Institut de radioprotection et de sûreté nucléaire (IRSN) BP n° 6 92262 Fontenay‐aux‐Roses cedex <philippe.hubertirsn.fr>
- Key words: risk assessment ; environmental health ; public health ; environmental exposure ; health status indicators
- Page(s) : 266-78
- Published in: 2003
The health impact associated with each of a variety of environmental hazards and lifestyles has been estimated in diverse efforts to assess and compare their risks and help define public health priorities. These estimates calculate the various components of what the High Commission for Public Health calls "avoidable mortality", a concept that groups together the deaths associated with human behaviors and activities that could be modified. The sum total is high, reaching one third of all deaths annually. We begin by noting that this operation, the addition of the "number of attributable deaths" is generally wrong, because it entails contradicting the models used to construct the estimates. We then describe the variety of approaches used to reach these estimates. Our heterogeneous inventory includes direct or indirect counts from national death statistics, from survey data, and from predictions by models based on uncertain knowledge. Next, we note the heterogeneity of approaches, which range from fairly simple observations to very complex models. This diversity engenders skepticism that on occasion results in a refusal to put these differences into perspective, or even to undertake the estimation. That is not the point of view we adopt here. On the contrary, we provide information that makes it possible to discuss the confidence that should be accorded to estimates. We start by characterizing the estimation process and, more precisely, the inferences it includes : extrapolation from high to low doses, transpositions from animals to humans or from one group of humans to a different population, and analogies between effects, substance and exposure routes. This description enables us to discuss the importance of these inferences and to deduce from them a qualitative, perhaps even subjective, assessment that characterizes the uncertainty associated with them. The contrast between situations then becomes clearer : some estimates are calculated for situations "close" to those on which the models are based, while other call for greater extrapolations, more distant analogies, or more daring transpositions. A procedure is proposed to systematize this critical reading : applied to an ever increasing number of datasets, it should lead to the gradual development of references and bases for judging the plausibility of the number of deaths attributed to various causes. It has already enabled us to avoid simplistic judgments of the credibility to accord to risk estimates.