Machine learning approaches to predicting no-shows in pediatric medical appointment

Area of Science:

• Healthcare Operations Research
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Patient no-shows disrupt treatment continuity and waste clinical resources.
• Predictive modeling for no-shows can optimize healthcare delivery and patient outcomes.

Purpose of the Study:

• To develop an interpretable deep learning model for predicting medical appointment no-shows at the time of scheduling.
• To address challenges like missing patient data and improve prediction accuracy using external factors like weather.

Main Methods:

• A retrospective study in a pediatric teaching hospital using a deep learning approach.
• Incorporated data imputation for missing records (77% of patients) and local weather data.
• Developed an interpretable model to explain individual patient risk predictions.

Main Results:

• The model achieved 83% accuracy in identifying no-shows at scheduling with a <17% false alert rate.
• Outperformed baseline models, demonstrating effectiveness even with incomplete patient data.
• Identified patients' past no-show history as the strongest predictor.

Conclusions:

• An interpretable deep learning model can effectively predict patient no-shows early in the scheduling process.
• This predictive capability allows for targeted interventions to reduce no-shows and improve resource allocation.
• Future research can focus on implementing and evaluating interventions for high-risk patients.