digital-health/behavior · applied
Reference snapshot
This snapshot describes the state of the editorial record, not a diagnosis or a recommendation to intervene.
Claim ledger
These are declared editorial claim-to-source mappings. A mapping does not establish independent verification of the claim or its support. Citation controls open the corresponding source record.
Signal
Passive smartphone and wearable signals can support research into sleep, activity, mobility, and relapse-pattern prediction.
Boundary
Internal validation, missingness, demographic diversity, device differences, and outcome definitions can make predictive performance look better than deployment reality.
Not established
A passive model does not justify autonomous diagnosis, coercive monitoring, or high-stakes decisions without external validation and governance.
Mechanism map (working model)
A simplified causal map for orientation; it is not a diagnosis or a deterministic prediction.
Input or intervention
Consent-based passive phone or wearable signals with a defined prediction target and human response pathway
Biological pathway
Behavioral and physiological proxies are modeled into individualized change or risk estimates
Measurable signal
External validation, missingness, calibration, intervention response, privacy, and user-reported burden
Boundary
A model prediction is not a clinical fact and should not silently determine access, treatment, or identity.
Research appraisal checklist
Evidence to prioritize
Human comparative studies, validated measurement, realistic adherence, dose or context reporting, and evidence that the effect transfers beyond the training setting. Regulatory status, contraindications, specialist oversight, and a credible stop or adverse-event pathway are part of the evidence—not optional extras.
Outcome anchor
Anchor the claim to a meaningful functional or clinical outcome rather than a single dashboard metric or biomarker.
Main uncertainty
The main uncertainty is whether the observed signal is causal, durable, and relevant outside the exact study context.
What would change the conclusion
Upgrade confidence only when independent studies reproduce the result and make harms, adherence, and durability visible.
Minimum diligence questions
- Are the intervention, dose or exposure, comparator, population, and measurement conditions clearly defined?
- Do the outcomes matter outside the experiment, and are missing data, adverse events, and conflicts visible?
- Can the result be reproduced independently, and does it justify the level of risk being discussed?
Application and translation pathway
Current application boundary
Research, monitoring pilots, and selected digital-health workflows; autonomous diagnosis remains outside responsible deployment.
Research and applied uses
- Relapse and deterioration prediction
- Personalized behavioral support
- Passive sensing and closed-loop intervention studies
A staged map for separating an interesting mechanism from a reproducible, useful, and responsibly deployable application.
Consent
Define data collection, inference, pause, deletion, access, and escalation rights.
Model
Use individualized baselines, missingness analysis, external validation, and calibrated thresholds.
Response
Specify who reviews an alert and what supportive action follows.
Governance
Audit bias, burden, security, false alarms, and downstream consequences.
Research and clinical readout
Research status
Passive digital phenotyping can estimate changing behavioral and physiological patterns from phones and wearables, with promising prediction signals but frequent concerns about internal validation, missingness, and generalizability.
Signal confidence
Grade C (Moderate for research-level pattern detection; low for autonomous diagnosis, crisis prediction, or population-wide deployment without external validation.)
Key question
Can a passive signal improve a human decision or intervention while preserving consent, privacy, calibration, and the person's ability to contest the inference?
Clinical perspective
Digital signals may support monitoring or a conversation, but they do not replace clinical assessment and should never silently label a patient.
Clinical uncertainty
How much continuous data collection is acceptable when the potential benefit is early warning rather than direct treatment.
Community patterns
Users track sleep, mobility, communication patterns, and mood proxies to identify overload, relapse, or recovery trends.
Common community failure mode
Treating a model's prediction as a fact about motivation, honesty, diagnosis, or future behavior.
Where this is heading
Prospective external validation, personalized baselines, transparent decision rules, and closed-loop trials will matter more than increasingly complex classifiers.
What to track next
Digital phenotyping uses smartphone, wearable, location, sleep, activity, and communication signals to infer changing behavior or health states over time.
Protocol and safety scaffold
Risk gate: Passive sensing requires meaningful consent, data minimization, an accountable human response, and a way to contest or pause inference.
Baseline first
- Define the prediction target, data sources, missingness plan, and who will act on an alert.
- Record baseline behavior and consent boundaries before collecting sensitive signals.
Implementation pattern
- Use individualized baselines and external validation rather than opaque population scores.
- Make intervention thresholds, escalation, and user feedback explicit.
Monitor
- Calibration, false alarms, burden, privacy, dropout, representativeness, and real-world outcomes.
- Whether a prediction improves support instead of increasing surveillance or stigma.
Pause or stop rules
- Pause collection for distress, coercion, security breach, excessive burden, or unexplained alerts.
- Do not let a model silently determine clinical, employment, insurance, or relationship decisions.
Clinician gate
Clinical use requires validated pathways, qualified review, secure governance, and clear escalation for urgent risk.
Glossary connections
Controversies to monitor
Controversies
- Always-on monitoring and coercive wellness programs.
- Opaque mental-health inference from location, communication, or device-use data.
Core reference classes
Research classes
- 2026 systematic review of digital phenotyping for psychiatric relapse
- 2025 review of digital phenotyping for mental health
Evidence timeline
Linked records are ordered by publication year. A newer record is not automatically stronger; design, population, and endpoint still determine weight.
Digital N-of-1 Trials in Experimental Physiology
Experimental Physiology / PubMed - Methods review
A 2026 review describes digital N-of-1 designs, statistical inference for individual participants, and opportunities to aggregate valid single-person trials.
View source recordDigital Phenotyping for Predicting Relapse in Psychiatric Disorders
BMC Psychiatry / PubMed - Systematic review of passive-sensing prediction studies
A 2026 systematic review synthesized passive smartphone and wearable signals for psychiatric relapse prediction, finding promising internal model performance but substantial risk of bias, limited diversity, and a need for prospective external validation.
View source recordMachine Learning for Predicting Stress Episodes Based on Wearable Sensor Data
Computers in Biology and Medicine / PubMed - Systematic review
A systematic review finds wearable sensor and machine-learning approaches can predict stress episodes in constrained studies, with generalization, labeling, privacy, and clinical utility still unresolved.
View source recordTopic assessment
These questions test evidence interpretation and safety boundaries, not personal medical decisions.
1. What is the most defensible first question when evaluating Digital Phenotyping and Passive Sensing?
Show answer and explanation
What outcome, evidence, and risk boundary are defined?
A clear outcome, evidence design, and risk boundary keep an encyclopedia claim from becoming an unsafe recommendation.
2. What would make a claim about Digital Phenotyping and Passive Sensing more trustworthy?
Show answer and explanation
Replication with appropriate controls, context, and meaningful outcomes
Reproducibility and patient-important outcomes matter more than novelty, complexity, or an isolated signal.
Editorial update history
This log records what changed in the encyclopedia record and why the update matters for interpretation.
A-Z topic scaffolding, safety framing, learning paths, and linked evidence architecture established for the living encyclopedia.
Added distinct dossiers for continuous glucose monitoring, GLP-1 and incretin therapeutics, metformin geroscience, rapamycin and mTOR, transcranial focused ultrasound, polygenic risk scores, digital phenotyping, and psychedelic microdosing, with evidence lanes and explicit medical and governance boundaries.
Related record update: Digital Phenotyping for Predicting Relapse in Psychiatric Disorders
Added indication-specific testosterone and male-fertility coverage alongside N-of-1 trial design and placebo, expectancy, and nocebo literacy, with explicit clinical gates and causal-inference boundaries.
Related record update: Digital N-of-1 Trials in Experimental Physiology
Added evidence-aware dossiers for dopamine and reward learning, neuroplasticity and cognition, allostatic load and chronic stress, and slow breathing with HRV biofeedback, separating mechanisms, subjective effects, functional outcomes, and safety boundaries.
Related record update: Machine Learning for Predicting Stress Episodes Based on Wearable Sensor Data
Linked evidence records
Digital N-of-1 Trials in Experimental Physiology
Experimental Physiology / PubMed • Methods review • Published: 2026 • Checked: Aug 17, 2026
A 2026 review describes digital N-of-1 designs, statistical inference for individual participants, and opportunities to aggregate valid single-person trials.
How to interpret this record
What it supports
Digital tools can support repeated crossover designs, individual inference, and aggregation of valid single-person physiology trials.
Major limitations
Technology does not remove bias, unsafe exposure, poor outcomes, missing data, or the need for transparent design and interpretation.
What it does not establish
More sensors or a digital platform do not make an unsafe or confounded personal experiment scientifically valid.
Machine Learning for Predicting Stress Episodes Based on Wearable Sensor Data
Computers in Biology and Medicine / PubMed • Systematic review • Published: 2025 • Checked: Aug 17, 2026
A systematic review finds wearable sensor and machine-learning approaches can predict stress episodes in constrained studies, with generalization, labeling, privacy, and clinical utility still unresolved.
How to interpret this record
What it supports
Wearable signals and machine-learning models can predict labeled stress episodes in some research settings.
Major limitations
Labels, sensor quality, model drift, behavior changes, demographic generalization, and privacy complicate real-world interpretation.
What it does not establish
A model alert is not a diagnosis, proof of chronic stress, or evidence that a specific intervention will improve health.
Digital Phenotyping for Predicting Relapse in Psychiatric Disorders
BMC Psychiatry / PubMed • Systematic review of passive-sensing prediction studies • Published: 2026 • Checked: Aug 17, 2026
A 2026 systematic review synthesized passive smartphone and wearable signals for psychiatric relapse prediction, finding promising internal model performance but substantial risk of bias, limited diversity, and a need for prospective external validation.
How to interpret this record
What it supports
The review maps passive smartphone and wearable signals used to predict psychiatric relapse and highlights recurring features such as sleep, activity, mobility, and communication patterns.
Major limitations
Most performance estimates rely on internal validation, definitions and devices vary, demographic diversity is limited, and high risk of bias constrains deployment.
What it does not establish
The review does not establish autonomous diagnosis, guaranteed early warning, or permission for coercive always-on monitoring.