Machine Learning Model Predicts No-Shows in Primary Care

Pennsylvania State University researchers have developed a machine learning model that accurately predicts no-shows and late cancellations in primary care, as reported in the Annals of Family Medicine. The study analyzed over 1 million appointments, including more than 77,000 no-shows and 75,000 late cancellations, using data from 15 family medicine clinics.

The research team compared four machine learning models—gradient boost, random forest, neural network, and LASSO logistic regression—to predict appointment outcomes. The gradient boost model performed best, achieving an AUROC score of 0.85 for no-shows and 0.92 for late cancellations. Importantly, the model's predictions were unbiased regarding patient sex or race/ethnicity.

Lead time, or the number of days between booking and the visit, emerged as a crucial predictor of missed appointments. Longer lead times, particularly those exceeding 60 days, were linked to a higher risk of no-shows. The study suggests that clinics could use these insights to prioritize shorter wait times for high-risk patients, thereby improving appointment adherence.

Annals of Family Medicine also published a related report discussing the need for data transformation to facilitate similar studies, emphasizing the importance of collaboration between AI communities and primary care sectors.