OBJECTIVES: Immature survival data can result in higher uncertainty over long-term extrapolations, and consequently in decision-making. Such uncertainty can be reduced with the use of external data. We aimed to review the use of external data to better inform survival extrapolations in submissions to the National Institute for Health and Care Excellence (NICE).

METHODS: All NICE technology appraisals (TAs) that dealt with any stage of cancer and had been completed after December 2001 were reviewed. All relevant documents available on the NICE website were reviewed with special attention to the reports developed by the independent Assessment Group (AG) or Evidence Review Group (ERG). We included TAs where external data was used to inform survival extrapolations. Then, in addition to general information about the TA, we extracted the following information: external data source (historical trial, registry data or general population mortality), methods explored in the sponsor submission, criticism and alternative scenarios proposed by AG or ERG.

RESULTS: Out of 143 reviewed TAs, ten TAs reported the use of external data to quantitatively inform survival extrapolations. These TAs were in two indications: melanoma (n=8) and chronic lymphocytic leukemia (n=2). Most frequently, a multistage approach to modeling survival based on different data sources was implemented. Key criticism included clinically implausible extrapolation, high risk of selection bias, a population not representative of the UK population, use of general population mortality for a metastatic disease and using methodologies not supported by evidence from the literature.

CONCLUSIONS: External data represent a sound source of information that may reduce uncertainties in the extrapolation of survival data. Nevertheless, modeling survival based on different data sources may result in clinically implausible extrapolations and increase the risk of bias. Therefore, exploring and applying novel methodologies that incorporate information from different data sources is recommended.