CONTINUOUS MULTIMODAL PASSIVE MONITORING OF DEPRESSIVE SYMPTOMS VIA SMARTPHONE: CLINICAL VALIDATION OF A NOVEL DIGITAL HEALTH TECHNOLOGY

Background

Standard episodic assessments of depressive symptoms place a high burden on patients and fail to capture critical inter-visit symptom dynamics. Passive, smartphone-based digital health technologies (DHTs) offer a continuous, objective, and lowburden alternative for tracking disease trajectories. This presentation highlights the clinical innovation and validation of EMOCARE, a novel passive multimodal DHT designed to continuously and automatically monitor depressive symptom severity in real-world settings. Methodology: The EMOCARE system operates on a patient’s smartphone to acquire multimodal behavioral data during routine use without requiring active patient input. The technology passively extracts novel mood biomarkers from opportunistic facial snapshots and short audio snippets, while also capturing actigraphy/motion signals and digital behavior patterns, such as screen unlock and session metrics. These behavioral and physiological proxies are transformed into a quantitative depressive symptom severity score (0–100 scale), which is updated daily through the server-side aggregation of encrypted data. To evaluate the clinical validity and real-world feasibility of this multimodal approach, an interim pooled analysis was conducted across 3 prospective observational studies involving adults with Major Depressive Disorder or Bipolar Disorder. The continuously generated EMOCARE scores were compared against established clinician-rated anchors (e.g., MADRS, HAM-D17) and self-reported scales (e.g., PHQ-9, GAD-7).

Results

In the pooled dataset, continuous passive monitoring demonstrated strong feasibility and robust clinical alignment. The algorithmically derived scores exhibited moderate-tostrong concurrent associations with reference symptom scales (Spearman’s rho ranging from 0.613 to 0.833). Within-person longitudinal concordance was notably strong against the clinician-rated MADRS (repeated-measures correlation r = 0.895, p = .016). Furthermore, the continuous passive scores accurately tracked dynamic symptom trajectories over time, demonstrating strong sensitivity to change when compared to consecutive-visit changes on the PHQ-9 (rho = 0.834, p < .001). Data density criteria (requiring ≥7 valid days in a 14-day window) confirmed measurement integrity, though adherence requirements remain a consideration for broader real-world scaling.

Conclusion

These findings provide strong preliminary evidence that continuous, passive multimodal smartphone monitoring can generate interpretable severity estimates that align closely with gold-standard episodic scales. By seamlessly capturing objective behavioral, vocal, and facial markers of depression, this technology addresses a critical unmet need for longitudinal, low-burden patient tracking. This scalable approach offers significant clinical value for the early detection of relapse, objective assessment of treatment response, and the extension of high-fidelity patient monitoring.