Cost-Effectiveness & Value-of-Information Analysis Using Machine Learning-Based Metamodeling: A Case of Hepatitis C Treatment
2022
This study compares machine learning (ML)-based metamodels with conventional metamodeling approaches in replicating the outcomes of a complex simulation model of hepatitis C virus (HCV) natural history. Three ML-based metamodels (random forest, support vector regression, and artificial neural networks) and a linear regression-based metamodel were developed from the simulation model's 40 input parameters. Performance evaluation included root mean squared error (RMSE) and Pearson's R2 on normalized data. ML-based metamodels generally outperformed linear regression, with random forest demonstrating the best performance. R2 values and RMSE indicated superior performance of ML-based metamodels in replicating societal costs, quality-adjusted life-years, and incremental cost-effectiveness ratios. Cost-effectiveness analysis curves and expected value of perfect information produced by the random forest metamodel closely matched simulation output. The study concludes that ML-based metamodels, particularly random forest, offer enhanced performance in replicating complex simulation model findings, suggesting their utility in decision-making processes and the development of decision-support tools for healthcare interventions.
Source:
McCandlish JA, Ayer T, Chhatwal J. Cost-Effectiveness and Value-of-Information Analysis Using Machine Learning-Based Metamodeling: A Case of Hepatitis C Treatment. Medical Decision Making 2022; 43 (1): 68-77. https://doi.org/10.1177/0272989x221125418