Approaches to modelling in decision-analytic health economic evaluations: Factors to differentiate between approaches

Abstract

BACKGROUND & OBJECTIVES Decision-analytic modelling can inform healthcare resource allocation and reimbursement decision-making, with modelling approaches adapted from a variety of disciplines. The objective of this thesis was to investigate the evidence surrounding when each approach should be used when conducting health economic evaluations. METHODS Project 1: A systematic review identified selection criteria, referred to as factors, through an evaluation of existing decision frameworks that aimed to differentiate between models. Project 2: Employing the factors identified from Project 1, a systematic review explored the extent to which empirical cross-validation studies agree on the importance of each on its impact to model selection. Project 3: A decision tree evaluating the cost-effectiveness of two vaccination strategies in children was reconstructed as system dynamics and agent-based models and compared. Scenario analyses assessed the situations whereby the model’s results would be sensitive to or be better handled by a particular approach. RESULTS Project 1: Among the eight frameworks identified; each involved a different set of structural or practical factors. Disagreements emerged between frameworks in the classification of the structural features specific to each modelling approach. Project 2: Nine exercises have been conducted, mostly focused on the criteria of interactivity (i.e., static vs. dynamic) and population resolution (i.e., aggregate vs. individual). Aggregate- and individual-level models were found to produce similar results with a practical trade-off between validity and feasibility. In the presence of large indirect effects, dynamic and static models often produced disparate results. Project 3: When calibrated, all three approaches reached consistent findings. Adaptation away from base-case assumptions led to different quantitative results on which vaccination strategy would be most optimal. CONCLUSION Despite disagreement among the frameworks on how to recommend modelling approaches, consistent conclusions were observed in empirical cross-validation studies. More empirical evidence is therefore needed to improve one’s understanding of the impact of the remaining factors on model selection.