Presentation (CTI)

OBJECTIVES: Survival extrapolation is used in health technology assessment (HTA) to estimate life-time treatment effects. Cost-effectiveness results are often highly sensitive to survival model choice. Long-term extrapolations are particularly important when treatments are potentially curative, but follow-up in clinical trials is often insufficient to provide evidence that supports cure assumptions. Publicly available registry data, combined with analyses of general population mortality rates, may provide valuable evidence on whether long-term survivors can be assumed to be cured. We demonstrate this using a case study in non-muscle invasive bladder cancer (NMIBC).
METHODS: NHS England’s Get Data Out programme was used to obtain survival data and derive hazard functions for patients diagnosed with NMIBC, providing up to 8 years follow-up. Office for National Statistics lifetables were used to estimate mortality rates in an age- and sex-matched general population. We compared mortality hazards in NMIBC patients with general population hazards to determine if excess mortality associated with NMIBC decreased over time.
RESULTS: NHS England data were available for 9,748 patients, split by disease stage (Ta/Tumour-in-situ (TIS); T1). Upon diagnosis, the hazard of death for patients with Ta/TIS NMIBC was 1.6 times higher than in the age- and sex-matched general population. After approximately 48 months, hazards converged. For patients with T1 NMIBC, hazards were initially 4 times higher than in the age- and sex-matched general population, converging after approximately 6 years.
CONCLUSIONS: Publicly available NHS England registry data can be used to provide real-world evidence on long-term survival, to support HTA modelling. In NMIBC, registry data suggests that long-term survivors have a similar mortality risk as an age- and sex-matched general population. Limitations include a lack of patient-level information on prognostic characteristics and treatments, and likely differences between real-world and clinical trial populations. However, analyses of registry data can be used to inform and justify cure modelling assumptions.