OBJECTIVES: Evaluating efficacy of novel treatments for rare diseases in pediatric populations presents a substantial challenge. These difficulties are particularly pronounced for histology-independent therapies trialed in pediatric basket trials. Bayesian borrowing methods to supplement limited pediatric sample sizes using adult trial data have recently gained traction, as have Bayesian hierarchical models (BHM) to model cross-histology heterogeneity in basket trials. We propose a combined method which allows for information borrowing both across histologies and from adult basket trials to estimate histology-specific overall response rates (ORR).

METHODS: We use a BHM which partially pools information across histologies, where the extent of pooling is dependent on the degree of variability in ORRs across histologies. We augment the BHM by partially borrowing information from adult basket trial data into the pediatric data via a power prior. The power prior down-weights the adult data based on a fixed discount parameter which we vary from 0 (no borrowing) to 1 (complete pooling) to conduct a tipping point analysis in which we assess how much borrowing is needed to meaningfully exceed an ORR threshold. We demonstrate the method using simulated data containing N_P=50 patients in the pediatric data and N_A=100 patients in the adult data, split across K=8 prognostically important histologies.

RESULTS: The method yielded reductions in the width the 95% credible intervals (CrI) ranging from 6.3% to 27.4% for the 8 histologies when varying the power prior discount parameter from 0 to 1. Figures showing the impacts of information borrowing on the precision and magnitude of ORR estimates will also be presented.

CONCLUSIONS: Assessing efficacy of histology-independent therapies in pediatric populations will require continued innovation to appropriately synthesize limited available data. By allowing for partial borrowing of information both across histologies and from adult data, our method can improve precision of ORR estimates for individual histologies.