OBJECTIVES: Health economic evaluations are essential for assessing the comparative costs and benefits of new health technologies. A critical aspect of these evaluations is accounting for heterogeneous treatment effects (HTE), which can greatly impact decision-making. Nevertheless, the modelling of HTE is usually discouraged when the supporting evidence is limited. This study aims to explore whether modelling HTE is warranted for population decision-making when the underlying evidence is still uncertain.

METHODS: We set up a simulation study where the analysis of a randomised controlled trial is used to inform the decision-modelling with a three-state state transition model. We compare five outcome modelling approaches under various outcome generating processes: unadjusted, adjusted, linear interaction, unrestricted spline, and monotonic spline models. The modelling approaches are assessed on their ability to predict population-level incremental net monetary benefits and their performance in probabilistic cost-effectiveness predictions in the target population.

RESULTS: Across all outcome generating processes, the two spline models consistently assign lower probabilities of the intervention being cost-effectiveness, mitigating over-confidence when the treatment effects decrease across the target population. The predictive accuracy of different outcome modelling approaches were comparable in all scenarios. Meanwhile, none of the methods provided sufficiently accurate point predictions.

CONCLUSIONS: Modelling HTE, even under limited supporting evidence, can mitigate over-confidence in decision-making. Modellers should adopt flexible modelling approaches with careful considerations of regularisation and extrapolations. The lack of accuracy in point predictions also highlights the importance of communicating the uncertainty when the supporting evidence is limited.