A causal framework for the estimation of attributable risks from aggregate data
vendredi 5 mars 2021, 9h30 - 10h30
The identification of the determinants of a disease is usually based on individual data from epidemiological surveys, such as cohort or case-control studies. In addition to the usual measures of association (relative risk, odds ratio) which help to identify risk factors for the disease, a measure of interest is that of attributable risk, which assesses the proportion of cases that would be prevented if exposure to the risk factor could be eliminated. In the context of public health surveillance, the attributable risk is often estimated from aggregate data, for which the association between the exposure and the disease must be modelled.
We first look at some simple estimators of the attributable risk from aggregate data, then propose a general framework derived from causal inference theory to define a counterfactual estimator, which can be shown as equivalent to a difference in quantiles between the estimated and interventional distributions. Finally, we explore how this estimator can be applied to Hawkes processes used for the modelisation of contagious diseases.