Towards anticipating panic attacks using a smartwatch: a pilot study

Two-month pilot: 19 enrolled, 14 completed people with frequent panic attacks wore a Fitbit (model not named). LSTM on the past 3 days predicted a next-day panic attack with AUC 0.69, F1 0.40. SHAP top features: heart rate, resting HR, light sleep, restless duration, minutes awake, REM sleep.

• Form factor: wrist (Fitbit)

Around 2-3% of the US population suffers from frequent panic attacks (PAs), though global prevalence remains poorly documented. Only a few studies have explored the potential of smart wearable devices in predicting panic attacks; however, the devices tend to either not be wearable daily or rely on regular psychological questionnaires to achieve adequate prediction. As such, this study aimed to track physiological features of individuals with PAs using a Fitbit smartwatch to anticipate episodes in near-real time to inform the design of a device or intervention that can support patients upon the onset of a PA event. Of the 19 enrolled participants, 14 completed or were close to completing the 2-month study. Preliminary findings show that LSTM achieved the best results for predicting a PA happening the next day (AUC = 0.69, F1 = 0.40) by assessing the past 3 days of data. A SHAP analysis identified heart rate, resting heart rate, light sleep, restless duration, minutes awake, and REM sleep as significant features for predicting a PA onset. Work is in progress to expand the sample size to validate these results and identify the important features and ideal timing of a consequent intervention that helps patients manage their PAs.

Verbatim from the publisher's DOI record (opens in a new tab) · Publisher licence: https://www.asme.org/publications-submissions/publishing-information/legal-policies

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.