Tobacco Documents Online

Lctccrs to thc ~dhor AN ESTIMATE OF ADULT MORTALITY IN THE UNITED STATES FROM PASSIVE SMOKING; FURTHER COMMENT Dear Editor: Rcpacc and Lowrcy (1991), Khoury (1991), and Wells (1991), in the ongoing discussion regarding the original paper of Wells (lgS8), raise a number of points which deserve further comment. Rcpacc and Lowrey attempt to explain the ap- parent discrepancy bctw¢¢n the cpidemiological data, which suggest the excess lung caner risk in ..relation m ETS exposure is 10-20~ or tha¢ from active smoking, and the dosimctri¢ data, which suggest a factor orders o£ magnRud¢ lower. They sta¢~ that there is no intrinsic reason for c~- ciaogenic dose.response relationships to b~ linear. This may be so ia cheery, but in practice one notes tha~ t~¢ cpid=miologi¢ data on active smoking c¢~l~y indicate an approximat¢lF linear relation- ship between number el cigarcRcs smoked and risk of lung ¢~c:r. Doll and Peto (1981), whom Repute and Lowtcy cite, sugg~s~ a quadratic com- ponent, bu~ this would on17 serve ~o heighten the discrepancy under discussion, not explain R. Fur- • crmore, one should also be aware that the EPA i~cl~, for whom Repute works, r~comm~nd for routine use in risk assessment a dos¢-rcs~ons~ r~lationship that is approximatcly linear at low doses (~dcrson ctal. 1983), and detailed argu- ments hav~ been prcscnmd in support of low-dose line~ity ~¢to 1977). It is of course ma~cmati¢al- ly possible alw8ys ~o find a suitable nonlinear equation to fit any obscrv~ dos¢-~sponsc relation- ship. The one proposed by g~pacc and Lowrcy to fit the relationship of lung can~er risk among activ~ and passive smokers to their estimated average intake of smoking-related par~culate mat- tot is, however, of no value at all as it totally fails to pr~ict rcport~d dos¢-r~spons¢ r¢lationships in active smokers. For ~eir model, the risk increases by I¢ss than a hctor 1.05 as cigarctt~ consumption rises from I0 ~o 50 ciga~cs a day. Inde¢d, R is co all intents and purposes completely invariant of consumption above 20 cigare~¢es a day. This bears no r~la¢ionship whazsoever ~o what is reported in numerous ¢pidemiological studies (USSG There appears to have b¢~n some misunderstand- Jug rcg~ding my commcnu on Wells' suggestion that variation in susceptibility to cancer might explain the unexpectedly high tung cancer risk associated with ETS exposure reported in epidemiological studies. I had n~ver indicated, as Wells implies, that I believed there was no varia- tion in individual susceptibility to cancer. Rather I found Wells' suggestion implausible because it would imply a degree of variation so large as to be inconsistent with the fact that.mathematical models involving no allowance for.susceptibility have been developed to describe!adequately,the variation in lung cancer risk in .smokers by age, dose, and duration of smoking. I also pointed out that the work oJ Khoury ctal. (1989}, which Wells cited, provided no evidence on susceptibility in me standard English meaning of the word. That a person dots not get cancer does not of itself imply absence of susceptibility to cancer. It is eminently possible for some members of a genetically homogeneous group exposed identically to get canner and for some not to get cancer. Wells" arguments concerning the purported feels related to the difference in phase distribution of semi-volatile compounds between ETS and mainstream smoke continue to be inconsistent with what is known about the chemistry and physics of ETS. His suggestion, that the effective retention og the semi-volatile compounds found in MSS particulat~ matter is 15 times higher than expected because of their appearance in the vapour phase in ET$, is simplistic and misleading. It is h'u~ that some semi-volatile compounds will evaporate as sidestream smoke is diluted by ~he surrounding air. However, one major effect of this is a greatly increased decay rate (as is seen in the behaviour of airborne nicotine) resulting in a substantially lower potential exposure. There is no evidence to suggest that polyaromatic hydrocarbons contain. ing four rings or more transfer to any extent to the vapour phase of ET$. The combination of a low exposure and a low retention would clearly out- weigh any relative increase in retention of the lower-weight semiovolatile corn.pounds; the in. consistency between the epidcmiologic and dosimetric data on ET$, therefore, remains unex- plained by Wells' arguments. Moreover, I find it impossible to believe his argument that suggests that the increase in retention or the semi-volatile compounds could account for both the disparity between the dosimctric and epidemiologic evidence and at the same time be responsible for "to an even greater extent" the potential for cancer at sites other than the lung. Repute and Lowrey, and also Wells, comment on my view that the American Cancer Society's (AC$) failure to publish results relating ETS to heart disease (or indeed to any disease except ATCo document for Mayo Clinic 27 March 02

lung cancer) has result:d in important publication bias. I retain my view and find their replies, and the continued failure o£ ~e ACS to publish, qui~ remarkable. Repace and Lowrcy claim that, ac- cording to Lawrence Ga~finkel. the ACS had never examined their da~ for h~t dis~as~ mortality from ETS.'Thi$ "s~mg su~rising sin~e som~ ~v~ y~s ago, in a visit w ~ ACS in N~w York, I was lold by Garfinkel ~at h~ had ~xmmin~d dala from their old million p~rson cancer pr~v~n~on study (CPS-I) and had found no relationship, but that ~y wer~ awaiting resul~ from lh~ir ~en ongoing I.~ million p~rson study (CPS-ID ~fo~ publishing. ~ later study finished in 1988 and KTS results ur~ still unp~bli~d. While I appr~ciat~ lhat, as Wells poln~s ou~, th~ la~c Dr. Cuyl~r Hammond saw problems in using CP$-I for estimation of risk of ~TS, thes~ problems apply ~o o~h~r published studi~$, often mor~ ~v~rely. Comp~ed with epidemlological studios on ETS, the ACS has numerous advantages not refvn~d to by Wells: it a very l~ge study wi~ dam on m~y ~ousands of heart dts~as~ deaths among nonsmokers; it or prospective d~sign;'i~ h~ ~cotd~d numerous pot~nfi~ ~nfoun~g v~ables; and it obse~es a US population. W~lls ~s to project ~hv view that studi~s in Orv~c~ ~d ~apan ~¢ more likely to produce a m~ningful r~sult in order to argu~ th~ AC$ da~a should not b~ published. From someon~ who has presented ~stimatvs of larg~ numbers of ETS-r~lat~d heart dis~ d~s in the US ~pula- fion b~ on m¢~-a~ys~ o£ da~ from ~nducted ne~ly ~1 in ~v US and ~ UK~nly on~ study b~ing in ~apan and non~ in repres~n~ a sudden shift of ground. I£ oth~r US smdi~ ~ wo~ ~nsid~g, ~S-I and CPS-II most c¢~inly are. Many nonsmokers whos~ spouses do not smoke hav~ m~asurable cotlnin~ in their urine, s~rum, or saliva; and Wells 0988) conceded his analysis relafiw ~sk b~d on $pous~ smo~ng up- ward so as to take account of ~i$ exposure of the "nonexpos~d'. In his la~sl l~tI~L h~ cites data by Cummings (1990) which finds a much smaller posed/unexposed cotinine ratio (1,55) than based on ¢~lier data (3.0), which if applied would result in a much larger upward risk ten,orion than ~fure. I fail to underhand why one should limit attention to spousal exposure data when using the epidemiology to ~sfimat¢ risk from ~TS and estimate risk from other exposures indirectly via corinth,. ~er¢ ~c by now quit~ substantial epidemiological da~a on workplac~ exposu~, and on childhood exposure, and a limi~d amount of data on other indices, which Wells ignores. As make clear elsewhere (Lee 1992) it is only for spousal exposure that an association is reported. Mats-analysis of available data for workplace posure gives a relative risk of 0.9g (95% limits 0.89-1.08); that for childhood exposure gives a relative risk of 0.98 (95% limits 0.86-I.1~). Wells believes ~hat no confounder has yet bvvn found to explain the cpidemiologically observed increase in lung cancer risk associ~od wi~ ~ousc smoking. Given lhat ~ero is ¢~dsnc¢ of bi~ du~ to misclassification of active smoking s~ms, of some publication bias, ~d of wha~ might be to ~ "poor studi~s bias"~studics wi~ sys~matic differences between cases and consols in ~c way in which thz data were collected show significantly is implauslblc that bias due to a singl~ uncon~olled confounding v~abic can explain ~ whole in- crease ~ec 1992). One should bc aw~c, however, ~at recent evldcncv ~c Ma~chand e~ al. 1991: Sidney ct al. 1989) has dcmonswated subs~ntial- ]y reduced dietary bc~a-caro~cn~ levels in non- smokers in relation to ET$ exposure and has estimated that confounding of the HTS~ung cancer r~lationship from ~is sou~ alon~ could bias the r~latiw risk by at least 1.10. Sinc~ hav~ bv~n reported between ETS-~gposed ~d non- exposed nonsmokvrs in other ~pvcm of diet and in exposur~ to occupational hazards, and sinc~ dlf- fcrenc~s in exposur~ to othcr risk factors ~or lung ~ncer may also exist (Lee 1992), it is cIe~ con- founding cannot be ignored. Wells (1990) calculated, based on data from ~S studios, that adjustment for bias duo ~ misclas- sificaflon of smokers as never smokv~s reduces epidemiologically observed relative risk of lung cancer for husband's smoking from 1.24 ~o 1.174. ~hus oxplaining 28~ of ~o increase. In my 1991 Ickier ~c~ 1991), I cstima~cd ~hat misclassifica~ion could explain 60% of the increase. Elsewhere 1992), I have reviewed available data on levels of misclassification and concluded that i~ is not reasonable to assume that, for US and European populadons, something like 5~ ot ever smokers deny smoking. Because ~esc ~cnd more to be smokers and particularly long-term and bemuse ~hosc cu~cnc smokers, who deny smok- ing. smoke fewer cigarettes than average cu~cnt smokers, the biassing effect is perhaps equivalent to about 2% of average cvcr smokers denying smoking. Based on data summarized elscwherc (Lee 199~), and adding in one reccnd~ ~poncd study (Fontham et al. 1991). I now estimate, using E ATCo document for Mayo Clinic 27 March 02

l..ctt~rs to t~e Bdltor a misclassification adjustment procedure which corrects study by study, that a 2~ mlsclassification reduces an observed relative risk of lag to 1.04 and that a 1% misclnssificafion reduces it to I.IL One reason Wells finds misclassification to be less impor- tant in US studies is that he starts with a higher uncorrected relative risk of 1.24 rather than 1.18. This is partly because he includes incorrect data from the Varel~Janerieh study, using a relative risk of 1.15 when 0.75 is more appropriate (see Lee 1992), and uses a v~eighting factor that is too low. Another reason is that Wells' prooednre implicitly assumes a bias equivalent to only about 1% of average ever smokers denying smoking. This is evident frem the similarity of the bias estimates of 1.24/1.174 = 1.056 for Wells' calculations and 1.18/1.11 = 1.063 for my 1% misclasslfication calculations. In view of the limited data available and the many uncertain- ties involved, it is impossible to be sure whether Wells' 1% or my 2% is more appropriate. However, iris clear that one should not argue, as Wells does, by his state- ment that misclassification may account for "up to 28% of the association observed in US women', that his estimates of bias are maximal corrections. They clearly are not. I agree with Wells that 1% or 2% misclassification will have very little biassing effect in Asian studies, where the reported proportion of smokers and. ob- served active smoking relative risk is rather low. As I noted in my earlier letter, "one would need much higher misclassification rates for relevant bias to arise, and though there are some reasons to believe these might exist, hard dam arc lacking. Wells cites Koo (1990) as reporting little difference between the Asian misclassification rates and such rat~ elsewhere in the 13 ~rea IARC study. Thi~ is misleading. Although detailed data area by area on this study have not yet been published, it seems clear, from the fact that it was only a study of nonsmokers, that Koo was talking about proportions of nonsmokers with high cotinine levels, and not about misclassifieation rates of smokers as nonsmokers. Equality of the propor- tions in Asian and We.stem studies would of course actually imply a substantially higher migclassification rate in Asian studies. In any case, the sample size in the ]ARC areas is not large enough to produce stable estimates. I am sure that this debate will continue, ly in the absence of critical data in a number of However, even the data currently available are con- gistent with the argument that the small association between spousal smoke exposure and lung cancer found in some ¢pidemiologic studies can bc explained by a combination of a number of sources of bias. Peter N. Lee P.N. Lee Statistics and Computing, Ltd. Sutton Surrey SM2 5DA U.K. REFERENCES Aederson, E.L. end the Carcinogen Assessment Group of the U.~. ~avirnnmenlal Protcofion Agency. P~iak Analysis 3:277-295; 1983. Dog, R.; Pete, R. The causes of caner. New York, R.Y.: O~ord Univcrsiw Press; 19~1. Cumm~/s, ~. S~=temcnt ~forc ~c Sclcn~c Advbo~ Board. Indoor Air Quality and Total Human Bxposu~c Com~tlee, U.S. En~ronmen~l Proration Agency, Wss~ngtoa. D.~.. Den. 4. 1990. Fonthtm, ~.T.H. etal. Lung cancer in nonsmoklng women: A muhtcentcr ~asc-con:rol study. Cancer ~pidcmiology markers and Ptcv~ion 1:35-43, 1991. Khou~. MJ.: ~tndcrs~ W.D4 G~nhnd, S,; Adams, MJ. On the measursment of susccptlb~ty in cpldemiololic studies. Am. J. ~pidemiol. 129:1~3-190; 1989. Khouty. M.L Ptsslvc smoking: t reply Getter]. Envirvn. 17:381; Ken, ~.; Ho, LH.-C. Worldwide cpid~iologieal pttte~s of lung cancer in nonsmokcn. Ira. L Epldemtol. 19:S14-23; 1990. ~c, P.N. An cstlmtte of sduh mo~sliW in the United States from passive smoking 0citer). ~n~ron. Int. 17:90; ~, P.N. En~mnmcntal tobacco smoke and mortality. A d~ailed ~view of cpidcmiologictl evidence ~latlng environmental tobacco s~oke to ~e risk of cancer, henn disease, and crusts of death in adults who have ncvct smoked. Basel: Ktrge~ 1992. ~ Man.and, L.L; Wilkins, ~R.; Htnkina, LM,; Hsley, DictaW patterns of female n~-smo~s wi~ and wi~out po~urc to environmental gobacco imoke. ~tnccr ~sutei ~on* trot 2:11-16, 1991. Peso, R. Spidem[olagy, muhl;tagc mo~cfs, and short-te~ metage~city tests. ~: Hht~ H.H.; Watson, LD.; W~tton, LA., cds. O~ins of human cancer, Cold Sp~ag Ha~r. ~Y: Sp~ng Harbor Ls~rato~, 1977:I~3.142~. Reptce, J.L.; Lowrey, A.H. Obse~sdoatl v* models in cst~s6n~ mo~lity f~om passive smok~t Environ. Ins. 17:386-387; 1991. Si~y, S,; Ct~, B3.; ~dedmtu, G,D. ~ett~ ~kc of carotene n~smokc~ wJtb and without passive amokjn~ az home. Am. Epldemiol. 124:1305.1309. 1919. U.S. Surgeon-GeneraL ~e h~lth consequences of smoking. ccr. s report of the Surgeon-General. Roekvillc, Md: DHHS. U.S. Public Health Se~ce; 19$2. Wells, A3. An cstimat~ of adult mo~ality in the Unhcd States from ps--ivc smoklns. ~nviton, Int. 14:249-265; Wells, A.J. Smokc~ misclassification does not tccount for observed passive smoking risk lot lung ctnccn Submission to Scientific Advi,oW Board. Ind~t Air Qutlhy and Total H~mtn ~xposurc Committee, U.S. ~nvironmcn~l Prmec~on Agency. Washingt~. D.~, Den. 4.1990. Wells. A.J. An estimtte of tduh mo~Hty ~ ~c Unhcd States from pa ssivc smoking: a rcs~nsc to c~tlcism 0cttcO. ~nviron. Int. 17:~S2-3$$: 1991. E ATCo document for Mayo Clinic 27 March 02

L,~rz to the Editor 32t l=o~them. I~.T.H. et el. Long cancer |u eenmmo~;-S women. Center 1992. ~ p~cu) and Sccondt~ tmblcnt air qua~iy standards, 40 C~ S0; July I, 19g7. Wtsblnlt~, D.~: USEPA; 19S7. L Oc*~. Mcd. 29:628-629; 1987. Turk, B.H.; Grimsrud, D.T.; Brown. ~'.T.; Gcis|ina-Sobozka, K.L.; Harrison. L; Pdll, RJ. Comm~rchl bulldog vca~lt- Zion rates lu~ psnlclc c~ccntrs~ons. ASHRAB Trtn~. ~$:422- 433: 1989. ~mcr, S.; ~f. L: Gross. A3. ~c m~surcmeet of tobacco smoke in 58S office cn~ro~cnts. Enema. Int. 18:19- 28; 1992. WHO ~orld He*l~ Orztni~don~ Rcpo~ on a WHO mecZia¢, August 21-24, I 9~4. ~d~r ~r ~ty ~cscs~. ~U~O tad Studies 103. Co~uhtgcu. D~ms~: WHO, Regi~ rice ~or Europ:; 1986. AN ESTIMATE OF ADULT MORTALITY IN THE UNITED STATES FROM PASSIVE SMOKING; A FURTHER REPLY Lee (1992) comments on my paper (Wells 1988) and on subsequent replies to his letters (Wells 1990a, 1991; Repace and Lowrey 1990,1991). Lee (1987) estimated that the upward bias from smoker misclas- sification in passive smoking studies on females is 1.24 . This estimate was relative to the worldwide relative risk data for passive smoking of 1.2 to 1.5, where most of the underlying data concerned females. Lee's estimate of 1.24 (Table 1) would decline to 1.02 i~ corrected for e~ors he has admitted in his treatment of the data, and should decline fur- ther to 1.015 if averaging of da~ on mal~ misclas- sification into the f~malc data arc discontinued in ord~ m d~ve his mi~sificadon factors. To his credit, Lee (1986, 1987) has suggested that the misclassified smokers be divided into current regular smokers, cu~ent occasional smokers, and former smokers, a system that allows more precise use of the available mi~J~si~cation da~. H~ dcvelop~ a de~Jled .mblo of marriage concordance for current and cx- smokers, and the ~rst formulae fur dealing wi~ mul- tilevel misolassifiuadon. Also, his input daze 1986, 1987) are ~onsisten~ whh data from other tigators and are used in Table I. However, his method. of nearing the data, namely 1) oonfusing nonourren~ users of tobacco with neversmokers, 2) basing m~s~lass~fiuation ra~es on neversmokers ra~er than eversmokers, and 3) using 1980"s United Kingdom active smoking risks and smoking for calculating bias ~n worldwide data, were serious eno~. These e~ors (Table I), already admitted (Lee 1990. 1991), in~ated the bias by a fautor of 24~ = 12; and, a~eptlng female-only da~a for oalculadon of ~e bias in female studies would increase his in,aden faotor to 24/1.5 or 16 ~imes the more likely value of 1.5%. Even ~is estimate of bi~ is probably high ~ells 1991). Th~ bias estimates in Table 1 are based on inputs tha~ come largely from work. The factors of 1/2 and 2/3 to discount the "non-user" misolassification ratestoneversmokcronly rates come from data derived from the codnine studies of Coulees (198g). Cummings (1990a). and Pierce (1987) on 1145 self-reported female nonsmokers where it was known how many said "never" and how many said "former". Also, the worldwide average smoking risks of 4.7, ] .6, and 1.4 are my estimates, but these can be partially derived from Lee (1988). My own work on smoker mlsclassification (Wells 1985, lPg0b, 1990c) indicates a steadily decreasing estimated bias for worldwide studies from 9% to 2% as more and better data and better methods have become available. Since my paper (Wells I988), my death estimate for lung cancer has been compared with estimates by other investigators CRepace 1990) and found to be near the middle of the range. The biological plausibility of my heart death estimate has been dis- cussed favorably and at length by Glantz and Pa~mley (1991). Staenland (I992) has made an assessment of heart deaths from passive smoking, namely 35 000 to 40 000 deaths per year and notes that it is remarkably close to my estimate of 32 000. In dealing with the U.S. studies, Lee (1992) has abandoned his earlier approach of dividing the misclas- sifted smokers into three levels. He simply says (Lee 1992) that "it is not unreasonable" ~o assume that over 5% of eversmokers deny smoking and this is "perhaps" equivaIent to 2% average eversmokers. The mlsclassiftcation factors that Stewart and I (Wells 1990c) have developed are based on 11 studies involv- ing 9431 females where either cotinine valnes or discordant answers were available, and where self- reported never- or former-smoker status was known or could be estimated. These are "community sur- vey" data, but are probably high relative to misclas- siflcatlon rates in epidemlologic studies aimed specifically at passive smoking (Wells 1991) The recent study of Fontham et el. (1991) was designed specifically to study passive smoking. It is the largest U.S. (or world) study to date with 420 cases from BATCo document for Mayo Clinic 27 March 02

L~ttcrs to t~c l~dhor Table 1. Rffccts of P.N. Let's crroncool u le of inputs in cslculstin| smoker miscIassifieatlon bias for overall world relsdw risk for lua$ cancer from passive smokin|. Lee's sd~tted errors (Lee~ 1990~ 1991) Should use Should use Should base Should do analysis ~le miscless£- never smoker mlsclass~f£- study by study, not flcat£on data, Let's m£sclasslf~ca- nat£on rates use U.K. smoker risk not on average or£g~nal tlon factors, on smokers, and smoker prevalence of male and inputs not never not on never throughout~ male data as Inputs Lee ~1987~ plus £ormer.a e~okers, the ~orld." Lee (1986~1987~ Mis~lass~fica~£on rates based on ~8.7Z never Xesuls~ smokers, ~ 1.~d l.~xl/2 - 0.70 0.70 0.70 1.Ix1/2 = 0.55 0cessions1 ~o~ers, ~ 1.1d 1.Ix2/3 " 0.7~ 0.7~ 0.76 1.1x2/3 " 0.~ Foyer ~okers, Z 10.0d 10.0 10.0 10.0 5.7 ~selass~cttLon races based on 51.3Z ~egular s~e=s, Z 0.665 0.665 0.52 Occasional smokers, Z 0.70 0.70 0.70 Foyer s~okecs, Z 9.5 9.5 bu~ based on uorld average of 30% ever smokers:~ Regular s=okers~ Z 0.285 O.2Z~ Occ~s¢on~l smokers, Z 0.30 0.30 Foyer smokers, Z ~.1 ~e= cen~ female ever smokers 51.3 51.3 51.3 30 30 ~sclassLfled s=okers' relsclve :lsks: Ze~ular smokers tO t0 10 ~.7 ~.7 Occas£onal smokers 2.5 2.5 2.5 1.6 Foyer s~kers2.0 2.0 2.0 1.4 1.4 Bias ~o ~orldvtde pass£ve ~isk: Lee's machems¢ics l.~ 1.1~ 1.0~ 1.02 Wells/Sceuar~ math~ 1.013 s The/'actors |~2 and 2/3 to convert "non-user" data to "never-user" data are based on Wells (1990~) usinS data ~rom Coulms (19~S), Cu~ngs (1990s), sad Pict~ (19S7) plus p~ond communtcs~ons flora ~hcm. ~ ~c bi~ ~an be c~ulat~ 8~dy-by-study and s~rcZst~, or the im~nant paramours can bc aggregated sepuately followed by t single ~ccdon ~lculadon to reach the asgregslcd bi~. The latter proc¢dur¢ is used here. ~ The val~s I.I, I.I, ~d 5.7 are from Lee (I986. 1987). 4 An avcr#~o of Let's m~e and fcm~c dat~ ~ ~csc v~ucs ~c obt~ncd by mul~plyin~ the ncvcrsmokcr rates by the ncvcrsmok~r[cvcrsmokcr prevalence rs~o ~c's ~c~dcs (~c I987) is based on misclassificafion rates expressed as ~ os ncvcrsmokcrs. Therefore, i~ is nccess~ to convert tb~ world-svcrszc cvcrsmokcr ~atcs by 30~0. the ~tio of cvcrsmokcr to ncvcrsmokcr prevalence. s Scc Wells (19~) for dct~ls of the method. Lc~'s ma~hcms~cs ~c~ 19S7) r~quirc the assump~on that the passive ~sk is 1.0 to calculate th~ bi~. The Wclls[Stcw~t method d~z not. E]ATCo document for Mayo Clinic 27 March 02

323 five pc.~ula'-'.:.n centers in the South and West repre- sentin~ lg.~,~ -~f the U.S. population. They used an elabor~.:¢ r=x.'.ti-level system of record examina- tion and in:~rviews to eliminate eversmokers from their ocher:: :otinine measurements were made as a final check: == the oases and controls. Of the 239 cas~ wMch ~..T.,-viv~I the interviews as never-smokers and which ~.: far have been analyzad for ¢otinine, only 2 or 0.-;% were found by cotinine level to be occasional x=okers and none as regular smokers. Table 2 sho.~'t these results along with Lee's (1987) misclassifi¢.~:ion factors for females from Table 1, his cur;eat "--eas (Lee 1992) and the Wells/Stewart "commanit.v .~,~rvey" results (Wells 1990c). It would appear tha.: ~_ determined effort by a qualified epidemiolo~.: group reduces the misclassification rate b.v int.-.-z'_,-..ws alone to about 0.3/1.6 = 1/5 of the ra;e fr=~ the community surveys and to about 0.24/1.6 = I.- :'f persons with high cotinine levels are also eli===ated. Of~.e 1.~ :.-..5. studies on passive smoking and lung cancer, on17 .-'our (Garfinkel I985; Janerich 1990; Kabat !990: ~-'~ntham 1991) were designed from the outset ".o m.-:L~-~e passive smoking effects and only those effec~z. Smoker misclassification was by then known :o be ~'~. issue, so, extra efforts were made to avoid stuck.-= classification. The weighted average relative risk ~om these four studies is about 1.21, Assuming th.~ the three studies other than Fontham have about "-~ the misclassification rate of the W¢lls/$tewa,--- community survey data or about 0,8% and keeping _==ntham at 1/7 or 0.24%, the weighted average co~.:'-.ed risk for these four studies is about 1.18 for a s.'~ker misclassification bias of 1.03 or 3%. This is a ~nuch more reasonable approximation to smoker m:.'~:lassification bias for recent wall-con. ducted U.S. s--,..udies than Lee's estimate of bias using 2% eve.'smo.~-..~ misclassification for each study. Lee sugge~. • that diet is a confounder in the lung cancer studie~ and refers to Le Marchand (I 99 I) and Sidney (195.-'b" who studied various dietary intakes among nonsm,=kers vs, tobacco gmok~ ~xpo~urei but, these icvesti_a:~tors did not simultaneously measure passive smo;'-~ug lung cancer effects. Five of the passive lung :~ncer studies (Dalager 1986, which is a rework of C.'~rrea I983: Hirayama 1989; Kalandidi 1990; Shimi:'.: 1988; Svensson 1988) also studied dietary factc~ including carotene, vegetables, fruit, vitamin C, am£ many others; and, they measured the effect of th.- .-'~et factor on the passive smoking risk. None found "_':'=: diet was a confounder of the passive smokin~ effe¢:.:. L¢ Marchand et al. (1991) also found that nor.smold=:g women married to smokers had about half the intaY~ of cholesterol and fat compared with nonsmok.:ng women married to ncversmokers. This would tend to confound the heart passive risk and make the observed risk lower than the true risk. Lee notes that recta-analysis of workplace data shows no association. Fontham (1991) found a statis- tically significant odds ratio of 1.34 (1.03-1.73) for workplace exposure with a p for trend of 0.02. This contrasts with Janerich (1990) who found a workplace risk of 0.9. The ~anerich study may be suffering from what appears to be a "latitude effect". If one aggregates the odds ratios from the "southern" U.S. studies (Correa 1983; Buffler 1984; Wu 1985; Humble 1987; Butler 1990; Fontham 1991; Stock- well 1991), one gets a combined relative risk of 1.43 with 95% confidence interval of 1.16-1.75. If one aggregates the odds ratios from the "northern" U.S. studies (Kabat 1984; Oarfinkel 1985; Brownson 1987; Kabat 1990; Ianerich 1990), one gets a combined relative risk of 1.11 or 0.99, depending on whether one uses the [anarlch risk of 1.0 that I prefer or the 0.75 that Lee prefers, Oarfinkel's study (1981) is not included since it was national in scope. Apparently, lung cancer signals were clearer from the southern studies than from the northern ones. This may reflect Cummings' (1990b) concerns mentioned in Wells (1990. In Erie County, New York, (similar to where ~'anerich's data were gathered), Cummings observed little difference between notinine levels of nonsmok- ing women whose husbands smoke (10.5 ng/mL) and those whose husbands do not smoke (6.8 ng/mL). Cummings has also observed (private communication) that cotinine levels among nonsmokers in Erie Coun- tyare higher in November to April (10.6 ng/mL) than in May to October (7.3 nglmL), reflecting probably less background ETS in the summer months. Thus, the northern summer may be more like the South is year- round. Lee hopes the American Cancer Society (ACS) will publish their passive smoking results from ACS- I and II on heart disease and ACS-II on lung cancer. I have discussed (Wells 1991) Hammond's reserva- tions about ACS-I. The ACS-II questionaires con- rained only one passive-smoking question for each sex. Thus, it is unlikely that data of the quality of Fonthsm et al. (1991) will be forthcoming from ACS-II. On the question of individual susceptibility, Harris (1978) found that human lung explants from $% of a population studied had binding energies between DNA andbenzo-a.pyren¢ that were five times the average of the other 95% of the population. The 5% were probably the potential ETS victims. Lee (1992) is now calling on a combination of biases to explain the small association between ETS ezposure and lung cancer, estimated for the U.S. by 3ATCo document for Mayo Clinic 27 March 02

Let~ra to the Editor Table 2. Female smoker miach.i~icatlon rates tram variooa sources. as per cent of ever seekers baled eo Lee (1992) based on b 75? vu~ene baaed on ? 943~'~o=en Oceas£on~l'n~okers~ % 0.7 Long cer~ exs=okece, • 5._.~ ~._.~ Total m£a¢less£f£ed, Z 6.6 $.0 Equivalent avscege ever seekers, g 1.$6e Foncha= (1991) basedc on 239 re=ale cases only tinine tese 0.6 0.0 0.0 1.3 0.2 0.0 5,6 2.~ 2.2 1,59f 0.29f 0.24f •Bascd on Let's data on females (Let 198fi, 1987) plus Wells/Stewart factors to convert from nonusers to ncvcrsmokers. t, TIcIs is what in the text is described as "community survey" da~. ~ Aeaurned to come from & cohort that is 85% cvcrsmokcrs bescd on U.S. statistics, and • smoker ~isk of 8. z Assumed to be 1/2 of the Wclls/$tewast "community sur~oy" data because ~,f more easeful multistage interviewing. • Based on a weighted average of Let's misclassified smoker excess relative risks of I0-I = 9 for regular smokers, 2.$-I = for occasional smokers, and 2-I = I for long-terra exsmokcrs vs. an assumed excess risk for aelf-repor:ed evecsrnokcrs orB-1 - 7. r Based on assumed 1985 excess retatlvc risks of 10-1 = 9 for regular smokers, 2.8-1 = 1.8 for occasional smokers, and L8-1 - 0.8 fo~ long-team cxsmokera vs. an assumed excess risk for self-reported evcrsrnokers of 8-1 = 7. him to be 1.18 before correction and by me to be 1.24 (Wells 1990c), 1.21 (recent studies, above) or 1.43 (southern studies, above). These arc passive smok- ing risks for exposure to eversmokers including many ex- and light smokers. Three of the four U.S. studies noted that were designed to test for passive smoking also report on more than one exposure level. ]:or the highest level of exposure, Garfinkel et al. (1988) found a risk of 2.11 for exposure to 20+ cigarettes per day at home by the spouse; ~ancrich et al. (1990) found 1.38 for exposure to 75+ smoker years in the household; and, Fontham ct al. (1991} found 2.06 for adenocarcinoma cases exposed to 40+ cigarettes per day by the spouse. The combined risk for these highest exposures is 1.84 (1.31 to 2.59 at 98%). Relative risks in the same range for highest exposures are also to be found in Europe (2.2 for three studies), Japan (2.0 for four studies), and Hens Kong (2.3 for two studies). Therefore, there is much less variation across regions at the highest exposures since background and the various possible biases have less effect. Also, the highest exposure data cited for the three U.S. studies are not a dcminimus fringe on the high end since they constitute almost 20% of the total statistical weight of all the U.S. studies combined. The U.S. combined result of 1.84, when corrected for smoker misclas- sification using the 1.03 factor noted above, is reduced to 1.79. That leaves a large risk to try to explain by a combination of remaining biases, none of which, so far, has been found to have an effect. In conclusion, the peer-reviewed literature pub- lished since Wells (1988) supports my positions, not Lee's, indicating that a passive smoking death toll in the S0 000 range for the U.S. is still the best estimate available. A. Judson Wells 41 Windermere Way Kennett Square, PA 19348 REFERENCES Brownson. R.C.; Rcif. iS. Keefc. T.L: Fcrzuson. S.W.; Pritzl. LA. Risk factors for adenocarclnorna of the lens. Am. J. Epidcmlol. 125: 25-34; 1987. Bufflcr. P.A.; Pickle. L.W.; Meson. TJ.; Content. C. The causes of lung cancer in Texas. In: Mizen, M. and Cortes, P.. cds. Lung cancer-" causes end prcvcmion. New York. Hy'. Vcrlas Chemic International; 1984: 83-99. Buder, T.L. The relationship of passive smoking to various heel~ outcomes among Seventh Day Adventists lo California. Abstract of paper presented at the Seventh World Confcrcnce on Tobacco and Health, Pooh, Australia, April 1990. Perth: Health Department of Western Australia; tgg0: 316. Cortes, P.; Pickle, L.W.; Fontham, E.: Ling. ¥.: Hsenszel, W. Passive ~mokins end long cancer. Lances ii: 595-597; 1983. Coultah D.B.; Howard, C.A.; Pcakc, G.T.; Skipper, BJ.; Samoa, J.M. Salivary codeine levels end ievoluettW tobacco smoke caposurc in children and adults in ~ew Mexico. Am. Rcv. Rcaplr. Dis. 136: 305-309; 1987. Cummings, K.M.; Mukclio, MeElroy, P.D.: Marshall, J.R. Messurcment of caracas exposure so environmental tobacco smoke. Arch. Environ. Health 45: 74-79; 1990e. Cummings, K.M. Statement before $ciend6c AdvitoP/ Board. Indoor Air Quality and Total Human Exposure Cornmiucc. U.S. Environmental Prntecden Agency, Wasklngten. D.C, Dee. 4,1990b. Dslager. N.A. ct el. The relation of passive smoking to lung cancer. Cancer Rcs. 46: 4808-4811; 1986. E ATCo document for Mayo Clinic 27 March 02

32~ Fnntham. E.T.H. ct aL Lung cancer in uonsmn]cing wcmcn: A mnlti¢¢mnr ease-control study. Cancer ~pidemlol. Biomatker~ P~ev. I: 3~-43; 1991. Gt~nkel. ~ Time trends ~ lung concur mo~di~y among non- #mokc~ and a note on passive smokes. 3. Nat. ~nccr Inst. ~: Gtd~kel, ~; Auerbach, O.; ~oubcn, L. Involun~ smoking and 463~69; 198~. Gltn~, S.A.; Pokey, W.W. Passive smokes and hea~ disctse: c~dcm[ololy~ phy~ololy, ~d bi~h~is~. ~reuh~on 83: 1-12; 1~1. Ha~s, C.: An~p, H.; Stoner, G.; T~p. B. ~oz~esls studics ~ h~ ~into~ epigram: An cz~6men~l m~cl rystc~ ~: Ha~s, C, ed. Pt~ogcncsis and ~cnpy of lung cancer. ~cw York, HY: Mar~el D~kec 197~: ~83. Hirtyama, T. Dinn~ habits s~ of l~hed im~nsnco in ~ucnc- ~g ~e l~g cancer ~sk among Itpancac ~slcs who never smok~ ~: Biers. C3.; Cou~ois, Y.; Govscm, M.. cds. Present ~umblc, ~G.; S~et, LM.; Pt~, U,R, Ma~sge to a smoker and lung concur Ezk. A~ L ~b. Heal~ ~: ~98-~2; 1987. Jtnedch, D.T. ct d, Lung cancer ted cx~sarc to tobacco smoke ~ the household, ~cw Enghnd ~. ~=d. ~: 632-636; 1990. Xtbt=, G.~.; Wyndcg. ~.~ ~ng esnccrln nonsmokers. Cancer ~3: 1214-1~1; 19~. Ktbtt, G.C, ~pldemiologic undies of ~e ~lstloship between o~ Fo~. Was~l~, ~: To~inly Fo~ 19~ l~-~t. Kalsndld~ A.: gtt~ouyan~, K.; Vovo~ulou, ~,; Bassos, O.; Stracci, K,; Tdchopoulot, D. Passive smokes sod ~et ~ the etiology of lunl cancer •mona nonsmokers. Cancer causes cont~oi 1: 15.21; 1990. ~c, P.~. Does b~st~S ~cr people's tobacco smoke cause lung cancer? Bc Mud. ;. 293: I~03-I$04; 1~86. ~e, P,S. Pssslvc smokes ~d lung cancer association: * resuh of bttt7 Hum~ To,coL 6: ~17-524; i~$7. ~e, P.N. ~:clsssifica~ of smo~nl habits and paesivc smok- ing: a r~cw of cvid~co. Intonational Archives of Oecupa- fi~ and ~tenmentsl Hc~. Berlin: Sp~gcr Voting: 1988. ~e, P~. ~ents on ~a~al submitted to EPA by A. Sudson Wells, U.S. Enviro~en~! Protcc~on Alcn~, Human H~th Asscssm~t Group, Was~gton, D~, N~. 8, 1990. ~e, P.S. An estate ef sd~t mo~a~ty ~ ~e Unltcd S~tcs from passive smokes Gc:mr}. Environ. ~t. 17: 379.381; 1991. Lee. P.N. An estate of sduh mortality in ~c Unltcd States from ~stive ~g: ~cr c~enc ~r~ ~L 18: 315-317; 19~2. Le M•rchando L.L.: Willdns. L.R.; H•nkin, J.H.; Haley.N3. Dietary patterns Of female nonsmokers with sad wlthous ezposure to envlmnmcutal tobacco smoke.. Cenccr Causes Control 2: ! 1-16: 1991. Picrceo LP. et al. Codninn valid•ainu of self.rcportcd smokin| in ccmmercitlly rnn communhy surveys. I. Chron. Dis. 40: 689- 695: 1987. Repacc, J.L.; Lowrcy, A.H. Risk assessment mcthodologics in passive smoking. Risk Anal 10: 27-37I 1990. Repsce, LL.; Lowrey, A.H. A febnt~.tl to Lee[/~atzenstcin corn- rectory on passive smoking risks 0.c~r). Environ. SoL 16: 183-184-" 1990. Shimizu. H. ct aS. A case control study of lung cancer in uoasmok- ins women. Tohoku ~. ~xpcr. Mcd. 154: 359-397; 1988. Sidnny, S; CaSh, B3.; Friedman, G.D. Dle,,ry intake of carotene in nonsmokers wi~ and without passive smoking at home. Am. L Epldemiol, 129: 1303-1309; 1989. Siceniando K, Passive smoking end heart disease. ]'AMA 267: 94-99; 1992. Stockwe]l, H.G.; Candclota, E.C.; Armstrong, A.W.; Pinkhsm, P.A. Environmental tobacco smoke in never unoking women. Am. 3. ~-pldcmloL 134: 724-725; 1991. Svcntson, D, Lung cancer etiology in women (dissertation). Stock- holm: Karolintka Institute; 19g8, Wells, A,J. Misclassifiuadon as • factor in passive smoking rhk (letmr). Lancet ii: ~qS; 1986. Wells, AJ. An estimate of adult monalhy in the United States from passive smoking. Environ. Int. 14: 249-265: 1988. Wells, AJ. An estimate of adult mnrtalhy in the United Sasses from passive smoking: a response to criticism. Environ. Int. 16: I f17-193: 1990a. Wells, A,L; Slower:, W.P. Is smoker misclasrification really en impotent factor in passive smoking risk7 In: Dorston, Jsmro.--~k, K.. cds, Tobacco and hcahb 1990, the global war. Perth: Health Department of Western Australia; 1990b: 220- 522. Wells. A.J. Smokcr misclassificetlon does not account for served passive smoking risk for lung cancer. Submission to Seienti~c Advisory Board, Indoor Air Qualhy and Total Human Exposure Cmnmluce, U.S. ~nvlronmen~1 Prolccdon Agency, Washington. DC, Dec. 4. 1990e- We|Is A-L An cst~natc of adult morttlityin the United Stales from passive tanking; a response so cri~cism (letter). Hnvlron. Inc 17: 382o385; 1991. Wu, A.H.; Henderson, B.I~.-" Pike. M.C.; Yn, M.C. Smoking and other risk factors for lung eencer in women. J. ~et. Cenccr In st, 74: 747-75h 1985. E ATCo document for Mayo Clinic 27 March 02