Cette thèse, qui s'inscrit dans la tradition méthodologique de l'épistémologie historique, porte sur l'histoire et la formation du concept de biais dans l'épidémiologie moderne. Elle montre que la fonction opératoire du concept de biais est essentiellement critique, au sens où ce concept, que les épidémiologistes opposent au cours de l'histoire aux concepts d'objectivité, de preuve et de causalité, joue un rôle décisif dans la constitution de l'épidémiologie comme science, mais aussi dans l'avènement d'une médecine scientifique. Un éclairage historique et critique est apporté à la définition actuelle du biais, conçu comme une erreur ou un écart systématique par rapport à la vérité, ainsi qu'aux différentes taxinomies des biais qui jalonnent l'histoire de ce concept, dont l'origine se situe chez les fondateurs de la statistique mathématique. Le biais apparait ainsi comme une menace aussi bien à la validité du plan d'expérience d'une étude épidémiologique qu'à la validité de l'inférence statistique et du raisonnement médical. En d'autres termes, ce sont les conséquences que la révolution probabiliste a eues sur l'épidémiologie et sur la médecine qui sont ici étudiées, et qui ont conduit les épidémiologistes et les médecins à une forme de scepticisme et même de criticisme envers leurs propres inférences, ce qui donnera naissance au mouvement de la médecine fondée sur des preuves. / This PhD thesis, belonging to the tradition of historical epistemology, deals with the history and the formation of the concept of bias in epidemiology. It shows that the operational function of the concept of bias is essentially critical, in the sense that this concept, used by epidemiologists throughout history as an antonym to both objectivity, causality and evidence, is central to both the construction of epidemiology as a scientific discipline and the advent of scientific medicine. An historical and critical account is given of the actual definition of bias, conceived as a systematic error or deviation from the truth, and to the various taxonomies of bias which marked the history of this concept, whose origin goes back to the founders of mathematical statistics. Bias thus appears as a threat to the validity of the design of an epidemiological study, and to the validity of statistical inference and medical reasoning. In other words, what is studied here is the consequences of the probabilistic revolution on both epidemiology and medicine, which led epidemiologists and physicians to a kind of scepticism or even criticism about their own inferences, which would ultimately give birth to the evidence-based medicine's movement.

The thesis consists of an introduction and two papers. In the introduction a brief historical survey of empirical investigations into the effectiveness of medicinal interventions is given. Also, the main ideas of the EBM (evidence-based medicine) movement are presented. Both included papers can be viewed as investigations into the reasonableness of EBM and its hierarchies of evidence. Paper I: Typically, in a clinical trial patients with specified symptoms are given either of two or more predetermined treatments. Health endpoints in these groups are then compared using statistical methods. Concerns have been raised, not least from adherents of so-called alternative medicine, that clinical trials do not offer reliable evidence for some types of treatment, in particular for highly individualized treatments, for example traditional homeopathy. It is argued that such concerns are unfounded. There are two minimal conditions related to the nature of the treatments that must be fulfilled for evaluability in a clinical trial, namely (1) the proper distinction of the two treatment groups and (2) the elimination of confounding variables or variations. These are delineated, and a few misunderstandings are corrected. It is concluded that the conditions do not preclude the testing of alternative medicine, whether individualized or not. Paper II: Traditionally, mechanistic reasoning has been assigned a negligible role in standard EBM literature, although some recent authors have argued for an upgrading. Even so, mechanistic reasoning that has received attention has almost exclusively been positive -- both in an epistemic sense of claiming that there is a mechanistic chain and in a health-related sense of there being claimed benefits for the patient. Negative mechanistic reasoning has been neglected, both in the epistemic and in the health-related sense. I distinguish three main types of negative mechanistic reasoning and subsume them under a new definition of mechanistic reasoning in the context of assessing medical interventions. Although this definition is wider than a previous suggestion in the literature, there are still other instances of reasoning that concern mechanisms but do not (and should not) count as mechanistic reasoning. One of the three distinguished types, which is negative only in the health-related sense, has a corresponding positive counterpart, whereas the other two, which are epistemically negative, do not have such counterparts, at least not that are particularly interesting as evidence. Accounting for negative mechanistic reasoning in EBM is therefore partly different from accounting for positive mechanistic reasoning. Each negative type corresponds to a range of evidential strengths, and it is argued that there are differences with respect to the typical strengths. The variety of negative mechanistic reasoning should be acknowledged in EBM, and presents a serious challenge to proponents of so-called medical hierarchies of evidence. / <p>QC 20150413</p>