Pcn240 evaluating partitioned survival model and response-based modeling approaches for use in cost-effectiveness analysis: estimating and validating survival outcomes

To assess long-term survival outcomes for nivolumab plus ipilimumab (N+I) and sunitinib in renal cell carcinoma (RCC) predicted by two models: a partitioned survival model (PSM) and a response-based model (RBM). The models were developed using either a three-state partitioned survival or a semi-Markov technique and data from CheckMate 214, a phase 3 trial comparing N+I to sunitinib in first-line RCC setting. In the PSM, the state occupancy was estimated by extrapolating progression-free survival (PFS) and overall survival (OS). In the RBM, transition probabilities were calculated based on extrapolated PFS, OS and time-to-progression (TTP) curves where curve fitting was done separately for patients with complete or partial response, stable disease, or progressed disease at a 6-month landmark. Curves were chosen using statistical fit, visual inspection, and clinical plausibility. In CheckMate 214, the 2-year OS rate was 64.8% and 52.5% respectively for N+I and sunitinib; 65.9% and 52.9% in the PSM, and 67.8% and 54.7% in the RBM. The five-year survival, conditional upon surviving one year, was 47.4% and 31.9% in the PSM for N+I and sunitinib respectively, versus 45.4% and 35.6% in the RBM. In comparison, the five-year conditional survival from real world data for sunitinib was 37.6% from an international, open label expanded-access trial. The PSM and RBM provided a good fit to the trial data, but different long-term survival was predicted by each model. Whilst the RBM allows for the treatment benefits on response to be explicitly modeled into patient transitions over time, it adds challenges to clinically validate multiple survival curves by response at different landmark points. Therefore, the use of multiple external datasets and clinical opinion will be important to validate long-term extrapolations of survival and justify the choice of the model structure.