Tobacco Documents Online

~alth :view 477 UNIVERSITY of PE~Y~YSYLPdNI.A rFm.AnELPFUa 19104 rhe Wharton School CC (:13) 243-8r2 CerUraxr ar SrAMrao Statement_of Richard J. Hickey, Ph.D. I am Richard J. Hickey, a senior research investigator in the Department of Statistics of The Wharton School, University of Pennsylvania, Philadelphia. I received a Ph.D. degree in biophysical chemistry and microbiology at Iowa State University, Ames, *_owa, in 1941. Thereafter, I performed research and develapment work on antibiotics, vitamins and other biochemicals. In the mid- -1960's, I became involved in research an the air pollution/public health prob- lem at the University of Pennsylvania. Since 1973, i have been investiqatinq the relationships of environmental chemicals and other variables to chronic dis- ease risk at the University's Wharton School. Many of the journal articles and book chapters published, of which I was author or co-author, pertain to anviron- mental pollution. " . The following comments concern both S.A. 4957 and S. 1929. A basic crit- icism of both proposed acts is the Implicit assumption that "findings" (1), (2), (3), and (4), which are used to justify both the labeling statements and the proposed actions, are known and proven. However, there are serious scientific questions about the scientific bases for the claims that cigarette smokinq causes the various conditions associated statistically with smokinq, such as lung can- cer, heart disease, and low birth weight. Documentary scientific evidence establishing causality is not given in the praposed bills. However, it is clear from the wording of tk:e "findi.zqs" that they rely heavily upon the alleged causality evidence presented in some publica-

478 tions and reviewed in recent U.S. Surgeon Generals' reports on smoking and health. Unfortunately, these reports do not represent an even-handed review of the available evidence. . There are several reasons for rejecting the conclu- sions of these reports, including: (A) Selective Reporting or Selective Ecclusion. Selective reporting is present in the Surgeon Generals' reports. Sources that conform to the prr -determined smoking-causality position of the reports are cited extensively, while studies that challenge or discredit this position are often selectively excluded. Warniags of the misuse of statistics have also been excluded. The Surgeon Generals' reports have the characteristics of briefs for the "prosecu- tion" onlyr the "defense" has been largely omitted even though these reports should have been unbiased. (9) Uses and xi.suses of Statistics. The Surgeon General's report of 1964 asserts (p. 182): "Statistical methods cannot establish proof of a causal rela- tionship in an association. The causal significance of an association is a matter of judgment which qoes beyond any statement of statistical probability." These two statements lie at the root of the probl®. The first sentence is a correct stat®ent; causality cannot be established by statistical associations. This is a standard caveat that is taught in elementary statistics courses. The second stat®ent is false. Subjective judgment or personal opinion cannot be used to interpret the causal implications of a statistical association, no matter how strong and no matter ahat level of significance. Judqment and opinion do not constitute substantive evidence. It should be recognized that statistical associations are mathematical quantities. When properly obtained, they may su est possible causal relation- ships, but they can never prove such relationships. Human conditions such as cancer, heart disease, spontaneous abortions, stillbirths and birthweight de- -2- 479 ficiencies are all basically biological and b: ical or statistical problems. When ill, one statistician. Properly used, statistical methods are cz to the extent that it is incompatible with oba can never establish that an hypothesis is cart aisunderstood. However, recantSurgeon Gener inferred causality from statistical associatic causes lung cancer if the two variables are pc tical sense. Statistical associations may be due to ck some third factor. For example, it was reporc sic,nificant correlation in the United States : ratio and the population to bed ratio and the cancer of the lung." [R.B. Rigdon and H. Eirc of the :'heories Relative to the Etiology and : aent. Biol. Med. 10, 76-91 (1952) (p. 88)]. . this correlation that physicians "cause' lung It is known that cigarette smokers tend positive correlation between measures of ciqa: some populations. is it to be concluded, the_ coffee drinking, or that coffee drinking "cau. Another fallacy is the presumption by so•. 5iased. This belief is especially hazardous ;eneous populations such as of humans. For a self-selected samples such as those of heavy and ex-smokers are likely to be biased. This -3- r

479 a- 3- )6J ela- ficiencies are all basically biological and biochemical problems, not mathemat- ical or statistical problems. When ill, one visits his physician, not his statistician. Properly used, statistical methods are capable of rejecting an hypothesis to the extent that it is incompatible with observed data, but statistics alone can never establish that an hypothesis is certainly true. This fact is widely misunderstood. However, receat Surqeon Generals' reports have rather routi.^.ely inferred causality from statistical association, e.g., that cigarette smoking causes lung cancer if the two variables are positively associated in the statis- tical sense. Statistical associations may be due to chance or to dual influences of some third factor. Fbr ezample, it was reported in 1952 that "there is ... a siqnificant correlation in the Onited States between the population to physician ratio and the population to bed ratio and the number of reported deaths from cancer of the lunq." [RH. Riqdon and H. Sirehoff, 'A Consideration of Some of the Theories Relative to the Btioloqy and Incidence of Lung Cancer," Texas s Rept. Hiol. Med. LO, 76-91 (1952) (p. 88)]. Surely no one would conclude from na. this correlation that physicians "cause" lung cancer. The It is known that cigarette smokers tend to be coffee drinkerst there is a :e t positive correlation between measures of cigarette and coffee consumption in at ar :o some populations. Is it to be concluded, therefore, that smoking "causes" :on- 3- eoffee drinking, or that coffee drinking "causes" smoking? .lnother fallacy is the praaumption by some that non-random samples are not biased. This belief is especially hazardous when working with genetically hetero- 9eneous populations such as of humans. For example, some studies that employ self-selected samples such as those of heavy smokers, light smokers, non-smokers, and ez-smokers are likely to be biased. This point was addressed by T.D. Sterlir.q -3-

480 ['Zhe Statistician vis-a-vis Issues of Public Health,' American Statistician L7, 212-216 (1973) (pp. 212-213)]: 'The manner of selecting populations to stady the effects of sswking has been criticized by some of the world's most respectd statisticians but to little avail (8erkson, 1958, 1960; Fisher, 1957, 19581 Mainland, 19561 just to mention a few). of concern to statisticians has been the major reliance on 'volunteers' in most of these studies and the conse- quent saltiple sources of serious bias. ordinarily, results from a study such as Doll's in which over 30 per cent of the queried population did not participate would be considered, if at all, with greatest suspicion, especially if, as is freely adsitted, the responders were quite different from the reference popula- tion. (Doll, 1964). Siailarly, the data collected by volunteers of the American Cancer Soci.ty and used to link smoking to a variety of diseases stem from a study population (alao of volunteer families) that dramatically differs in most significant comparable measures fron the U.S. population (Sterliag, 1971a, 1972a)." (C) The Genetic/Constitutional .llternative. In the late 1950's, Sir l0onald Fisher, a former President of the Royal Statistical Society, suggested the following alternative to the smoking-causality hypothesis: "Cigarette smok- ing and lung cancer, though not mutually causative, are both influenced by a co®on cause, in this case the individual genotype." The problem of resolving the conflict between the smoking-causality hypothesis and the genetic/constitu- tional hypothesis was recently reviewed in depth by P.R.J. Surch ["Smoki.-:g and Lung Cancer: The Problem of Inferring Cause (With Discussion," Journal of the Roval Statistical Society, Ser. A (General), Vol. 141, Part 4, pp. 437-477 (1978)]. Hurch concluded (p. 476): "The discussion has allowed me.to'under mine one causal model and it has revealed divergent opinions that might surprise those nurtured on the reports of the Royal College of Physicians and the Surgeon -4- 481 General of the United States. *_ am also happy to rec tional theory, proposed by a fcrmer President of this Fisher, is alive and well." . To thi.s date the genetic/constitutional hypothee it has, however, been denigrated subjectively and wic fact that Burch's paper appeared in a major statistic cited in any Surgeon General's report. Further, the has been "criticized' by some who would seem to equat Some criticism has been of the "do not like" variety. Unless the genetic/constitutional alternative h* valid scientific grounds, it would be risky to base : smoking-causality hypothesis. If more anti-smoking legislation is passed and lished that much habitual smoking is in fact symptom trait that both influences smoking behavior and cont of developing those diseases associated statisticall government would be placed in a totally untenable go R.A. Fisher, considered by some to be the father of correct all along is very real and cannot be ignored (D) Air Pollution and Chronic Disease. In add lications such as that of Hurch, the Surgeon General sively excluded from consideration published epidemi plex relationships between a number of ooaon urban aortality rates for several categories of cancer, hE of death in U.S. cities. Some reports showed that concentrations of sev. -5-

481 aost , 1957, ns has conse- ysuch rticipste - as is popnla- dserican a a in .ost , a, 1972a)." S.r .ated te ssdc- 5y a 3lvinq sstitu- 3q and of the 77 ndsr- surprise Sur?eon General of the United States. : am also happy to record that the constitu- tional theory, proposed by a fcrmer President of this Society, the late R.A. Fisher, is alive and well." . To this date the qenetic/constitutional hypothesis has not been discredited. It has, however, been denigrated subjectively and widely ignored. Despite the fact that Eurch's paper appeared in a major statistical journal, it bas not been cited in any blLrgeon General's report. %rther, the constitutional hypothesis has been "criticized• by some who would seem to equate criticism with rejection. some criticism has been of the 'do not like" variety. Qnless the genetic/constitutional alternative hypothesis is rejected on valid scientific grounds, it would be risky to base public policy upon the smokinq-causality hypothesis. If more anti-smoking leqislation is passed and if eventually it is estab- lished that much habitual smoking is in fact symptomatic of an underlying genetic trsit that both influences smoking behavior and contributes to increased risks of developing those diseases associated statistically with smoking, then the O.S. qovenment would be placed in a totally untenable posi ion. The prospect that d.a. Fisher, considered by some to be the father of modern statistics, has been correct all along is very real and cannot be ignored. (D) Air Pollution and Chronic Disease. In addition to the omission of pub- lications such as that of Burch, the Surgeon Generals' reports have also exten- sively excluded from consideration published epidemioloqical reports on the com- plex relationships between a number of cosmon urban air pollutant chemicals and mortality rates for several categories of cancer, heart disease, and other causes of death in O.S. cities. Some reports showed that concentrations of several ccaon air pollutants -5- ..

482 auch as sulfur dioxide, sulfate, nitrogen dioxide, and trace metals in the air of 38 cities were significantly associated statistically with lung cancer mort- ality rates for these cities. In fact, over 70% of the variance iz lur.a cancer mortality rate was explained statistically by air pollutant characteristics of these cities. Further, over 50% of the variances in mortality rates for several other cancer classifications were also explained statistically by air pollution data. Also, over 50% of the variances in mortality rates for certain heart dis- ease classifications was explained statistically by air pollutant concentration data for these cities. Significant relationships were also found between air pollutant concentration data and mortality rates for infants under one year of aqe and mortality rates for congenital malformations. rsee, for exampla, R.J. Hickey et al., "Fcoloqical Statistical Studies Concerning Environmental Pol-'u- tion and Chronic Disease," IEEE Transactions on Geoscience Electronics, Vol. ~, 186-202 (1970)r R.J. Hickey, "Air Pollution," pp. 189-212, in: Enviroffient: Resources, Pollution s Society (W.P1. Murdoch, Ed.), Sinauer 1lssociates,-Stamford, Conn., 19717 R. Mendelsohn and G. Orcutt, "An Ekapirical Analysis of Air Poliut'_on DosrRespons. Curves," Journal of Enviror.mental Economics and Management, 4.1. ~, 85-106 (1979)]. (E) Comments on Four "Findinqs" of H.R. 4957 and S. 1929. The following coam<ents refer to "findinqs" (1), (2), (3), and (4) that appear on page 2 of both bills. The greatest detail is offered regarding finding (4) which pertains to pregnancy, prematurity, low birth weight, and related problems of infant health. Finding (1). H.R. 4957 states that "cigarette smckinq is the largest pra- ventable cause of illaess and pr®ature death in the United States and is assoc- iated with the unnecessa=y deaths of over three hundred thousand Americans ann- ually." This statement pertains in large part to deaths from cancer and from 483 cardiovascular diseases. However, it says noth'_ ;7oothesis nor about epidemioloqical and other e of air pollution to chronic disease. Findinq (2). S. 1929 states that "smokinq cancer" in the United States, and "is associatec c:aim has been made in various ways by the Surge as stated previously, this claim is defective be :,e reporting, misuse of statistics, ignores evi cor.stitutional influences, and fails to consider strong statistical relationships between air pol i:.d cancer mortality rates. It is also a fact t _::nq cancer has never been induced in experiment st~dies by fresh tobacco smoke when experimental :ataral exposure. As mentioned above, deficiencies in the the -t_yer cancers were examined at length by Burch i ;"A aeview of 'Smoking and Health'," Journal of !;tion 60, , 722-739 (1965)] examined the 1964 Sur ard aealth, and found it defective in biochemist -ortrf that reports claiming that the smokinq-mc appear largely in medical and biomedical journal ;ournals. In contrast, reports in statistical j t:e inference of causality from statistical asso _."at Brownlee's review was aot mentioned in seve reports. Finding (3). 5. 1929 states that "heart di Of the deaths in the United States and one-thir2 -6- -7- .

483 cardiovascular diseases. However, it says nothing about the constitutior.al hyoothesis nor about epidemioloqical and other evidence regarding the relation of air pollution to chronic disease. Finding (2). S. 1929 states that "smoking is the primary cause of lung cancer" in the United States, and "is associated with other cancers." This claim has been made in various ways by the Surgeon Generals' reports. However, as stated previously, this claim is defective because it involves in part select- ive reporting, misuse of statistics, ignores evidence regarding the genetic/ constitutional influences, and fails to consider published evidence regarding strong statistical relationships between air pollutant ch-i.cal concentrations and cancer mortality rates. It is also a fact that, despite many attempts, lung cancer has never been induced in experimental animals in well-designed studies by fresh tobacco smoke when experimental conditions resembled those of natural exposure. As mentioned above, deficiencies in the theory that smolcinq causes iung and other cancers were examined at length by Burch in 1978. In 1965, E.A. Brownlee ["A Review of 'Smoking and Health'," Journal of the Bmerican Statistical Assoc- iation 6yo, 722-739 (1965)] examined the 1964 Surgeon General's report, Smoking and Health, and found it defective in biochemistry and statistics. It is note- worthy that reports claiming that the smokinq-®ortality associations are causal appear largely in medical and biomedical journals rather than in statistical journals. In contrast, reports in statistical journals do not accept as valid the inference of causality from statistical association. It is also noteworthy that Brownlee's review was not mentioned in several subsequent Surgeon Generals' reports. Findinq (3). S. 1929 states that "heart disease accounts for nearly one-half of the deaths in the United States and one-third of the deaths attributable to -7-

heart disease are associated with smoking." Death rates from heart disease in- crease rapidly with increasing age, especially beyond 40 years. Thus, heart diseases appear to be in large part diseases of aging, a fact not evident in this fiading. _ . The statemant that one-third of these deaths are "att--ibutable" to smoking implies cause. 'Associated with" does not. There is no question about the assoc- iation in some populations, but anyone who then argues for a causal relationship makes a huge leap based on faith alone. Any causal implication is defective because it involves in part those factors discussed earlier. .In addition, gu.ite_ strong statistical relationships between concentrations of a number of conmon air pollutant chemicals and mortality rates for several classifications of heart disease have been reported.- The air pollution - heart disease relationship has , often been overlooked or minimized in Surgeon Generals' reports. Finding (4). S. 1929 states that "the risks of miscarriage, stillbirths, premature births, and child weight deficiencies for pregnant women who smoke are higher than for pregnant women who do not smoke." The tera, 'risks", is strictly statistical and is not biological or biochemical. The implication is that because of statistical associations, the smoking-causality explanation is true. This is the same misuse of statistics, and of science, noted previously. Causality cannot be inferred from statistical association. The causal im- plications or claims, as stated in the Surgeon Generals' reports, are also de- fective because of selective reporting. Among the published reports that add- ress the problem are: R.J. Hickey, R.C. Clelland, and E.J. Bowers, "Maternal Smokinq, Birth yteight, Infant Death, and the Self-Selection Problem;" American Journal of Obstetrics and Gynecology, Vol. 131, 805-811 (1978). Misuse of . statistics is pointed out along with other errors and fallacies, including the . . ' S.o -8- 485 overlooking of or subjective denigration of fessor J. Yerushalmy. That low birth weight has been associat smoking during pregnancy in some populations implications or claims in the Surgeon Generz means smoking-causality constitute misuse of smoking also correlate with low birth weight :ables, maternal height and weight, geograpt tions. The 2,500 gram birth weight criteric :s unhealthy, ignores the large variations i :s evidence that birth weights of large, tal .eiqhts of children of small, shorter women. 1. 7:^ er, Rithin-Family Standards for Birt =.rely constitutional factors influence birt 'c.ns. An informative but controversial kind -c -causality versus the constitutional hypothe .ro later take up smoking ("future smokers") c:.ildren even before they started smo)cing. essor J. Yar+,"h"li.["Infaats with Low Birth 5tarted Smoking Cigarettes," American Journa 277-284 (1972)]: He found that, indeed ac;cers tended to be lower than bizth weight a_:r.dinq is in accord with.the constitution -ckinq-causality hypothesis. Y"*^"h"lmy's 'p°rted by D.T. Silverman Maternal Smokinq 0 (r.~ 0 ~ ~ Q CA -9-

S 7 esoc- iip ite art a x a 1, is is :ly. 3e- idd- can the 485 overlooking of or subjectiw denigration of the publications of the late pro- fessor J. Yerushalmp. That low birth weight bas been associated statistically with maternal smoking during pregnancy in some populations is not in question. However, the implications or claima in the Surgeon Generals' reports that the association means smok;.ag-causality constitute misuse of statistics. Factors other than smoking also correlate with low birth weight, such as certain socio-eccnoaic var- iables, materna.l height and weight, geographic differences, and altitude var'a- tions. The 2,500 gram birth weight criterion, implying that less than this weight is unhealthy, ignores the large variations in these factors. For example, there is evidence that birth weights of large, tall women tend to be heavier than birth weights of children of small, shorter women. [See J.:l. Tanner, L. Lajarraga, and G. Turner, Nithin-Family Standards for Birth-Weight," Lancet 2, 193-197 (1972)]. Surely constitutional factors influence birth size and weight in many popula- ti::ns. An informative but controversial kind•of evidence that addresses the smoking- -causality versus the constitutional hypothesis pertains in part to whether women who later take up smoking ("future smokers") tended to have low birth weight children even before they started smoking. This question was examined by Prof- essor J. Yerushalm;•["Infants with Low Birth Weight Born Before Their Mothers Started Smoking Cigarettes," American Jour~al of Obstetrics and Gynecology, Vol. 1~12, 277-284 (1972)]: 8e found that, indeed, birth weights of children of future smokers tended to be lower than birth weights of children of never smokers. Such a finding is in accord with.the constitutional hypothesis rather thaa with the smoking-causality hypothesis. Yerushalmy's findings were consistent with those reported by D.T. Silverman MaterR al fmoking and Birth Weight, Thesis, Johns -9-

486 Sopkins University, Baltimore, 1972]. These _indings have not been widely pop_ ular. It is also notevort4y that Yersshalmy found that the smaller babies of smoking mothers were healthier than the smaller babies of non-smokir.g ;+omen. 1b summarize, in view of selective reporting in the Surgeon Generals' re- ports, the evidence supporting the constitutional hypothesis, the misused stat- istics and the sometimes biased samples, it seems desirable to table anv legis- lative actions until the Congress has received a more complete analysis of the .ot evidence. At the present time, statements about deleterious effects of smoking are often put for+ard supported by evidence that is scientifically unconvincing and possibly invalid. Richard J. Hickey~ ~ March 2, 1982. _ ,. "4'R$ ~ -10- 48? Bioqraphical Sketch: Ricbsrd J. Nickey .,- Ilichard Jamq Nickey sirtl .~,A.1dress: 43 L. Clearfield Road, sirt: ;•.avertovn, Fs. 19083. s_~s^ess Address: - :..parcnent of Statistics, n. Wharton School, Lhiversity of Yennsylvania, philadelpt:ia, Ps. 19104. Mari .a•or Areas of Research Interest: - Epidemiological relationships between environmen environmental variables and wrious life forms, selectionr ecological genetics; mutsgeneaisr gan ecoloqyr ecological statistical analysisr analys biologicsl systems, such as multi-reqional, sarlt systems that pertain to elucidation of stioloqie sucn as cancer, cardiovascular diseases, and bi: pollution and public hulthr theoretical biolog} the biology of behaviorr the molecular biologicz senescence and of the developaent of diseases a: ology, valid and invalids social eonsequences o: istics and of science generally. Also hormesis. _":ut.on. College Level: . . Onivereitv of Illinois, Drbana, 1931-1935, B.S. minor. Vaiversity of Buffalo (N.Z.)(now S.O.N.7.), gra chestistry and cheaieal physics, 1936-1936. . Iova State Oniversity, lmes, 1936-1941, Ph.D., minor in microbiology. toeorarv Societies: Sigma xi, pi Nu Epsilon, Phi Lambda IIpsilon, Yt P"•essional Organizations: American Chemical Societyr American Associatiot Science (AAAS Fellow); Ner York Academy of Sci, Nutagen Society. 'each:no 17a erience: - Taught undergraduate chemistry at the Universi State University. rrofessional ENoerianee: t ~ September, 1975 to oresents Department of St. University of pennsylvania, Philadelphia, 3a. xugust, 1973, to September..1975,