Keywords: non-proportional hazards
A joint modeling approach for progression-free survival (PFS) and overall survival (OS) is introduced as an extension of "Luo et al. 2018", with updated R package for efficient theoretical power calculation and intuitive visualization of the results. A cases study is presented for a phase III immuno-oncology trial in non-small cell lung cancer with primary analysis based on regular log-rank test and Cox model for PFS and OS. As study results from similar drugs were disclosed, a study design modification was considered, especially to address the potential presence of NPH. In the control arm, it is anticipated that high percentage of patients in the control arm would receive medication for the treatment arm after disease progression. Hence complex modeling was needed to more accurately capture the impact of crossover on the power of OS. A three-state model was built to describe patients’ transition from enrollment, disease progression and death. Based on multiple piece-wise exponential survival functions, together with fine-tuned hazard rates as well as transition probabilities, the jointly simulated PFS and OS data could closely resemble what had been reported for previous studies, including survival functions, hazard ratio and crossover rate. Other relevant operational and regulatory concerns were also considered. The modeling results assessed the power for OS in difference scenarios determined by factors such as sample size, expected crossover rate and analysis timing. From statistical perspective, it provided critical and interpretable results supporting strategical decision making. This method can also be applied broadly to other complex study designs involving time-to-event endpoints.