Bayesian Network Meta Analysis of Survival Data Using a Near Ignorance Dirichlet Process with Pseudo IPD

Among the few approved therapies for advanced renal cell cancer, Sunitinib is a common active comparator in most trials. The pembrolizumab-plus-axitinib and nivolumab-plus-cabozantinib combination therapies have shown better efficacy compared to Sunitinib in different studies but there is no direct head-to-head study between the two combination therapies. Network Meta Analysis is employed to compare the treatments indirectly. Usually, the aggregate Hazard ratio-based approach and endpoints are used in Network Meta Analysis. Matching Adjusted Indirect Comparison has been reported for checking these two sets of combination therapies. Proportionality assumption violation is an issue with using Cox proportional hazard ratio while Matching Adjusted Indirect Comparison is not free of bias. We therefore employ pseudo-Individual patient level data generated from digitized survival curves and apply a new prior near-ignorance Dirichlet Process. We then compare the results of cox-regression based method and a Bayesian approach based on near ignorance imprecise prior Dirichlet Process with Pseudo-IPD based Network Meta Analysis. Based on both the Cox-regression and pseudo-IPD based-approaches, there is no statistically significant difference between the two groups based on efficacy. Both the combination therapies perform significantly better than Sunitinib arm in term of efficacy when using the Overall Survival and Progression Free Survival endpoints. Since the Bayesian prior near-ignorance Dirichlet Process based method does not assume proportionality, it is a better choice. In the current work, a R software-based analysis is done with an example dataset to compare and present the results from the two methods.