Classics in epidemiology: should they get it right?

Extract

One of the best ways to learn epidemiology is through reading the classic texts. Excellent anthologies of such classics have been published, and Snow on cholera, Goldberger on pellagra, Doll and Hill on smoking and lung cancer, Morris on exercise and coronary heart disease, and Keys and Stamler on cholesterol and heart disease are tremendous models for how to get things right. In the International Journal of Epidemiology we have reprinted several early (and often under-appreciated) papers in which novel empirical or methodological advances, that now would be considered central to epidemiological thinking and methodology, were advanced.^1–14

There is a natural tendency to apply asymmetrical criteria to papers that reached what, with hindsight, were correct or incorrect conclusions. In the former case we look for the exemplary design, analysis and interpretation, and in the latter case for the flaws that should have been detected at the time. There is, however, at least as much to be learnt from studies that have reached what now appear to be the wrong conclusion as those that got it right. In many cases these will have been carried out to the highest contemporary standards, and yet somewhere along the line the authors were misled. In this issue of the IJE we reprint a highly cited and influential meta-analysis and systematic review of observational studies on hormone replacement therapy (HRT) and coronary heart disease risk, first published in 1991.¹⁵ This paper is a model of clarity, yet with the findings of randomized controlled trials (RCT) of HRT the conclusion it reached—that the apparent protective effect of HRT was ‘unlikely to be explained by confounding factors’—appears wrong. A series of commentaries^16–21 debate why this situation arose. While the commentators are not unanimous in their opinions it is, at the very least, clear that observational epidemiology may be more fallible than some have suggested. Other examples of apparently convincing findings from observational studies failing to be confirmed though RCTs exist with respect to associations between several dietary factors, in particular anti-oxidant vitamins, and disease.²² However, in only a small minority of cases can the findings of observational epidemiological studies be compared to those from randomized trials. Surely assuming that the various factors that mitigated against a correct interpretation of the effect of HRT or anti-oxidants on disease do not apply in these cases is not a sensible way to proceed?