Machine Learning for Health Services Researchers

Area of Science:

• Health Informatics
• Biostatistics
• Machine Learning Applications

Background:

• Machine learning (ML) is increasingly vital for predicting healthcare outcomes such as cost, utilization, and quality.
• Healthcare outcomes research benefits from ML's predictive capabilities.

Purpose of the Study:

• To provide a high-level overview of ML for healthcare outcomes researchers and decision-makers.
• To introduce key concepts for applying ML methods in healthcare outcomes research.

Main Methods:

• Describes standards for rigorously learning an estimator (ML algorithm for prediction).
• Covers data preparation, estimator selection, parameter learning, regularization, and evaluation.
• Compares decision tree, deep learning, and ensemble methods for predictive modeling.

Main Results:

• Demonstrates ML application using a simulated insurance claims dataset.
• Provides R and Python code for developing and evaluating estimators.
• Focuses on predicting heightened risk for hospitalization from ambulatory care-sensitive conditions.

Conclusions:

• Healthcare outcomes researchers must understand rigorous evaluation standards for ML-developed estimators.
• Different ML approaches are suited for distinct research problems in healthcare outcomes.