Sleep, physical activity and panic attacks: A two-year prospective cohort study using smartwatches, deep learning and
Chan-Hen Tsai1, Mesakh Christian2, Ying-Ying Kuo3
1Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei City, Taiwan; Department of Psychiatry, En Chu Kong Hospital, New Taipei City, Taiwan.
Sleep Medicine
|December 28, 2023
Summary
Deep learning models accurately predict panic attacks and anxiety symptoms. Adequate sleep and physical activity can reduce the occurrence of panic attacks in individuals with panic disorder.
Area of Science:
- Psychiatry and Behavioral Science
- Computational Neuroscience
- Digital Health
Background:
- Panic disorder (PD) management lacks sufficient data on sleep and physical activity's predictive role.
- This study addresses the need for predictive models for panic attacks (PA) and anxiety symptoms.
Purpose of the Study:
- To predict panic attacks (PA), state anxiety (SA), trait anxiety (TA), and panic disorder severity (PDS) using deep learning.
- To identify lifestyle factors influencing PA occurrence in PD patients.
Main Methods:
- A two-year prospective cohort study involving 114 PD patients.
- Utilized deep learning (RNN, LSTM, GRU) and SHAP explainable AI on data from clinical questionnaires and wearable devices.
- Collected data on sleep, physical activity, heart rate, and anxiety/depression scales.
Main Results:
- LSTM model achieved 7-day prediction accuracies of 92.8% for PA, 83.6% for SA, 87.2% for TA, and 75.6% for PDS.
- Higher baseline anxiety/depression scores and comorbidities predicted increased PA risk.
- Specific sleep durations (6h 23min–10h 50min), deep sleep (>50min), and physical activity (e.g., climbing stairs) were associated with decreased PA.
Conclusions:
- Deep learning models demonstrate high accuracy in predicting recurrent PA and anxiety symptoms.
- Optimizing sleep and physical activity levels can be a viable strategy for reducing PA frequency in PD.
