Digital Phenotyping of High-Risk Gaming Behavior Using Wearable Devices.

Abstract

Background: Gaming addiction is a serious mental health issue. Early identification of high-risk gaming behaviors (HRGB) is crucial for prevention, yet research using wearable data for detection remains limited. This study analyzed the temporal dynamics of gaming behavior and health indicators using 28-day summary statistics derived from longitudinal data in young adults with HRGB, thereby providing a basis for gaming disorder prevention and intervention.

Methods: This 28-day longitudinal study employed a self-designed system implemented on the Huawei Research Platform for participant recruitment and data collection. Participants completed daily mobile app questionnaires between 18:00 and 23:00 assessing gaming time, mood, and sleep, while a smart band continuously collected physiological and activity data. Emotional states were assessed weekly using the Patient Health Questionnaire-9 (PHQ-9) and the Generalized Anxiety Disorder-7. Machine learning models were applied to compare the predictive performance of active (self-report), passive (wearable), and combined data in identifying HRGB, and to determine key predictive factors relevant for intervention.

Results: The final sample comprised 37 participants in the HRGB group and 23 in the non-HRGB group. Significant between-group differences were observed across multiple indicators. Combined data yielded the highest area under the receiver operating characteristic curve (AUROC) in five of six models (logistic regression: 0.86 vs 0.81). Wearable-derived data enhanced early classification of HRGB, with depression severity (PHQ-9), heart rate, and sleep metrics emerging as potential markers. A one-standard-deviation increase in PHQ-9 scores was associated with a 3.51-point increase in Gaming Disorder Screening Scale (GDSS) scores (89% credible interval [CrI], 1.85-5.20), while a higher heart rate index was associated with a 1.36-point increase in GDSS scores (89% CrI, 0.13-2.60). The population attributable fraction was 15.51% for PHQ-9.

Conclusions: Wearable data enhances early HRGB classification. PHQ-9, heart rate, and sleep metrics are key predictors, showing clinical utility.