P. Krotka, M. Posch, M. Bofill Roig
Platform trials efficiently evaluate multiple treatments, allowing late entry of new arms and sharing controls. For the comparison of late-entering arms, concurrent and non-concurrent controls (NCC) can be used. Using NCC in the analysis can increase power, but may introduce bias due to time trends. To leverage NCC while ensuring valid inference, model-based methods, adjusting for time trends by including a factor period — the intervals bounded by the time points an arm enters or leaves the platform — have been proposed for continuous and binary endpoints.
We extend these approaches to time-to-event endpoints and consider Cox models including the factor period to adjust for time trends. However, a challenge is that survival trials have two timescales: the individual patient time relative to study entry and the calendar time of the patient's recruitment. Through simulations, we assess the type I error rate and power of the proposed Cox model and outline assumptions for valid inference.
Keywords: Platform trials, Non-concurrent controls, Survival analysis, Statistical inference, Statistical modeling
Scheduled
Biostatistics I
June 13, 2025 11:00 AM
MR 3