Predictive Machine Learning Models Help Reduce Missed Appointments in Primary Care

Researchers at Pennsylvania State University have developed and tested machine learning models aimed at predicting the likelihood of patients missing or canceling appointments in primary care settings. Published in The Annals of Family Medicine, their study explores how integrating patient history, socioeconomic factors, and environmental data can improve appointment adherence.

The team analyzed data from over 109,000 patients across 15 family medicine clinics, encompassing more than 1.1 million appointments. They categorized outcomes into three groups: completed visits, no-shows, and late cancellations (within 24 hours). Using four machine learning approaches—gradient boost, random forest, neural networks, and LASSO logistic regression—they identified the most effective model for predicting no-shows and last-minute cancellations.

The gradient boost model emerged as the top performer, achieving AUROC scores of 85% for predicting no-shows and 92% for late cancellations. Notably, the models found no bias based on patient demographics such as gender or ethnicity. Key predictors included the time from appointment request to scheduled date, prior missed appointment rates, socioeconomic status, and health indicators.

The findings highlight important barriers to appointment adherence, such as longer scheduling lead times and socioeconomic challenges faced by certain patient groups, including younger, female, and ethnic minority populations. Recognizing these factors enables healthcare systems to tailor interventions—like reminder texts or transportation support—to improve attendance rates.

This innovative approach demonstrates how personalized, data-driven strategies can enhance primary care efficiency and patient care. Implementing such predictive models could significantly reduce appointment no-shows, optimize clinic operations, and improve health outcomes.

source: https://medicalxpress.com/news/2025-07-machine-primary-clinics.html