Machine Learning Models to Identify Clinically Significant Anxiety in Short-Term Insomnia Using Accelerometers

Area of Science:

• Sleep Medicine
• Artificial Intelligence
• Psychiatry

Background:

• Clinically significant anxiety (CSA) frequently co-occurs with short-term insomnia.
• Understanding the interplay between anxiety and sleep parameters is crucial for effective treatment.

Purpose of the Study:

• To investigate the relationship between CSA and subjective/objective sleep parameters in short-term insomnia.
• To develop machine learning (ML) models using accelerometer data to identify CSA.
• To explore the utility of accelerometer features in identifying anxiety in insomnia patients.

Main Methods:

• 205 participants with short-term insomnia were categorized into groups with (N=33) and without (N=172) CSA.
• Linear regression analyzed interaction effects between accelerometer features, CSA, and sleep problems.
• Multiple ML models (4 feature sets, 8 algorithms) were constructed; SHAP values assessed feature importance.

Main Results:

• CSA was associated with more severe subjective sleep issues.
• Significant interactions were found between anxiety, physical activity duration, and insomnia severity (P<0.05).
• Anxiety and interdaily stability interacted with sleep hygiene (P<0.01).
• The XGBoost model with weekday accelerometer data achieved an AUC of 0.777 for CSA identification.
• SHAP analysis highlighted circadian rhythm features' importance.

Conclusions:

• Machine learning effectively utilizes complex accelerometer data to identify CSA in short-term insomnia.
• Accelerometer-derived features, particularly circadian rhythm metrics, are valuable for detecting anxiety.
• SHAP-based visualizations offer practical, personalized insights for clinical decision-making.