Ai-enabled wearable and digital biomarkers for stress and mental health monitoring: a systematic literature review

Abstract

Stress, anxiety, depression, and post-traumatic stress disorder (PTSD) carry a heavy burden. Yet the usual way we assess them still leans on episodic, self-reported instruments, which don't always reflect real life. Now wearable biosensors, smartphones, and artificial intelligence (AI) are opening the door to digital biomarkers. In this systematic literature review, we looked at how AI takes physiological and behavioural signals from wearables and smartphones to detect, predict, and monitor stress and mental-health conditions. We followed PRISMA 2020. A Scopus search turned up 403 records. After 14 duplicates were removed, 389 records went into screening. From there, 88 full texts were assessed, and 15 studies published between 2018 and 2026 made the final cut. Each eligible study combined digital sensing with AI or machine learning and linked those signals to outcomes. Three main themes came out of the synthesis. Heart-rate variability remains the dominant biomarker. Multimodal signals, combining electrodermal, cardiac, and electroencephalographic data, tend to improve classification. And smartphone phenotyping pushes monitoring further into depression, social anxiety, and trauma. The reported performance is promising, but small samples, device heterogeneity, and limited interpretability still hold these models back. Future work needs to prioritise larger cohorts, standardised reporting, explainability, and proper clinical validation.