M Arias1, F J Díez. 1. M. Arias, Dept. of Artificial Intelligence, UNED, Juan del Rosal, 16, 28040 Madrid, Spain, E-mail: marias@dia.uned.es.
Abstract
BACKGROUND: Cost-effectiveness analysis (CEA) is used increasingly in medicine to determine whether the health benefit of an intervention is worth the economic cost. Decision trees, the standard decision modeling technique for non-temporal domains, can only perform CEA for very small problems. OBJECTIVE: To develop a method for CEA in problems involving several dozen variables. METHODS: We explain how to build influence diagrams (IDs) that explicitly represent cost and effectiveness. We propose an algorithm for evaluating cost-effectiveness IDs directly, i.e., without expanding an equivalent decision tree. RESULTS: The evaluation of an ID returns a set of intervals for the willingness to pay - separated by cost-effectiveness thresholds - and, for each interval, the cost, the effectiveness, and the optimal intervention. The algorithm that evaluates the ID directly is in general much more efficient than the brute-force method, which is in turn more efficient than the expansion of an equivalent decision tree. Using OpenMarkov, an open-source software tool that implements this algorithm, we have been able to perform CEAs on several IDs whose equivalent decision trees contain millions of branches. CONCLUSION: IDs can perform CEA on large problems that cannot be analyzed with decision trees.
BACKGROUND: Cost-effectiveness analysis (CEA) is used increasingly in medicine to determine whether the health benefit of an intervention is worth the economic cost. Decision trees, the standard decision modeling technique for non-temporal domains, can only perform CEA for very small problems. OBJECTIVE: To develop a method for CEA in problems involving several dozen variables. METHODS: We explain how to build influence diagrams (IDs) that explicitly represent cost and effectiveness. We propose an algorithm for evaluating cost-effectiveness IDs directly, i.e., without expanding an equivalent decision tree. RESULTS: The evaluation of an ID returns a set of intervals for the willingness to pay - separated by cost-effectiveness thresholds - and, for each interval, the cost, the effectiveness, and the optimal intervention. The algorithm that evaluates the ID directly is in general much more efficient than the brute-force method, which is in turn more efficient than the expansion of an equivalent decision tree. Using OpenMarkov, an open-source software tool that implements this algorithm, we have been able to perform CEAs on several IDs whose equivalent decision trees contain millions of branches. CONCLUSION: IDs can perform CEA on large problems that cannot be analyzed with decision trees.
Authors: Wade P Smith; Patrick J Richard; Jing Zeng; Smith Apisarnthanarax; Ramesh Rengan; Mark H Phillips Journal: Transl Lung Cancer Res Date: 2018-04