Extrapolation in HTA: Background reading

The problem of trying to predict future outcomes has long been a problem in HTA. The below provide some key reading on extrapolation, from a variety of different perspectives. All of the below should be open access, but if you have any problems please contact me.

NICE Decision Support Unit Technical Support Document 14 (2013)

This is a detailed report into the methods available for the analysis and extrapolation of time-to-event outcomes (also known as survival data). It was authored by the NICE Decision Support Unit (DSU), who provide support to NICE and their HTA work (the DSU have a series of Technical Support Documents, which are required reading for anyone working in HTA).
This report had two main components: a review of extrapolation methods currently used, and guidance on producing extrapolations. This guidance includes both a model selection process algorithm, and clarification points. Since its publication, this report has been highly cited and used, both in the literature and in HTA submissions. It is a key reference for extrapolation in HTA.

The link also provides references to a journal article (which is a shorter version of the full report), along with subsequent discussions on this.

Special collection of articles on extrapolation (2017)

This special issue of Medical Decision Making provides eight articles on extrapolation. Whilst they are not all in the domain of HTA, the issues and approaches considered are all relevant. The topics covered include: how to incorporate external evidence, how to account for uncertainty, and advanced modelling methods. Many of the methods described are cutting-edge, and there is an accompanying editorial which provides a useful overview.

How to Appropriately Extrapolate Costs and Utilities in HTA

All of the other articles linked to focus on survival outcomes. This article considers instead the extrapolation of costs and health-related quality of life (as measured using utilities). Example approaches from the literature are described and critiqued with regards to advantages and disadvantages.
One approach not considered by this article is to jointly-model outcomes. The obvious example is to jointly model survival and quality of life, as one may provide information about the other. This was originally considered almost 20 years ago, but has been largely neglected since.

Case-study Highlighting the Issues of Extrapolation (2016) 

This is a talk that I gave at the Royal Statistical Society in 2016. It uses an economic evaluation of screening for ovarian cancer, and considers different plausible methods for predicting the future impact of screening on cancer-related mortality. It is shown that different plausible methods can result in drastically different results. It has some novel methodological applications; with flexible spline (Royston-Parmar) models, time-series models and model discrepancy methods used for extrapolation in HTA.

Cautions Regarding the Fitting and Interpretation of Survival Curves (2011)

This article discusses two HTA submissions in depth. It provides good illustrations of what is now known to be a common occurrence: competing statistical models give similar fits to the observed data, but markedly different extrapolations. This article also reinforces the importance of both clinical plausibility and of considering alternative statistical models in sensitivity analyses.


Any thoughts on these? Any other articles you think should be added? Comment below or continue the discussion on Padlet!