Meta-analysis of preclinical measures of efficacy in immune checkpoint blockade therapies and comparison to clinical efficacy estimates

Abstract

Abstract Background Despite the successes of checkpoint inhibitors targeting T-cell receptors, clinical efficacy is highly cancer-dependent and subject to high inter-individual variability in treatment outcome. The ability to predict the clinical success in different cancer indications is therefore an important capability for successful clinical development. In this meta-analysis, the main goal was to identify factors that modified the clinical efficacy estimates of checkpoint blockade therapies derived from preclinical animal data to improve the robustness and reliability of such estimates. Methods To this end, animal studies testing checkpoint inhibitors (anti-PD-1, anti-PD-L1, anti-CTLA-4) were identified in PubMed ranging from 1.01.2000 to 31.12.2018. The eligibility criteria included the reporting of the Kaplan–Meier estimates of survival and the number of mice used in each experiment. A mixed-effects model was fitted to the preclinical and clinical data separately to determine potential sources of bias and heterogeneity between studies. Results A total of 160 preclinical studies comprising 13,811 mice were selected, from which the hazard ratio (HR) and the median survival ratio (MSR) were calculated. Similarly, clinical Phase III studies of checkpoint inhibitors were identified in PubMed and the ClinicalTrials.gov database ranging from 1.01.2010 to 31.12.2020. This resulted in 62 clinical studies representing 43,135 patients subjected to 8 therapies from which overall survival (OS) and progression-free survival (PFS) hazard ratios were obtained. Using a mixed-effects model, different factors were tested to identify sources of variability between estimates. In the preclinical data, the tumor cell line and individual study were the main factors explaining the heterogeneity. In the clinical setting, the cancer type was influential to the inter-study variability. When using the preclinical estimates to predict clinical estimates, the cancer-type specific estimates of treatment effect using the MSRs better approximated the observed clinical estimates than the HR-derived predictions. Conclusions This has strong implications on the design of ICB preclinical studies with respect to sample size determination, selection of cancer cell lines and labs to run the experiments and the choice of efficacy measure.