ISPOR Abstract and Poster

When is a simple model enough? How to handle structural uncertainty in cost-effectiveness models.

Abstract

AUTHOR(S): Bryning Sam1, Grimsey Jones Frank2, Pollit Vicki2, Brennan Alan1

1University of Sheffield, Sheffield, UK, 2Symmetron Ltd, London, UK

OBJECTIVES: Whilst methods to formally account for structural uncertainty are not mandated by health technology assessment (HTA) guidelines, appreciation of their importance is growing and was raised in a recent HTA methods consultation. Health economists and decision makers need to understand if developing a more complex model is worthwhile to reduce decision uncertainty. This research tests and enhances published methods to assess structural uncertainty.

METHODS: A published discrepancy approach that assesses structural uncertainty was modified to be more efficient and pragmatic. This modified approach was applied in a simple Markov model to determine the value of a more complex model, given structural uncertainty as to the impact of disease progression. Discrepancy terms reflecting structural uncertainties were located within the model and were included within probabilistic sensitivity analysis and value of information analysis. To validate the approach, expected results of the simple model (with structural uncertainty accounted for) were compared to those of a more complex version.

RESULTS: The modified method when used in a simple model approximated the conclusions of a more complex model, suggesting the method could be used to indicate the value of building a more complex model. However, absolute costs and QALYs did not converge to the same extent as incremental outputs, suggesting that there is scope to increase robustness. Measures of the importance of the structural uncertainty were highly sensitive to assumptions about correlation. Specifying and accounting for correlation remains a key challenge in robustly accounting for structural uncertainty.

CONCLUSIONS: The modified discrepancy approach has potential for investigating the source, and impact, of structural uncertainty on model results and indicating the value of reducing such uncertainty. Further research is needed to establish the method’s robustness and further understand the impact of correlation.

Find out more about the leading author and their motivations for the project in our blog discussing this research.