Comparative Efficiency Analysis of OECD Health Systems: FDH vs. Machine Learning Approaches with Efficiency Analysis Trees (EAT and RFEAT).

Abstract

Background: As health expenditure continues to rise due to income growth, technological advancements, and an aging population, it has become increasingly important to accurately measure and improve the efficiency of health systems. This is because financial resources are limited, and the allocation of resources can significantly influence the quality of health systems and health outcomes.

Methods: This study applies machine learning techniques-Efficiency Analysis Trees (EAT) and Random Forest for Efficiency Analysis Trees (RFEAT)-to evaluate the efficiency of health systems in 36 OECD countries, comparing the results with those from the traditional free disposal hull (FDH) method.

Results: Analysis shows high discrimination power in the order of RFEAT, EAT, and FDH. The correlation in efficiency rankings shows more than 80% similarity between RFEAT and EAT, while both show less than 80% similarity with FDH. According to RFEAT estimates, the countries with the highest efficiency are South Korea, Switzerland, and Costa Rica, whereas the United States, Lithuania, and Latvia are identified as the least efficient. The group-level analysis reveals that Asian countries, on average, perform more efficiently followed by Oceania, Europe, and the Americas. The groups with higher out-of-pocket healthcare expenditures per capita tend to show slightly better efficiency and the group with the smallest elderly population proportion exhibits the highest average health system efficiency.

Conclusion: Traditional methods like FDH are prone to inefficiency underestimation, especially in small samples with multiple variables. This study demonstrates the potential of machine learning approaches like EAT and RFEAT to provide more reliable efficiency estimates. These methods can help policymakers make better resource allocation decisions by mitigating inefficiency underestimation and offering greater discrimination power.