PRM200 - Parameter Importance Assessment in a Health Economic Evaluation Model for Heart Failure

Abstract

Along with uncertainty around the parameters and the initial parameter value assumptions used in health-economic evaluation models, an analysis of the uncertainty around the model inputs/outputs is essential. Parameter importance analysis (PIA) provides an explicit framework to quantitatively identify the contribution of each uncertain input to the output uncertainty. There are several methods available to be used in PIA. The objectives of this research were to investigate different PIA methods with the pros and cons of each method and identify the most robust method with respect to different initial parameter value assumptions. A health economic model for heart failure is developed to serve as a basis to implement different PIA methods. Six alternative methods are applied: One-way sensitivity analysis, rank correlation analysis, analysis of covariance (ANCOVA), dominance analysis, standardized regression analysis and expected value of perfect parameter information (EVPPI) analysis. Initial parameter assumptions are varied and the robustness of each method is assessed with respect to how close the parameter importance rankings are with different initial parameter assumptions. Each technique/initial parameter values’ assumption combination generates a different ranking for the importance of the parameters that explain the uncertainty around the expected net monetary benefit with £20,000/QALY. EVPPI is the most robust method with respect to different initial parameter assumptions. However it is the most demanding method in terms of computation time. On the opposite side, one-way sensitivity analysis is the least computation time demanding method; however the importance rankings are very susceptible to change with different initial assumptions. Other Monte-Carlo simulation based methods (e.g. ANCOVA, dominance, standardized regression and rank correlation analysis) are alternative PIA methods, which generate rather robust rankings with different initial parameter assumptions. These alternative methods require substantially less computation times compared to EVPPI with high consistency and robustness to different initial value assumptions.