Extrapolating Parametric Survival Models in Health Technology Assessment Using Model Averaging: A Simulation Study

Previous work examined the suitability of relying on routine methods model selection when extrapolating survival data in a health technology appraisal setting. Here we explore solutions to improve reliability restricted mean time (RMST) estimates from trial by assessing plausibility and implementing averaging. We compare our previous selecting for extrapolation using Akaike information criterion (AIC) Bayesian (BIC). Our averaging include equal weighting across models falling within established threshold ranges AIC BIC BIC-based weighted averages. apply assessment implement output simulations, where 10,000 runs 12 trial-based scenarios were examined. demonstrate that removing implausible consideration reduces squared error associated with estimate each method increases percentage RMST 10% parameters sampling distribution. The superior single optimal extrapolation, aside some exponential already selected model. wide criterion-based thresholds outperformed BIC-weighted majority scenarios. conclude approaches should feature more widely technologies is influential considerable uncertainty present. Where complicated underlying hazard rates, funders account additional these extrapolations their decision making. Extended follow-up trials be encouraged used review prices therapies ensure fair price paid.