Tobacco Documents Online

530 =son, :e:a 0.: Management of the _ _n_ Herman Klein _, 57r1=;`te1d, 1973. Chap. 10, _.. __ ==__tcrc Trc n?na tin the ?ostdoctarai . ; ____ =-s.r =a;ort of the Advisory :f _ost'_:ccoral Fellovships _ioae_icsl Sccences, ::ational Academy sear:-- _:__ci:, _.73. =.i :;as-=. 7`e =f:ect of total- ::e:i. =-sx-_ca = ribits. In The =-.- -Iin S_=tzeeis, Proceedings of the ?racY. aug•.'st 10 - 13, 1970. _- ie^3... -11 ve z-.:3 'L;rlova. 1971. e nte_-s a_. i rsei=. 'rs I.. sf--ition and classification ---~ i.d ~3ce~cc. -e=-.= Z:ein (ed. ) . of -=.e _olycythenias. '-`-e^_-s :er-:. tlein (ed. ) . .-. Ive:view. t = --ve. Pers?r_-ves _z _zalogy and Medicine __ _ax >sss. =Zi;ago, Il. (in cress) 531 STATEMENT OF LAWRENCE L. KUPPER, Ph.D. 1. Introduction I am Lawrence L. Kupper, Ph.D., a biostatistician specializing in epidemiology and environmental health, currently Professor of Biostatistics in the School of Public Health, University of North Carolina at Chapel Hill. I have been a consultant to the U.S. Environmental Protection Agency and a member of the Instructional Staff of that Agency. I am currently doing research under a grant from the National Institute of Environmental Health Sciences. Since 1967, I have been engaged in biostatistical research with particular reference to the analysis of epide- miological data on occupational health and cancer. I arn an elected member of the Regional Committee of the Eastern North American Region of the Biometric Society and am a member of the American Statistical Association and the Society for Epidemiologic Research. I have published or presented approximately 100 scientific papers and am on the Editorial Board of the Journal of Chronic Diseases. My curriculum vitae and list of publications are attached.

530 -16- Leon Lcon 0.: Hanagemenc of the _ Mananement. Herman Klein :ringfield, 1973. Chap. 10, 'ostdoctoral Training in the :Zuaticn of P17C,NS PostdoctoraZ 'rograma. Report of the Advisory CA1S Postdoctoral Fellovships :edical Sciences, National Acaden,v ~.h Council, 1973. :aston. E.O.: The Effect of total- n production in rabbits. In The "aemoglobin Synthesis, Proceedings of the eticum, Prague, August 10 - 13, 1970. ds.), Universita Karlova, 1971. and research. Leon 0.: Definition and classification zd Management. Herman Klein (ed.). :1., 1973, pp. 12-24. J.: Management of the polycytnenias. and :4anagement. Herman Klein (ed. ) , 1., 1973, pp. 209-237. _ernal Medicine: An Overview. . 1975. ~e. Perspectives in Biology and Medicine I of Chicago Press, Chicago, 11. (in press) 531 STATEMENT OF LAWRENCE L. KUPPER, Ph.D. 1. Introduction I am Lawrence L. Kupper, Ph.D., a biostatistician specializing in epidemiology and environmental health, currently Professor of Biostatistics in the School of Public Health, University of North Carolina at Chapel Hill. I have been a consultant to the U.S. Environmental Protection Agency and a member of the Instructional Staff of that Agency. I am currently doing research under a grant from the National Institute of Environmental Health Sciences. Since 1967, I have been engaged in biostatistical research with particular reference to the analysis of epide- miological data on occupational health and cancer. I am an elected member of the Regional Committee of the Eastern North 'American Region of the Biometric Society and am a member of the American Statistical Association and the Society for Epidemiologic Research. I have published or presented approximately 100 scientific papers and am on the Editorial Board of the Journal of Chronic Diseases. My curriculum vitae and list of publications are attached.

532 My appearance at this hearing is voluntary and the opinions expressed are personal, not representing those of any organization. 2. Discussion The widely-held belief that smoking "causes" lung cancer stems mainly from associations encountered in a wide variety of epidemiologic studies on human populations. Accordingly, it is important to discuss some of the limita- tions inherent in the use of epidemiologic research studies to assess the purported smoking-lung cancer relationship. First, epidemiologic studies are generally observa- tional rather than experimental. Epidemiologists are not able to conduct controlled experiments in laboratory-type settings where environmental conditions can be strictly monitored. Epidemiologists generally have no control over what their experimental units (i.e., human subjects) may be exposed to over a lifetime. A human being is exposed to a myriad of environmental impingements over a lifetime; these impingements take many forms (e.g., air pollution, diet, stress, exposure to occupational hazards, etc.), and they vary both in intensity and in the sequence in which they -2- occur. Furthermore, th such agents may act, ei to produce an adverse h If we want to comp smokers and non-smokers development over time o how can we best deal wi are different with regar impingements and so fo we knew that these two susceptible to the dise is introduced, regardle respect to all other en try to achieve such a d study would be to rando: two groups. This proce attempt to insure compa two groups with respect which vary from individ factors for the disease is by no means a 100: be achieved, it is cert the distribution of suc: exposures, genetic diff logical factors, etc.)

3mokiag "causes"_lung 3 encountered in a wide are generally observa-, )idemiologists-,are,.not a ia laboratory-type , )ns can be_strictly ,over,a;,lifetime;.these . air pollution, diet,_, :rds, etc.), and they uence in which they 533 occur. Furthermore," the biological mechanisms by which such agents may act, either separately or in combination, -~n ~ to produce an adverse health effect are generally unknown. If we want to 'compare two groups of individuals (e smokers and non-smokers) with regard to their rates of development over time of some disease (e.g., lung cancer), how can we best'deal with the fact that study populations ire different with regard to genetic background, environmental impingements and so forth? We would be fairly content if we knew that these two groups were, on the average, equally susceptible to the disease before the suspected causal agent . is introduced, regardless of previous exposure histories with respect to all other environmental impingements. One way to try to achieve such a desirable starting place for a follow-up study would be to randomly assign individuals to one of the V two groups. This process of randomization represents an attempt to insure comparability, on the average, between the two groups with respect to both measured and unmeasured factors which vary from individual to individual and which may be risk factors for the disease under study. Although randomization is by no means a 100% guarantee that exact comparability will be achieved, it is certainly one reasonable way to balance the distribution of such factors (e.g., previous unknown exposures, genetic differences, diet, psychological and socio- logical factors, etc.) between the two groups. -3-

534 Unfortunately, the technique of.randomization is not available to the epidemiologist in most situations. For example, people who smoke have chosen to smoke, and people who do not smoke have chosen not to smoke. Thus, any epidemiologic study comparing smokers and non-smokers suffers from the basic flaw that the individuals have self- selected themselves into these two distinct groups. Thus, any observed differences between two such groups with respect to health outcomes may not, in fact, be the result of smoking, but may instead be due to other more basic factors (e.g., those of a genetic origin) which are neces- sarily different between the two groups because of the self- selection process itself. In other words, one can reasonably maintain the view that smokers are constitutionally different from non-smokers, and that such constitutional factors cause such individuals both to smoke and to develop lung cancer. Under this plausible hypothesis, then, smoking is an outcome variable just like lung cancer, being a manifestation of the constitutional factors uniquely possessed by people who choose to smoke. ~ 'r Actually, ther smokers are, indeed, a of people from non_smc many studies have shov from non-smokers with and genetic markers. of monozygotic twins , have much more simila: dizygotic twins;'thus tendency to smoke. With regard t differences'among`3nd disease susceptibilit demonstrated.by'sophi existence (for U.S., data) of a particulai individuals born aroi which appears to`be r than do other biirth time. The existence that genetic; prenat determinants of 'subs ad '-... ._ns.:., ..

ot . to ,smoke.. 4Thus, - :ters and non-smokers individuals have self- stinct groups. Thus, 3uch groups with +~ i fact, be the result other more basic ;in) which are neces- )s because of the self- ..._~_. , _ _ . . - )rds, one can reasonably ~ istitutionally different, :tutional factors cause develop lung cancer. , smoking is an outcome a manifestation of the ssed b1y people who choose -;,i. _....._ - , -- ' 535 Actually, there is considerable evidence that smokers are, indeed, a constitutionally different group of people from non-smokers (Burch, 1978). For example, many studies have shown that smokers differ, on the average, b from non-smokers with respect to morphology, personality, , and genetic markers. Studies of twins have shown that pairs of monozygotic twins (whether brought up together or apart) have much more similar smoking habits than do Pairs of like - dizygotic twins, thus suggesting a genetic link with the tendency to smoke. With regard to the relationship between genetic differences among individuals and differing levels of disease susceptibility, Janis and Kupper (1982) have recently demonstrated,by sophisticated statistical methodology, the existence (for U.S., and also for British, male lung cancer _ data) of a particular birth cohort (i.e., that group of individuals born around the end of the nineteenth century) which appears to be much more prone to develop lung cancer than do other birth cohorts born either before or after that time. The existence of such high-risk birth cohorts suggests that genetic, prenatal, and neonatal factors can be important determinants of subsequent disease susceptibility. ' -5- sex

+ 8 . ~ '. F:. 4. :536 Based on all these findings, the self-selection bias problem is a very real one, which casts serious doubt on the ability of epidemiologic research studies to provide final answers regarding the causal relationship between smoking and lung cancer. .:-e-,y , -- . _ _ z;.X - - -In addition, there are several other sources of bias .that typically affect the validity of epidemiologic studies designed to address the smoking-lung cancer causality issue. These are: l.. Various studies (e.g., several of those referred to in the latest Surgeon General reports) involve particular sub-groups of individuals,.which are in no way representative of the general population and which are,excellent illustra-, _ tions of the self-selection syndrome.. Examples of such studies are the British male doctor study of Doll and Hill (1964), and the studies of Seventh Day Adventists (e.g., see Lemon and Walden, 1966) and Mormons (see Enstrom, 1975). It is well known that doctors differ from the average person in very many respects; and, members of religious sects which prohibit the use of alcohol and tobacco generally are much different personality-wise from the general population. Clearly, fipdings regarding such unique sub-groups cannot be applied to the general population. -6- : 2.'It lung cancer is fact, the diagn error rates hav changes can'und smoking and lun studies. ,f3.~Am < bias, has been < specifically-ivi- . versy. General= knowledge'of exT the objectivity knows a person : for lungcaIIcer smoker or avnon- -obviously tend t tion, as has bee real-life data. 4. -'It i accuracy associa on questionnaire

ral of those referred a).involve,_parttcular ~{b no way representative,_ excellent.illustra-.,,.-._5J 3xamp1es of such _ _ ,1't:).a 2 ~ X9 sE; iy_ of Doll and Hill ldventists (e.g., see : finstrom,- 1975 ) . It. :he average.person in ,„ _gious sects which y^1; generallK are much ~ral population. Clearly, )s cannot be applied •-- 2. It is well documented that the diagnosis of -lung cancer is a difficult clinical problem and that, in fact, the diagnostic criteria and associated diagnostic error rates have changed considerably over the years. Such changes can undermine the validity of associations between smoking and lung cancer reported in various epidemiologic studies. 3. A more subtle form of bias, known as detection bias, has been discussed by Feinstein and Wells (1974), specifically with regard to the smoking-lung cancer contro- versy. Generally speaking, such a bias can arise when the knowledge of exposure or non-exposure to an agent alters the objectivity of the scientist. For example, an M.D. who knows a person is a heavy smoker may be more on the lookout for lung cancer in that person than he would be for a light - smoker or a non-smoker. This form of detection bias would obviously tend to exaggerate the smoking-lung cancer associa- tion, as has been demonstrated by Feinstein and Wells with real-life data. 4. It is well-known that there is considerable in- accuracy associated with the use of exposure information based on questionnaires and interviews with next-of-kin to assess -7-

538. habits of deceased relatives with regard to smoking or drink- ing, even though this is often the only source of such informa- tion in many epidemiologic studies. _ t_ .~^ss'J _ -5. With regard to the control of confounding factors when studying the smoking-lung cancer controversy, probably the single most common and important factor is age. And, unfortunately, the effect of age on cancer causation processes is little understood. Eminentt cancer researchers believe that the aging process, in and of itself, is causally associated with the development of cancer (e.g., see Blumental, 1978). 6. There are two main types of epidemiologic studies: the follow-up study and the case-control sthdy. Such studies have reported an association between smoking and lung cancer. However, both kinds of studies can seriously suffer from the presence of biases. The case-control study is probably the epidemiologic study design most susceptible to bias; in fact, an entire issue of the Journal of Chronic Diseases (1979, Volume 32, No. 1/2) is devoted exclusively to discussing the biases and problems associated with case-control study research. The follow-up study, in addition to the general list of biases already discussed, is subject to a special form of selection bias, namely, loss-to-follow-up; in particular, we are refer- -8- 0 ring to indiv: study at the e during the sut death from otk etc. Such lo; up studies (e. lung cancer i; from such stuc seriousness oi examples.. The be can be questic attendant witt cancer relatic Statis cancer, as re; Surgeon Gener~ a causal relat designed animz meaning of the 95-077 0-82-3

539 to smoking or drink- source of such informa- 4 confounding factors )ntroversy, probably , aor is age. And, :er causation processes ~searchers believe that causally associated :e Slumental, 1978). epidemiologic studies: 1 study. Such studies oking and lung cancer. ously suffer from the tudys is probably the ible to bias; in fact, ,ic,Diseases (1979, -ely to discussing the ;e-control study research. ) general list of biases :ial form of selection -ticular, we are refer- a ring to individuals who, although initially enrolled in the study at the start of the follow-up study period, are lost during the subsequent follow-up period because of migration, death from other causes, refusal to continue participation, etc. Such losses are characteristic of all long-term follow- up studies (e.g., like those designed to study the smoking- lung cancer issue), and can seriously bias the conclusions from such studies. Greenland (1977) has illustrated the seriousness of this problem with some well-chosen numerical examples. . Conclusions The belief jthat smoking is a cause of lung cancer can be questioned in light of the documented sources of bias attendant with epidemiologic studies of the smoking-lung cancer relationship. Statistical associations between smoking and lung cancer, as reported in the various studies described in the Surgeon General reports, have been interpreted to mean that a causal relationship does exist. In the absence of well- designed animal and laboratory studies elucidating the meaning of these reported associations, such a quantum -9- 0 w CM 0 00 ~ ~ w 95-077 0-82-35