Prm247 - Meta-Analyses Using Real-World Data: A Systematic Literature Review of Existing Recommendations

Abstract

OBJECTIVES: A systematic review on 180 NICE oncology submissions identified three main network meta-analysis (NMA) methods; (1) Hazard ratio NMAs (HRNMA); (2) parametric survival NMAs (PNMA); and (3) fractional polynomial NMAs (FPNMA). This study aims to compare these NMA methods. METHODS: For comparing the NMA methods, three datasets with two trials were considered. The first NMA dataset considered trials in which proportional hazard (PH) holds. In the second, PH-assumption was violated in one trial. In the third, the PHassumption was violated and there was a sudden change in hazard. For the PNMA, Weibull, Lognormal and Loglogistic distributions were assessed. Secondorder FPs were assessed by varying P1 and P2 from -2.0 to +2.0. Best fitting models were selected on DIC. Outcomes were compared in terms of restricted mean survival (RMST) over the trial period and predicted mean survival. RESULTS: In the second dataset, the observed RMSTs of the first trial were 27.12 and 19.90 months. The RMSTs with HRNMA were 27.90 and 20.48 month; PNMA 28.00 and 20.44 months; and FPNMA 28.66 and 21.13 months, respectively. The observed RMSTs in the second trial (PH violated) were 12.35 and 10.51 months. The RMSTs with HRNMA were 13.22 and 10.51 months; PNMA 12.96 and 10.57 months; and FPNMA 12.52 and 11.24 months, respectively. Despite visual fit, RMSTs were similar. Mean incremental survival with HRNMA and PNMA in first trial were