Abstract
PURPOSE: Factors influencing missed appointments are complex and difficult to anticipate and intervene against. To optimize appointment adherence, we aimed to use personalized machine learning and big data analytics to predict the risk of and contributing factors for no-shows and late cancellations in primary care practices. METHODS: We conducted a retrospective longitudinal study leveraging geolinked clinical, care utilization, socioeconomic, and climate data from 15 family medicine clinics at a regional academic health center in Pennsylvania from January 2019 to June 2023. We developed multiclass machine learning models using gradient boost, random forest, neural network, and logistic regression to predict appointment outcomes, followed by feature importance analysis to identify contributing factors for no-shows or late cancellations at the population and patient levels. We performed stratified analysis to evaluate the prediction performance by sex and race/ethnicity to ensure the fairness of the final model among sensitive features. RESULTS: The analysis consisted of 109,328 patients and 1,118,236 appointments, including 77,322 (6.9%) no-shows and 75,545 (6.8%) late cancellations. The gradient boost model achieved the best performance with an area under the receiver operating characteristic curve of 0.852 for predicting no-shows and 0.921 for late cancellations. No bias against patient characteristics was detected. Schedule lead time was identified as the most important predictor of missed appointments. CONCLUSIONS: Missed appointments remain a challenge for primary care. This study provided a practical and robust framework to predict missed appointments, laying the foundation for developing personalized strategies to improve patients' adherence to primary care appointments.