OBJECTIVES: Conducting indirect treatment comparisons (ITC) between novel histology-independent therapies (HIT) assessed in basket trial settings presents a unique challenge. Basket trials of HITs enroll patients with a targeted mutation or biomarker regardless of their tumour histology and usually lack a concurrent control arm. As histology may be prognostic for the outcomes of interest, failure to adjust for imbalances in the composition of histologies between comparators could confound treatment effect estimates when performing an ITC. However, the severely limited sample sizes in typical basket trials can make it difficult or infeasible to conduct a traditional population-adjusted indirect comparison.

METHODS: We implement a novel variant of a Bayesian hierarchical model (BHM) ITC approach for unanchored comparison of two treatments assessed in separate basket trials. The model mitigates risk of confounding due to differences in histology composition by incorporating modelling for heterogeneous responses by histology. This approach allows for partial pooling of information, assuming exchangeability in (i) absolute response across histologies and (ii) relative treatment effects across histologies. We apply the modelling approach to an ITC of larotrectinib vs. entrectinib for NTRK-fusion-positive solid tumours for an overall response rate endpoint.

RESULTS: We find that the posterior probability of superiority in response for larotrectinib exceeds 80% for all tumour types considered, ranging from a low of 80.9% for hepatocellular tumours to a high of 95.1% for thyroid tumours. However, these estimates fall short of conventional thresholds for statistical significance.

CONCLUSIONS: Our analysis provides some evidence that larotrectinib may yield superior response compared to entrectinib for multiple types of NTRK-fusion-positive solid tumours after adjusting for differences in histology composition. However, this relies on potentially strong exchangeability assumptions, and the possibility that residual confounding remains cannot be ruled out. Nonetheless, use of BHMs provides a potential way forward for the particularly data-scarce setting of conducting ITCs for basket trials.