Transforming Clinical Trials for Elderly and Rare Cancers through Intelligent and Robust Information Borrowing - PROJECT SUMMARY/ABSTRACT Innovative clinical trial designs, such as basket, hybrid controlled, and platform trials, are transforming treatment development, particularly for early cancers, rare tumors, and elderly patients. Over the previous funding period, we made significant advances in enhancing the generalizability of randomized clinical trial findings, leveraging real-world data to improve external validity, developing methods to detect heterogeneity in treatment effects using auxiliary real-world evidence, and integrating machine learning and AI approaches for greater analytical accuracy. These advances have strengthened sta- tistical inference in modern trials, yet emerging designs introduce new challenges. Small sample sizes and patient heterogeneity continue to limit the reliability of treatment effect estimation, necessitating further methodological innovation. This renewal application aims to develop, validate, and implement robust, flexible, and interpretable statistical methodologies to address these challenges and enhance causal inference of treatment ef- fects in complex clinical trials. We propose three specific aims: Aim 1. Develop quasi-randomization inference methods for single-arm basket trials to improve the validity and efficiency of treatment effect estimation. By integrating quasi-randomization with synthetic control arms and pretraining techniques, this approach better accounts for tumor heterogeneity while maintaining statistical rigor. Aim 2. Ex- tend randomization inference to hybrid controlled trials using model-free selective borrowing. This method leverages conformal inference to selectively incorporate unbiased external controls, ensuring strict Type I error control and improved statistical power. Aim 3. Create a novel tensor completion framework for causal inference in platform trials. This approach models potential outcomes as a 3- dimensional tensor, employing Tucker decomposition, baseline covariates, treatment clustering, and propensity score weighting to enhance the accuracy and interpretability of treatment effect estimation. This project is highly innovative, introducing new statistical frameworks that merge causal inference with machine learning to improve the efficiency and robustness of modern clinical trials. Expected out- comes include validated methodologies and publicly available software to support the design and anal- ysis of basket, hybrid controlled, and platform trials. By building on prior discoveries and addressing new methodological gaps, this renewal will improve treatment evaluation, accelerate drug develop- ment, and ultimately enhance patient outcomes in oncology and beyond, furthering NIH’s mission to advance scientific knowledge and improve human health.
