Methods for Selecting Survival Extrapolations in Practice: A Systematic Review of Health Technology Assessments

OBJECTIVES: Although several guidelines exist for conducting time-to-event (survival) analysis for health technology assessment (HTA), extrapolation preferences remain a major point of contention. We aimed to review how marketing authorisation holders (MAHs) and HTA bodies justify and validate survival models to ascertain if accepted practices exist beyond those stated in guidelines.

METHODS: A systematic review was conducted on HTA appraisals using survival analysis from NICE (England), TLV (Sweden), NoMA (Norway) [2019–2023], and DMC (Denmark) [2021–2023]. Data extracted included base case model selection and justifications by the MAH and HTA bodies.

RESULTS: A total of 272 assessments were included. Both MAH and HTA bodies cite guidelines for selecting curves (e.g., statistical and visual fit, fit to external data, or clinical plausibility), however fit to trial data was more commonly used for curve selection by MAH (>93%), with 25-75% of submissions also referencing clinical plausibility and/or external data. In case of disagreement, TLV often cited statistical fit, followed by clinical plausibility and external data. For NoMA and DMC, statistical or visual fit and clinical rationale were cited most, with external data being less commonly referenced. In NICE assessments, multiple reasons were cited for changes. However, across assessments the metrics of ‘clinical plausibility’ (e.g., landmark survival or biological plausibility of hazards), the source of input (HTA expert committee vs. MAH external experts), and the weight this took in assessments (vs. trial data or external evidence) were divergent.

CONCLUSIONS: Statistical fit is the most cited rationale, despite recent guidance that it is a less relevant criterion for curve selection. The clinical plausibility of an extrapolation is pertinent for curve choices. However, there lacks a clear approach on how to systematically determine clinical plausibility. Greater use of external data and improved use of structured expert elicitation may lead to greater consensus between MAHs and HTA bodies.