EPIDEMIOLOGIC EVIDENCE AND CAUSAL INFERENCE - 04/09/11
Résumé |
Fifty years ago, a paper entitled Observation and Experiment appeared in the New England Journal of Medicine.13 Its author, the British-born medical statistician Austin Bradford Hill, 15 traced the development of passive observational study designs that had characterized medical science during the nineteenth and early twentieth centuries and contrasted their key characteristics with a new study design that coupled an active intervention with random allocation. This new approach was called the randomized controlled clinical trial.12 First used in agricultural research to evaluate the efficacy of fertilizers, the randomized trial was the product of decades of biostatistical thinking.5, 16 One of the earliest published examples in medicine appeared in the British Medical Journal in 1948. The trial was sponsored by the Medical Research Council (MRC) of Britain20; Hill was its Director of Statistics. Since then, the randomized trial has flourished as a methodology in investigations of therapeutics and has been used increasingly for prevention studies. During this same period, the randomized trial's not-so-distant cousins—observational case-control and cohort studies—became firmly established as the primary methodologic tools of the modern science of analytical epidemiology.22, 28
Hill is credited with bringing the randomized trial methodology into the forefront of Western biomedical science, and he also holds a special place in the history of epidemiology.19 In 1965, about 15 years after his early efforts to promote the randomized trial, he published an influential paper on causal inference.14 Hill's approach to this key problem in public health and preventive oncology grew out of a conversation in the public health literature traced back to the 1950s30 and is closely related to the methodology used in the 1964 Report of the Surgeon General declaring cigarette smoking a cause of lung cancer.26 Hill discussed how several criteria can be used to judge the presence of a disease-causing agent or exposure from available scientific evidence. Hill's criteria-based approach is, even several decades later, the central methodologic approach used to interpret causation from scientific evidence.
Causal inference methodology—as it has evolved from Hill's now-classic 1965 paper—is the primary focus of this article. Other interpretative methods have been added to Hill's criteria-based method, including meta-analysis and a systematization of the narrative review process within which Hill's criteria typically appear. Any method of causal inference will also have connections with theories of disease (e.g., cancer) causation, the logic and epistemology of causal hypotheses, and the ethics of preventive interventions.31 In the interest of space, these more philosophical concerns are not addressed here. Nevertheless, the practitioner of causal inference should remember that causal judgments can be—and almost always will be—influenced by underlying philosophical beliefs and by personal and institutional values. There is good reason, in other words, to think of causal assessments as judgments.27
The focus of this article is more practical than philosophical. The practice of causal inference is described. Cancer epidemiology is one—perhaps the only—discipline within which this practice has been systematically examined.32 Judging causation from scientific evidence is a common practice among cancer epidemiologists, preventive-oriented physicians, and public health professionals alike. Individuals take up this challenge when they are asked to review the literature for publication in journals or textbooks. Causal inference is also an important goal of organizations charged with making public health policies, such as the Institute of Medicine, the Office of the Surgeon General, and the Environmental Protection Agency. The International Agency for Research on Cancer, with its long history of published assessments of chemical and environmental carcinogens, is another institution for which causal inference is a central concern.29
Le texte complet de cet article est disponible en PDF.Plan
| Address reprint requests to Douglas L. Weed, MD, PhD, Office of Preventive Oncology, Division of Cancer Prevention, National Cancer Institute, 6130 Executive Boulevard, Suite T-41, Bethesda, MD 20892–7105, dw102i@nih.gov |
Vol 14 - N° 4
P. 797-807 - août 2000 Retour au numéro
Article précédent
- HYPOTHESIS TESTING IN CLINICAL TRIALS
- Sylvan B. Green
- PRIMARY CANCER PREVENTION TRIALS
- Ernest T. Hawk, Scott M. Lippman
Bienvenue sur EM-consulte, la référence des professionnels de santé.
L’accès au texte intégral de cet article nécessite un abonnement.
Déjà abonné à cette revue ?