Random survival forest is a popular machine learning tool for modeling censored survival data. In this talk, we give a brief introduction to this model and discuss several theoretical results and recent applications to personalized medicine. First, we show that the current methodologies of random survival forests are potentially biased in the sense that the splitting variables may be falsely selected due to censoring. This is mainly because of the marginal splitting rule, which does not fully utilize the conditional independence properties. Based on this intuition, we then introduce two models for personalized medicine, where two potential treatments are being considered. The first model is based on a randomized trial setting where we impute the censored observations to its conditional expectations and utilize the outcome weighted learning approach to solve the optimal treatment decision rule. The second approach is under an observational study setting, where we calculate a double robust conditional expectation of the conditional survival time after adjusting for the propensity score. This allows us to provide a valid confidence interval of the estimated treatment effect. Future research directions are discussed.