Using Machine Learning to Predict Primary Care and Advance Workforce Research.

Area of Science:

• Health Informatics
• Machine Learning
• Medical Workforce Research

Background:

• Accurate physician specialty data is crucial for healthcare workforce planning.
• Existing methods for specialty identification can be time-consuming or lack real-time accuracy.
• Medicare claims data offers a rich, albeit complex, source for physician practice pattern analysis.

Purpose of the Study:

• To develop and validate a machine-learning model for predicting physician specialties using Medicare claims.
• To assess the model's performance in identifying primary care physicians and other specialists.
• To provide a tool for near real-time assessment of the primary care workforce.

Main Methods:

• Utilized 2014-2016 Medicare prescription and procedure claims data.
• Trained three sets of random forest classifiers: prescription-only, procedure-only, and combined.
• Evaluated model performance using F1 scores and area under the receiver operating characteristic curve (AUROC) on a held-out testing cohort.

Main Results:

• The combined model achieved a superior aggregate F1 score (0.876) compared to prescription-only (0.745) or procedure-only (0.821) models.
• Mean F1 scores for the combined model ranged from 0.533 to 0.987, with primary care achieving a mean F1 score of 0.920.
• The combined model demonstrated a high mean AUROC of 0.992, with primary care specifically at 0.982.

Conclusions:

• A machine-learning approach using Medicare claims data effectively predicts physician specialties with high accuracy.
• The developed model provides a valuable tool for near real-time assessment of primary care practice.
• This method has significant implications for improving primary care workforce research by overcoming data limitations.