A two-stage super learner for healthcare expenditures

Area of Science:

• Health Economics
• Biostatistics
• Machine Learning

Background:

• Accurate estimation of healthcare expenditures is crucial for policy and resource allocation.
• Traditional methods struggle with the inherent skewness and zero-inflation common in healthcare cost data.
• Developing robust statistical methods is essential for reliable healthcare expenditure analysis.

Purpose of the Study:

• To introduce and evaluate a novel two-stage super learner method for improved healthcare expenditure estimation.
• To address challenges posed by skewed and zero-inflated data in healthcare cost analysis.
• To compare the performance of the proposed method against existing techniques.

Main Methods:

• Proposed a two-stage super learner ensemble machine learning approach.
• Separately estimated the probability of healthcare expenditure and the mean expenditure amount.
• Incorporated various regression and machine learning algorithms (e.g., random forests) for each stage.
• Evaluated performance using Mean Squared Error (MSE) and R-squared (R²) on simulated and real-world datasets (MEPS, BOLD).

Main Results:

• The two-stage super learner demonstrated superior performance compared to one-stage super learner and individual algorithms.
• Significant improvements were observed in healthcare cost estimation across diverse simulated and empirical settings.
• The advantage of the two-stage approach was particularly pronounced in datasets with high zero-inflation.

Conclusions:

• The proposed two-stage super learner is an effective method for enhancing healthcare expenditure estimation.
• This approach offers a robust solution for analyzing complex healthcare cost data.
• The findings support the adoption of advanced machine learning techniques for more accurate health economic analyses.