OBJECTIVES: Clinical trial data is often limited in terms of follow-up time while HTA submissions require assessment of long-term outcomes, thus survival extrapolations are required. Recent guidance from NICE has suggested employing a relative survival approach to incorporate general population background mortality (GPM) into survival extrapolations to better account for increasing GPM as patients age. However, alternative methodologies exist to incorporate GPM within survival extrapolations, and knowledge on the appropriateness of these methods across situations is lacking. This research uses a full-factorial simulation study to compare the performance of different approaches for incorporating GPM in survival extrapolations.

METHODS: 288 scenarios with different patient characteristics (age, survival, survival distribution and heterogeneity) and trial characteristics (trial size, level of censoring and covariate knowledge) were assessed with the aim to determine patterns in accuracy. For each scenario, 2,500 datasets were simulated. The performance of the relative survival approach was compared to a) adding GPM hazards to fitted parametric hazards (additive hazards), b) applying GPM hazards when fitted parametric hazards are lower (converging hazards), and c) non-GPM extrapolations. Models were fit using the standard parametric distributions recommended by NICE, plus the generalized F distribution. Accuracy was assessed based on the extrapolated mean survival, restricted mean survival time (RMST) and survival probability at time t, compared to the simulated known values.

RESULTS: Significant differences in accuracy were found between the methods. The methods that incorporate GPM outperformed non-GPM extrapolations in 228 scenarios (79%) on average across the three estimands. The distribution used to simulate the datasets was not always identified as the best performing distribution.

CONCLUSIONS: Adjusting for GPM results in more accurate survival extrapolations, with all three methods outperforming non-GPM extrapolations. Our results provide guidance for selecting between the methods to incorporate GPM within survival extrapolation, given patient and trial characteristics.