Doctor: Emmanuel Caruana
Title: Development of a new equilibrium measure to aid variable selection in a propensity score model
Supervisors: Sylvie Chevret, Romain Pirrachio
Doctoral school: ED 393 Epidemiology and Biomedical Information Sciences, Université Paris Cité
Date of thesis defense: 03/2017
Jury: Romain Pirracchio, Sylvie Chevret, Gilles Potel, Yohann Foucher, Denis Frasca, Enrique Casalino, Didier Journois.
Propensity score (PS) methods have become increasingly used to analyze observational data and take into account confusion bias in final estimate of treatment effects. The goal of the PS is to balance the distribution of potential confounders across treatment groups. The performance of the PS strongly relies on variable selection in PS construction and balance assessment in PS analysis. Specifically, the choice of the variables to be included in the PS model is of paramount importance. In order to priorize inclusion and balance of variables related to the outcome, a new balance measure was proposed in this thesis. First, a new weighted balance measure was studied to help in construction of PS model and to obtain the most parsimonious model, by excluding instrumental variables known to be related with increasing bias in final treatment estimate. Several balances measures are proposed to assess final balance, but none of them help researchers to not include instrumental variables. We propose a new weighted balance measure that takes into account, for each covariate, its strength of association with the outcome. This measure was evaluated using a simulation study to assess whether minimization of the measure coincided with minimally biased estimates. Secondly, we propose to apply this measure to a real data set from an observational cohort study.