This is the project I did during my intership at Eli Lilly in the summer of 2021. The problem is about a method to leverage historical data to improve decison making in early phase clinical trials. The method in question integrates two methodologies: power priors and propensity scores. The first is a Bayesian method that can dinamycaly weights previous information to avoid the, typically much larger sample size, historical information dominating the inference of the current study. The weaknes is that it does not takes into account the covariates different distribution, at least not in the version of the power prior I am refering in this project. On the other hand propensity scores with matching are desing to take into account different distributions of the covariates in the historical and current studies, however it discards a lot of the information from the historical study.
The goal of this project is to assess if a method that uses both methodologies will earn more of their strengths than its weaknesses and thus be more robust to umeasured confounders. Please go throught the slides and shoot me an email if you want to talk/learn more about it.