What is passive depression monitoring?
Passive depression monitoring is a technology approach that continuously measures depression-related biomarkers from a patient’s smartphone, without requiring the patient to do anything after initial setup. Unlike PHQ-9 surveys or app-based journaling, passive monitoring runs silently in the background, capturing objective physiological and behavioral data across four signal streams.
The term “passive” is the operative word. Depression monitoring has historically required active patient participation: completing questionnaires, logging moods, or answering check-in prompts. Each of those requirements introduces the same failure point: a depressed patient who most needs monitoring is also the least likely to complete it. Passive monitoring removes that dependency entirely.
The clinical rationale is straightforward. 1 in 4 TMS patients never completes the acute course (Sackeim 2020), 37.5% of completers relapse inside 12 months (Dunner 2014), and 79.6% of Spravato patients discontinue within the same window (ESKALE). Most of that decline is invisible to the clinic until the patient stops showing up. A monitoring approach that depends on patient action will always fail at exactly the moment it matters most.
24%
never complete the TMS acute course (Sackeim 2020)
37.5%
of TMS responders relapse within 12 months (Dunner 2014)
79.6%
of Spravato patients discontinue within 12 months (ESKALE)
How it works: the four signal streams
Advanced passive monitoring platforms, like Emobot, process multiple signal streams simultaneously. No single signal is sufficient. Depression manifests differently across individuals and across time. Multimodal fusion (combining multiple data types) produces far more accurate estimates of depression severity than any single biomarker alone.
Four signals → one continuous score
Facial analysis is on-device; voice is processed in a short ephemeral window and discarded. Only the fused 0–100 index is retained: daily, continuous, objective.
Front camera · on-device
Facial microexpressions
Brief, involuntary muscle movements (microexpressions lasting less than 1/25 of a second) during normal phone use. Deep neural networks extract depression-relevant features: reduced smile intensity, slower affect recovery, flat emotional reactivity.
Microphone · on-device
Vocal biomarkers
Speech prosody, rate, pause intervals and formant energy. Reduced prosodic variation (monotone speech), slower speech rate, and longer pauses are measurable well before subjective mood changes appear.
Accelerometer · on-device
Actigraphy (movement)
Daily step count, activity intensity and circadian rest-activity patterns. Psychomotor retardation shows up in objective movement data days before a patient would report it on a PHQ-9. Sleep/wake patterns correlate strongly with severity.
App usage · on-device
Digital behavior
Screen-on time, app usage, and communication frequency. Social withdrawal, app-use drops and routine disruption are early behavioral signals, grounded in real-world action rather than self-reported perception.
All four streams are processed entirely on the patient’s device by an on-device neural network. Raw facial video, audio, and sensor data never leave the phone. What reaches the clinician’s dashboard is a single numerical index: a depression severity score between 0 and 100, updated every 24 hours. Think of it as a continuous, objective vital sign for psychiatric state.
Passive monitoring vs. PHQ-9 check-ins
The PHQ-9 is a validated, widely used screening tool. But using it as your primary monitoring mechanism between visits has well-documented limitations that passive monitoring directly addresses.
12-week monitoring window · same patient
Patient enters at PHQ-9 ~18 (severe). TMS works: by W6 the patient is in remission, the PHQ-9 visit confirms it. Then the relapse: at W8 Emobot detects the upward trend break, triggers an in-app rebooking nudge — and the patient is recaptured weeks before the next scheduled PHQ-9 at W12 would have surfaced the deterioration.
| Dimension | PHQ-9 by email | Passive monitoring |
|---|---|---|
| Completion rate | 30–40% | 100% (no action required) |
| Frequency | Every 6–8 weeks | Continuous (daily score) |
| Objectivity | Self-reported, subject to recall bias | Objective physiological signals |
| Early warning | Reactive; captures decline after the fact | 48h lead time on relapse detection |
| Patient burden | High: must remember, open email, complete 9 items | Zero: install once, then nothing |
| Data resolution | Single point in time | 84-day trend with daily granularity |
The most important difference is not accuracy; it’s coverage. A PHQ-9 that achieves 35% completion means 65% of your patients have no monitoring between visits. Of the 35% who do respond, many complete the survey during a relatively stable window that doesn’t reflect their worst moments. Passive monitoring captures every patient, every day, including the days when they’re too depressed to open an email.
The appropriate framing is not “passive monitoring vs. PHQ-9”; it’s “passive monitoring plus PHQ-9.” Passive monitoring fills the gap between your PHQ-9 assessments with continuous objective data. PHQ-9 remains a validated, structured clinical instrument for formal severity assessment at key timepoints.
Clinical evidence
Passive monitoring is not theoretical. The approach has been validated across 10+ clinical studies conducted in collaboration with academic partners including Yale, Harvard Medical School, Johns Hopkins, UCSD, McGill, Charité Berlin, and GHU Paris.
r=0.89
Correlation with MADRS
Clinical gold standard
r=0.83
Correlation with PHQ-9
Self-report standard
48h
Early warning lead time
Relapse detection
10+
Clinical studies
US, France, Canada, Germany
The r=0.89 MADRS correlation is particularly significant. MADRS (Montgomery-Åsberg Depression Rating Scale) is a clinician-administered scale; it requires a trained evaluator conducting a structured interview. Achieving r=0.89 concordance with a passive, fully automated system represents a meaningful advance in objective psychiatric measurement.
The 48-hour early warning capability for relapse detection has been the most clinically actionable finding. In a TRD population, 48 hours is enough time for a clinic to reach out, adjust a dosing protocol, or expedite a visit. In the absence of passive monitoring, the same relapse might not be detected for 4–6 weeks, at the next scheduled appointment, by which point significant deterioration has already occurred.
Regulatory posture: Emobot is today a patient-facing wellness app; no device-class gate applies. In parallel, the CE Mark pathway is in flight (EMC1 is 80% complete under Pr Schwan at CHU Nantes/Rennes; EMC2FR and EMC2-BD are 50% complete), alongside a dedicated US FDA clearance study (EMC1-MDD-US, N=92, Q2 2026 → Q1 2027) and a French reimbursement RCT (REMOOD, N=504).
Privacy and HIPAA considerations
Privacy is the first question every clinician asks about passive monitoring, particularly given that the technology involves a camera and microphone. It is the right question. Here is the architecture that makes passive monitoring privacy-safe by design.
Data architecture · minimum footprint by design
Patient's phone
Raw video never leaves the phone.
Emobot cloud
voice in 30-min ephemeral window
scored, then discarded
HIPAA · BAA available
Clinician dashboard
On-device processing only
The neural network that analyzes facial expressions and vocal biomarkers runs entirely on the patient's phone. The phone's CPU and neural processing unit (NPU) perform all computation locally. Raw video frames, audio samples, and sensor readings never leave the device.
Only numerical output transmitted
The only data transmitted to Emobot's servers is a single numerical index per 24-hour period: a number between 0 and 100. No audio. No video. No identifiable biometric data. The transmission is equivalent to sending a lab result value: clinically meaningful, but containing no raw biological data.
Zero raw biometric storage
Emobot does not retain any raw biometric data: not on-device, not on servers. There is nothing to breach, subpoena, or misuse. The data minimization is absolute.
HIPAA-ready architecture
Transmission uses TLS 1.3 encryption. Patient records are isolated from EHR systems by design, simplifying compliance. No BAA complexity beyond standard software-as-a-service agreements.
Patient transparency and consent
Patients install the app knowing it monitors their wellbeing. They can view their own depression trend in the EmoDTx app at any time. The monitoring is disclosed, consented, and actually valued by most patients; many report feeling 'held' by the system rather than surveilled.
The privacy principle in one sentence:
Emobot’s AI is like a blood test: you send the result (a number), not the blood sample (raw biometric data). The analysis happens on the patient’s device, and only the output reaches the clinician.
Implementation in your practice
Passive monitoring is designed for zero-friction deployment. There is no EHR integration required. There is no IT project. Implementation follows a consistent three-step pattern across all practice types.
Enroll patients during a scheduled visit
During your next visit with a TRD patient, show them EmoDTx on your own phone or the clinic tablet. Explain that it monitors their wellbeing passively between visits: no surveys, no check-ins. Walk them through the 3-minute setup. Proposing it and installing it with the patient in the visit drives an 80% activation rate, vs. 30–40% for PHQ-9 email.
Access your dashboard before each visit
Log in to portal.emobothealth.com from any browser. Review the 84-day trend for each patient before their appointment. Patients with downward trends get a different conversation than patients who have been stable. You arrive at every visit already informed.
Act on alerts between visits
When the AI detects early relapse indicators, the dashboard sends an alert 48 hours before the signal typically becomes clinically apparent. Your clinical coordinator calls the patient, adjusts the next appointment, or triggers a protocol review. This is the primary ROI mechanism: preventing the patient from disappearing.
Measured ROI in interventional psychiatry clinics
60%
Undetected relapses prevented
Up to $19k
Revenue recapture per patient
4×
Documented net ROI in IP clinics
The future of psychiatric monitoring
Passive monitoring is not a niche research curiosity; it is on track to become the standard of care for psychiatric monitoring between clinical visits. Several convergent trends make this trajectory clear.
First, regulatory posture is clarifying. CE Mark and US FDA clearance frameworks are increasingly accommodating digital mental health tools with demonstrated clinical validity. The combination of objective biomarker correlation (r=0.89), privacy-by-design architecture, and a patient-facing wellness app today positions passive monitoring as a pragmatic, regulatory-friendly approach.
Second, payer recognition is building. Value-based care contracts increasingly reward outcome measurement, and passive monitoring produces the objective, continuous outcome data that those contracts require. Clinics that can demonstrate objective patient monitoring have an advantage in payer negotiations.
Third, patient expectations are evolving. Patients who wear continuous glucose monitors, use sleep trackers, and check their HRV data daily expect the same continuity from psychiatric care. The question is no longer whether passive monitoring will become standard; it is which platforms will get there first with the clinical evidence to support adoption.
Ready to implement?
See passive monitoring live in your workflow
30-minute demo. We’ll walk through a real patient case and show you exactly how the dashboard works.