In many observational studies, a natural quantity to estimate is the difference in mean survival time between two treatments. Unfortunately, individuals choosing one treatment may be prognostically different that those choosing the other treatment. In this talk we will review the assumptions required to estimate the average causal effect. We present a weighted estimator that is equivalent to the Kaplan-Meier estimator and extend the idea of inverse-probability-of-treatment-weighted (IPTW) estimators to the setting of censored data. Using simulated data, we will compare the properties of this estimator to the true underlying survival distributions.
Return to Biostatistics Working Group