Ctni-16. use of predictive score-based historical control data in single-arm trials

Abstract Even though the randomized controlled trials (RCTs) are still the preferred approach to evaluate the effects of treatments, single-arm study designs have been traditionally used in Phase II oncology clinical trials. The single-arm studies using historical control or external control might be inevitable and even fatal when concurrent controls are impractical or unethical in rare and high-risk diseases. Due to recruiting difficulty, ethical and economic reasons, the single-arm trials can be a good option to test the experimental efficacy with the external control data of historical control rates for early efficacy phase 2 trials in glioblastoma. The FDA guidelines on external control data focus on the use of patient-level data from other clinical trials or real-world data (RWD), and so there is no guidance for summary-level historical control. Furthermore, there exist methodological issues to use historical data like identification of relevant data, amount of discounting (time difference and dissimilarity), prior robustness or operating characteristics. In this study, we propose a synthetic data based predictive scoring method optimally to match the historical control data with the characteristic conditions of the experimental studies. We consider both a binary (PFS on six months) and survival (median PFS) endpoints as surrogates for Phase 2 single-arm trials in newly diagnosed and recurrent glioblastoma patients. We perform meta-analysis to summarize the endpoints and the related socio-demographic, treatment and genetic factors from the summary-level historical data. Then we generate the synthetic patients based on the meta-summarized data. The generalized boosted regression models will be used to identify significant characteristics affecting the survival of GBM patients using the synthetic data. Finally the predictive score-based historical data model will be developed to provide the optimal matching with the characteristics of the experimental studies. This study will be validated using existing patient-level data as well as cross-validation methods.