OBJECTIVES: Economic models are frequently complex and the published results are often of limited value in other settings due to differences in parameter values. We explored the potential of using regression to summarise a model as a formula, enabling sensitivity analyses without access to the underlying model.

METHODS: We used a case study of an economic model of mechanical thrombectomy for the management of acute ischaemic stroke, compared with intravenous thrombolysis. The Markov model incorporated 40 parameters to estimate cost-utility. Regression was used to predict the model-generated incremental costs and benefits from the input parameters. We reran the economic model 1,000 times, changing the mean values of between one and ten randomly selected input parameters. The model outputs were compared with those generated using the regression formula with the updated mean parameter values. The simulation was also run using only extreme parameter value changes.

RESULTS: The ICERs estimated using the two approaches were highly correlated: 99% of simulations resulted in the regression formula ICER being within 5% of the model generated ICER. The accuracy of the regression formula estimates reduced when more extreme mean parameter values were used, with only 81% of simulations within 5% of the model-generated ICER. However, accuracy remained high when less than five parameters were set at extreme values at once (96% within 5% of the model-generated ICER).

CONCLUSIONS: Summarising a model using a regression approach can facilitate rapid sensitivity and scenario analyses without rerunning the model. This approach could enable HTA agencies with limited resources to explore the impact of alternative parameter values specific to their setting without access to the original model. The success of this approach may depend on the complexity of the underlying model and whether the alternative parameter values are markedly outside the confidence bounds used in the model.