Online Program

Identify Interactions and Distinct Risk Groups of Time-to-Event Data Using Survival Tree Approach (302813)

Keywords: survival tree, time-to-event, interaction

The majority of conventional studies focus on the main effects of factors in time-to-event or survival data modeling. Evaluating interactions or combinations of multiple factors may help increase the prediction power of modeling. For a time-to-event outcome, the Cox regression is the most common method for modeling. However, Cox regression is not flexible in defining sub-groups with a distinct risk using multiple factors. The conditional inference tree, a survival tree approach, can effectively apply to identify risk sub-groups. This tree method uses binary recursive partitioning in a conditional inference framework to estimate a regression relationship between candidate factors and outcome. The identified sub-groups can be included in a Cox model to obtain hazard ratios and their 95% confidence interval. The survival curves of the sub-groups also can be generated using the Kaplan-Meier method, and the log-rank test is used to assess the differences among the various subgroups. In this study, we demonstrated and compared the two approaches (Cox regression vs. survival tree) using epidemiology data. The “party” r package was used for the survival tree analysis.