Statistical approaches to interim analysis : a critical appraisal

Abstract

Several important questions have been raised about decision of stopping a trial early and on what basis to reach such a decision. It seemed therefore of interest to investigate the forms of monitoring used in cancer clinical trials and to gather information on the role of interim analyses in the data monitoring process of a clinical trial. The project addressed the following issues: - what is the performance of different interim analysis approaches; - how often interim analyses are used in cancer clinical trials; - which types of statistical analyses are more frequently adopted; - how the data monitoring is organised and which is the weight of statistical analyses in the decisional process. Analysis of performance of different statistical analysis approaches has been conducted by comparing the probability of stopping and the estimation bias on clinical scenarios based on real data of trials performed in ovarian and colorectal cancers. The project also focused on the prevalence of different types of interim analyses and data monitoring for both safety and efficacy in cancer clinical trials. Sources of investigation were the literature data and the protocols of cancer clinical trials included in the in the Italian registry of clinical trials.Results of our research indicate that the more widely used statistical approaches reduce the risk of “incorrect “ early stopping, compared with the adoption of no stopping rule, with similar performance. Analysis of protocols and early reports suggests that the implementation of these procedures in a monitoring strategy is not satisfactory. Use of interim analyses is still limited to the frequentist approach of the alpha-spending function, while the Bayesian is not considered. Interim analysis plans are still scarcely described, even in more recent protocols, denoting a not yet sufficient attention to this issue not only by the researchers, but also by regulatory bodies.