Presentation (CTI)

OBJECTIVES: Economic evaluations frequently require extrapolation of long-term survival beyond the period captured by observed Kaplan-Meier estimates. The reliability of these extrapolations is influenced by the duration of follow-up and the assumptions made about underlying hazard functions. Because cancers differ substantially in their natural history, progression patterns, and aggressiveness, no single parametric survival distribution is universally optimal across all cancer types. The objective of this study was to assess the projected survival based on parametric extrapolations compared with observed survival. Specifically, we aimed to examine how the choice of survival model affects restricted mean survival time, overall survival estimates, and statistical model fit across varying follow-up durations. These findings may support more robust survival predictions in health economic-evaluations, cost-effectiveness analyses, and payer decision-making.
METHODS: SEER data from 2011-2023 were analyzed to validate the performance of parametric survival models, including Exponential, Weibull, Gompertz, Log-logistic, Gamma, and Generalized-Gamma, in extrapolating long-term outcomes for breast-cancer(BC), colorectal-cancer(CRC), and gynecological-cancer(GynC). These cancers were selected to represent long, intermediate, and short survival horizons. Kaplan-Meier was fitted on actual data from 2011-2023, while parametric models were fitted using 2011-2018 data and extrapolated to 2023. Performance was evaluated by comparing extrapolated restricted mean survival time with Kaplan-Meier estimates.
RESULTS: Among all parametric models, the best mean absolute error between the projected and observed RMST was 0.16 for BC with exponential, compared with 0.55, and 0.19 for CRC and GynC respectively using Gompertz. These results suggest that Gompertz may be better suited to aggressive cancers whereas exponential fits longer-survival cancers.
CONCLUSIONS: These findings highlight the importance of model choice in health economic evaluations, as survival extrapolation directly influences cost-effectiveness estimates, reimbursement decisions, and long-term policy planning. Researchers and decision makers should therefore apply flexible modeling approaches, transparently report uncertainty, and consider sensitivity analyses to ensure robust evidence for payer and policy use.