greater. An obterved value of 1.20 or greater Is generated by 10.7% of the combinations while a value of 1,22 or greater Is generated by only 4.0% of the oombinations. Using the methodology and range of values selected by the USEPA, relative risks as high as 1.20 are not unlikety to be observed when the true relative risk is 1.00. To adjust for thk potential magnitude of bimc due to misclassifictitlon, an observed relative risk for spousal smoking and IunY cancer shoulcd be compared to a value of at least 1.20 (and not unity) to assess the presence of an aasodation. Thut, the USEPA's conditlon that no identtfied bias be a possible explanation for the observed epideatiologic association Is not met for U.S. populations. (USEPA,19S6, pg 33999J 4. The Potenttet Role of OonlbuAdtag Variables '1'tte summary estimates of the relative risk discussed In the preceding sections measure the assodation between spousal smoking and lung cancer. Spotual amoking status is an inexact measurement of ETS exposure sinoe it does not Indude Information on either the amount the husband smoka ln the presence of his wife or the non-spousgl sources of E'i'S exposure. FLrtber. epoussl Iaroktrg status Is not a :urragate measure apedfically for M'S exposure sinoe It Is expected to be correlated with a number of other lifestyle variables. If these other variables are also associated with lung cancer, then the observed associatlon between spoutal tmoking and lung cancer could reflect the risk associated with these other lifestyle factora ather than a rlsk due to ETS eacpos<tre. The degrec to which this is true depend: o0 the magnitude of the associations of these ltfestyle factors with both lung cancer and spcwsat smoking. P.10 . ~taN" °tO1~«N~ ~N~~rdd wot~d f2~ET 06. Ti 10O

OCT 11 '90 16:54 SdCS&R The USEPA states that theIr 'review and analyses of the epidemIologic studies w have not indicated a oorrelate of ETS that may explain the observed association between ETS and lung canoer." [USEPA,1990, pg 4-1] Obvioudy, it is oorrolatet of spousal smoking which would constitute oonfounders since it is ipousal smoking (not BTS expaure) which the USEPA has used as itt exposure variable. The USEPA does not present any support for this conclusion, either in the form of a review of the literature or an analysis of available data. As required by their own ritk assesameut guidelines [USEPA, 198b, pg 33994j, the USEPA should indude an analysis of the potential impact of confounding variables at least as thorough as that which was presented for mitclauiflcadoa M is partioularty important since, given the:mall relative risks for lung cancer and spousat smoking (see Sections 2 and 3), it would not require a very strong confounder (or set of confounders) to explain the observed magnitudes of association. A quantitative expression for the observed tutadficstad relative rlsk between a disease and a dichotomous exposure variable in the presenee of a dichotomous confounder was developed by Kupper et al. [1981j. Under certain general auumptioas6, It can be shown that the unadjusted estimate of the relattNe risk calculated from a cohort study design is a fttnctioA of three perametera: 1) the relative risk of disease wlth respect to the presence or Absanee of the confounder. 2) the proportion of exposed itldividuafs who have the confounder, and 3) the proportion of unexposed Individuals who have the ooafounder. Letting RR(D/F), P(FIE), and P(F) E) represent these three parameters, respectively, the expected value for the observed unadjusted relative risk, cRR, Is eacpres:ed as followt: RR(D/F) P(F(E) + (1- P(FIE)) RR(D/F) P(F1 -E) + (1- P(F1 'E)) Tbe unadjusted relative risks whicb are calculated from this expression for spedW values of the three paruneters are listed in the foUowiqg table. The magnitude of the association between lung cancer and a potential confounder Aeed not be exceedingly large to cause observed relative risks In the rattge of 1.1Q For example, a relative risk of 1.08 between tpousal rmoking and lung putoer would be observed If there were a confounder which had a relative risk of 2.0 with lung cancer and was I 6 The fanowieg artwaptloat wete msdt ta dWdbpie9 tA4 ePrOS{W 1) TLe r+dative tW !or the diwm wM rtspect to tbe peGSeaoe or abwnc4 of evoulrta it 2.00 Whew #AeOtd for the pmeaoo of tbe ooolowndor; 2) 7bere K no Interadlw dfoct botwaea the experun aad eoaround.r varl+iW.L P.11 IaA u~e u~e ~$~ S I S~lIdNli 3af11I k!d WOad C2 tE i OB~~ 1O0

present for 30% of those with a spouse who :maleed and oAly 20% of those with a spouse who did not smoke. wAwa.ca Ra.cwe Risk ottWw c.mo f&SPMt sMoft ro~de~. tn " rreseam ofaDkaaomow c= 'T1ro~e bcpo~ed~ta~.~ VA&tivd Rht a[ Dium for ths Coato~ aT8 No BTS 1.10 1.70 L3o L/0 %.SO 1.7S ?+00 -;m~ .OS A2 1M 01 1.01 1.01 1101 la 3A3 .10 AZ 1.01 LM L02 1M 1A lA6 LOS .10 ,OS IM 1,01 L01 1.02 3Aa 1.04 IM .20 A3 L01 L03 IM 3.06 L07 1.11 1.34 .20 .10 1.01 1.02 L03 1,04 1AS 1A7 1Z JO .10 1A3 L04 lA6 1A6 L10 I•14 l.ld X 1.01 LC2 im L04 1AS 1AT 1A .10 ~ 1.CI 1.04 10 1M 149 1.U L17 AO X LO1 lA1 ha3 L04 1.04 1.06 LOs .S0 .3o IM L04 IM to Lu L1S The variables to be considered as potential oonfounders of the auod'uloo between tpousal smoking and luttl cancer would be selected, preferably, because they are known to be assodated with both spousal arnokiog aad lung cancer. 'Iltis Is a particulariy severe restrtaion since linmited informatlon is avatlable on such variables. In practice, It Is neceaary to consider variables which either are known to be assoc;ated with spousal smoking and might be rlsk facton for lung cooer or are known to be rislc btctors for lung cancer and might be associated with spousal smokit;g. With rtspect to the former, there is limited Information on the differenoea in lifestyle faetors between married, noqimoldttg women whose husbands differ on smoking status. Freedmatt et aL t1983] noted that many "health-related cbaracteriitla' were asaocated with self-reported hours per week of BTS eacposure. Since they fouM that reported BTS expowre fs associated wttr spouse's tmoldng status, the factors they found to be associatcd with Ei'PS exposure might also be associated with spouul smoking status. V.O •enuJ P.12 ...+ !~ 4iSLIMNM aLIfl1Ik1j WOlIi 6atC1 e6. I1 120