Journal of Clinical Epidemiology 54 (2001) 325–331
Reflections on the limitations to epidemiology
George Davey Smith*
Department of Social Medicine, University of Bristol, Canynge Hall, Whiteladies Road, Bristol BS8 2PR, UK Received 4 July 2000; received in revised form 9 August 2000; accepted 30 August 2000
1. Introduction Epidemiology provides a good example of the dictum that the smaller and more specialised a particular field is, the more ferocious are the internal battles fought by the initiates. Over the past few years the “epidemiology wars” have intensified considerably, in ways which, I am sure, would have entertained and engaged Petr Skrabanek’s interest and elicited further contributions from him. There are several separate dimensions to critiques of current epidemiological practice, with diametrically opposed views being expressed on several key issues by different commentators. As is often the case, however, the various criticisms are sometimes linked in ways that suggest they are mutually reinforcing, when on occasion they point in different directions. The apparent crisis in epidemiology is, in fact, a series of crises—although most commentators have been engaged with the particular aspect of epidemiology’s malaise that interests them most. The failings of epidemiology that have been identified in recent critiques embrace most aspects of the epidemiological enterprise. First, the propensity of “risk factor epidemiology” to become the indiscriminate identification of particular aspects of daily life as dangerous to health has been widely condemned. Second, critics censure the willingness of epidemiologists to accept “black box” associations between exposures and disease as serious contenders for causal relationships, with no understanding of biological mechanisms. Third, many preventive medicine campaigns—regarding diet, drinking, smoking, recreational drugs, forms of sexual behaviour, tanning, sloth, obesity— are seen to be driven more by moralism than by science. Epidemiology is, from this perspective, not value free, but a tool used to support predetermined goals. Fourth, epidemiology as it is currently practiced is seen to be excessively
concerned with individual risks and inadequately engaged with the social production of disease. Why should problems within a small academic discipline—a subspecialty within public health, itself a minor part of the health and health care enterprise—be of interest to other than professionals earning their living within the field? Epidemiology certainly elicits emotional responses. In his recent pop-history The Rise and Fall of Modern Medicine , the medical journalist James Le Fanu went so far as to suggest that the solution to the fall of his title was to close all departments of epidemiology. If it would work, this certainly sounds like a cost-effective approach to problems within medicine (except to those of us who would be looking for alternative employment opportunities). Can epidemiology really be responsible for the current malaise in the medical enterprise? The suggestion that it might is certainly helped by the high profile epidemiology receives in the popular media. In Britain the broadsheet daily newspapers tend to pick one medical story out of the weekly medical journals (the BMJ and the Lancet), which are published on a Friday. More often than not the highlighted story will be an epidemiological one—some aspect of daily life has been shown by epidemiologists to be bad for people. 2. Risk factor epidemiology A major component of Petr Skrabanek’s attack on epidemiology relates to what he termed “risk-factorology”. This is the practice of performing studies that relate a myriad of potential exposures to disease risk and identifying those that are positively associated with disease as being “risk factors” for the condition (and those that are negatively linked as being protective factors). Although it is continuously pointed out (by the risk-factorologists among others) that association does not mean causation, the major reason for doing such studies is to identify causes of disease. There is not a great deal of excitement in showing that cabbage consumption is linked to a lowered risk of cancer if it were not thought that by encouraging people to eat more cabbage you would lower the burden of cancer in the population . A particular weakness of risk factor epidemiology is that different investigators often report contradictory findings
This paper is adapted from the third Skrabanek lecture, given at Trinity College, Dublin on 26th March 1999 in honour of the late Petr Skrabanek. The discussion of the limits of epidemiology focused on Petr Skrabanek’s work, given the nature of the occasion. A collection of Petr Skrabanek’s papers “False premises, False promises” (Whithorn: Tarragon Press, 2000) contains an earlier and shorter version of the lecture. * Corresponding author. Zetkin@bristol.ac.uk
0895-4356/01/$ – see front matter © 2001 Elsevier Science Inc. All rights reserved. PII: S0895-4356(00)00 3 3 4 - 6
6] is partly explained by the above reasoning. The situation is probably worse than even this reasoning suggests. For cause of death C the relative rates were 2. and measure a wide range of potential outcomes. the men who smoked 1–19 cigarettes per day (less than 1 pack per day) had 1.05 level” finding.05.5 times that of nonsmokers.e.. Without previous positive associations having been reported. Remember that this example relates to a case where 1 in 10 hypotheses is meaningful. especially when each potential association can be looked at within many subgroups (men.e.000
from such studies [3–5]. which shows steadily increasing risk with increasing smoking.000 associations are examined and that 90% of the associations that are tested reflect hypotheses that are not valid (i.05. the men who smoked 40–59 cigarettes (2 to less than 3 packs per day) had 2. The published epidemiological literature certainly contains reports of more positive associations than null results—an example of “publication bias. We consider the situation where the power of studies that are carried out is 0. several false-positive findings are virtually guaranteed [3. the men who smoked 20–39 cigarettes per day (1 to less than 2 packs per day) had 1.D.3 (2–3 packs). for cause of death A. those with high blood pressure. 2.e. with your knowledge of the health consequences of smoking. but little increase after this.7 (3 or more packs). with a significance level set at 0.). A second set of associations can be uncovered that does exist in the real world.326
G. which causes of death do you think are represented by A. If only the positive findings are published then they may be mistaken for meaningful links between exposure and disease. For cause of death B the equivalent relative rates were 1. This is not a chance findings. These supposed “findings” do not reflect links in the outside world. 450 out of 9000 times the statistical test will have wrongly rejected the null hypothesis (which is referred to as a type I error). which also shows a steadily increasing risk with increasing smoking. since the nature of many epidemiological studies means that a great magnitude of potential associations can be examined. where 50% of the time the null hypothesis (of no association) is rejected (i. it is assumed that there is a real effect).5.5 (2–3 packs). it almost always seems.3 (1–2 packs).4 times the death rate of nonsmokers (i. as by conventional reasoning examining 20 associations will find one “statistically significant at the P . which means that there is a 50% chance of detecting real effects if they exist. it has been reported from a series of studies  and. the young. not as the necessary chance associations which they are.. and 2. 1. it is suicide . The bottom line.” whereby only apparently exciting findings reach the light of day. but to a final risk about 3. 2. the elderly.000 associations. Cause of death A. those with normal blood pressure. Since many studies contain long questionnaires collecting information on hundreds of variables. As Petr Skrabanek points out. This demonstrates the association between cigarette smoking and the risk of dying from three causes of death in a follow-up study of a third of a million men in the U. is coronary heart disease. but contains associations that may have a different meaning than is conventionally attached to them. perhaps a genuine cause is being identified?) while negative findings may be controversial and thus of interest . which shows an increase in risk from nonsmokers to less than a pack a day smokers.6]. By this means alone a host of purely chance findings will be published. to a final risk about 2. those with existing disease.5. B. with the baseline (relative risk 1) being for nonsmokers and a relative risk of 2 indicating twice the risk of dying for the group that experiences it.05 effects commonly reported. this apparent weakness of the method can be turned to an advantage. The actual situation is probably less favourable than this. the level of statistical significance is
. 2. At the conventional significance level of P . are cases when the null hypothesis should be accepted). they are constructed by the epidemiologist’s methods of examining data and reporting results. The percentage of apparent findings that are “false positives” will depend on the proportion of associations that are examined that are meaningful associations. Now. The vertical axis represents the relative risk of dying.5 times that of nonsmokers.5. 2.4 times the risk.S. however. In other words 450/ 950 times—on nearly half of the occasions—apparently statistically significant findings are type 1 errors .000 Null hypothesis false 500 Type 2 errors 500 1.3 times the risk and the men who smoked 3 packs a day or more had 3. etc.3 (1–2 packs). unlike the P 0. is stroke. since 450 divided by 9000 0. The near-random nature of some epidemiological associations that have been identified in the literature [3. of the 950 cases where the null hypothesis is rejected. 40% higher rate). Imagine that 10. Smith / Journal of Clinical Epidemiology 54 (2001) 325–331 Table 1 Hypothetical outcome of examining 10. is not (as you may have thought) lung cancer. and 2. 500 times it is rejected correctly and 450 times it is rejected incorrectly. To illustrate this I would like the reader to inspect Fig. those without existing disease.
women.0 (less than a pack). Thus. This is reflected in the “null hypothesis false” column.000 in the table). B and C? C. Thus.9 times the risk.05 and a power of 50% How the world is Decision Accept null hypothesis Reject null hypothesis Total Null hypothesis true 8550 450 Type 1 errors 9. is that “more research is needed”—a conclusion comforting to epidemiologists working in the field.4 (3 or more packs).5 (less than a pack). studies finding no association between a potential risk factor and a disease are unlikely to be of great interest. Table 1 represents the scenario in which the proportion of hypotheses tested in epidemiological studies that reflect real associations is 1 in 10 (1000 out of 10. since once a positive association has been identified further positive studies are noteworthy (replication being the hallmark of science..
during which they shoot smokers. 60 3 or more packs per day). but unfortunately it is of limited use. depression) but also the variables that are measured (e. in an observational epidemiological study yellow fingers would be associated with smoking and a naïve interpretation would consider yellow fingers as a potential cause of lung cancer (Fig. this association is very unlikely to be causal.g. The conventional exposition of confounding refers to the cause of death not included in Fig. The problems of chance and particularly confounding are exacerbated when the exposure–disease associations are weak. in this case. In theory this sounds an excellent solution. 1. 2). as the causal mechanism between exposure and disease is unknown and may even be considered irrelevant . The “black box” nature of risk factor epidemiology may be particularly evident where low-level risk are being investigated. and this had almost no effect on the finding (the relative risk for three or more packs a day changed from 3. What is likely to be happening here is a mixture of confounding and reverse causation.D. In our example. Exposure A (e. Black boxes Petr Skrabanek considered risk factor epidemiology to be a form of “black box” thinking. exposure B (yellow fingers) is associated with exposure A — and hence to the outcome. Certainly to Petr Skrabanek epidemiology without biology was nonsense. all of which are potential confounders. 20–39 1 but less than 2 packs per day. Relative rate of three causes of death in relation to smoking (0 nonsmoker. 3.G. 1—lung cancer.g. if people who are depressed. 40–59 2 but less than 3 packs per day. Smith / Journal of Clinical Epidemiology 54 (2001) 325–331
Fig. 2. smoking suppressing brain serotonin levels) have been advanced. Smoking causes people to acquire yellow fingers. leaving wide scope for residual confounding to exist after statistical adjustment .3 on such adjustment). thus. smoking) is associated with the outcome (lung cancer risk). since it is here that it may stumble upon novel findings of public health importance . 1–19 less than 1 pack per day.
.... Not only are there missing variables (e. he himself moved from neurotoxicology research (particularly relating to substance P) to his particular blend of “natural scientist. poor.4 to 3. reduce their risk of suicide (indeed for depressed people who smoke to relieve stress it may increase their risk). income) are measured with error. since a small risk applied to a large proportion of the population can have important public health implications .
considerably more stringent ( p 0. conversely. In the example of smoking and suicide we controlled for a variety of other risk factors (income.” leaving an estimate of the “independent effect” of the risk factor of interest. with a relative rate of more than 10 among smokers of two packs of cigarettes a day . The strategy of epidemiologists to deal with confounding is generally statistical “adjustment” or “control. Unless the provisional wing of the health education lobby has moved on to a direct action phase.g. By confounding we mean that the potential risk factor of interest—smoking—is linked to another factor that in turn is a cause of the outcome. doctor of medicine and connoisseur of the absurd” (as the footnote to one of his papers read ). Lung cancer shows a much stronger (and certainly causal) dose–response association with cigarette smoking. Many epidemiologists. smoking will be noncausally associated with suicide. Taking away the cigarettes from smokers will not. To address this we looked at an even less plausible cause of death—homicide. chronic disease). However. Given the strength of antismoking feeling in public health—considered below in the section on epidemiological
moralism—and also the justified outrage at the tactics of the tobacco companies in covering up the genuine health-damaging effects of smoking.g. feel that the future of epidemiology must lie in a rapprochement with biomedical science. alcoholic or chronically ill are more likely to be smokers then. Biological mechanisms through which this could act (e.0001). forensic toxicologist. Some epidemiologists consider that epidemiology is most useful when it ventures into territories unconstrained by biological knowledge. investigating these associations could be
Fig.. ethnicity. as these states increase the risk of suicide. it is perhaps not surprising that even though the paper reporting the smoking–suicide association was subtitled “illustration of an artefact from observational epidemiology”  it has been widely taken to illustrate that smoking does indeed cause suicide. Smokers of two packs or more per day were more than twice as likely to be murdered as nonsmokers.
which then hampers the entry of HIV into the uterine cavity. making it a less successful barrier to HIV. The author of this study was. and the epidemiological work has enormously stimulated this now vibrant field of biological enquiry . The authors again found a biologically supported mechanism for this effect: oral contraceptives thicken the cervical mucus. The epidemiologists should inform.” Another moral duty emphasised by public health is the duty to be well. Health as duty can lead to unattainable aims. they have no right to tell people what they should do. they had no old age. however. and dementia are some of the final rewards. they become
.” Certainly. most states have been much concerned . as though in sleep. but equally epidemiology without rapprochement with biology can produce long lists of spurious associations and make no contribution to furthering our understanding of disease aetiology and strategies for disease prevention. found the exact opposite: a protective effect of oral contraceptives on HIV transmission . If they start advising government on “population measures. such as the World Health Organisation definition of health as “Not merely the absence of disease or infirmity” but “A state of complete physical and mental and social well-being. Not everybody hopes for a long life followed by death from boredom. The advantage of installing the habit of expecting state interference in (at worst. dependence. Biological knowledge of the effects of oral contraceptives support both these possibilities. if it feels good. Here initial ecological associations between indices of poor early life development earlier this century and current disease rates indicated the possible importance of in utero and early childhood growth for later health.D. or when high on drugs. Health and longevity are.328
G. incontinence. “Health for all by the year 2000” remains a bad joke for the large majority of the population of the world.” they declare themselves as agents of social control. 4. now that we have reached the new millennium and most people still live in poor and unhealthy places. Strictures against smoking. Thus. Plato in The Republic recalled the gymnastic teacher Herodicus whose skills enabled him to reach old age in a prolonged death struggle. An example of the malleable nature of biological plausibility is demonstrated by two epidemiological studies that appeared within a few weeks of each other in 1991. Although there was some existing basic research on fetal development and later health this was limited. smoking. it is of concern that only a small degree of measurement error in a confounder can generate residually confounded effects of the magnitude under consideration for low-level radiation. however. of Kenyan prostitutes. Why be afraid of sudden death from coronary heart disease if you cannot regret it the day after? The function of epidemiology is to study associations and to provide hypotheses for experiments. To quote Petr Skrabanek: There are fates worse than death. Poverty. of course. Hesiod’s golden race died swiftly. These ecological studies were followed by analytical epidemiological studies showing that at an individual level low-birthweight and other indices of suboptimal fetal development were related to cardiovascular disease risk profiles in adults. sloth and obesity [5. Smith / Journal of Clinical Epidemiology 54 (2001) 325–331
of potential importance. electromagnetic fields around power lines or mobile phones and cancer. In many cases these associations may also be considered biologically implausible . although it must be recognised that what is biologically plausible and what is not changes over time. only self-damaging) aspects of private life and private behaviours—through the Trojan horse of beneficence—is also not lost on those who want their leadership accepted. often diet. oral contraceptive use was independently related to risk of future infection with HIV . Petr Skrabanek quoted a study of cervical cancer which suggested that masturbation was a risk factor .21]—and thus the focus of health improvement schemes end up endorsing the message “if it tastes good. forms of sexual behavior. The authors suggested several mechanisms through which oral contraceptives could facilitate HIV transmission. Clearly apparent “biological plausibility” adds little strength to epidemiological conviction.” The targets of a mainstream strand of epidemiology are. loneliness. don’t eat it. The second study. drinking. tanning. not the only aims of life. historically. In one study. capable of producing an apparent biological mechanism: the pathways from the nervous system to cervical tissue could promote cancer if stimulated .13. this is “the sort of feeling ordinary people may achieve fleetingly during orgasm. recreational drugs. as mentioned. Mencken’s comment that hygiene—the public health of its day—”is the corruption of medicine by morality. It is impossible to find a hygienist who does not debase his theory of the healthful with a theory of the virtuous. a fitter army and a more fertile population: things with which. This example—of epidemiology and basic research feeding off each other—illustrates the weaknesses of
both “black box” epidemiology and of attempts to deny that epidemiology can contribute in a fundamental way to our understanding of disease. L. Public health moralism Petr Skrabanek used to like quoting H. or an immunosuppressive action that increases susceptibility to HIV. Longevity drags if you have buried your children.” In Petr Skrabanek’s words . for example. An important recent example of how epidemiologists and biologists working together have produced important advances in understanding both epidemiological and biological aspects of disease is seen in the “fetal programming” field . As one of his examples of moralism misinterpreted as science. don’t do it. including a direct effect on the genital mucosa. drinking and sloth can also be ways to ensure a more productive workforce.
on a whole host of characteristics of their lives. Both nondisclosure of interests or censorship of viewpoints by suppressing publications by those with such interests prevent readers getting an informed and rounded view of the issues. he considered that the “fact that these achievements have had little or no bearing on the lives of all those millions of our fellows which are still ‘nasty.000 Reichsmark donation Hitler made to the Institute for the Struggle against the Dangers of Tobacco supported the first methodologically strong epidemiological study showing that smoking causes lung cancer [24. Throughout the long history of the war against smoking—from James I pamphleteering against tobacco. With respect to public health he considered that “the issues of preventive medicine have little to do with science. it has been shown that the expressed viewpoints of experts regarding the beneficial effects (or otherwise) of calcium-channel antagonists is strongly related to the financial relationships the authors had with manufacturers of these drugs . However. poor. and risk factors. Transparency in this regard recognises the impossibility of separating facts and values. could produce high-calibre science is a demonstration of the impossibility of reading the truth or falsity of claims from the values underlying them. Others have strongly taken issue with this view . however . for example. Rothman’s belief that scientists can be objective. whether psychotherapy (known as “moral treatment” in the past) or somatic treatment predominates as the ruling paradigm depends less on therapeutic results than on the dominant ideology of the profession. He was unwilling to allow this admission of the important influence of social circumstances on health to influence public policy decisions. through to Hitler’s vigorous opposition to smoking—the combination of puritanism and self-interest seem clear .25]. Indeed. Kenneth Rothman. Epidemiology as an asocial science There has been considerable recent debate regarding the individualistic focus of much epidemiology on the lifestyles or psychological profiles of people abstracted from their social context [31–33]. These authors—in my view correctly—point out that there are broader social determinants of the risks to health that people suffer. For example.D. The fact that researchers working under an appallingly inhumane dictatorship. is remarkably similar to the relative risk of 16. has suggested that forcing the declaration of conflicts of interest is akin to McCarthyism in science.
5. solitary and short’ is an indictment of our selfish world” . and objectivity is not something which is brought or sold. They could be more profitably debated within the framework to which they belong—ethics. politics. 3) . Take. beta-carotene increased cardiovascular disease risk (see Fig. Believing that these differences could be summed up. there is also strong evidence that financial interests influence what people say. However. a leading epidemiological methodologist.G. and vested interests” . the extensive research on beta carotene consumption and the risk of cardiovascular disease. psychological or therapeutic—from the complexity of the life and
. is an opinion I feel Petr Skrabanek would have shared. of 16. Perhaps less widely acknowledged is that the abstraction of such associations from their particular context can lead to severely misleading conclusions. What these examples have in common is that the groups of people who were apparently receiving protection from these substances in the observational studies were very different from the groups not using them. brutish. the impossibility of separating ethics or morality from science is highlighted by realising that the 100.6 . My own view is that transparency—through full declaration of potential sources of conflicts of interests—is the best policy in
such cases . The problematic interpenetration of facts and values has recently been highlighted in the epidemological world in a debate regarding whether or not potential conflicts of interest should be declared. but long-term randomised controlled trials found that.3 among male heavy smokers that can be calculated from the classic Doll and Hill case-control study . Similar self-interest is. It would be gratifying if the refutation of observational studies by randomised controlled trial evidence in these areas led to a critical evaluation of approaches that abstract single elements—which are almost always behavioural. of course. relative risks. if anything. vitamin E and vitamin C intake in relation to cardiovascular disease among them. Smith / Journal of Clinical Epidemiology 54 (2001) 325–331
preventionists for whom the interests of the state override the interest of the individual . There are now a series of similar examples: hormone replacement therapy. As Petr wrote with respect to prevailing views in psychiatry. The frequent concordance of morality and what passes at any particular time for the science of public health must surely be more than coincidence. during which the most basic elements of human morality and decency were abandoned wholesale. shown by those who profit from the tobacco trade and deny the evident harmful effects of smoking. political and social context have been convincingly elaborated on [31–33. Petr Skrabanek recognised the important role of poverty in determining health—referring to the fact that medicine brought many benefits. The many weaknesses of epidemiological approaches that fail to locate exposure–disease associations within their historical. Observational studies revealed strong apparently protective effects of beta carotene. and risks mislabelling everyone as potentially unethical .35]. in measures of a few “potential confounders” and adequately adjusted for in statistical analyses fails to recognise the complexity of the reasons why people differ with regard to particular and general characteristics of their lives. the relative risk of lung cancer amongst smokers that can be calculated from the 1943 German study. and that attempts to reduce these risks should recognise this fundamental social determination.
a clear demonstration that the cleansing act of criticism is required as much today as ever. A collection of 56 topics with
. In my view epidemiology will only progress if it combines a detailed understanding of the ways through which the historical. Our conscientious objection to the epidemiology wars.  Mayes LC. health-related behaviours and mood). Risk-factor epidemiology: science or non-science? In: Anderson D.37] does not abstract individual life trajectories from these contexts. 1999. editor. Petr Skrabanek’s self-termed “destructive criticism” can feed into the process of formulating better questions and developing the necessary methods for answering these questions.D.”  An illustration of the importance of such criticism is provided by the fact that the British newspapers have recently been full of reports—based on a recent Lancet paper — that breast cancer screening with mammography may not be a useful preventive strategy. modes of self-presentation. . J Epidemiol Community Health 1998. Horwitz RI. New York: Little Brown. Fifteen years prior to this paper Petr Skrabanek provided a devastating critique of breast cancer screening. of crumbled clay from the legs of the clossal cacademics carrying the dogma on their shoulders. . indeed in a Lancet editorial he dismissed a methodological paper I coauthored about improving the design of epidemiological studies —which made the modest proposal that more rigorous measurement of expo-
Acknowledgments Many thanks to Liz Humphries and Claire Diamond for helping in preparing this manuscript and to Nancy Krieger and John Lynch for comments on an earlier version. that as in many decaying research programmes auxiliary hypotheses will be mobilised to explain each apparent “mistake. indeed.47–56. London: Social Affairs Unit. Smith / Journal of Clinical Epidemiology 54 (2001) 325–331
Fig. . lifestyle and environment. Meta-analysis of results of observational cohort studies of beta-carotene intake and cardiovascular mortality and of randomised controlled trials of the same issue. however. Low birthweight. 6.33]. and relate these to single health outcomes. lung function. 1991. which then allowed the building of useful knowledge. Others have been similarly (but less entertainingly) dismissive of attempts to improve epidemiological methodology . social and geographical location. the long-term and sometimes irreversible outcomes of social circumstances at different stages of life are seen to become literally embodied .52:613–4. may seem to fall within different categories.
times of people. also in the Lancet . The rise and fall of modern medicine. It is likely.  Le Fanu J.” on a case-by-case basis.330
G. economic and political constitution of how the world is influences the health of populations—and thus the individuals within these populations—with the appropriate development of methodology and concepts to deal with this complexity [32. Such a lifecourse approach [36. Adapted from . Health.
 Poole C. He saw such criticism as a necessary cleansing act.
sures on smaller samples facilitates the production of meaningful findings than poorer exposure measurements on larger samples—by quoting the Irish country saying “you cannot make a pig grow by weighing him” . growth in childhood (and final adult height). and ways of dealing with misfortune. . Feinstein AR. of dust from the “dry-run” research. A critical appraisal of a medical theory takes so much time and effort that most people (and some of them wise) do not bother. persistent infections acquired in early life (such as H pylori) or the failure to acquire infections (leading to immunological programming increasing the risk of atopy). degree of adiposity. An epidemiology that appreciates the necessary interconnections between these different domains of life will avoid the dead-ends that research strategies based on abstraction and narrowing the focus of attention can lead to. In my view a more promising approach is to see the development of health potential within its particular concrete historical. but they are all essential components of life trajectories that influence health. Petr Skrabanek was not interested in playing this game. however. of debris left by those who would rather publish than perish in the rat race.  Skrabanek P. Criticism and the advance of knowledge Some recent writings about the “limits of epidemiology” have raised ways in which we could do things better than we have in the past. Rothman KJ. 3. a habitus which embraces particular dispositional characteristics (including attitudes. It is like an archaeological dig— the slow painstaking search for nuggets of truth by removing layer after layer of deposits left behind by fly-by-night scholars.
False promises and false premises of breast cancer screening.  Pearce N. Epidemiology and the web of causation: has anyone seen the spider? Soc Sci Med 1994. 338:101–6. Egger M. editors. Plummer FA.  Smith R. 2000. BMJ 1952. 1992.II:1271–87. Am J Public Health 1967. JAMA 1993. Lynch JW. Skrabanek P.  Najman JM. Hill AB.342:1502. Davey Smith G.173–83. 310:396. Rose G. Detsky AS. Carpenter KM. Adami HO.88:216–22.54:261–9.  Skrabanek P. Piot P. Cameron DW. Lungenkrebs and Tabakverbrauch.39:887–903.  Doll R.57:815–29. Waiyaki P.355:129–34. Aust N Z J Public Health 2000.  Stelfox HT.  Koopman JS. 1986.317:291–2. Perspect Biol Med 1992.41:577–82. Egger M. Ndinya-Achola JO.48:220–3. Z Krebsforschung 1943. Egger M. McCormick J.
. Nicolosi A. Chichester: Wiley. Davey Smith G. In: Leon D. p. Life in the womb: the origin of health and disease. Davey Smith G. Oxford: Oxford University Press. Arch Intern Med 1991. J Clin Epidemiol 1988. babies and health in later life. Phillips AN. Schöniger E.i:143–4. Lancet 1985. Other people’s smoke. New York: Promethean Press.5:550–2. Transparency is the key.5: 553–5. Ben-Shlomo Y.151:2411–6. Skrabanek P. in press)
   
  
   
       
 Davey Smith G. Smith / Journal of Clinical Epidemiology 54 (2001) 325–331 contradictory results in case-control research. Sexual characteristics of a cervical cancer population. Davey Smith G.D. Bringing context back into epidemiology: variables and fallacies in multilevel analysis.  Kuh D. Davey Smith G. Nathanielsz PW. A study of the aetiology of carcinoma of the lung. In defence of destructive criticism. Smoking and society. The new McCarthyism in science.51:643–6. Cofactors in male-female sexual transmission of human immunodeficiency virus type 1.24:3–4. Oxford: Oxford University Press. Barker DJP.163:233–9. Ben Shlomo Y. 1991. J Epidemiol Community Health 1994. Lancet 1986. J Infect Dis 1991.  Rothman JK. Oakes M. Berkeley: Galen Press. The epidemiology of errors. London: The Social Affairs Unit. Schairer E. Follies and fallacies in medicine.  Krieger N. The poverty of epidemiology.  Diez Roux AV. The embodiment of class-related and health inequalities: Australian policies. Musicco M. Cheang M.  Rothman KJ. Spurious precision? Metaanalysis of observational studies.30. Trichopoulos D. Saracco A.340:709–12. N Engl J Med 1998. Ströbele S. Skrabanek P. The emptiness of the black box.  Davey Smith G.89:1175–80. Lazzarin A.  Gøtzsche PC. 18:725–7.352:810–3.46:1203–11. Smoking and health promotion in Nazi Germany. editors. Statistical inference. McKinlay JB. Phillips AN. Back to the future in epidemiology and public health: response to Dr Gori. Ronald AR. Man-to-woman sexual transmission of the human immunodeficiency virus. 3rd ed.  Skrabanek P. BMJ 1998. BMJ 1998.316:140–4. Withorn: Tarragon Press. Davey Smith G. Is screening for breast cancer with mammography justifiable? Lancet 2000. 35:182–5.” Int J Epidemiol 2001. BMJ 1992. 305:757–9. Epidemiology 1994. Ströbele SA. A life course approach to chronic disease epidemiology. Chua G. Oxford: Oxford University Press. BMJ 1995.269:2782–4. Should the mission of epidemiology include the eradication of poverty? Lancet 1998. Lancet 1993. London: Churchill Livingstone. Smoking as “independent” risk factor for suicide: illustration of an artifact from observational epidemiology? Lancet 1992. Int J Epidemiol 1989. The design of prospective epidemiological studies: more subjects or better measurements? J Clin Epidemiol 1993. Gakinya MN. Kreiss JK. editor. Mothers. 1999. (translated as “Lung cancer and tobacco consumption. Skrabanek P. Phillips AN.ii:316–20.313:929–33. Walt G.  Schwartz S. Cervical cancer in nuns and prostitutes: a plea for scientific continence. Smoking and death. O’Rourke K.  Davey Smith G. Gunnell D. Health inequalities and public policy. Olsen O. Confounding in epidemiological studies: why “independent” effects may not be all they seem. Skrabanek P. Am J Public Health 1999. In defense of black box epidemiology. The right answer for the wrong question: consequences of type III error for public health research. Am J Public Health 1998.  Skrabanek P. Perspect Biol Med 1986. J Clin Epidemiol 1998. 15–41. Am J Public Health 1999. Lifecourse approaches to socio-economic differentials in cause-specific adult mortality. Skrabanek P. Scheider M. Skrabanek P. 1998. 1997. Rotkin ID.89:1170–4. Conflict of interest in the debate over calcium-channel antagonists. In: Armitage A. Simonsen JN. Individual causal models and population system models in epidemiology. 1994. Preventive medicine and morality. The death of humane medicine.G. Conflict of interest. Beyond conflict of interest. 1998. Epidemiology 1994.30:19–26. The strategy of preventive medicine.  Phillips AN. Passive smoking and health: should we believe Philip Morris’s “experts”? BMJ 1996. Neaton JD. Savitz DA.