OBJECTIVES: Clinical evaluation of histology independent therapies is commonly conducted using open-label, uncontrolled basket trials, including multiple tumor types with a specific biomarker. Published literature recommends that without an internal control arm or published comparative data for the relevant biomarker, patients treated with intervention therapy who do not respond to treatment may be used as a proxy for comparator outcomes.

This study explored the validity of this approach by comparing pembrolizumab non-responder overall survival (OS) data from the KEYNOTE-164 and KEYNOTE-158 clinical trials for five previously treated MSI-H/dMMR tumors (endometrial, gastric, small intestine, biliary, colorectal) with published data for relevant comparators.

METHODS: A systematic literature review was conducted to identify published sources of comparator OS. Where multiple studies were available for individual comparators, survival data were pooled. The Kaplan-Meier (KM) curves of comparator OS were visually compared with pembrolizumab non-responder OS curves for each of the five tumour sites to assess the appropriateness of the proxy.

RESULTS: Visual assessment showed that the pembrolizumab non-responder OS did not consistently align with comparator OS. For most tumor sites pembrolizumab non-responder OS was superior to comparator OS, although the magnitude of difference varied by site. In some sites, a plateau was observed in OS for pembrolizumab non-responders that was not present for comparators.

CONCLUSIONS: The non-responder surrogate approach requires the strong assumption that non-responders derive benefit similar to that received from current standard of care. These analyses suggest that previously treated MSI-H/dMMR patients receiving pembrolizumab who do not respond to therapy obtain additional clinical benefit compared with existing treatment options, and therefore may not be a suitable surrogate for comparator OS. Where available, published data for comparator OS outcomes provide more reliable estimates than a non-responder analysis. Key limitations included small patient numbers and high response rates resulting in few non-responder patients.