How is retrospective independent review influenced by investigator-introduced informative censoring: a quantitative...

A reliable determination of progression is of key importance in determining progression-free survival in oncology trials. An independent review of tumour assessments made by investigators is often implemented with the aim of reducing a possible bias. Often, the independent review is performed in a prespecified but retrospective fashion by reviewing a patient after all assessments have been performed. It has been discussed that this mechanism can lead to informative censoring with respect to independent review. This is caused by the fact that often no further assessments are available after the investigator has declared the patient to be progressive, possibly leading to a considerable amount of patients being judged progressive by the investigator and being censored by independent review. We introduce and investigate a model for the error in assessment with the aim of quantifying the bias in independent review. The model is based on single error probabilities at each assessment time-point that are independent from each other but dependent on the time to the true progression time-point. The bias introduced for the independent review is described and quantified. We show that the investigator assessments of progression can lead to less bias for progression-free survival than the results for independent review. Results show that a within-arm discordance rate is not necessarily correlated with the bias in independent review. Finally, we propose an approach for a sensitivity analysis that is a useful tool to sandwich the true underlying distribution by the results for independent review itself and the described sensitivity analysis.