Volume 150

Number 9

November 1, 1999

American Journal ofEPIDEMIOLOGY

Copyright 1999 by The Johns Hopkins UniversitySchool of Hygiene and Public Health

Sponsored by the Society for Epidemiologic Research

COMMENTARY

Invited Commentary: Circular Epidemiology

Lewis H. Kuller

Circular epidemiology can be defined as the continuation of specific types of epidemiologic studies beyondthe point of reasonable doubt of the true existence of an important association or the absence of such anassociation. Circular epidemiology is an extreme example of studies of the consistency of associations. A basicproblem for epidemiology is the lack of a systematic approach to acquiring new knowledge to reach a goal ofimproving public health and preventive medicine. For epidemiologists, research support unfortunately is biasedtoward the continued study of already proven hypotheses. Circular epidemiology, however, freezes at one pointin the evolution of epidemiologic studies, failing to move from descriptive to analytical case-control andlongitudinal studies, for example, to experimental, clinical trials. Good epidemiology journals are filled with verywell-conducted epidemiologic studies that primarily repeat the obvious or are variations on the theme. Am JEpidemiol 1999; 150:897-903.

association; epidemiologic studies; heart diseases; preventive medicine; public health

Epidemiology is the basic science of public healthand preventive medicine. Good epidemiologic studiesprogress from descriptive to analytical to experimentalepidemiology and then to studies of effectiveness lead-ing to prevention programs. Unfortunately, there is atendency in epidemiology to perseverate at one level ofevidence, for example, on one type of study design,without moving forward or, worst yet, to regress to orrediscover the evidence documented from many paststudies. Circular epidemiology can be defined as thecontinuation of specific types of epidemiologic studiesbeyond the point of reasonable doubt of the true exis-tence of an important association or the absence of suchan association. Circular epidemiology is an extremeexample of studies of the consistency of associations(1).

There are at least two major reasons for circular epi-demiology. First, epidemiology is subcategorized intospecific topics. Methodology in one specialized area is

Received for publication March 19,1999, and accepted for publi-cation April 12, 1999.

Abbreviation: LDL, low density lipoprotein.From the Department of Epidemiology, Graduate School of Public

Health, University of Pittsburgh, 130 DeSoto Street, Room A526,Pittsburgh, PA 15261. (Reprint requests to Dr. Lewis H. Kuller at thisaddress).

often rediscovered in a newly described category ofepidemiology. Second, subcategories of epidemiologymature. The tendency is to further rediscover themethodology, that is, the study designs, and the resultsof past studies.

A basic problem for epidemiology is the lack of asystematic approach to acquiring new knowledge toreach a goal of improving public health and preventivemedicine. Our research support unfortunately is biasedtoward the continued study of already proven hypothe-ses. A new hypothesis for which there is a lack of sub-stantial prior data is unlikely to be successful in termsof peer review.

Replication by investigators, both believers and non-believers of a hypothesis, of studies among differentpopulations is obviously the pillar of good epidemio-logic research. However, circular epidemiologyfreezes at one point in the evolution of epidemiologicstudies, failing to move from descriptive to analyticalcase-control and longitudinal studies, for example, toexperimental, clinical trials. Instead, investigatorsoften identify a population or a methodology and thengo round and round in circles. Specific errors or diffi-cult, insoluble methodological issues are suddenlyrediscovered. The new study repeats the errors of pre-

897

by guest on September 19, 2015

http://aje.oxfordjournals.org/D

ownloaded from

898 Kuller

vious studies. Fortunately, after a period of learning,the investigators rediscover the errors and then amaz-ingly reach the truth. Sometimes, investigators maynot rediscover the methodological problems, which ismore worrisome. They continue to use the same inap-propriate methods, leading to probably erroneousresearch conclusions. Their questionable results thensuddenly become dogma and are the basis for newhypotheses that generate even more studies.

Good epidemiology journals are filled with verywell-conducted epidemiologic studies that primarilyrepeat the obvious or are variations on the theme. Howmany studies have to show a relation between smokingand lung cancer? How many studies do we need toshow that weight gain is a risk factor for diabetes? Ahigh level of low density lipoprotein (LDL) cholesterolis clearly a risk factor for heart attack. Early age at firstpregnancy is associated with a reduced risk of breastcancer. Should we be continuing to report observa-tional studies of the relation between specific vitaminsand risk of cancer or cardiovascular disease, whenongoing experimental clinical trials are testing thehypothesis about whether the same specific vitaminsare related to disease? We do the public a disservicewhen we publish a paper saying that a specific risk fac-tor is or is not related to a disease, while at the sametime that investigator is testing the hypothesis in clini-cal trials (2).

There is likely an excess expenditure of resources, interms of both personnel and money, for these circularepidemiologic studies. These costs drain the assetsneeded to move the field forward, that is, to go fromobservational to experimental to application of epidemi-ology as a basis for public health action and practice.

Certain aspects of cardiovascular epidemiologyprobably represent some of the best examples of theproblems associated with circular epidemiology.There have been extraordinary advances in under-standing the etiology and prevention of coronary heartdisease and stroke. Epidemiologic studies have playedan extremely important role in determining the riskfactors and pathogenesis of atherosclerosis and themethods for reducing the incidence of coronary heartdisease (3). Since the 1960s, cardiovascular mortalityhas fallen more than 50 percent in the United States aswell as in many other countries (4). Much of thisdecline is certainly related to the reduction and treat-ment of selected risk factors.

The association of blood cholesterol levels anddietary intake of cholesterol and saturated fat with riskof coronary artery disease began as a result of animallaboratory studies and clinical epidemiologic studiesof genetic familial hypercholesterolemia conductedprior to World War n. Descriptive epidemiologic stud-

ies by Keys demonstrated the striking geographic vari-ations in coronary heart disease mortality in relation todietary saturated fat and to total cholesterol levels inthe 1950s (5). The International AtherosclerosisProject further documented the geographic variation inatherosclerosis among countries (6) and its associationwith cardiovascular risk factors. Animal and short-term human feeding experiments in the 1960s showedthe relation of elevated blood cholesterol levels withdietary intake of specific-chain-length saturated fats,cholesterol, and polyunsaturated fatty acids (7). Thesestudies resulted in equations such as those by Keys etal. and Hegsted of the relation of the nutrients withblood cholesterol levels (7, 8). Numerous subsequentfeeding studies have verified the Keys et al. andHegsted equations.

Longitudinal studies such as the FraminghamStudy documented the association between highblood cholesterol levels and blood pressure levels,smoking, and risk of coronary heart disease as earlyas the 1950s, 40 years ago (9). The National Diet-Heart Study clearly demonstrated the association ofdiet with cholesterol levels (10). This study alsoshowed that changes in blood cholesterol levels inrelation to diet were a function of adherence to theexperimental diet. The reduction in blood cholesterollevels was greater in closed than in open populations(10). International migrant studies, such as those con-ducted in Japan, Hawaii, and California, also docu-mented that an increase in saturated fat and cholesterolin the diet resulted in high blood cholesterol levels andrisk of coronary heart disease (11).

Criticism of the diet-heart hypothesis in the 1960swas based on data from the Framingham Study thatfailed to show an association between diet, as mea-sured by various instruments in observational studies,and either blood cholesterol levels or coronary heartdisease mortality (12). In a seminal paper, Jacobs et al.demonstrated that difficulties in measuring dietaryintake, and the limited variability of blood cholesterollevels in homogeneous human populations, resulted ina null relation in spite of the substantial, proven exper-imental association of dietary variables, blood choles-terol levels, and atherosclerosis (13). The Oslo StudyDiet and Antismoking Trial documented that a reduc-tion in saturated fat and cholesterol in the diets of per-sons with high blood cholesterol levels resulted indecreased blood cholesterol levels and risk of coronaryheart disease (14). Other dietary trials focusing on bothreducing saturated fat and increasing polyunsaturatedfat demonstrated reductions in the risk of coronaryartery disease. In the United States, there was a sub-stantial decrease in consumption of cholesterol primar-ily from eggs and saturated fat, specifically a decrease

Am J Epidemiol Vol. 150, No. 9, 1999

by guest on September 19, 2015

http://aje.oxfordjournals.org/D

ownloaded from

Circular Epidemiology 899

in butter and whole milk products and an increase inmargarine consumption, and a substantial decrease inblood cholesterol levels and coronary heart diseasemortality (15). Subsequent clinical trials including,most recently, the very powerful Statin Drug Trialshave clearly documented the great benefits of loweringLDL cholesterol levels to reduce the risk of new andrecurrent heart attacks and the extent of atherosclero-sis, and at LDL cholesterol levels that include almost50 percent of the US population (16).

In a circular epidemiology fashion, a new crop ofstudies in the 1980s and 1990s rediscovered the lack ofan observational association between diet and coronaryheart disease morbidity and mortality. The studies hadall of the pitfalls previously identified by Jacobs et al.(13) and others. Not surprisingly, these studies overtime ultimately concluded that, yes, saturated fat, cho-lesterol, and polyunsaturated fat are important determi-nants of heart disease. The problems with these studieshad been noted for generations. The inability to mea-sure individual exposure to a common source, in thiscase diet, within a homogeneous population had beenrecognized from the studies by Goldberger on pellagra(17), in which the primary units of comparison werecomprised of groups of persons. Investigation of acommon source was the method used in prior studies ofair pollution and disease. The unit of measurement wasoften a city or geographic area rather than a specificindividual exposure. In occupational epidemiologicstudies, which also examine common-source exposure,the unit of measurement is often a class of workers, asthese studies have a limited ability to measure individ-ual exposure within specific groups of workers. Thestudies by Keys et al. (7) were based on the methodsused to study common-source epidemics. They com-pared nutrient intake and blood cholesterol levels inpopulations and the subsequent risk of heart attack. Theinvestigators needed a well-defined marker of exposurewith less variance within persons as compared tobetween persons, that is, body burden.

The use of ecologic correlations not based on soundbiologic plausibility can result in fallacious associa-tions, often referred to as the ecologic fallacy. Keys, onthe other hand, tested the hypothesis of a possiblecommon-source epidemic related to specific dietarynutrients that was based on sound animal experimentaland human high-risk, genetic models; he and othersthen tested their hypothesis in longitudinal studiesamong these populations that varied in their exposureto the common-source nutrients and then in nutrition-experimental studies that firmly established the diet-heart hypothesis (5).

Many investigations of common-source epidemicstherefore depended on the next step to prove the

hypothesis, experimental epidemiology. The experi-ment could be a true randomized trial or often was anatural experiment, such as determining changes in airpollution levels in a community (18), conductingmigrant studies, or studying changes in manufacturingprocesses. Unfortunately, the model of investigatingcommon-source exposures within persons moved tocancer. In spite of numerous studies over many years,there still is no consistent evidence of an associationbetween any major nutrients and cancer risk. This cer-tainly does not mean that the nutrients are not related tocancer but rather that the case-control or longitudinalstudies of the association of nutrients with chronic dis-ease and cancer have been conducted inappropriately inpopulations in which the nutrient intake is homoge-neous, thus resulting in erroneous interpretation of thedata. Recently, epidemiologists have rediscovered theapproaches to studying common-source epidemics andhave called for more "macro epidemiology" (19).

Previous studies also stressed that a low blood cho-lesterol level and decline in blood cholesterol overtime is often a manifestation of chronic disease, espe-cially liver disease, infections, and cancer. Suddenly,however, investigators rediscovered low cholesterollevels (20). The adverse effects of low cholesterolbecame a new research topic. Again, little new infor-mation has been generated over time, and the likeli-hood that lowering the blood cholesterol level causes aperson to be killed in an accident seems rather remote(21). The basic problem with these studies is the lackof recognition of the comorbidity, namely, the associ-ated chronic diseases and other health behaviors, asnoted in the Commission on Chronic Illness studies ofthe 1950s (22).

Some epidemiologists decided that studying thediet-heart hypothesis was rather dull and that theyneeded to find new risk factors. Not surprisingly, theyfound that people who swallowed all kinds of pills,from beta-carotene to vitamin E to garlic, had areduced risk of heart attack. It is likely that at leastsome of these pills have some effect on coronary arterydisease, although the study design used to measurespecific pill consumption might have been severelylimited by selection bias. Previous studies clearly doc-umented that adherence to almost any therapy is asso-ciated with a decreased risk of disease. Canner andothers, in the Coronary Drug Project conducted in the1970s, showed that good adherence to consumingplacebos was associated with a substantial reduction inthe risk of coronary heart disease death (23). Onlywell-conducted clinical trials will substantiate whetheran association exists between risk and benefit regard-ing many of these agents, often considered the "pill ofthe week." It is probably still too early to drink herbal

Am J Epidemiol Vol. 150, No. 9, 1999

by guest on September 19, 2015

http://aje.oxfordjournals.org/D

ownloaded from

900 Kulier

tea while eating quarter-pound hamburgers withcheese on buns sprinkled with folic acid and lycopene-laced ketchup.

Another problem related to circular epidemiology isthat epidemiologists appear to have forgotten that mostcausal associations are generalizable beyond the studypopulation(s) (1). They began to argue that reportedassociations were limited to White men only, espe-cially those in Framingham, Massachusetts, and not towomen, minorities, or older people. They capitalizedon this observation by continually replicating studiesby age, race, and sex groups. Much to the surprise ofsome but hopefully not most people, their results wereconsistent across age, race, and sex groups.

These epidemiologists found that results of clinicaltrials that did not include every race or ethnic group,women, or older age groups would have to be repli-cated. This approach obviously had major ethicalimplications, since it required a "placebo" high-riskgroup of people of a specific race or ethnicity or ofwomen. It is very unlikely that most interventions arespecific to an entire race or ethnic group.Epidemiology is the study of host susceptibility. Therewill always be large variations in response to an expo-sure but, as noted, not by a whole class of persons.Within any population, response to therapy certainlywill vary in relation to the existing genetic polymor-phisms and other comorbid conditions and to other riskfactors. We forget that one of the false arguments forconducting the Tuskegee experiment was that treat-ment of syphilis with penicillin would be different forBlacks than for Whites, which again defines circularepidemiology.

Can you conceive of repeating a successful vaccinetrial with Blacks because the original was conductedprimarily with Whites, for example, the polio trials?Can we identify any specific, successful interventions,except those for which there clearly is a fairly obvioussex-specific or racial or ethnic association (e.g., hor-mone therapy to prevent breast cancer in women ortreatment of sickle cell anemia), whose effects havebeen limited to specific race and ethnic groups or tomen and women? Unfortunately, attempted replicationof well-conducted studies and trials according to raceand ethnic groups only delays implementation of goodpublic health practices and often is a detriment ratherthan a help to the populations at risk.

Failure to recognize the incubation period of diseasealso often leads to misinterpretation and circular epi-demiology. For example, the lack of an association ofrisk factors such as blood cholesterol level with heartattacks among older persons has generated the hypoth-esis that blood cholesterol levels are not an importantrisk factor for older persons (24). These investigators

failed to appreciate the incubation period for develop-ment of atherosclerosis, a marker of subclinical dis-ease, and the association of measures of subclinicaldisease with cardiovascular risk factors. If, during theincubation period, development of atherosclerosis is afunction of the level and duration of exposure to cer-tain blood cholesterol levels, then, in older persons,lower blood cholesterol levels and a longer duration ofexposure to these levels will result in atherosclerosis,for example, higher blood cholesterol levels, and ashorter duration of exposure will lead to atherosclero-sis at a somewhat younger age.

Infectious disease epidemiologists clearly recognizethe importance of the incubation period. Few epidemi-ologic studies of chronic diseases have defined theincubation period of disease, its relation to the risk fac-tors, and, in particular, measurements of subclinicaldisease (25). If a risk factor is measured before onsetof clinical disease, the study is prospective. The patho-physiologic changes, those caused by the subclinicaldisease, are not recognized and are assumed to be acause of the clinical disease. We have an outpouring ofpapers in which markers of inflammation, endothelialfunction, hormonal changes, and psychologicalchanges are identified as causes of disease rather thanas consequences of "subclinical disease." Neither can-cer nor atherosclerosis has a 1- or 2-year incubationperiod.

The classic occupational epidemiologic studies rec-ognize the importance of the incubation period in rela-tion to the defined exposure (26). Measurements ofrecent exposures are not likely to predict disease. Doesit make sense for epidemiologic studies to demonstratethat blood cholesterol level is not related to risk ofheart disease in older persons, even though we knowthat blood cholesterol level is an important determi-nant of atherosclerosis and that atherosclerosis is amajor pathophysiologic process leading to clinicalcoronary artery disease? The results of clinical trials(i.e., experimental epidemiology) are consistent withthe fact that reduction of LDL cholesterol levels inolder persons results in a decreased risk of clinical car-diovascular disease (16, 24).

Epidemiologic studies have suddenly begun toswitch their enthusiasm to genetics. They estimate that60-70 percent of a specific lipoprotein or disease isdue to "genetic factors" (27). Unfortunately forgottenare the tremendous international and national varia-tions in diseases and risk factors and the results of themigrant studies and of clinical trials and natural exper-iments. In infectious disease epidemiology, specificmeasures of host susceptibility have been inferred butnot identified clearly, and epidemiologists have recog-nized that genetic variations in both the host and the

Am J Epidemiol Vol. 150, No. 9, 1999

by guest on September 19, 2015

http://aje.oxfordjournals.org/D

ownloaded from

Circular Epidemiology 901

agent can contribute to "cause of disease." The specificreason that only one in a thousand persons exposed tothe poliomyelitis virus developed clinical disease isstill not understood but, at least in part, must havesome genetic basis. Nevertheless, the development ofthe polio vaccine was remarkably successful. Host sus-ceptibility was considered a continuum from subclini-cal disease (infection and immunity) to clinicaldisease (paralysis) to bulbar polio and finally death(28).

The need for good measures of host, agent, and envi-ronment has come full circle in the generation of newgenetic epidemiologic studies of chronic diseases. Inthis "new epidemiology," better definitions of host sus-ceptibility, for example, genotype, can be determined.The interrelations between environmental exposuresand specific agents have often been lost. Thus, we havegone around in a circle, from studies in which we couldmeasure the environment and the agents fairly well byusing relatively crude determinations of host suscepti-bility to studies that attempt to define in great detail thegenotype and host susceptibility but not measures ofenvironment and specific agents.

Are we now at the threshold of being able to providea genetic host susceptibility scorecard for each personthat will define the exact dose of a specific agent thatthe person can consume per day (e.g., 80 vs. 100 mEqof sodium/day) to prevent elevated blood pressure?Too bad that we cannot measure the agent (sodium) orthe outcome (individual "blood pressure") with theprecision required for this careful definition of a hostsusceptibility scorecard. Failure to recognize the broadgradient of susceptibility to most agents and of envi-ronmental exposure to disease is generating a newround of studies that may delay implementation ofeffective preventive medicine programs.

Should Salk have waited for the specific host-susceptibility genetic markers to be identified beforedeveloping polio vaccines? Similarly, is it necessary toidentify the specific genes associated with elevatedLDL cholesterol before intervening successfully at theindividual and population level to reduce LDL choles-terol levels and the risk of atherosclerosis? We mustavoid going around in a circle from prior good studiesof agents and environment and poor measures of hostsusceptibility to good measures of host susceptibilityand poor measures of agent and environment. The pub-lic health approach still has the greatest impact onreducing morbidity and mortality.

To survive as an important discipline, epidemiologymust be the basic science of preventive medicine andpublic health (29). The emphasis in epidemiologicresearch must be to link the causal chains to provide ascientific foundation for preventive programs (30).

Epidemiology cannot succeed by describing newmethods of analyzing the same types of data sets or byrediscovering the important social ills of society (31).

Epidemiologic studies should cover the breadthfrom phase I exploratory studies, often called descrip-tive epidemiology, to phase IV clinical trials and eval-uation of the application to policy and programs.Public health and preventive medicine not based onsolid science, especially epidemiology, is very likely tobe unsuccessful and even hazardous to the populationsinvolved. Taking the fruit-, vegetable-, vitamin-, ortoxic-chemical-of-the-month-club approach does notbenefit public health.

On the other hand, epidemiology that does not focuson how it applies to public health and preventive med-icine will not be supported in the future. The need todefine the implications of an epidemiologic study inpublications should be encouraged rather than discour-aged. Epidemiologic papers should clearly describehypotheses and implications of results in terms of thedirection of future studies and the potential importanceof results to the practice of public health, prevention,and clinical medicine. Papers that include only caveatsabout more limitations of the implications of theirstudies should receive a very low priority. The role ofa current study in the evolving process of epidemiol-ogy, from descriptive to experimental epidemiology,should be clarified. In addition, the paper should pre-sent the future direction of epidemiologic research, notjust replicate the current study in different population,race, and sex groups.

Alternative hypotheses and criticism of existingdogma clearly should be encouraged. Epidemiologicstudies should be based on sound scientific methodol-ogy and not just replication of previous errors.Previous study design problems that have already beensolved should be documented carefully in the newstudies. Replication of the same methodological prob-lems and errors is circular epidemiology and will notmove the field of epidemiology forward.

There is a very substantial risk that erroneous resultsof epidemiologic studies based on repetition of poormethodologycircular epidemiologywill be used todeter effective public health and preventive medicineprograms. For example, industry has been very effec-tive in using observational epidemiologic studies thathave failed to demonstrate a relation between theamount of salt in the diet, especially from processedfoods, and blood pressure levels to prevent a reductionin the amount of salt in processed foods (32).

Epidemiology is in part dependent on new technol-ogy and concepts of pathophysiology provided byother disciplines. The use of new technology mayovercome some of the limits of epidemiologic studies,

Am J Epidemiol Vol. 150, No. 9, 1999

by guest on September 19, 2015

http://aje.oxfordjournals.org/D

ownloaded from

902 Kuller

and these studies should not hesitate to include infor-mation on new advances in scientific methodology.Examples include molecular genetics, noninvasivemethods of measuring atherosclerosis and cancer, andnew techniques of measuring hormone levels, growthfactors, immune function, clotting, and thrombosis.Epidemiologic studies can also point other scientificdisciplines in new directions (33).

We should continually monitor the successes andfailures of epidemiologic studies in improving the"public health." The heros of epidemiology identify aproblem (descriptive epidemiology), develop methodsto test specific hypotheses (observational epidemiol-ogy), experiment (clinical trials) to prove or disprove ahypothesis, and then apply good public health and pre-ventive medicine strategies to enable the informationthey have acquired to be used to reduce morbidity andmortality. They do not stop at one point in their searchfor information and replicate the same study over andover again. Their solution is not to develop moreextensive regression analysis that adds yet anothervariable to a long list of variables or to define a newmethod that might reduce the variance and change therelative risk from 2.3 to 2.4 but rather to succeed indocumenting the benefits of their research in terms ofimproving the public's health.

The future of epidemiology is very bright if we con-tinue to stress that it is an important basic science ofpreventive medicine and public health (34-40) and thatthe study designs that link host, agent, and environmentare as important today as they were in the past. Newtechnologies do not change the basic concepts of epi-demiology but just improve the methodology and studydesigns. However, epidemiology will be in trouble ifwe continue to do the same things over and over again,that is, to practice circular epidemiology.

REFERENCES

1. Smoking and health: report of the Advisory Committee to theSurgeon General of the Public Health Service. Washington,DC: Public Health Service, 1964. (Public Health Service pub-lication no. 1103).

2. Holmes MD, Hunter DJ, Colditz GA, et al. Association ofdietary intake of fat and fatty acids with risk of breast cancer.JAMA 1999;281:914-20.

3. Kannel WB, Doyle JT, Ostfeld AM. Optimal resources for pri-mary prevention of atherosclerotic diseases. AtherosclerosisStudy Group. Circulation 1984;70:157A-205A.

4. National Heart, Lung, and Blood Institute. Morbidity and mor-tality: 1996 chartbook on cardiovascular, lung, and blood dis-eases. Bethesda, MD: National Institutes of Health, 1996.

5. Keys A. Seven countries. A multivariate analysis of death andcoronary heart disease. Cambridge, MA: Harvard UniversityPress, 1980.

6. McGill HC Jr. The geographic pathology of atherosclerosis.Baltimore, MD: Williams & WilMns, 1968.

7. Keys A, Anderson JT, Grande F. Predictors of serum choles-terol responses of man to changes in fats in the diet. Lancet1957;2:959-66.

8. Hegsted DM. Unanswered questions. In: Proceedings from theScientific Conference on the Effects of Dietary Fatty Acids onSerum Lipoproteins and Hemostasis. (Sponsored by theNutrition Committee and the Council on Thrombosis,American Heart Association). Washington, DC, May 1989.

9. Dawber TR. The Framingham Study. The epidemiology of ath-erosclerotic disease. Cambridge, MA: Harvard UniversityPress, 1980.

10. The National Diet-Heart Study final report. Circulation1968;37:I-1-1-428.

11. Robertson TL, Kato H, Rhoads GG, et al. Epidemiologic stud-ies of coronary heart disease and stroke in Japanese men livingin Japan, Hawaii and California. Am J Cardiol 1977;39:244-9.

12. Mann GV. Diet-Heart: end of an era. N Engl J Med1977;297:644-50.

13. Jacobs DR Jr, Anderson JT, Blackburn H. Diet and serum cho-lesterol: do zero correlations negate the relationship? Am JEpidemiol 1979; 110:77-87.

14. Hjermann I, Holme I, Leren P. Oslo Study Diet andAntismoking Trial. Results after 102 months. Am J Med 1986;80:7-17.

15. Ernst ND, Sempos CT, Briefel RR, et al. Consistency betweenUS dietary fat intake and serum total cholesterol concentra-tions: the National Health and Nutrition Examination Surveys.Am J Clin Nutr 1997;66:965S-72S.

16. Miettinen TA, Pyorala K, Olsson AG, et al. Cholesterol-lowering therapy in women and elderly patients with myocar-dial infarction or angina pectoris: findings from theScandinavian Simvastatin Survival Study (4S). Circulation1997;96:4211-18.

17. Terris M. Goldberger on pellagra. Baton Rouge, LA: LouisianaState University Press, 1964.

18. Higgins ITT. Effects of sulfur oxides and particulates onhealth. Arch Environ Health 1971;22:584-90.

19. Susser M. Does risk factor epidemiology put epidemiology atrisk? Peering into the future. J Epidemiol Community Health1998;52:608-ll.

20. Jacobs D, Blackburn H, Higgins HM, et al. Report of theConference on Low Blood Cholesterol: Mortality Asso-ciations. Circulation 1992;86:1046-60.

21. Lindberg G, Rastram L, Gullberg B, et al. Low serum choles-terol concentration and short term mortality from injuries inmen and women. BMJ 1992;305:277-9.

22. Kuller L, Tonascia S. Commission on Chronic Illness follow-up study. Comparison of screened and nonscreened individu-als. Arch Environ Health 1970;21:656-65.

23. Influence of adherence to treatment and response of choles-terol on mortality in the coronary drug project. N Engl J Med198O;3O3:1038^1.

24. Corti MC, Guralnik JM, Salive ME, et al. Clarifying the directrelation between total cholesterol levels and death from coro-nary heart disease in older persons. Ann Intern Med1997; 126:753-60.

25. Armenian HK, Lilienfeld AM. The distribution of incubationperiods of neoplastic diseases. Am J Epidemiol 1974;99:92-100.

26. Monson RR. Occupational epidemiology. 2nd ed. Boca Raton,FL: CRC Press, Inc., 1990.

27. National Heart, Lung, and Blood Institute. Report of the expertpanel on genetic strategies for heart, lung, and blood diseases.Bethesda, MD: National Heart, Lung, and Blood Institute,1993.

28. Evans AS. Causation and disease. A chronological journey.New York, NY: Plenum Medical Book Company, 1993.

29. Shy CM. The failure of academic epidemiology: witness forthe prosecution. Am J Epidemiol 1997; 145:479-84.

30. Frost WH. The age selection of mortality from tuberculosis insuccessive decades. Am J Epidemiol 1995;141:49.

31. Rothman KJ, Adami HO, Trichopoulos D. Should the missionof epidemiology include the eradication of poverty? Lancet

Am J Epidemiol Vol. 150, No. 9, 1999

by guest on September 19, 2015

http://aje.oxfordjournals.org/D

ownloaded from

Circular Epidemiology 903

1998;352:810-13.32. de Wardener HE. Salt reduction and cardiovascular risk: the

anatomy of a myth. (Commentary). J Hum Hypertens 1999;13:1-4.

33. Morris JN. Uses of epidemiology. 2nd ed. Baltimore, MD:Williams & Wilkins, 1964.

34. Hutchison GB. Experimental trial and program review. In:Clark DW, MacMahon B, eds. Preventive and communitymedicine. 2nd ed. Boston, MA: Little, Brown & Co., 1981.

35. Stolley PD, Lasky T. Investigating disease patterns. The sci-ence of epidemiology. New York, NY: Scientific AmericanLibrary, 1995.

36. McGinnis JM, Foege WH. Actual causes of death in the UnitedStates. JAMA 1993;270:2207-12.

37. Rose G. The strategy of preventive medicine. New York, NY:Oxford University Press, 1992.

38. Stamler J. Lectures on preventive cardiology. New York, NY:Grune and Stratton, Inc., 1967.

39. Wynder EL, Graham EA. Tobacco smoking as a possible eti-ologic factor in bronchiogenic carcinoma. A study of sixhundred and eighty-four proved cases. JAMA 1950;143:329-36.

40. Levin ML, Goldstein H, Gerhardt PR. Cancer and tobaccosmoking. A preliminary report. JAMA 1950;143:336-8.

Am J Epidemiol Vol. 150, No. 9, 1999

by guest on September 19, 2015

http://aje.oxfordjournals.org/D

ownloaded from