Presentation (CTI)

OBJECTIVES: In many oncology indications, overall survival (OS) data are increasingly immature at the time of health technology assessment as trials shift to earlier treatment lines and demand for accelerated assessment grows. Immature OS data typically yield highly variable parametric extrapolations, resulting in substantial decision uncertainty. We outline a practical framework for robust long-term OS extrapolation.
METHODS: We reviewed established and emerging methods for handling immature OS data within partitioned survival models (PartSMs), the dominant cost-effectiveness model type in oncology, in which OS extrapolation typically drives cost-effectiveness results. Beyond traditional extrapolation validation criteria, we organized these methods into a triangulated framework drawing on three complementary sources of evidence to identify and reject implausible long-term extrapolations.
RESULTS: First, alternative outcome data from within the trial (i.e., time to progression [TTP], pre-progression survival, and post-progression survival) can be leveraged to model OS through state transition modeling. Although this requires evidence of a surrogacy relationship between TTP and OS, it reduces reliance on immature OS data and, when compared with PartSM outputs, can identify clinically implausible extrapolations. Second, evidence beyond the trial (e.g., long-term follow-up from comparator or later-line trials and real-world data [RWD]) can validate extrapolations or directly inform OS through multiparameter evidence synthesis, anchoring extrapolations to external benchmarks. Third, where empirical evidence is limited, structured expert elicitation (SEE) quantifies experts' uncertainty around landmark OS proportions and can validate extrapolations, inform priors, or populate probabilistic sensitivity analyses. Robust extrapolation should therefore incorporate insights from alternative modeling approaches, later-line trials and RWD, and SEE.
CONCLUSIONS: Reliance on direct extrapolation of immature OS data can lead to substantial decision uncertainty and delayed, negative, or suboptimal reimbursement decisions. A triangulated approach integrating alternative modeling approaches, external evidence, and SEE provides a transparent and robust framework for addressing immature OS data in oncology cost-effectiveness modeling.