gms | German Medical Science

Objective: Economic evaluations are increasingly important to help decision makers to efficiently allocate scarce resources. Despite this importance, the use of economic evaluations by decision makers is still largely limited in part due to their low quality and the mistrust between decisions-makers regarding the validity of the methods used. Indeed, several biases can occur when performing economic evaluations. It is therefore important for decision makers to be able to assess potential biases and for researchers to minimize them. Earlier research revealed, which biases exist in trial-based economic evaluation. This article aims to identify biases that are specifically related to model-based economic evaluations.

Methods: Using the Philips guideline for good practices in decision-analytic modelling as a framework, we identified sources of bias in model-based economic evaluation. Biases were identified through literature review, systematic reviews using the Philips checklist, working-group meetings and discussion with experts. The different biases identified are illustrated using evidence from earlier studies.

Results: Several specific biases for model-based economic evaluations were identified. These biases related to structural assumptions, model type, time horizon data selection (such as treatment effects), assessment of uncertainty and internal validation were discussed and illustrated by examples from the literature. Evidence form the literature illustrates that the impact of these biases can be massive, changing the outcomes from being highly cost-effective to not being cost-effective.

Conclusion: In this study, we identified several biases that are related to model-based economic evaluations. Such biases could be helpful for decision-makers and for researchers that have to minimize them The development of a checklist addressing biases in model-based economic could be the next forward leading to a better understanding and quality of economic evaluation. Next to that transparency is needed both in the reporting as well as open access to data and models.