PhD student: Florie Bouvier
Title: Estimation of individualized treatment effects using an individual participant data meta-analysis (temporary title)
Supervisors: Raphaël Porcher
Doctoral school: ED 393 Epidemiology and Biomedical Information Sciences, Université Paris Cité
Personalized medicine aims at tailoring treatment to the individual characteristics of each patient.
A key aspect of personalized medicine is to identify subgroups of patients who benefit from one intervention, or more from one intervention than another. Several approaches exist to determine these subgroups, in particular the estimation of the individualized treatment effect (ITE). The ITE represents the expected treatment effect for an individual with specific characteristics and is estimated using predictive models or machine learning methods.
Most prediction models for estimating the ITE are developed using data from a single randomized controlled trial. However, these randomized trials are generally undersized for these purposes (they are sized to test the treatment effect in the whole study population), which can lead to over-fitting of the models or, on the contrary, the failure to capture the effect of relevant variables. One solution is the use of meta-analyses on individual data (IPD-MA), which both provide data from a larger number of subjects and improve the generalizability of results. In a first project, we investigated the methodology of ITE estimation in an IPD-MA. In particular, we investigated the performance of different methods accounting for the heterogeneity between the studies included in the meta-analysis, combined with two methods for estimating ITE.
Identifying subgroups of patients who benefit from an intervention enables the development of individual treatment rules (ITR). In the last few years, many machine-learning methods have been proposed to create such rules. However, it is not clear to what extent these methods lead to the same ITRs, i.e. whether they recommend the same treatment to the same people. In a second project, we compared ITRs created from commonly used methods in two randomized clinical trials: the International Stroke Trial and the CRASH-3 trial.
In order to develop efficient rules, ITRs need to be able to correctly distinguish patients who benefit from taking a certain treatment from those who don’t i.e. the rules need to discriminate well. In my latest project, we’re interested in which distributions of treatment effects enable good discrimination through a simulation study involving various scenarios.