r3l 16/12 '94 13-11 FAX e8l 642 2135 P N LEE A- kikr, .I- JTrL and lung cancer Weakresses of re!cent review tiavers by J TrOdaniel and hin golloalues Author : P N Lee Data : 114.12.94 rartly biujiliuLLad by EUROPASS (the turopean Concerted Action on Passive Smoking of cite Commission of the European CcmmtatItLes) , Dr jean Trddanial. in collaboraclon with Dr Paolo Boffetta and Dr Rodolfo Saracci of IARC, and other colleagues have produced a series of review papers on ETS: A. -Enviruricental tobacco smoke and the risks of cancer in adults". European Journal of Cancer. 1993, _Z2A, 2056-2068; B. "Exposure to environmental tobacco .9moke and adult tion-neoplastic ze-upiracory dise&B&sl, CUrapacafj RespiraLory Journal. 1994, Z, 173-185; G. "Exposure co passive smoking during pregnancy and childhood, and cancer risk; the epidemiclogical evidence", Paediatric and ?erLnaval Epidemiology, 1954, JJ, ZJU-Zj); and D. "Exposure to environmental tobacco smoke and risk of lung cancer-- the epidemiological evidence", European Respiratory Journal, 1994. 2, 1877-1888. ~7 CD, CD Lil "0 CD BATCo document for Legal Services: Health Canada 19 October 1999 16/12 ' 94 13: If FAX 081 642 2135 . P N LEE 11003 .2- Of these papers. D is most rplevant to ETS and lung cancer. The Section In papef A concerning lung cancer is essentially identical Co a similar section IZ3 paper D and need nor be considered saparaLely, No data relevanC Lo lung cancer are Siven in paper 9, but Paper G. which mainly considers child'hood cancer, does have a section citing studies Ln which lung cancer risk in adulchood is related to ETS exposure Ln childhood. As the iARC has a very good rapuratton scientifically, these review papers will inevitably carry considerable weight and no doubt the conclusions of Ukeir lung cancer paper (0), as summarized in the dbXLract below, will be widely cited: "ABSTRACT: Exposure of nonsmokers to environmental tobacco smoke (M) is widespread in European countries, the most serious exposures occurriTig ac home and in the workpLace. Epidemiological atudies available up to 1986 have been reviewed by sereral internaLional and national auLhorities, which agreed In concluding chat ETS axpvuuro is causally related to luitg cancer. A number of epidemiological studies have been published slac" then, and have confirmed this aasuclation. The possibility of positive, results due to bi4a has been envisaged; iL seems, however, chat such bias could not explairt the whole excess of lung cancer. Few d4ta are available on confounders, such as dier and previous history ur lung disease, chat might be responsible tar the association: however, there is no evidence that they play an important role. moreover, the biological plausibility of a causal association is SUPpOZLed by the similarity of the composition of ETS and active amoke. The causal dsuouluLlon betwedn ETS exposure and Lung cancer now seems well-established; however, lLs public health impact in sLill debated. Estimates are available from the United States, Canada. Australia, New Zealand and England." (-;-I CD QD Cn CD Ul IL:D BATCo document for Legal Services : Health Canada 19 October 1999 16.,12 ' 94 13 - 12 FAX 061 642 2135 . P N LEE Q (104 Unfortunately, for those JC&mLliar wirh the data, the revieve art, highly inadequate with regard co lung cancer and give a completely distorted view of the evidence, as I will attempt to Mdkc clear. In my comments I will, at various points, contrast the conclusions of Trddaniel e al wizh those I have reached Vt- - independently by a detailed review of the epidemiological evidence. This review will be submitted separately to the SCUrK committee as 3upport for ~he 3t&teMeIIL3 I make. and 1 referred co it as document RLE (review o.C lung cancer epidemiology). 2. Comments on Paver C In a single paragraph un pp 244 and 21-5, Trddaniel Al refer to three studias that have investigated the risk of lung cancer according to childhood ETS exposure, bv Correa, by Wu and by Janerich. Aa Mo of there were reporLud to find some elevated risk of lung cancer in STS exposed Individuals, the authors conclude in the summary that oThere is some consistency of association between ETS exposure III 4~hildhood and che risk of lung cancers in adulthood.* This is a, totally superficial and inadequate summary of the &WildblW dbLh. As shown in KU there are 11 studies that have reporLed evidence relating risk of lung cancer in nonsmokers to ETS expoaure in childhood. A meta-analyaLs of all the data shows no trace of a positive relationship (Relative risk - 0.96, 95% confidence interval 0.87-1.06). For two of the studies they did cite (Correa and Wu) they reported evidence fnr smokers and nonsmokers (Do N) BATCo document for Legal Services : Health Canada 19 October 1999 16/12 *94 13:12 FAX 081 642 2135 - P N LEE 2 L) Z) Z-- -4- autabLned rather than, as is normal and more appropriate, evidence for nonsmokers only. For the other scludy (Janerich) thev reported high exposure data only, failing to note that this study found no signific2nc relationship of Lung cancer risk with childhood ETS expus-ire as a whole, nor indeed with any other Index of ETS expuoure, Had the authors considered the totai data, chore is simply no way they could have reached chair conclusion, tentative as it 3. a 3.1 Selection ariteria Egr studies unsgared- KLZ considers data from 38 studies. 11%is list includes all 30 in the EPA report, together with eight that have recently presented data, Tredartiel g_t al omit a number of studies because of major methodological limitations but, in a totally unscientific and arbitrary way, fail to xLata vhac their criteria for inclusion of studien were, nor the limitations of the omitted studies. I noce thac four of the omitted studies (Butler, Geng, Inoue and Liu 1) were included not only in RLE but also in the EPA's 111sc. There are also five recent studies (Kabar 2, Joeckel, Du, gang, Layard) ;-n RLE noc mentioned by Tr6danial It &I. 3.2 Faillare to UDdnce meta-anakrsa Cn V 1879, paper D cites a number of early meza-analyses of daLa relating risk of lung cancer in aonsmokluS women in relation to smoking by the husband. Including Chat published by Saracci and U- Riboli of rARC in 1989.None cita data post 1987 and all present a Co k-n CD BATCo document for Legal Services: Health Canada 19 October 1 999 15/12 '94 L:-13 FAX 081 642 2135 P N LEE r_ 006 -5- cotabincd relaCive risk of Lung cancer of about 1.35. Tr4daniel j_r jj. then go on to refer to more iecatiL evidextue, but: fail to give any uFaa~,wd meta-unaLlystis figure for this assoclattLon, mere-,y noting (wLrh absclutely no indication of how it was arrived ac) that "the Increase in r!sk seems to be no higher than 40*." They thus totally conceal the fact that, ftruLly, (~he Moca-analysis escimace for smoking by the husband has dropped dramatically since the early "ta, and that, secondly, the more recenr- evidence abows no significant relationship at all. To support this view I cite various meta-analyses I have carried OuL based an the relative risk estimates given in Table I of M. Husband's MOUDZ - meCa-arislysis relaclve rigks (with 951 CD I. All studies in RLE (n-38) 1.13(l.05-1.22) la. Studies published in 1981-SS (n-22) 1.36(2.21-1.53) lb. Scudies published L989-94 (n-16) 1.00(0.90-1.10) 2a. Studies considered by Saraccl and Riboli. (n-14) 1,30(l.12-1.51) 2b. Studies published since zien (n-24) 1.08(0.99-1.18) 3. All studies considered bv Tredanial (n-29) 1.13(l.05-1.23) 3.3 F"lure CQ sevarate 2Mt findings for differgpea exRgsure indices Having cited results of the meta-analysis by Saracci and Riboli 1>ased an studies up :;o 1987, Triidaniel jLt_ S1 then refer to a rumber of rew case-control studies, that have been published since 11387, pointing out that they pro-.ride data, noL only on risk in relation to husband's smoking, but also on risk from other exposuzes (e.g. 1 il (Z) ~10 BATCo document for Legal Services : Health Canada 19 October 1999 16/12 ' 94 13 - 13 FAX Q81 642 2135 , P N LEE Q oo-, -6- i during childhood and ac wurk) and in relation to extent and duration of exposure. Vhat is staced to be on `vvLLview" of resultS iS preserred in Table 5. The cext and Table 5 are grossly misleading for a number of (i> :t is noc made clear chat many of the earlier studies considered by Saracci and Riboli (and others) a 1sc provided data on indices other than husband's smoking and on extznt and duration of exposure. Any overview should clearly consider all rolovant daLa. (ii) 'to attempt whatsoever is made Lo bring together resulto for r~omparable indi.ces to obtain a valid summary of the data. Many bLudlv4 pi.uvLde results for multiple indices of exposure and lc is obvious that, whether or not ETS is carcinogenic, the impression can easily be gained that it is by selectLng out a relatively high estimate from each srudy in rho "averview". 7ridattiel Z.% Al give no Indicacion of the crIlteria. they used to select out the results in Table 5, which is very unscienzific. I am not saying that they deliberately tried to give an overstated impression of the association, but there are carzainly some examples where the effect. has been Clio same. e.g. (a) Brovrzon 11 - why select "heavy axposure during adulthaud' estimated using a semi-quantitative evaluation, rather Lhan spouse smoking, as is most commonly used and which shows a relative risk of 1.07 U7 LM ---0 ~~I BATCo document for Legal Services : Health Canada 19 October 1999 16/12 '94 13-14 FAX 091 642 2135 - P N LEE Q ow (b) Mim zu - why include resulLs for nother and husband's fattier vilen other such indices are not selected in ot;her studies, and when the source paper gives a number of lower relative risk escimates thaL are not cited (father 1.1. sons or daughters 0.8, someone else ar- workplace 1.2)? (C) Svensson - why give results for sic home and aL work rather than ac home only? (d) Cao. Li - why give results for iteavy or long-term spousal oxposu2a when the normal index used is any spousal exposure, and a separate taUle (6) deals with dose-rexponse relationships? (1-ii) No attempt lis made to present mera-analyses of resulca reLating to diftering indices of exposure. Had this been done. it would have become clear (see RLE) that there is no relationship of lung cancer with eiLhez childlivud exposure (k(R w 0.96, 95W CI 0.87-1.06), or workplace exposure (RR 1.00, 95% CT 0.91-1.08). and Lhal; U10 association WiLh spousal exposure is. as noted already, now much weaker than reported In Lhe early reviews. 3.4 Inadeguare digcussion of methodological 11robigms This section of the paper starts, on p 1881, by noting that it is highly unlikely that chance alone can explaits Lite ausuclatiov, bet7dean ETS acid lung cancer. It should ha-ve been made clear that zhis only applies to the association baLween sp ousal smoking and lung cancer. Because there is an a5soclaCion of one Index and nOL U1 t*jo others with lung cancer, Lhere should have been a dtsC'J5310[1 BATCo document for Legal Services : Health Canada 19 October 1999 16., 12 ' 94 13 - 14 FAX 081 642 2135 , P N LEE 1Z U.)g considering whether this i s due to the tirst indox being particuldcly susc~ptible to bias. or to it being a beuter marker of true ETS expos-.zre. As Trddan!eL et . jLj fail to summarize theLr data property, and so do not even seen to realize there is no associscion of lung cancer wich childhood and -workplace RTS exposure, their ability to LULerprec the data validly is much weakened, The discussion of methodological problems, though raising many of Lite importanc issues, is limited in that little or no attempt; at quantication is made, i.e. the likely magnitude of bias from a particular source is not compared with the magnitude of the association to be explained. Without such an attempt, though difficulc for some sources. one can have little confidence In Lhe views of Trddaniet er Ll that "bias could not explain the whole excess of lun6 cancer." Coments relacing Lo some specifLe biases are as fo.Llows: Misclassificati2n of self-reported smoking stata. I have published widely on this issue and have concluded that mLsclassification of smoking Is an important decerminant of the slighL excess risk observed in nonsmokers married ta smokers. Irk a ruccuLly cujitylek,*d paper to be submitted to Seazisellas In Medicine (actachad as Annex A) I make It clear that the adjuSt.joent procedures used by ErA are unsound, being based on methodologically Incorrect formulae and unzealluk.Leally low esti=are5 of rho mLaclassikication rate. I find U1 CD BATCo document for Legal Services : Health Canada 19 October 1999 16/12 '94 13:15 FAX 081 642 2135 . P N LEE laolo -9- it unsatisfactory chat Trddanial Lr Al have not even pointed out there is a dissenting view, let alone put forward any reason why L11IS vLuw iu wi;ut%. Publication bias. Treeanial et " make no attempt to Use Lhe current available data co tesc for publication Was. Kdd Choy done so, they would have found some, As shown below, there is significant (p-0.05) evidence that risk estimates are higher In smaller rather than larger studies, consistent with che probability that the studies most likely noc to be published are small studies which find low relacive ria" . sttl~dx size RR (95% Cl) At: least: 100 cases lu 1.06 (0.97-1.16) 50-100 cases 14 1.35 (1.14-1.59) Less than 50 cases 14 1,26 (0.97-1.63) Ccnfoundlnz factors. The :eeancly published paper by Thornton SL in Journal of Clinical Epidemiology has sbuwu LhdL, for a wide variety Of liresLyte LL~ok fauLurs, prevalence is not only higher in smokers Chan In nonsmokers. buc is also somewhat higher in never smokers living with a smoker thin in those not living with a smoker. This paper presented q-.L%tLLiLdL1ve estimazes; which demonstrated the potential LInportance of confounding to the association baLween lung cancer and smoking by the husband, especially in view of its weakness, with a relative risk of only 1.13 fur the 38 studies. 00 co BATCo document for Legal Services: Health Canada 19 October 1999 13-15 FAX 0,11 642 2135 - P N LEE la(il .10. 3.3 "Proof of causac_ion CV101bt-gul;.1 s&.Ld -3Li:gcixQ ujC AbbuL;iitiua Tredaniel pS &I admit cbe association with smoking by the husband is weak, They do not point ro various Inconsistencies in LL; Lur Unst2ple L110 variation In relative Lisk over time. with no association reporLed In BLUdLeS CuliducLed in recenc years, or Che fact that relative risk& are mucts higher lit sLudiets clabulfied as *inferior" char in those classified as "superior" by certain deflued criteria (see RLE). Nor do t1sey point to the fact that the overall eviJence sLows no association Witt% or-he:' Indices of M exposure. such as workplace, childhood or social exposure, Weak and inconsistent would be a fair summary of the evidence. The discussion with regard to histoloSical c~rpe is obscure and the paper nowhere addresses the key issup. as to whether the evidence suggests that ETS is related to squamous call carcinoma (strongly associated with active smoking) or adencearclinoma (weakly associated with active smoking) or both. As disci-ased further In RLE, the evidence is its fact conflicting, thus weakening the 'proof of causation". Coherence Though the association betveen husband's smoking and lung cancer is evident in various different countries and conLittents. aLrosig rdVidence of a cause and effect this of itself IN E1U relationship. Various sources of bias, such as misc Lass if icac-lon of cc smokLng habi'zz and contoundLng. could be expected to apply widely. 1~0 BATCo document for Legal Services : Health Canada 19 October 1999 I" ' 94 12 - 16 FAX 081 642 2135 - P 3 LEE Q012 61U .il- D!2se - reRans e ejaclotishIR The evidence presented in Table 6 is neicher compreherLuLve nur systematically examined. Though, as made clear in RLE. there is a =andency for nonamoking women married f-o hedvy smokera or to smokers of long duration to have an increased relaLive r131K Of lung cancer, zhere are a number of factors unmencioned by Tredaniel et gl which Limit. Interpretation. '11tus they fall to nota strong evidence that tudies uhich provide dose-response data are highly selected, with he overall relative risk estimate Cor huuband's w;.Aing 1.37 (95% C1 1.23-:.52) for the studies chat do provide data, and 0.92 (95% CI 0.83-1.03) for those that do not. Nor do they point to various sources of bias (recall bias, conf ounding, misclassLfication of smoking) that may cause an artificial dose-response. Biological 2latisibili No attempt Ls made to compare relatLve ex=osure tu smoke constituents from ETS and mainstream smoke. The receat study by Phillips 11 Al suggests thar, on average, eXPOSU.-e LO parLLculaLe matter and nicotine from ETS is some thousands of -_'imes lower than that fron active smoking. It is difficult to see IM-W one can assess plausibility widicut taking into account the magrituda of exposure Ln. relation to rite magnitude of the claimed effect. ~he reference to tile HULOpSy study of Trichopoulos Sj Al is grossly misleading in failing ED point out the fact chat the inde-A U-1 of lung damage used (epithetLal, possibly 7recancerous lesions) r_-% f CZ) showed no relationship with active smoking. be:.ng similar lit 00 I CD Lr CD BATCo document for Legal Services : Health Canada 19 October 1999 16/12 '94 13 17, FAX 081 642 2135 - P Y LEE 4) 13 -12- lifelong never smokers as III smokers or luote Ulan '40 CL&dCetLeS a day. How can an Index Lhat is unassoulat-ed wILIL accivu smoking be used to suggest harm from ETS exposure? Artinal evidence It seems surprising that Ttedaniel rj Al d6 ME 2646MABd cond-acting lifelong animal Inhalation studies, and recommend furLher scud.es in pat dogs. ReferrLng to the Reif study as findinS a weak relatiorLship seems misleading in LhaL LIM relMLIvel titik ubservq~d was not even close Lo being sLaLlstically signifirant. Tredaniel it al do not make it clear tnat; linear extrapoLaLiort from active smoking data would suggest a very much lower risk resulting from ETS exposure than that suggested by the ur the association between huBband's smoking and lung cancer. No do they consider the possibility of a threshold dose. Ceisation-of exposure When examining whether an association is causal 0: nUL, evidence relating to reduction of effect given reduCLIJU of exzcsure ts of:en considered very Important. Tredaniel g-t- 11- fail to =ske LL clear that no such evidence exists here. Sg= El ry It is clear from the arguments put forward here and from the e-iderce presentod in RLE chat the available data do not, as c.aimad CD C~3 (JI CD BATCo document for Legal Services : Health Canada 19 October 1999 '94 1317 FAX eal 642 2135' P N LEE Q014 -13- by Tradanial g; al "neen to fulgil ar least to a rosanoTuable uorrso, the criteria needed to accept a esuaal link between ETS and lung cancer among li felong nonmitakera. 3.6 Public Realth Ig2agr. elsewhere (refs) L have shown quite clearly that the escimates of numbers of deaths per yoar in relation Co ETS nre of no sclen-.ifie value. For #xample, the EPA's figure of 3,06n lung Cancer deaths contains a hi6h pr oportion of deathe estimated by extrapolating risk estimates tor spousal, smaking to estimate risk in relation to workplace exposure, torally ignorin& substantial evidence that lung cancer risk is noc associated with workplace U-1 '13ATCo document for Legal Services: Health Canada 19 October 1999