Zhaowen Sun of the Department of Biostatistics defends her dissertation on "Power and Sample Size Determinations in Dynamic Risk Prediction".
Graduate faculty of the University and all other interested parties are invited to attend
Dynamic risk prediction has recently attracted attention because of its ability to incorporate time-varying information such as repeatedly measured covariates and intermediate event status into the estimation of the probability of failure. Using a landmark data set, the prediction is updated by sub-setting the data with left-truncation at the landmark time and enforcing administrative censoring at the prediction horizon time. The landmark Cox model provides a valid estimation of the probability of failure at the horizon time under single event setting and the landmark proportional sub-distribution hazards model for the cause-specific cumulative incidence function under competing risks setting. Risk difference, defined by the difference in conditional probabilities of failure, serves as an accessible, easily interpreted measurement of effect size when comparing two treatment groups.
In this study, we proposed a test statistic that could be used to compare two conditional probabilities of failure. We derived an analytic formula to calculate the sample size needed to reach the desired risk difference, significance level, and power. We also investigated factors that can affect the power and sample size of the test and conducted simulation studies under various settings to investigate their impact.
Public health significance: This study aims at introducing relatively new risk prediction methods that could incorporate time-dependent information and update risk estimation during the time course of study follow-up; also, providing researchers with references on the power and sample size issues when planning studies involving dynamic risk prediction.