Panic Attack Prediction Using Wearable Devices and Machine Learning: Development and Cohort Study

Chan-Hen Tsai; Pei-Chen Chen; Ding-Shan Liu; Ying-Ying Kuo; Tsung-Ting Hsieh; Dai-Lun Chiang; Feipei Lai; Chia-Tung Wu

doi:10.2196/33063

Panic Attack Prediction Using Wearable Devices and Machine Learning: Development and Cohort Study

JMIR Med Inform. 2022 Feb 15;10(2):e33063. doi: 10.2196/33063.

Authors

Chan-Hen Tsai^{1

2}, Pei-Chen Chen¹, Ding-Shan Liu³, Ying-Ying Kuo², Tsung-Ting Hsieh¹, Dai-Lun Chiang⁴, Feipei Lai^{1

3}, Chia-Tung Wu³

Affiliations

¹ Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei City, Taiwan.
² Department of Psychiatry, En Chu Kong Hospital, New Taipei City, Taiwan.
³ Department of Computer Science and Information Engineering, National Taiwan University, Taipei City, Taiwan.
⁴ Financial Technology Applications Program, Ming Chuan University, Taoyuan City, Taiwan.

PMID: 35166679
PMCID: PMC8889475
DOI: 10.2196/33063

Abstract

Background: A panic attack (PA) is an intense form of anxiety accompanied by multiple somatic presentations, leading to frequent emergency department visits and impairing the quality of life. A prediction model for PAs could help clinicians and patients monitor, control, and carry out early intervention for recurrent PAs, enabling more personalized treatment for panic disorder (PD).

Objective: This study aims to provide a 7-day PA prediction model and determine the relationship between a future PA and various features, including physiological factors, anxiety and depressive factors, and the air quality index (AQI).

Methods: We enrolled 59 participants with PD (Diagnostic and Statistical Manual of Mental Disorders, 5th edition, and the Mini International Neuropsychiatric Interview). Participants used smartwatches (Garmin Vívosmart 4) and mobile apps to collect their sleep, heart rate (HR), activity level, anxiety, and depression scores (Beck Depression Inventory [BDI], Beck Anxiety Inventory [BAI], State-Trait Anxiety Inventory state anxiety [STAI-S], State-Trait Anxiety Inventory trait anxiety [STAI-T], and Panic Disorder Severity Scale Self-Report) in their real life for a duration of 1 year. We also included AQIs from open data. To analyze these data, our team used 6 machine learning methods: random forests, decision trees, linear discriminant analysis, adaptive boosting, extreme gradient boosting, and regularized greedy forests.

Results: For 7-day PA predictions, the random forest produced the best prediction rate. Overall, the accuracy of the test set was 67.4%-81.3% for different machine learning algorithms. The most critical variables in the model were questionnaire and physiological features, such as the BAI, BDI, STAI, MINI, average HR, resting HR, and deep sleep duration.

Conclusions: It is possible to predict PAs using a combination of data from questionnaires and physiological and environmental data.

Keywords: lifestyle; machine learning; panic attack; panic disorder; prediction; wearable device.

©Chan-Hen Tsai, Pei-Chen Chen, Ding-Shan Liu, Ying-Ying Kuo, Tsung-Ting Hsieh, Dai-Lun Chiang, Feipei Lai, Chia-Tung Wu. Originally published in JMIR Medical Informatics (https://medinform.jmir.org), 15.02.2022.