OBJECTIVES : Nivolumab+ipilimumab is the only immuno-oncologic combination showing significant and stable long-term clinical benefit compared with standard of care (sunitinib) in first-line advanced/metastatic intermediate- or poor-risk renal cell carcinoma. Follow-up data up to 4 years from CheckMate-214 (NCT02231749), the phase III trial comparing these treatments, have previously indicated substantial improvement in overall survival (OS). This study assessed the stability of the lifetime OS extrapolations considering additional follow-up with a minimum of up to 48 months.

METHODS : Hazard ratios (HRs) for OS of nivolumab+ipilimumab versus sunitinib were compared for database locks (DBLs) from CheckMate-214 with a minimum of 18, 24, 30, 37, 42 and 48 months of follow-up data. Parametric functions, including traditional distributions and cubic splines, were used to extrapolate observed OS to lifetime, corrected for British population background mortality. Model performance was evaluated based on the NICE DSU guidance (TSD-14). The stability of lifetime OS estimates was assessed by comparing extrapolation functional forms, mean OS, and landmark OS across the different DBLs.

RESULTS : OS HRs were stable across DBLs, including the 42- and 48-month DBLs. The proportional hazards assumption was upheld in all DBLs, thus dependent models were used for OS extrapolations. A log-normal distribution consistently provided the best fit in all DBLs, although fits to earlier DBLs slightly underestimated observed OS rates in later DBLs. Mean lifetime OS for nivolumab+ipilimumab was 7.13 years in 18-month and 7.73 years in 48-month DBLs. Predicted OS rates at 5, 10, and 20 years were respectively 39.9%, 23.0%, and 11.1% in the 18-month DBL, and marginally increased to 42.1%, 25.4%, and 13.1% in the 48-month DBL.

CONCLUSIONS : CheckMate-214 4-year follow-up data confirmed robustness of previous DBL survival extrapolations, including the choice of a log-normal distribution. Due to emerging signs of an OS plateau, original extrapolations may have slightly underestimated OS.