Incremental value of wearable digital phenotypes for anxiety under real-world missingness
Abstract Wearable sensing may enable low-burden digital phenotyping of anxiety, but its incremental value beyond brief symptom questionnaires remains uncertain under real-world missingness. We analyzed a public 4-week dataset from 49 adults with two 14-day participant-windows aligned to week-2 and week-4 GAD-7 assessments (98 windows; 22 with GAD-7 ≥ 5). XGBoost models based on demographic/lifestyle, wearable, questionnaire, and combined feature sets were evaluated by participant-level leave-one-subject-out cross-validation, with uncertainty estimated by participant-level cluster bootstrap. QUESTIONNAIRE models achieved the highest classification performance (AUC 0.888; average precision 0.826; Brier 0.085), whereas WEARABLE alone remained near chance (AUC 0.501). Adding wearables to QUESTIONNAIRE did not improve performance (ΔAUC − 0.045; ΔBrier 0.025). In the high-compliance subset (34 participants; 68 windows), wearable-only discrimination remained weak; questionnaire-plus-wearable point estimates were more favorable, but confidence intervals included the null. Continuous GAD-7 analyses showed improved prediction beyond demographic/lifestyle variables but little added value beyond questionnaires. These findings indicate baseline-dependent incremental value: wearable-derived features may provide adjunctive information when questionnaire data are unavailable or sparse, but their added value was limited once baseline symptom questionnaires were available.
Authors
- Xiangming Meng (ORCID: https://orcid.org/0000-0001-8250-9887)
- Mingjing Cai
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-12
- DOI
- https://doi.org/10.1038/s41598-026-80666-z
- Primary Topic
- Mental Health Research Topics
- Type
- article
- Field-Weighted Citation Impact
- 0.00