Keywords: Oncology, event projection, early drop-offs, informative censoring, competitive studies
In recent years, oncology treatment landscape has been changing dramatically, with new immunotherapies and target therapies approved for various cancer types and biomarker populations. For existing studies with slow and long recruitment targeting at a relatively small disease population, enormous challenges may occur when other drugs are under investigation in competitive studies with the same target population or in either earlier or later lines that may affect patient profile or overall survival. For example, recruitment and event accumulation might be compromised by evolving patient pool and early withdrawals. Statistical methods conditional on collected data, like parametric Poisson-Gamma + exponential/Weibull model, non-parametric model, piecewise model, Empirical Bayesian model, etc., may help the trial team to estimate when required number of events can be achieved with a confidence interval and determine timeline accordingly. Also, during study conduct, there may be need of adjusting, e.g. minimally statistically significant difference and hence sample size assumption, or go / no-go decision rule for phase II studies. Another perspective is to evaluate the impact of early drop-off on study power and results assuming either non-informative but only unbalanced censoring or informative censoring when more early drop-off in one treatment arm is suspected, especially in open-label studies.
Key questions for discussion: How to determine which model to use for event projection conditional on collected data during study conduct? What factors need to be taken into consideration when evaluating the impact of unbalanced or even informative censoring? What kind of overall strategy can be considered when an ongoing study is facing challenges from new competitive drugs and trials?