Beyond Efficiency Scores: Explaining Health System Performance Using Two-Stage Bootstrap DEA and Machine Learning.

Background/

Objectives

Health systems involve numerous stakeholders interconnected through nonlinear relationships. While Data Envelopment Analysis (DEA) has been widely used to measure health system efficiency, conventional estimates may exhibit finite-sample bias. An important question, therefore, concerns how health system performance can be measured more reliably, and what factors explain cross-country differences in efficiency. This study introduces an integrated framework that combines Two-Stage Bootstrap DEA with machine learning to assess the performance of the health systems of 26 OECD countries using 2022 data.

Methods

In the first step, technical efficiency scores are computed using an output-oriented constant returns to scale (CRS) DEA model. Subsequently, bias-corrected efficiency estimates are derived using the Bootstrap procedure proposed by Simar and Wilson. In the second step, truncated regression analysis and machine learning-based partial dependence analysis, the latter validated through leave-one-out cross-validation, are employed to investigate the determinants of efficiency.

Results

The Bootstrap procedure reveals statistically significant differences from conventional DEA results, and bias-corrected results indicate that South Korea, Canada, and the United States achieve the highest efficiency levels. The findings show that tobacco use prevalence has a significantly negative association with health system efficiency and alcohol consumption exhibits a negative, threshold-type pattern, while GDP per capita and out-of-pocket health expenditure display more complex, nonlinear effects. Furthermore, the scenario analysis indicates that a 10% reduction in tobacco use yields the largest predicted single-intervention improvement, while combined interventions produce additional but sub-additive gains.

Conclusions

The proposed framework presents a transparent and validated approach for assessing and explaining health system performance, generating findings relevant to the development of evidence-based health policy.