Mesurer l’impact du travail sur la santé : du longitudinal, oui, mais lequel ?

Abstract

This paper intends a systematic comparison of the performances of the “naive” epidemiologic models, explaining the prevalence of health issues through existing characteristics of workers. This would be compared to more rigorous models, that will include a history of some former professional exposures (retrospective static models), or will study the incidence (instead of the prevalence) of the disorders according to either the exposure from the initial date (longitudinal standard models), or the evolution of the exposure (dynamic models).We first show the weakness of the impact of the individual factors of confusion such as the consumption of alcohol and tobacco, living conditions or medical history. However, the correlations between health and employment characteristics highlighted by the naive models remain practically unmodified in the more thorough models. That doesn’t mean that these factors are without effects on health. On the contrary, the analysis shows the importance of some of these effects. But the effects of the professional and the personal factors act in a largely independent way.We then examine the respective merits of “standard” and “dynamic” models. The first explain the incidence of health disorders between two dates by the exposure from the initial date. The second take the evolution of the exposure between the two dates as explanatory variables. Concerning the (infra)pathologies related to stress, the “standard” models appear under-efficient, insofar as they clearly underestimate the impact of the exposures on health disorders. This lack of efficiency is probably due to the greatest reversibility of the disorders in the case of disappearance of the exposure, a phenomenon which the “standard” model confuses with negative correlation between the exposure and the disorder.

Keywords

• statistical models in epidemiology
• health and working conditions
• individual factors of confusion