OBJECTIVES: Many cost effectiveness analyses require extrapolation of survival data. These extrapolations can be improved by incorporating external evidence, but it is unclear what methodology should be used.

A case-study of nivolumab in previously untreated advanced melanoma was used to compare the impact of different methods on the accuracy of, and uncertainty in, extrapolations.

METHODS: The CheckMate 066 and CheckMate 067 trials were used as the internal and external evidence, with 51 and 66 months follow-up, respectively. Longer follow-up for CheckMate 066 (38% 72-month overall survival) was used for validation.

Five approaches were considered: (1) No use of external evidence, (2) Piecewise, (3) Bias-adjustment (applying a hazard ratio to the internal data), (4) Informative prior, and (5) Pooling with an M-spline model. The first four approaches used a log-normal model as this provided the best fit to the internal data.

Point-estimates and 95% confidence interval widths for 72-month predictions were compared, as were twelve-month estimates to assess the impact on within-trial fit.

RESULTS: Extrapolation without external evidence led to survival estimates that were too low (72-month predicted survival of 34% compared to 38%). The incorporation of external evidence led to increased estimates (range 35% to 41%) with the most accurate estimates for the informative prior.

Confidence intervals are the widest (15.2%) at 6-years from the bias-adjusted method, which is the only method that led to more uncertainty than no external data (14.4%). The smallest confidence interval width is for the informative prior method (11.5%).

With the exception of the piecewise method, incorporating external evidence also leads to different within-trial estimates.

CONCLUSIONS: Incorporating external evidence can increase the accuracy of survival extrapolations. For this case-study the informative prior method led to the most accurate 72-month predictions, with the smallest uncertainty. However, it resulted in slightly worse within-sample predictions at 12-months.