Simulation‐Based Power Analysis for Time‐Dependent Area Under Receiver Operating Characteristic Curve Using Approximate Bayesian Computation

In this study, we propose a simulation-based power analysis framework for study designs that assess the prognostic accuracy of a biomarker for time-to-event outcomes using the time-dependent area under the receiver operating characteristic curve. The proposed method consists of two primary components: (1) generation of pseudo censored survival data by integrating Approximate Bayesian Computation (ABC) to estimate failure and censoring distributions based on available information, and (2) iterative Monte Carlo simulations to estimate the required sample size, biomarker effect size, and statistical power. The framework accommodates common complexities encountered in clinical trials, including single and staggered entry enrollment designs and both continuous and dichotomized biomarkers. Simulation studies demonstrate that the proposed framework accurately and consistently estimates these quantities across a range of study designs and biomarker settings. Finally, we illustrate the practical utility of the proposed approach through an application to a real clinical trial of relapsed/refractory large B-cell lymphoma.