Machine Learning Prediction of Healthcare Utilization Using National Open Primary Care Clinic Data

Gulnura Kalbagayeva

2026

Full text Open access: Hybrid Metadata (CSL-JSON)
DOI 10.4103/wkpj.wkpj_21_26
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

A BSTRACT Background: Healthcare utilization is an important indicator of health system performance and demand for healthcare services. Accurate prediction of healthcare utilization may help healthcare organizations improve planning, workforce allocation, and distribution of healthcare resources. Objective: The objective is to predict annual primary care healthcare utilization measured by total encounters using clinic staffing and geographic variables from a national open dataset. Methods: A retrospective secondary analysis was conducted using the “Primary Care Clinic Annual Utilization Data” dataset obtained from Data.gov. The outcome variable was total annual primary care encounters (TOT_ENC_PCP_ENC). Predictor variables included physician full-time equivalents (FTEs), family nurse FTEs, registered nurse FTEs, percentage of nonEnglish-speaking patients, total staffing involved in primary care encounters, physician encounters, physician assistant encounters, and county. Missing numerical values were handled using median imputation, and the county was converted into numerical variables using one-hot encoding. The dataset was divided into training (80%) and testing (20%) sets. Linear regression and random forest models were developed in Python using scikit-learn. Model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination ( R 2 ). Results: Both models demonstrated high predictive performance. Linear regression achieved an RMSE of 6366.23, an MAE of 4076.90, and an R 2 of 0.93. Random forest demonstrated better performance, with an RMSE of 5758.00, an MAE of 3187.44, and an R 2 of 0.94. Physician encounters were the strongest predictor of healthcare utilization, followed by total staffing involved in primary care encounters. Conclusion: Random forest accurately predicted annual primary care healthcare utilization and performed better than linear regression. Physician encounters and staffing-related variables were the strongest predictors of healthcare utilization. These findings suggest that machine learning models may be used to forecast future demand for primary care services and support healthcare planning and resource allocation. However, external validation in other geographic regions is needed before broader implementation.