Presentation (CTI)

OBJECTIVES: Survival extrapolation is used in cost-effectiveness analysis (CEA) of health interventions to estimate long-term benefits when clinical trials have restricted follow-up. As mentioned in the NICE-DSU-TSD-21 (2020), Bayesian may constitute a flexible approach for improving survival predictions by incorporating external information. This work aims to report how Bayesian is used for extrapolating long-term survival in CEA.
METHODS: We conducted a systematic review up to October 2024 to identify methodological and non-methodological studies using different electronic databases (PubMed, Scopus) and ISPOR conference database, completed by a manual research.
RESULTS: Of the 51 selected studies (78% automatically and 22% manually) of which 51% were published since 2022 and 90% (n=45) focused on oncology, 53% (n=27) represented articles and 39% (n=20) were methodological works. 88% (n=45) used external data and 14% (n=7) experts elicitation. We propose a classification of studies into four non-mutually exclusive categories of Bayesian modeling (C1-C4). The first three categories combine, in order of increasing complexity, both survival modeling and approaches for incorporating external information. C1 (n=14, 27%) includes standard parametric models (SPMs) with prior of parameters informed by historical data. C2 (n=24, 47%) includes Bayesian multiple parameter evidence synthesis and non-SPMs. C3 (n=14, 27%) groups complex hazard regression models (poly-hazard, relative survival) that account for both disease-specific and expected mortality from general population mortality. These models mainly use non-informative prior distributions on parameters. The last category (C4, n=7, 14%) represents Bayesian model averaging, which addresses structural uncertainty in model selection by using posterior model probabilities.
CONCLUSIONS: This review highlights the broad spectrum of Bayesian survival models and the different ways in which external information can be incorporated to reduce uncertainty in survival extrapolation for CEA. A comparison of the different approaches is required in order to propose recommendations based on the intervention mechanisms and available external information.