Personalized Physician-Assisted Sleep Advice for Shift Workers: Algorithm Development and Validation Study.

Area of Science:

• Sleep Medicine
• Artificial Intelligence
• Occupational Health

Background:

• Shift work disrupts circadian rhythms, leading to shift work sleep disorder.
• Cognitive behavioral therapy for insomnia (CBT-I) principles guide effective management strategies.
• Personalized interventions are crucial for addressing individual sleep disturbances.

Purpose of the Study:

• To develop and evaluate machine learning (ML) algorithms for predicting physician-provided sleep advice.
• To create a system offering individualized sleep and behavior recommendations for shift workers.
• To leverage wearable and survey data for personalized sleep intervention.

Main Methods:

• Collected 5 weeks of Fitbit, survey, and sleep advice data from 61 shift workers in Japan.
• Engineered physiological and behavioral features, then used hierarchical clustering to identify participant clusters.
• Explored random forest, light gradient-boosting machine, and CatBoost models with various data-balancing techniques to predict frequent advice messages.

Main Results:

• Participant clusters were distinguished by work shifts and distinct behavioral patterns, correlating with sleep quality.
• ML models significantly outperformed baseline predictions (P<.001), achieving high area under the precision-recall curve.
• Feature importance analysis revealed that models accurately used relevant data, e.g., bedroom brightness for "darken the bedroom" advice.

Conclusions:

• An accurate ML algorithm shows potential for automated, trustworthy sleep advice generation for shift workers.
• The current system, while promising, is limited to predicting the 7 most frequent advice messages.
• Further research is needed to expand predictive capabilities to less common sleep advice recommendations.