Can AI predict no-shows before they happen? This new model says yes

A new study out of Penn State College of Medicine demonstrates that machine learning (ML) can accurately predict which patients are likely to miss or cancel their primary care appointments, potentially arming practices with tools to reduce care gaps and workflow disruptions.

Published this week in Annals of Family Medicine, the study analyzed more than 1.1 million appointments from 15 family medicine clinics across Pennsylvania between 2019 and 2023. Researchers developed and tested four ML models using integrated clinical, socioeconomic and environmental data to classify appointments as completed, no-show or late cancellation. The gradient boost model outperformed others, correctly identifying no-shows with 85.2% accuracy and late cancellations with 92.1% accuracy.

Lead time emerges as top predictor

The single most influential factor in missed appointments was lead time — the number of days between booking and the visit itself. Longer lead times (over 60 days) were strongly associated with a higher likelihood of missed visits, the study found.

“Given the strong effect of lead time, clinics could prioritize shorter wait times for high-risk patients,” the authors wrote.

Other key predictors included the physician’s years of practice, patient age, prior no-show history and whether the visit was scheduled with the patient’s usual physician. Area deprivation index, distance from home to clinic and even weather data like daily temperature also played a role in the model’s risk assessments.

Equitable and personalized predictions

Notably, researchers tested the gradient boost model for fairness across sex and racial/ethnic subgroups and found no significant performance bias. The tool’s predictions remained consistent across demographic lines, reinforcing its potential for equitable deployment.

The model also allowed for patient-specific insights. Using Shapley values — a technique drawn from game theory — the research team could quantify how each factor contributed to an individual’s predicted risk.

Implications for clinical practice

For physicians and practice managers, the study could open the door for targeted interventions like customized reminder systems, transportation assistance or prioritized scheduling for at-risk patients.

Wen-Jan Tuan, DHA, MS, MPH., lead author of the study — along with co-authors Yifang Yan, MS, Bilal Abou Al Ardat, M.D., Todd Felix, M.D., and Qiushi Chen, Ph.D. — wrote that the framework could help practices design “personalized strategies to improve patients’ adherence to primary care appointments.”

Though the model requires further validation beyond a single academic health system, the research underscores how predictive analytics can enhance continuity of care, reduce resource strain and minimize health disparities.

Also in the July 2025 issue of Annals of Family Medicine is a related special report calling for improved infrastructure to support more studies like this. It recommends better automation of data collection, integration of fragmented datasets and closer collaboration between the artificial intelligence (AI)/ML community and primary care professionals.

“These types of cross-sectoral collaborations are key to realizing the transformation of primary care data into a treasured resource that can unlock the true potential of artificial intelligence and machine learning in primary care,” the authors wrote.