Development and Temporal Validation of an Electronic Medical Record-Based Insomnia Prediction Model Using Data from a...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Sleep Medicine Research

Background:

• Insomnia is a prevalent sleep disorder impacting public health.
• Early identification of insomnia risk is crucial for timely intervention.
• Electronic Medical Records (EMR) offer a rich data source for predictive modeling.

Purpose of the Study:

• To develop and temporally validate an EMR-based prediction model for insomnia.
• To assess the model's performance in identifying insomnia risk before clinical detection.
• To evaluate the scalability and longitudinal viability of EMR-based insomnia screening.

Main Methods:

• A nested case-control study using EMR data from 2011-2018.
• Trained machine learning models (including XGBoost) on demographics, diagnoses, and medication data.
• Validated model performance on holdout sets and subsequent years to assess temporal validity.

Main Results:

• An extreme gradient boosting (XGBoost) model achieved an AUC of 0.80.
• Models demonstrated strong predictive performance, with AUCs of 0.80 and 0.70 for different time windows.
• Temporal validation showed minimal performance drop (≤4% AUC) even with a five-year data gap.

Conclusions:

• EMR-based prediction models can identify insomnia risk up to six months prior to clinical detection.
• The developed models offer an inexpensive, scalable, and longitudinally viable screening method.
• This approach facilitates early identification and potential intervention for individuals at high risk of insomnia.