Document LJxmr2bvwy6O2q4YQ2zxQrMLd

FILE NAME: Smoking (SMOK) DATE: 1956 Feb DOC#: SMOK007 DOCUMENT DESCRIPTION: Journal Article - Epidemiological Tests of Theories on Lung Cancer Etiology V olum e 71, N um ber 2 C O N TJEX TS I'.pidemiology of U-vikcmia --P1IK review Gilcin F. Meadors. Clues l<>suicide........................................................................ F<1win S. Shneidmnn and Xnrninn /,. Fnrhrrow. Cardiovascular diseases and public liealtli............................ John IF. Ferrer. The epidemic climate 1'IIR review..................................... IFinston II. Frier. Unfinished business in maternal and child nutrition.......... Marjorie M . Heseltine. The use of insecticide treated cords for housefly control. . John IF. Kilfioliich- and II. F. Sell oof. Man's emergence toward health The (,'.-l\. A. \\ inslow lecture, 19.75.......................................................................... Henry van '/.He Hyde. Public health in Petms\l\ania................................................ Ileruwn F. M alii son. Kpidemiological tests of theories on lung cancer etiology . IFi/liani Ilnenszrl. Health services in ei\il defense A symposium................... Biological-medical considerations in atomic defense. . Edwin G. II illiarns and Samuel G. Ingraham / /. Radiological delense........................................................ Simon Kinsman. ( '.hemieal 'weapons........................ Hurry F. Kramer. Fa#! 103 109 I IS 12.', 139 II1 IbfT 103 173 173 l!U '*1 <Ion Iim iod f r o n t i s p i e c e ........ .............. --..., -- A laboratory technician determines blood-dotting time, informa tion necessary to the safe use of anticoagulant drugs. (See story of the American Heart Association dn page 115.) Epidemiological Tests of Theories on Lung Cancer Etiology By WILLIAM HAENSZEL Q UESTIONS on the etiologies of lung eaneer cannot be resolved by a single, deeisive experiment or set of observations, and a single study is not the. natural unit, for dismiss ing the. epidemiology of 1his disease. The major eonrern must be with the interrelationships of findings and their collation to see whether a chain of evidence stronger than any single com ponent part can be forged. Information on a proposeu etiology cannot be assessed without relating it to both positive and negative findings on alternate possibilities, drawn from animal experimentation and other laboratory work as well as from epidemiological studies. Also, nith a subject under such active scrutiny as lung cancer, it is tlillicult to attempt to divorce facts already established from interpretations which may either be stated explicitly or merely be implied by investigators through their choice of specific studies to be pursued. Under these conditions, a better perspective on epidemiolog ical issues in lung cancer studies may be gained by viewing them in the persjx'ctive of possible future work. Since interrelationships of indi vidual findings are the main concern, this paper Mr. Haenszel is head of the Biometry Section, Biametrr and Epidemiology Branch, National Cancer Institute, Public Health Service. He pre sented this paper at a joint session of the Biometrics Society and the Institute, of Mathematical Statistics, Chapel Hill, N.C^ April 23, 1955 deals more with the ideas underlying a set, of studies than with the tactical details o f 'any single study. Increase in Lung Cancer . If no real increase in lung i-ancer existed, all associations between htngr ,, j><i- ;jj11 expo.-mc to atmospheric poll lit ion, cigu. .lie smoking, or other exogenous agents recent ly introduced into the environment would he suspect, as not point ing to possible etiological relationships. A few persons still maintain that the recorded increase in lung cancer incidence and mortality is liftitiousund has resulted from improved diagnosis ( /) , hut. this is no longer a popular position (2). Among the reasons for accepting a real increase in lung cancer mortality are the sex dill'erences in rate of increase, and the magni tude of Ihe current. emit ilining increase {>). One purpose, for inquiries on (he, increase in risk would he to provide, more precise quan titative estimates than are now available from mortality data which may later prove useful for reconciliation with numerical measures of exposure to suspected agents. For this reason, anyone proposing to study the. elfeet of possible misdiagnosis of lung cancer as tuberculosis t.j) or some other cause slum Id relate, these effects to data produced by the death registration sys tem. Samples of deaths from lung cancer and other causes taken from vital statistics sources could he compared with autopsy protocols and other diagnostic cvidency. Evaluation of tho Yol. 71, No. 2, F ebruary 1956 i r3 Iwo sources of error in oltlcial statistics, ((f) deal Iis from oi her causes erroneously recorded as limy cancer and (//) lung cancer deaths at tributed to other causes, would l>e rendered dillieull. by the. selective character of autopsy statistics. Ilowever, the etl'ort should be made. A useful minimum objective would be to deter mini' how many lung cancer eases remain un diagnosed at present. Farber reported that in a series of 1,070 morphologically proved eases of bronchogenic carcinoma, (11 percent were, not posit ivc1y diagnosed prior to autopsy (/>). U n doubtedly, many of these deaths were not re corded as luny cancer on the. death certificate. Environmental Agents The v.-orkiny hypothesis for most investiga tors lias been that the probable causes of limy cancer are to he, found among environmental agents (2). The association of (lie luyh pre valence of cancers for several other sites, partindarly the mout h, with sped (it*customs found, only in certain parts of the world sfrenythens the belief that direct contact with known or suspected eareinoyens may bo involved ((>'). On the assumption that agents with opportu nity for direct contact with limy tissue are the most likely candidates, ell'orts to establish etioloyical relationships between limy cancer and environmental agents have been directed into three main areas: (a) airborne ayents en countered in special occupations: (h) atmos pheric pollutants, particularly those, rcsult.iny from combust ion of hydrocarbons: and (a) smoking of tobacco. To assert, that an agent is e( ioloyieally re lated to limy cancer simply implies that its re moval from a defined population will result ultimately in a marked reduction of limy can cer. The population inust.be defined in a man ner which safeyuanls ayainst an ell'eet, beiuy produced merely by the.process of selection and classification. The. concept does not require, de tailed specification of a mechanism produciny the ctl'ccf. Goldheryers lindiny that, pellayra may be prevented by diet was no less valid be cause subsequent, work led to the identification of the vitamin B complex and, finally, to nico tinic. acid as the effective ayent. This state ment does not imply that ideas on mechanisms 164 are not helpful: to the contrary, they may suyyest situations in which data can be collected and tested. Occupational Carcinogens There is yenernl nyreement on the etioloyical siyniticanee of several occupational exposures, and such findings have been important in establishiny the principle of multiple etioloyies for luny cancer. Long-term observations of closed populations have convinced most people that the Sclmeeherg and Jonehimsthal mi urn's ( '/, A) and certain groups of chromate workers (0, 10) were, subjected, by virtue of their occu pations, to excessive limy cancer risks. The evidence on chromate workers is purely epi demiological, not con firmed by animal experi mentation, and was first suspected on the basis of clinical observations. Although many inves tigators have deduced from the Schneeberg and Joachimsthnl observations that radiations emit ted from radon were responsible, Loren/, felt that the case had not been firmly established and that other possibilities, such as dust pneumoco niosis, bad not been conclusively ruled out. (11). Excess luny cancer risks detected by exami nation of occupational mortality data, such as data on workers employed in the preparation of coal tar products and producer yas (12). have also been generally accepted as reflecting an etioloyical relationship between luny cancer and some suspected carcinogenic ayent in these products. Nor is there any great disposit ion to question the status of ayents detected by clinical observations and retrospective studies of ayents such as arsenic dust, nickel carbonyl, asbestos dust, and coke oven fumes (12). When the study results are reviewed to pick up occupa tional exposures which appear repeatedly, the danger of being misled by a highly unusual re sult in a single study is minimized. Additional occupational risks will no doubt be demon strated in the future. Ifowever, because of the small numbers of workers exposed, the further denumeration of very restricted types of risks will not be crucial in working-out the epidemio logical details of luny cancer. This statement would not apply to the common and widespread ayents contributing to atmospheric pollution, including hydrocarbon combustion products Public H ealth Reports such as industrial and heating plant wastes and mol nr vehicle exhausts. Atmospheric Pollutants To date, atmospheric pollution theories have not been well documented by demonstrations of excess risk among workers with heavy exposures to common atmospheric pollutants such as may exist in railroad yards, oil refineries, .and con gested urban ureas. With the advent of group insurance and pension plans, the methodological problems of such inquiries have become much simpler. Several cohorts can be located for study. The National Cancer Institute is fol lowing a group of 1,100,000 railroad employees who have been in the industry 10 years or longer, including station, oflice, yard, and shop em ployees, as well as train crews. Ilather detailed occupational histories are available and, with the accumulation of mortality data over several years, such specific points as the efi'ects of dieseli/alion, prolonged exposure to smoke in railroad tunnels, and metal dust exposures in shops may be looked into. The important feature is that work histories and exposures be specified in some detail. Significant efi'ects may be diluted and overlooked, by considering only gross com pany or industry experience. The widespread use of petroleum products makes it very desirable to gather data on work ers in that industry. Some medical departments of the major oil companies have stall's well equipped to undertake such studies. The major problem here, as elsewhere, is to secure access to company records and to rearrange data or ganized to meet other administrative needs. lit areas where petroleum operations are concen trated, data might be collected on a geographic rather than an industrywide basis. At least, one study is under way in Oklahoma which may yield some information on lung cancer risks among oil industry employees. Those advancing atmospheric carcinogen theories have pointed to the sizable urbanrural differentials in lung cancer risk (1,)). al though some further assumptions are required to reconcile this hypothesis with the great excess risk oltserved among males. The presence of carcinogenic substancesin urban atmospheres has been repeatedly demonstrated bv experi- meiits with mice (/.]. 1C). Kurt her epidemio logical work 1ms been hampered by the inability loclassify individuals in the general population quantitatively with respect to atmospheric ex posures. Aieaningful histories are hard to col lect., and in mobile populations, problems of classification by residence histories become com plex. Wliim leads on specific types of atmos pheric pollutants are obtained from studies of special situations, the way may then be opened for return to the problem of reclassifying the general population with respect to special ex posures with greater chance for success. Smoking of Tobacco Although at lcu>t I I retrospective clinical studies (17) hare reported on the association o f, smoking and lung cancer, these studies have been criticized on a variety of technical grounds. The objections advanced cover such points as interviewer bias (interviewers in some studies knew the, identity of lung cancer patients and cont rols at the time ok interview), and bias arising from selection factors associated with hospital admission (in most studies only hos pitalized lung cancer patients wore inter viewed). However, the, objection based on knowledge of Ilie identity of patients and con trols ignores the evidence, of Doll and Ilill (IS) that males erroneously interviewed as lung can cer cases, as established by later events, showed smoking histories characteristic of the control rather than of the lung cancer series. Because of possible selection factors involved in taking up the smoking habit, some feel that the tobaccolung cancer associations rellec.t associations with other common, but still unidentified factors. The present data on associations bet.ween smoking and lung cancer are the most extensive and afford the most opportunities for discussing additional studies. For oriental ion, the type of evidence now on hand may be, compared with the epidemiological findings for cholera, pel lagra, and dental caries, all diseases in vyhicli animal experimentation played a minor role, at least in the early phases of the investigation (19-25). On review, the basic core of associations on smoking and lung cancer must, appear as im pressive as any produced in these,'other investi- V,,l. 71. No. 2, F ebruary 1956 163 370SG6-- 50----- 5 * gations. However, lung cancer does not present the leaf tires of geographic localization charac teristic for cholera, pellagra, and the absence of dental caries, which made it possible to elab orate in a rather straightforward manner the basic associations with water supplies suspected of contamination (cholera), diet (pellagra), and fluoridated water (dental caries) with corroltorative detail and combinations of isolated special situations. Such detail is helpful in ruling out some association as not significant etiologically; in this manner, for example, the associations between cholera and altitude (willin' .i'!on) and bet ween pellagra and mill \iiiage occupations could he discarded as not -igiiibcnnl. For lung cancer, there are as yet no comilerparls, for example, lo the cholera patient in an isolated rural area, who regularly sent 'or water from the famous Broad Street pump {19) ; to the absence of pellagra among doctors, nurses, and attendants in mental insti tutions, although the disease was common among pal ients; to the peculiar age distribution for pellagra in a children's institution where the disease was limited to children between the ages of f> and 12; to differential attack rates for dental caries by gradations of exposure to lluoridated supplies among natives and in-migrants; and to the sequence of dental events fol low! ng shifts in the source of a community water supply. Collection of such details generally results from the study of populations. This is one of the reasons why forward studies on lung cancer among defined population cohorts are so important. Funhermoro, the epidemiological models for cholera, pellagra, and dental caries have been put to the test in Ihe successful application of control measures. For pellagra, only the addi tion of meat and milk lo the diet, was involved {'!(>). Fluoridation effects could bo observed under coni rolled conditions because of tho happy accident of public water supplies; if private wells were the only source of water, tho problem would have been complicated by self selection of families fluoridating their own supplies. Tho test of Snow's conclusions on cholera {19), which required implementation through administrative decisions on methods for control of public water supplies, occurred 166 much later, more than 10 years after he had formulated them. The right combinations of circumstances to permit such straightforward tests of most of the proposed lung cancer etiologies are not present now. Thus, the immediate lines for further work on the nature and meaning of the smokinglung cancer associations are to develop reason able facsimiles to tests by control measures and to assemble supporting, corroborative details from population studies, not overlooking any negative evidence which may appear. Substitutes for Direct Tests As a substitute for a direct test by control measures of the smoking-lung cancer model one may-look for groups in which smokihg is either proscribed for religious reasons or does not otherwise, form'part of the cultural pattern. Two of the four basic study elements required are at hand: data on the distribution of smoking habits in the United States population (col lected as part of the. Current Population Survey for February 1955), and data on lung cancer mortality (from publications of vital statisticsoffices). The major problems would be to. secure the cooperation of the groups selected and to develop procedures to obtain the coun terparts of these data for their memberships. To avoid artifacts which might be introduced by classification procedures, the primary test must be whether the overall lung cancer mor tality in the group studied was markedly less than for an appropriate segment of the popula tion of the United States and commensurate with differences in the smoking patterns. To this, other refinements may be added, such as observing whether mortality differences be tween tlie study and control populations dis appear when specific comparisons are made by amount of tobacco smoked. The limited number of countries which can provide reliable diagnostic data on causes of illness and death has discouraged similar stud ies of groups abroad which have unusual smok ing patterns. However, Special study situations may be encountered. One illustration is cited, without judgment as to its intrinsic merit. The director of the Iladassah medical organization recently reported tho absence of a single case Public H ealth Reports of lung cancer among Yemenite Jews in Israel, during the past 15 years, and ho further ob served that they did not smoko cigarettes but used a form of Oriental water pipe (27). Israel is a country w ith western standards of m edical care and this lead may be worth pursuing. In' these matters, the importance of reviewing tho primary sources of population, morbidity, and mortality data should be stressed. Other clues may lead indirectly to population groups with unusual smoking histories. For example, Steiner (G), on the basis of necropsy evidence in Los Angeles, has found a possible exception among Mexicans to the usual sex ratio for lung cancer; tho proportion of Mexican women with lung cancer at autopsy approaches the proportion for Mexican men. This suggests the need for further studies among Mexicans to confirm the facts and to uncover possible reasons for the aberrant sex ratio for lung can cer. This might conceivably lead to the finding of a group of unusually heavy smokers among women, a useful contrast to the experience of abstainers. Corroborative Details Additional details on the nature of the smok ing-lung cancer relationship should be assem bled from large-scale forward studies. These possibilities were obviously in the minds of those who planned tho American Cancer Society ami the National Cancer Instilute-Vcterans Administration studies. Both study groups consist mainly of men between tho ages of 50 ami GO, the ages of highest lung cancer inci dence. One approach is to expand the evidence on the nature of the association by detailed crossclassilieaticn. Epidemiologists have long ac cepted the principle that the etiological signifi cance of an association is enhanced if it can be shown to persist within a variety of sub-uni verses. The National Cancer Institute-Veter ans Administration cohort can bo subdivided with respect to occupation and industry. By lengthening the observation period to several years, further tests can be applied in both studies, including checks on the consistency of regression relationships between lung cancer mortality and amounts of tobacco smoked for soveral occupational groups and for urban and rural residents. Additional tests can be pursued by relating age at death to "age started smoking" to see if any consistent pat tern emerges. Studies Among Women Unfortunately for study purposes, the spread of "age started smoking" is not too great among men, and it may be hard to assemble sizablo cohorts for the more unusual combinations of year of birth and "age. started smoking." Women would he more promising subjects, since women of all ages began smoking in large num bers during tho lOJO decade, and duration of smoking by women is not so closely tied to chronological age. For this reason, more pro spective and retrospective studies on smoking and lung cancer among women would he re warding. In addition to the one series of studies on smoking among women iiuiorted lx Doll and Hill (28), two v o .ctdve studies are known to be under wav. The relative rarity of lung cancer among women imposes severe study handicaps and has undoubtedly deterred other eForts. Potential cohorts for followup would include employed women covered by group life insur ance policies. To minimize turnover and loss to observation it would be desirable to restrict the cohort to employees with several years' service. Studies of lung oancur among women should establish what part, if any, of (lie sex dilFerential in lung cancer mortality (or incidence) dis appears when comparisons are made specific for smoking class. The fragmentary Doll-Hill data suggest .that a. good part, but not. all, of the sex dill'erenlial may be accounted for in this manner. Indirect Checks Retrospective and prospective, studies may be. viewed as ell'orls to secure etiologieally mean ingful data, which official vital statistics sources cannot provide, since they are limited to classi fication by standard demographic variables. There is another transitional bridge by which vital statistics data on lung cancer deaths can Vol. 71, No. 2, February 1956 167 In- utilized. By taking in format ion on smoking I>aHerns in the general population (collected in Current, I'opulat ion Survey for February l!t,V>) and applying to them (lie data on ab solute. and relative lung cancer risks by smok ing class as reported in the preliminary results of the American Cancer Society study (29) and in several retrospective studies (20), the "ex pected" distribution of lung cancer deaths can be computed by sex, urban or rural residence, geographic region, and broad occupational groups, on the assumption that risks by smoking class in the groups studied hold also in the total population. Comparison of the observed and "expected" numbers of deaths would then show how much of the prevailing variation in lung cancer mortality can ho accounted for by considering smoking histories. In this manner, some useful consistency checks on the smokinghmg cancer model would be provided. .Such consistency checks need not be confined to smok.ng histories. The Current Population Survey material illustrates an economically feasible means for remedying some deficiencies in standard population classifications. A l though many investigators have expressed the need for procuring information on the. popula tion distribution of characteristics understudy, they have often beeii deterred by the cost. Probability sampling methods for gathering such data can provide a powerful tool to the epidemiologist. Other Possible Factors Turning from suspected environmental .agents with properties which permit direct con tact with lung tissue, several possible causes of lung cancer remain. There is the. class of environmental agents which do not come in di rect contact with lung tissue. Closely related would be possible tissue changes induced by virus and other infectious agents. Constitu tional susceptibility might also be considered. If some persons have a predispos'd ion to attacks on lung tissue by some morbid process, the rise in lung cancer might bo due to the suppression of "competitive" respiratory causes, such as tuberculosis and pneumonia. Finally, there is the possibility of the association of smoking with other physical, psychological, and emo tional factors which may be engendered by the processes intluoncing persons to take up smoking. fn dealing with lung cancer, where prospects for tests of etiological Relationships by trial of control measures are remote, other hypotheses, however unlikely, and the collection of evidence to support or discredit them should he en couraged. The pursuit of other lines of in vestigation will be useful. I f results are posi tive, another effect of lung cancer will have been discovered; if negative, they will buttress the interpretations applied to effects already known. Gilliam, in a general discussion of chronic disease epidemiology, alluded briefly to a va riety of items investigated by him in a study of lung cancer patients and controls, using the retrospective, case-history method (SI). These items ranged from color of eyes and use of den tures to histories of illnesses and use of general anesthetics. As he hastens to say, such indi vidual findings cannot be used for any general izations. However, it. may be worth while to inquire systematically into other findings col lected but never deemed worthy of report. Several of the, other early retrospective studies were "shotgun'' inquiries and collected in form al tion on a variety of suspected agents and ex posures. Collectively, these results might prove useful for testing and discriminating between hypotheses and the planning of further studies. Several studies exploring characteristics d if ferential mg smokers and nonsmokers may ho expected. The question of differences between the two groups arises because smokers select themselves; smoking is not a treatment which can be studied experimentally by randomized application to a population. There is a wide range of possibilities for retrospective studies on the physical, psychological, and emotional attributes of smokers and nonsmokers. The demonstration of an item as being associated with smoking history is essentially a screening device. The further test is whether the item, when applied to a population, produces differ ences in lung cancer risks equal in magnitude to those yielded by smoking history classifications (if smoking history is uncontrolled), or whether an effect is observed when smoking history is controlled by classification. 168 Public H ealth Keporti. Association With Other Illnesses Some theories on (lie subsequent increased risk of limy caneer among persons with histories of respiratory illnesses such as influenza, pneu monia, bronchitis, hmd tuberculosis, based usu ally on clinical and pathological impressions, have been reviewed by Poll (32), who continues on to point out some observations inconsistent with these, theories. Xo associations of lung cancer with the presence or absence of an ante cedent. illness have yet been established by oiner vations on populations through the. use of retrospective or prospective studies. Retro spective studies are handicapped in this field, because patients witli lung cancer and other respiratory disorders apparently recall previ ous respiratory ailments more readily than do other persons. There are some possibilities for forward studies on cohorts known to have re covered from specific illnesses, free from the "recall" bias in the retrospective approach, which may he exploited. For example, records can he assembled of "World War I veterans who had influenza and/or pneumonia in 191S. These veterans can he traced forward in time to determine their lung cancer mortality expe rience. (A study of British veterans (.(0) has been published recently.) From civilian life, one could draw on recovered pneumonia cases reported to health departments in the 1930 dec ade, when these agencies had active programs based on pneumococcus typing anil distrilm! ion of serum. If excess lung cancer mortality among such groups were demonstrated, this would not neces sarily he a result of eil'ects of antecedent ill nesses. If constitutional predisposition of lung tissue were a factor, the rise in lung cancer mortality might be due to the suppression of i uberculosis and pneumonia as causes of death. I f a genetic basis for constitutional predisposi tion to lung cancer were postulated, other tests could be devised. One approach would be to assemble lists of relatives of diagnosed cases of hum' cancer for followup to determine lung e.-meer mortality. Lists of familial contacts could be assembled from records maintained in iuberculosis control programs. Investigations of this character might be undertaken more readily in some European countries, where pop ulation mobility is low and more comprehensive population record systems exist. Retrospective and Forward Studies The preceding sections lone, emphasized the wide range of possibilities for the use of retro spective and forward study techniques. Retro spective, ease-history studies use as their point of departure diagnosed eases of a disease and matched controls and compare antecedent events in their previous histories. Prospective, or forward, studies start with the assembly at a fixed point in time of defined cohorts classi tied with respect to certain attributes, trace, them forward in time, and note events occurring subsequently. Perhaps the major criticism of retrospective studies for lung cancer has been that diagnosed cases have generally been drawn from hospital populations. Posit ive assoeial ion between two diseases, not present in the general population, may be produced when hospital admissions alone are studied, because persons witli a com bination of complaints are more likely to re quire hospital treatment 3.',). Smoking is not an illness, and for lung cancer it is difficult to see how the smoking history could have any inlluence on hospital admission. Imng cancer is a serious disease normally requiring hospital ization, and roughly four-fifths of all diagnosed cases are hospitalized (33). The proportion is even higher when microscopically confirmed cases alone are considered. There would have to be extraordinary dill'erciiees in smoking his tories bet ween hospitalized and noiihospitalized patients and controls to upset inferences drawn from hospitalized cases (36). So far, the re sults of forward population studios on excess lung cancer risks among smokers as compared to nonsmokers have agreed in general with tliu-o of the retrospect ive s( udies. This suggests (lint biases entering into the -election of hospital ized cases and controls studied ret ros|X`efi velv are probably not. (he source of the. associations noted. 'Tim fundamental assumption underlying ret rospective studies and the estimation therefrom of differences in risks between population groups is that the. diagnosed rases and controls each be representative of the i,{inverse chosen Yol. 71, No. 2, F ebruary 1956 169 for investigation. Although (hero, are precau tions which can be taken in devising a sampling plan, this representative property cannot be guaranteed merely from internal examination of a single set of data. One must be guided on this point by the comparison of results from several studies and judgment as to the possible biases operating in any setting. Forward population studies have been quostioned on the grounds that the cohorts have not been selected by probability sampling methods no i i.at, imb.vuiuui -tudy results cannot, be. geni-rali/.ed to t:;e dual population. The expense involved in tracing cohorts drawn from the gen eral population by probability sampling meth ods would Ire great and most investigators have been forced to look for populations followed routinely for other purposes, such as persons covered by employment or insurance benefits. The criticisms of unrepresentative sampling would appear to disregard the experience avail able. from actuarial sources. There- are two common types of insurance: "ordinary" poli cies. which require applicants to pass a physical examination, and "industrial" policies, which undergo some underwriting selection but re quire no physical examination. Policyholders would scarcely meet the usual criteria for rep resentative samples. The'Metropolitan Life Insurance Co. has published extensive data on the mortality experience of their indus trial policyholders and finds rather close, agree ment- with the mortality experience of the. gen eral population, after appropriate adjustments for age. sex. and race composition (-V). Data drawn front actuarial experience and reported * in su^li sources as Length of Life (/S) and The. Shitixtirt'il JlvUctih, a monthly publication of the Metropolitan Life Insurance Co., indicate that the relative patterns of mortality with re spect. to cause, age, sex, race, and geographical region in insured populations have their coun terparts in general mortality experience, despite some, di fl'erences in magnitude of rates. The co horts assembled for smoking and lung cancer investigations do not appear to have been sub jected to more rigorous selection than that en countered by industrial policyholders, nor is there reason to believe that the selection effects which did exist discriminated between smokers and nonsmokers. Even when stronger selection 170 effects on mortality experience exist, as in the case of ordinary policyholder's, it is well known that the selection effects of physical examina tions on mortality experience wear off quickly: very conservatively, in 10 years or less (3d), usually in about 5 years. I f there is any ques tion about the prospective study results on the association of smoking and lung cancer being affected by selection, the conservative course would be to discount the first few years' experi ence and to require that the early results be con firmed by later experience after selection effects have worn off. The collation of experience from several studies is also a safeguard against error in gen eralizing from forward studies on lung cancer. Data from prospective studies may be sub classified by other population characteristics and further checked for internal consistency. "With these precautions, the dangers of drawing inferences from forward and retrospective studies inapplicable to the general population seem rather remote. Summary ..The size of the lung cancer problem, as indi cated by the number of lives at stake and the economic implications of any potential control measures which might later be advanced, should not blind investigators to the many possibilities for studies similar to those- on other chronic diseases winch have yielded some of their secrets to the epidemiological approach. Diseases showing pronounced variations in risk among population subgroups are the. more amenable to epidemiological study; and lung cancer falls into this category. Studies on human popula tions should continue to play an important role in delineating possible etiological relationships for lung cancer, the mechanisms for which could then be elaborated by animal experimentation. Progress must come by the cross-checking of the several results of epidemiological studies, animal experimentation, and other laboratory findings. The issues will probably be settled by an evolutionary process, as was the fluoridedental caries relationship. Gradually, the meaning of cei'tain associations will become accepted. Dissent will die off and the debate will shift to the many other points in any Public H ealth Reports epidemiological models still r e q u i r i n g elaboration. ^ REFERENCES (/) (2) (31 (.}) (5) (it) (7) (8) (0) (10) {It) (12) (1,1) (/.}) (15) (IS) Itigdon, H. H., and KirehofC, II.: Smokiug and cancer of the lung--Let's review the facts. Texas Itep. Biol. & Med. 11: 715-727 (1053). Clenmiesen, J.,, Nielsen, A., and Jensen, E . : Mor tality and incidence of cancer of the lung in Denmark and some other countries. In Cancer of the L u n g; A symposium, edited by Dr. J. Clemmesen, reprinted from Acta Unio Contra Caneruru. Paris, Council for International Organizations of Medical Sciences, 1953, 210 pp. Dunn, II. L . : Lung cancer in the twentieth cen tury. J. Internnt. Coll. Surg. 23: 320-312 O'. >55). Gilliam, A. G .: Some aspects of the lung cancer problem. ML Med. 11G: 1G3-171 (1955). Earlier, S. M .: Lung cancer. Springfield, 111., C. 0, Thomas, 1951, p. 1. Steiner, I'. E.: Cancer: Uaee and geography. Baltimore, Williams & Wilkins Co., 1954, 303 pp. itosotoski, O., Saupe, E., and Schmorl, G .: Die liergkraiikheit der erzbergleute in Schneelierg in .Sachsen. Zeitsehr. F. Krebsforschung 23: 300-384 (192G). Poller, S . : Lung cancer among mine workers in Jouehimsthal. Human Biology 11: 130-143 (1930). Brinton, H. P,, Frasier, E. S., and Koven, A. L . : Moriiidity and mortality ex'iterieuce among chromate workers. Pub. Health Itep. 07: S35S 17 (1952). Maelile, W., and Gregorius, F . : Cancer of the res piratory system in the United States chromateprodtioing industry. Pub. Health Bop. 03: 1114-1127 (1948). Lorenz, E . : Radioactivity and lung cancer; A critical review of lung cancer in the miners of Schneeberg and Joachimstlial. J. Nat. Cancer Inst. 5: 1-15 (1944). Kennaway, E. L., and Kennawny, N. M .: A fur ther study of. the incidence of cancer of the lung and larynx. Brit. J. Cancer 1: 200-298 (1947). ilueper, W. C .: Recent developments in environ mental cancer. Reprinted, with additions, from the A. M. A. Archives of Pathology 58: 360399, 475-523, G45-GS2 (195-1). Hooper, W. C .: Environmental lung cancer. Indust. Med. and Surg. 20 : 49-02 (1951). Leitor, J. Shiinkin, M. B., and Shear, M. J . : Pro duction of subcutaneous sarcomas in mice with tars extracted from atmospheric dust. J. Nat. Cancer lu st 3: 155-105. Kolin, P,, Falk, n . L,, Muder, I'., and Thomas, M .: Aromatic hydrocarbons. Arch. Ind. Ilyg. ami Occup. Med. 9*: 153-163 (1954). Vol. 71, No. 2, February 1956 (Pi) I ( / 8) (10) (20) (21) (22) (2.!) (2.)) t- (2.7) (2S) (27) (28) (29) (30) (31) (32) Dorn, H. F . : The relationship of cancer of the lung anti the use of tobacco. The American Statistician 8: 7-13 (1954). Doll, U,, and Hill, A. it .: Smoking uml carcinoma of tile lung. Brit. M. .1. 2: 745 (1950). Snow, .7. : Snow on clinU'i-n. Now VovU, Tlu> Commonwealth Fund, 1900, 191 pp. Goldberger, J., Wheeler, G. A., and Sydcnstricker, E . : A study of the relation of diet to pella gra incidence in seven textile-mill communi ties of Smith Carolina in 1910. I'ub. Health Rep. 35: 048-713 (1920). Goldberger, J., Wheeler, G. A., and Sydcnstricker. E . : Pellagra incidence in relation to sox, age. season, occupation, and "disabling sickness" in seven cotton-mill villages of South Carolina during 1910. Pub. Health Rep. 35: 1G50-1GG3 (1920). Goldberger, J,, Wheeler, G. A., and Sydcnstricker, E .: A study of the relation of factors of a sanitary character to jadlagra incidence in seven cotton-mill villages of South Carolina in 1010. I'uli. Health Itep. 35: 1701 -1714 (1920). Goldberger, J.. Wheeler, G. A., and Sydcnstricker, E . : A study of the relation of family income and other economic factors to pellagra in- ' efdencc in seven cotton-mill villages of South Carolina in 1910. Pub..Health Itep. 3,5 20732714 (1920). P.ihby, B. G., and Brudcvold, K. : 4.- external action of fiuorides and oilier agents on tinteeth in the prevention of tooth decay. In Fluoridation as a public health measure, edited by J. II. Slinw. Washington, It. C.. American Association for the Advancement of Science. 1951, pp. 148-17.8. U. S. Public Health Service: Epidemiological studies of llunride waters and dental caries. Collection of reprints from Public Health Re ports, 193S-44. Goldberger, .T., Waring, C. If., and Tenner. W. F . : Pellagra prevention by diet among insti tutional inmates. Puli. Health Itep. 28: 230123GS (1023). New York Times. Feb. 1, 1955. Doll, It., and Hill, A. B . : A study of the aetiology of carcinoma of the lung. Brit. M. J. 2: 1271128G (1952). Hammond, E. C.. and Horn, I ) . : The relation ship between human smoking habits and death rates. J. A. M. A. 155: 1310-132.8 (1954). Cutler, S. J., and Loveland. D. B . : The risk of developing lung cancer and ils relationship to smoking. J. Nat. Cancer Inst. 15: 201-211 (1954). Gilliam, A. G .: Opjiortunities for application of epidemiologic method to study of cancer. Am. J. Puli. Health 43: 1217-1257 (1953). Doll, It.: Bronchial carcinoma: Incidence and aetiology. Brit. M. .1, 2: 521-527 (1953). 171 i Hickson, .1.: Limitations i>t' ;is*- application n[ f iuirnid table analysis to hospital data. Bio metrics Unit. 2: 47-3.'5 (1940), i.i;l White, (*. : Sampling in medical research. Brit. .\l. .1. So. -IS Hr. 128 I-1288, Dec. 12, 1953. M i) II. S. Titblie Health .Service. National Cancer Institute: Cancer illness in ten urban areas of United States. Cancer Morbidity Series Nos. I IP, ipna-.io 1.Pi) Kraus. A. S . : The use of hospital data in study ing the association between a characteristic and a disease. I'ub. Health liep. (!!): 1211IlHt (1954). (.17) llubliii. I... ini<) .St>tesetnian, M .: Health progress amoaa' industrial policyholders 194b to 1950. See. Actuaries Tr. 9: 294-3118 (1951). (33) Ibiblin. L. I., Lotkn, A. .T., and Spieaelmnn, M .: Length of life. Rev. ed. New York, Ronald Tress Co., 1919, p. 1^)3. (.i.'i) Jordan, C. \V .: I.ife contingencies. New York, Society of Actuaries, 1952. p. 30. ( jfl) Case, It. A. M., and Lea, A. J , : Mustard gas poi soning. chronic bronchitis, and lung cancer. An investigation into the possibility that poi soning by mustard gas in the 1014-18 war might ho a factor in the production of neoplasia. Brit. J. Tree, & Social Med. 9: 62-72 (1955). Ninth Annual Seminar on Seafood Sanitation The. ninth animal Seminar on Seafood Sani tation, sponsored by the Virginia and Mary land State Health Departments, was held in Solomons, iMd., September 27-29. 1955. More than 50 persons attended (lie 5 sessions, which included an inspection of an o\slur-packing house and a visit by boat to a soft-clam harvest ing area on the Patuxent River. The first Seminar on Seafood Sanitation was held in 19'16 because seafood sanitation spe cialists found that their work required broad up-to-date technical knowledge. In addition to the common food plant sanitation problems, they were concerned with the construction and operation of municipal and private sewage disposal works; they had to make bacteriologi cal. hydrographic, and sanitary surveys of shellfish-growing areas; and llicv were obliged to apj ly the results of these studies In decisions on the suitability of an area for shellfish cul ture. The seminar lias provided a forum where bacteriologists, sanitarians, and sani tary engineers working with seafood sanitation in the Chesapeake Ray area can exchange in formation on such problems of mutual interest. Many seafood sanitation problems en countered by county, State, and Federal per sonnel were discussed at the 3-day meeting. These included sanitary surveys of shellfish growing areas, general food plant sanitation, sanitation of oyster-shucking plants, sanitation problems in the crabmcat industry, sanitary problems in the new Chesapeake Bay soft-clam industry, preservation of food by radioactive sterilization, and a discussion of the relative merits of State certification of processors of breaded seafood. Included in tire agencies represented in the discussions were State health departments of Maryland. Virginia, South Carolina. Georgia, and the District of Columbia; New York State Department of Conservation; Maryland county health departments which have seafood sanitation programs; the Food and Drug Administration and the Public Health Service: and the Fish and Wildlife Service. The seafood industry was repre sented by officials of the National Fisheries Institute, the Oyster Institute of North Amer ica, and tlie Oyster Growers and Dealers Asso ciation of North America, Inc. William Ballard, president of the Oyster Growers and Dealers Association of North America, Inc., and operator of one of the world's largest oyster-shucking and packing plants, told of his industry's dependence on sanitation mid expressed the opinion that the future of the seafood industry rested with the accomplishments of the fisheries research scientists and with the seafood sanitation specialists. ' The participants recommended that the health departments of Maryland and Virginia sponsor another seafood sanitation seminar in 1956. 172 Public H ealth Reports