Online Program

Dynamic Prediction and Causal Inference

Sangbum Choi, University of Texas MD Anderson Cancer Center 
*Xuelin Huang, MD Anderson Cancer Center 
Jing Ning, University of Texas MD Anderson Cancer Center 

Keywords: Causal inference, dynamic prediction, survival analysis, time-dependent covariates

Causal inference and prediction are closely related. At any time point during a patient’s follow-up visits after treatment, we need a good prediction of future disease prognosis, so that they can make decisions on whether or not to initiate extra treatments or interventions at that moment. This prediction/decision should use not only the baseline information, but also all the information, such as patient response, up to the time point of making prediction/decision. Often the models used for such prediction/decision have a causal interpretation. A difficulty is that this kind of real-time dynamic prediction models are usually very complicated. In this talk, we first try to provide some parsimonious models with strong assumptions, and then gradually relax them to be more realistic.