OBJECTIVES: Survival models are often the back bone of cost-effectiveness models. Although goodness-of-fit data are increasingly used to select the most appropriate survival model, this selection is typically uncertain and a key driver of outcomes. Nevertheless, this structural uncertainty is not routinely included in cost-effectiveness models, which potentially biases the estimated cost-effectiveness. Therefore, we aimed to develop methods to incorporate uncertainty related to survival model selection in the probabilistic sensitivity analysis of cost-effectiveness models using model averaging.

METHODS: A cost-effectiveness model with three survival models (progression-free survival (PFS), overall survival (OS), and time to treatment discontinuation (TTD)) was used. Seven different distributions were fitted to the PFS, OS, and TTD data. This resulted in (7×7×7=) 343 scenarios to use in model averaging. Three methods to obtain weights for model averaging were compared. Method 1 represents current practice (deterministic weight of 1 for the ‘best’ model), Method 2 uses Akaike weights, and in Method 3, bootstrap cross-validation is used to compute mean Akaike weights across bootstrap samples. Incremental net monetary benefit (iNMB), probability of cost-effectiveness and expected value of perfect information (EVPI) were calculated for each method using a willingness-to-pay threshold of €50,000 per quality-adjusted life year.

RESULTS: The iNMB for Methods 1-3 were €11,191, €3,489 and €6,956 respectively. The probability of cost-effectiveness decreased by respectively 23% and 13% in Methods 2 and 3 compared to Method 1. The individual EVPI in Methods 2 and 3 were respectively 3 and 2 times higher than in Method 1.

CONCLUSIONS: This paper provides methods to incorporate the uncertainty surrounding the selection of survival models for cost effectiveness analyses based on goodness-of-fit data. Our results demonstrate that ignoring this structural uncertainty leads to biased iNMB estimates, and an underestimation of the uncertainty surrounding cost-effectiveness results. This has important implications for decision making.