Wearables provide life- saving data: when the signals are accurately interpreted

A veteran's smartwatch records every heartbeat, every step, every hour of sleep. That data currently sits in an app, reduced to a daily summary no one reads. URSA turns that passive stream into an active lifeline by detecting the behavioral and physiological signal shifts that precede a mental health crisis, often days before the wearer recognizes the danger.

The science supporting this approach is no longer speculative. Zhang et al. (2025) built an explainable anomaly detection framework using Fitbit data from over 2,000 participants and achieved an adjusted F1-score of 0.80 in identifying clinically meaningful increases in depression and anxiety symptoms. Their LSTM autoencoder learned each participant's normal patterns of sleep duration, step count, and resting heart rate, then flagged deviations from that individual baseline. Resting heart rate ranked as the most influential feature in 71.4% of detected anomalies, followed by reduced physical activity and shortened sleep. The model performed even better when depression and anxiety worsened simultaneously, reaching an F1-score of 0.84. URSA operates on the same principle of individualized baseline comparison, but extends the architecture beyond retrospective analysis into real-time continuous monitoring with adaptive intervention pathways. The critical distinction is timing. Zhang's framework proved that detection is possible. URSA makes detection actionable before the crisis window closes.

The sensor infrastructure required for this capability already exists in consumer devices that millions of veterans own. Sheikh, Qassem, and Kyriacou (2021) cataloged the full landscape of wearable, smartphone, and environmental sensors available for mental health monitoring. Their review confirmed that heart rate variability, electrodermal activity, skin temperature, accelerometer data, and sleep metrics all correlate with psychiatric symptom changes across depression, anxiety, bipolar disorder, and PTSD. The authors noted that combining multiple physiological parameters produces the highest classification accuracy for emotional states, a finding that directly supports URSA's convergent signal architecture. URSA does not require veterans to purchase specialized medical equipment. The watch on their wrist and the phone in their pocket already generate the raw data. URSA supplies the intelligence layer that transforms those signals into early warning.

Shen et al. (2025) reinforced these findings in a scoping review of 42 studies spanning a decade of passive sensing research. Their analysis identified heart rate, movement index, and step count as the three most frequently used digital biomarkers, appearing in 67%, 60%, and 40% of included studies respectively. Deep learning models such as CNN-LSTM architectures achieved anxiety detection accuracy of 92.16%. Yet the authors identified a persistent gap that defines URSA's opportunity. Only one of forty-two studies conducted external validation. Sample sizes remained small. Monitoring periods stayed short. The technology works in controlled research. Nobody has built the bridge to continuous real-world deployment for the populations who need it most.

Veterans deserve a system that watches their back the way they watched ours. URSA builds that bridge.