PCN62 COMPARISON OF NETWORK META-ANALYSES USING STANDARD, MIXTURE CURE AND SPLINE MODELS

Abstract

Novel immune therapies are potentially curative options in oncology. A parametric network meta-analysis (PNMA) is commonly used to assess their effectiveness compared to other therapies in the absence of head-to-head trials. However, NMA based on standard parametric curves does not adequately capture cure. The aim of this study is to explore three methods for extrapolating survival: PNMA, spline NMA, and a novel mixture-cure PNMA (MCPNMA) methodology. To compare the methodologies, two melanoma trials including potentially curative treatments were used; 1) nivolumab versus dacarbazine (Ascierto, 2018), and 2) ipilumumab + dacarbazine versus dacarbazine (Robert, 2011). Weibull distribution was fitted on the trial data using the standard PNMA, spline NMA and MCPNMA methodologies. The survival extrapolations were compared in terms of fit and mean (incremental) survival. MCPNMA and spline NMA had better fit over the trial data than PNMA. For dacarbazine, PNMA, spline NMA and MCPNMA resulted in mean survival of 2.01, 3.72, and 5.75 years, respectively. For ipilumumab + dacarbazine, the three methodologies resulted in 2.76, 5.64, and 7.93 years, respectively. For nivolumab, these resulted in 5.24, 9.81, and 11.17 years, respectively. Cure rates predicted by the MCPNMA were 20%, 28% and 41%. MCPNMA and spline NMA both provided a good visual fit on the observed data, but resulted in different lifetime (mean and incremental mean) survival estimates. The MCPNMA predicts longer mean survival, and the predicted mortality hazards cannot go lower than the general population mortality (GPM), as it is incorporated in the fit of the cure fraction. Spline hazards, on the other hand, may become lower than GPM over time. The choice between the models should be informed by the long-term survival observational data and clinical expert opinion on clinical plausibility of long-term survival and cure.