When making decisions under uncertainty, it is reasonable to choose the path that leads to the highest expected net benefit. Therefore, to inform decision making, decision-model-based health economic evaluations should always present expected outputs (i.e. the mean costs and outcomes associated with each course of action). In non-linear models such as Markov models, a single 'run' of the model with each input at its mean (a deterministic analysis) will not generate the expected value of the outputs. In a worst-case scenario, presenting deterministic analyses as the base case can lead to misleading recommendations. Therefore, the base-case analysis of a non-linear model should always be the means from a probabilistic analysis. In this paper, I explain why this is the case and provide recommendations for reporting economic evaluations based on Markov models, noting that the same principle applies to other non-linear structures such as partitioned survival models and individual sampling models. I also provide recommendations for conducting one-way sensitivity analyses of such models. Code illustrating the examples is provided in both Microsoft Excel and R, along with a video abstract and user guides in the electronic supplementary material. Supplementary file 6 (MP4 20900 kb).
© 2021. The Author(s), under exclusive licence to Springer Nature Switzerland AG.