Machine Learning Predicts Missed Primary Care Appointments Using Clinical, Geosocial, and Environmental Data

This study developed multiclass machine learning models using integrated clinical, demographic, geosocial, and climate data to predict appointment outcomes at primary care clinics, finding that the gradient boost model outperformed others with an overall accuracy of over 80%. Lead time was identified as the most important predictor of missed appointments, followed by factors like patient sex, age, socioeconomic status, and distance to the clinic, with the model demonstrating little bias toward sex or race/ethnicity. The study analyzed a large dataset and utilized SHAP analysis to understand individual patient risk factors, offering a framework for personalized interventions to improve appointment adherence and continuity of care.

Study limitations include the single center population, potentially limiting generalizability. The study period encompassed the COVID-19 pandemic, which may have affected appointment adherence. Future research should explore ensemble learning to optimize prediction accuracy and robustness. The analysis predicted the outcomes of each patient’s appointments separately instead of treating them as time-series data.

This analytical framework lays a foundation for health systems to assess individual risk of missed appointments and design personalized strategies to help patients adhere to primary care appointments.

Study limitations include the single center population, potentially limiting generalizability. The study period encompassed the COVID-19 pandemic, which may have affected appointment adherence. Future research should explore ensemble learning to optimize prediction accuracy and robustness. Researchers can consider more sophisticated deep learning approaches, such as recurrent neural networks, to assess individuals dynamic risk of missed appointments.

This analytical framework lays a foundation for health systems to assess individual risk of missed appointments and design personalized strategies to help patients adhere to primary care appointments.

Study limitations include the single center population, potentially limiting generalizability. The study period encompassed the COVID-19 pandemic, which may have affected appointment adherence. Future research should explore ensemble learning to optimize prediction accuracy and robustness. Researchers can consider more sophisticated deep learning approaches, such as recurrent neural networks, to assess individuals dynamic risk of missed appointments.

This analytical framework lays a foundation for health systems to assess individual risk of missed appointments and design personalized strategies to help patients adhere to primary care appointments.