Keywords: meta-analysis, Bayesian hierachical model, immuno-oncology, cancer immunotherapy, checkpoint, indirect comparison, PD-1, PD-L1
The first PD-1 checkpoint inhibitor, pembrolizumab, was approved in 2014, and subsequently the medical literature regarding cancer immunotherapy has rapidly expanded. The PD-1/PD-L1 treatment axis is characterized by: 1) the same pathway with 2 counterpart binding-sites, 2) a large number of molecules in development, 3) broad range of indications being explored, and 4) similar efficacy results within an indication. This provides a unique opportunity to assess many important clinical questions, e.g. whether antibodies targeting PD-1, the receptor located on T cells, have a different efficacy and/or safety profile from the antibodies targeting PD-L1, the ligand for PD-1. Despite the sudden increase of publications shedding light on this and other critical questions, there is only a limited focus on systematic statistical methodologies and analyses which may provide a more formal quantitative characterization. We used a meta-analytic procedure to compare efficacy and safety data between PD-1 and PD-L1 checkpoint inhibitors across tumor types. The procedure uses a Bayesian hierarchical model to synthesize efficacy and safety data between molecules across indications. The model was applied to public data collected from approximately 70 clinical studies through Dec 2016, across 31 solid tumor types and involving 12,025 patients. This yielded interesting insights including that treatment with PD-1 inhibitors resulted in slightly higher numerical response rates, but the magnitude of difference was not clinically relevant. In conclusion, it is critical to understand the ever-changing PD-1/PD-L1 landscape and optimal treatment regimens. Direct comparison of these molecules is difficult, but our proposed method provides an approach to address this critical question, and potentially others, through efficient use of publically available data.