OBJECTIVES: Assessing lifetime treatment effects based on trial data often necessitates survival extrapolation in economic evaluation with the choice of model impacting the outcomes. Our aim was to assess accuracy and variability between alternative approaches to survival extrapolation.

METHODS: Data on HER2-positive breast cancer patients from the National Breast Cancer Register of Sweden were used to fit standard parametric distribution models (SPD models) and excess hazard models (EH models) adjusting survival projections based on general population mortality (GPM). Models were fitted using 6-year data for stage I and II, 4-year data for stage III, and 2-year data for stage IV cancer, reflecting early data cut-offs while maintaining sufficient events for statistical analysis, We compared model projections of 15-year survival and restricted mean survival time (RMST) to 15-year registry data, and explored variability between models in extrapolations of long-term (50 year) survival.

RESULTS: Among 11,224 patients, the 15-year RMST from the registry was 12.7, 11.4, 9.3, and 4.8 for stages I to IV, respectively. Compared to the registry estimates across the disease stages, the AIC-averaged projections varied as follows: -8.2% to +5.3% for SPD models, -4.9% to +5.2% for EH models without a cure assumption, and -19.3% to -0.2% for EH models with a cure assumption. EH cure models provided the most accurate estimates in patients with stage I – III disease while EH models without cure most closely matched survival in patients with stage IV disease where cure assumption is less plausible. EH models significantly reduced between-model variability in the predicted RMSTs over a 50-year time horizon compared to SPD models. EH models generally predicted lower RMSTs whereas SPD models produced higher RMSTs especially among early stage patients.

CONCLUSIONS: EH models may be considered as alternatives to SPD models to produce more accurate and plausible survival extrapolations that account for general population mortality.