Temporal extrapolation in the cost-effectiveness analysis of healthcare interventions

Abstract: Cost-effectiveness models are constructed in order to aid decision making regarding resource allocation and the need for further research in healthcare. The evidence that informs input parameter values is often short-term in nature, whereas the appropriate model time horizon is often long-term. It is therefore common that significant ‘temporal uncertainty’ exists in relation to several parameters, and indeed structures, within a model. It may be possible to extrapolate the available short-term evidence over time in order to estimate outcomes over the unobserved period. However, it is unlikely that our true temporal uncertainty can be characterised by using the short-term evidence alone.

Based on the tenets of cost-effectiveness analysis, current best practice regarding decision modelling and the implications of a long-term time horizon, this work proposes a set of broad principles for appropriate temporal extrapolation in cost-effectiveness models.

A case study is then outlined (a re-analysis of a cost-effectiveness model using data from the RITA 3 trial), where these principles are interpreted and applied in this specific decision context. In particular, a modelling approach is selected where the two parameters of interest (baseline risk and treatment effect) are modelled independently. Alternative plausible temporal scenarios are generated for each parameter with a view to incorporating the scenarios into the decision model through model averaging or parameterisation in order to characterise the full breadth of temporal uncertainty.