In the development of a new therapy, it is common that late phase trials offer the first opportunity for in-depth investigation of potential safety issues of the new therapy, given the larger number of patients being studied in these trials than in early phase trials. There is often, if not always, much interest in identifying subgroups of patients with elevated risk, as evidenced by large number of pre-specified as well as post hoc analyses regarding subgroups routinely performed by both trial sponsors and regulatory reviewers. In addition, these analyses frequently form the basis for warnings for, or even restrictions in, patient populations within drug labels. We believe that both prospective and retrospective analyses play useful roles in risk assessment of patient subgroups, and recent developments in statistical methodology can improve both types of analyses. In this presentation, I will describe a framework to systematically utilize both prospective and retrospective risk assessment in late phase trials. Specifically, the proposed approach seeks to leverage new statistical methods to address multiplicity issues and effectively identify patient subgroups of interest. Our goal is to perform rigorous risk assessment that enables proper interpretation of findings and sensible decision-making.