Digital phenotyping and passive smartphone sensing

1. Introduction

Version followed: the Onnela Lab (Harvard T.H. Chan School of Public Health) definition and Beiwe research platform; the open-source mindLAMP platform, version 2, as described by Vaidyam, Halamka and Torous (2022); the digital-navigator model of Wisniewski et al. (2020); and the ethics frameworks of Martinez-Martin et al. (2018) and Resnick and Appelbaum (2019).

Digital phenotyping is not a psychotherapy. It is a measurement approach. The Onnela Lab defines it as the moment-by-moment quantification of the individual-level human phenotype in situ, using data from personal digital devices, especially smartphones [source: Onnela Lab Beiwe page]. Data are of two kinds. Passive data are collected in the background from phone sensors and logs, without the person doing anything. Examples are GPS-derived distance travelled, accelerometer and gyroscope movement, screen on/off, call and text logs (metadata), Bluetooth and Wi-Fi proximity, and, in some platforms, health-kit data such as activity or heart-rate variability. Active data are what the person enters: surveys, brief cognitive games, journals, diary cards [source: Vaidyam et al. 2022; Onnela Lab Beiwe page]. Insel (2017) also named keyboard interaction (typing and scrolling patterns) and voice and speech features (prosody, sentiment, coherence) as candidate signals [source: Insel 2017, JAMA].

The hoped-for route to change is indirect. Continuous, objective measurement might support measurement-based care, earlier detection of relapse, and better-timed help. Insel (2017) noted that no studies yet showed that better measurement improves outcomes, and called the step from data to care the "last mile" problem [source: Insel 2017]. For a psychotherapist, the practical value today is mainly collaborative: client and therapist look at the data together, as one more source alongside the client's own account, to spark reflection on routines, sleep, activity and social contact [source: Vaidyam et al. 2022; Nestor et al. 2024] [synthesis].

Research versus clinical status (as of October 2026, per the sources listed here):

Beiwe is presented by its developers as a research platform run by study investigators. Using it for interventions is possible but study-specific and outside the lab's main focus [source: Onnela Lab Beiwe page].
mindLAMP is an open-source platform used both in research and in a clinical "digital clinic" setting alongside ongoing care at Beth Israel Deaconess Medical Center. Its 2022 platform paper is a design description, not an outcomes trial [source: Vaidyam et al. 2022].
A 2023 systematic review described the field as feasible but still at an initial stage of validation. Most studies were small and short, and several reported modest predictive accuracy [source: Bufano et al. 2023].
Mindstrong Health, a commercial digital phenotyping company co-founded by Thomas Insel, scaled back to text-based therapy and announced closure in early 2023. A co-founder described pressure to commercialise too early [source: STAT, 28 February 2023].
As of October 2026, none of the sources cited here describes any smartphone passive-sensing tool for diagnosing mental disorders or predicting relapse as FDA-cleared, NICE-recommended or DiGA-listed [uncertain: not verified]. Treat any product's regulatory claim as something to check at the regulator's own database on the date you use it.

2. Framework

Because digital phenotyping is a measurement layer, not a treatment, it has no developer-defined phases or session count. The stages below are a practical sequence for a therapist who uses (or is asked to use) passive sensing within ongoing therapy. They draw on the mindLAMP clinical workflow and the ethics literature; the sequence itself is [synthesis].

Decide whether to use it at all. Goal: confirm a clear clinical question that data could help with (for example, "Does my sleep slip before low weeks?"), that the client wants it, and that the setting is not coercive. Move on only if the client freely agrees and there are no red flags (see 4.3) [source: Martinez-Martin et al. 2018; Resnick and Appelbaum 2019] [synthesis].
Evaluate the tool and agree consent. Goal: know what the platform collects, where data go, who sees them and how to stop. Use a structured app evaluation (privacy and security, evidence, ease of use) and a staged, plain-language consent [source: APA Ethics Committee 2021; Martinez-Martin et al. 2018]. Move on when the client can explain back what is collected and how to switch it off.
Set up with the minimum necessary data. Goal: switch on only the streams tied to the agreed question. mindLAMP lets sensors be turned on or off week by week according to clinical goals [source: Vaidyam et al. 2022]. A digital navigator, where available, does setup and troubleshooting [source: Wisniewski et al. 2020; CIBHS handout].
Baseline. Goal: a few weeks of data to learn what is typical for this person, plus active self-report alongside. Passive data are still being validated against self-report, so keep both [source: Bufano et al. 2023]. The length of baseline is not standardised [uncertain].
Collaborative review in sessions. Goal: look at summaries together, let the client add context, and link patterns to therapy goals. mindLAMP shows charts to patient and clinician; in one example, DBT diary cards were reviewed at every session [source: Vaidyam et al. 2022]. Navigators can prepare weekly summaries [source: CIBHS handout]. Typical format: 5 to 10 minutes at the start of a session [synthesis].
Periodic re-consent and adjustment. Goal: check that the client still wants monitoring, especially when their state changes. People who agreed while well may object when symptoms return [source: Martinez-Martin et al. 2018]. Suggested interval: every 4 to 6 weeks and at any change in symptoms or circumstances [synthesis].
Ending. Goal: stop collection, agree what happens to stored data, and fold the useful insights into the relapse-prevention plan [synthesis].

Criteria for stopping at any stage: the client withdraws consent (they keep their care), the data cause distress, paranoia or family conflict, the data are not being used, or the setting turns coercive [source: Martinez-Martin et al. 2018; Nestor et al. 2024] [synthesis].

3. Guiding principles

Measurement, not treatment. Practical meaning: the data are a conversation aid; the therapy is still the work you do together [source: Insel 2017] [synthesis].
No clinical decisions from unvalidated signals. Practical meaning: never change a diagnosis, risk level, treatment plan, or report to a third party on the basis of a passive signal alone; confirm with the client and with standard clinical assessment. Models give probabilities with false positives and negatives, yet may be used for binary decisions such as hospitalisation [source: Martinez-Martin et al. 2018; Bufano et al. 2023].
The client is the expert on context. Practical meaning: a drop in GPS movement may mean a holiday, a broken bus pass or a phone left at home. Ask before you interpret. Adolescents and parents feared misleading data from phones left in bags or apps running in the background [source: Nestor et al. 2024].
Minimum necessary data. Practical meaning: switch on only the streams that answer the agreed question; social-communication and keyboard data are the most sensitive [source: Nestor et al. 2024; Martinez-Martin et al. 2018].
Consent is staged, plain-language and revocable without penalty. Practical meaning: explain each stream, check understanding, and make clear the client can stop monitoring and keep their care [source: Martinez-Martin et al. 2018; Vaidyam et al. 2022].
Transparency about limits. Practical meaning: tell clients what the tool cannot do, including that it is not a crisis service and may not be watched in real time [source: Martinez-Martin et al. 2018] [synthesis].
Privacy is clinical, not just technical. Practical meaning: "content-free" metadata can still identify a person and reveal sensitive facts, and may sit outside HIPAA-style protections [source: Martinez-Martin et al. 2018].
No coercion. Practical meaning: do not make monitoring a condition of care, and refuse to use it as a tool of control by courts, employers, schools or families. Choices are never fully free when refusal carries a penalty such as incarceration [source: Resnick and Appelbaum 2019; Martinez-Martin et al. 2018].
Know who else can reach the data. Practical meaning: in forensic and legal settings, police, prosecutors and courts may seek the data, and privilege may not protect them [source: Resnick and Appelbaum 2019].
Clear duty-to-act rules before you start. Practical meaning: agree in writing who looks at data, how often, and what happens if a worrying signal appears, because these duties are unsettled for apps [source: Martinez-Martin et al. 2018] [synthesis].
Equity and bias awareness. Practical meaning: models trained on narrow samples (often students, US and Europe) may misread other groups [source: Bufano et al. 2023; Martinez-Martin et al. 2018].
Stay within your competence and route medical questions. Practical meaning: if data raise medication or medical questions, the client takes them to the prescriber or physician; the therapist does not advise changes [synthesis].

4. Intervention inventory

Part A describes how the platforms work, at a descriptive level. Part B gives the psychotherapist's own interventions. Example language is illustrative wording written for this guide.

4.1 How research and clinical platforms work (descriptive)

Purpose: so you can explain the technology accurately to clients and colleagues.
Steps (what happens, per developers):
- Beiwe (Onnela Lab, Harvard; NIH-funded): a native Android and iOS app, a cloud back end for study management and data collection, and an analysis pipeline that turns raw data into daily summaries per participant. Investigators configure surveys and passive streams for each study. Participants must consent and should understand what is collected and why [source: Onnela Lab Beiwe page].
- mindLAMP (Learn, Assess, Manage, Prevent; BIDMC Division of Digital Psychiatry): a patient- and clinician-facing app plus web dashboard, data store and the Cortex analysis toolkit. Active activities include surveys, cognitive games, journaling, breathing, meditation and tips. Passive streams include accelerometer, gyroscope, GPS, screen state, call and text logs, Bluetooth and Wi-Fi, and health-kit data. Clinicians or coordinators schedule activities and switch sensors on or off. Patient and clinician see charts in a "Prevent" view. Data are encrypted in transit and at rest; organisations can host their own back end; the code is open source [source: Vaidyam et al. 2022].
- Digital navigators: non-licensed team members trained to evaluate apps, help clients set them up, troubleshoot, and summarise data for the clinician (for example, a weekly report). A proposed 10-hour curriculum covers smartphone skills, troubleshooting, app evaluation, clinical terminology and data, and engagement [source: Wisniewski et al. 2020; CIBHS handout].
Example language (illustrative): "The app counts things like how far your phone moves each day and when the screen is on. It does not read your messages. You and I would look at simple charts of those counts together."
Watch for: platform-specific claims you have not checked (for example, whether message content or audio is captured). Check the platform's own data list and privacy policy before describing it [synthesis].

4.2 Responding when a client asks about digital phenotyping or tracking

Purpose: give balanced psychoeducation when a client mentions a mood-tracking app, a wearable, a research study invitation, or news about phones "predicting" mental illness.
Steps:
1. Ask what they have heard or are using and what they hope it will do.
2. Explain active versus passive data in plain terms [source: Onnela Lab; Vaidyam et al. 2022].
3. State status honestly: promising, still being validated, not a diagnostic test [source: Bufano et al. 2023; Insel 2017].
4. Name privacy and data-use questions to ask the provider (see 4.4) [source: APA Ethics Committee 2021].
5. Offer, if useful, to look at any data they choose to bring, on their terms (see 4.8).
Example language (illustrative): "These tools can spot patterns, like sleep shifting before a low week, but they are not yet good enough to tell you or me what is wrong. Your own account still counts most."
Watch for: anxiety or shame triggered by scores; magical thinking that the phone "knows best"; the reverse, distrust that spills into the therapy. Working if: the client can name one realistic use and one limit [synthesis].

4.3 Suitability screening before any monitoring

Purpose: decide whether passive sensing could help or harm this client.
Steps:
1. Name the clinical question the data would answer. No question, no monitoring [synthesis].
2. Check for coercion: is anyone (court, probation, employer, school, parent, partner, insurer) requiring or pressing for monitoring? Non-medical actors may want behavioural predictions [source: Martinez-Martin et al. 2018; Resnick and Appelbaum 2019].
3. Screen for features that make monitoring risky: persecutory beliefs about being watched, current or past intimate-partner control or stalking (shared phones, partner access to accounts), unstable housing with shared devices, or immigration or legal fears about data [synthesis].
4. Check capacity and voluntariness for this specific decision; comprehension and voluntariness are hard to guarantee in mental health [source: Martinez-Martin et al. 2018].
5. For minors, plan confidentiality with the young person and parents before anything is switched on (see 4.11).
Example language (illustrative): "Before we even look at an app, I want to check: is anyone else asking you to do this? And how would you feel knowing your phone is recording where you go?"
Watch for: hesitant yeses, glances at a parent or partner, statements like "I have to". Working if: the client's decision is clearly their own, and you have written down the question the data must answer.

4.4 Evaluating the platform or app with the client

Purpose: choose a tool you can stand behind, using a structured approach.
Steps:
1. Use the APA app evaluation approach: a hierarchical set of questions to discuss with the client, centred on privacy and security, evidence, and ease of use [source: APA Ethics Committee 2021]. The full model is at the APA's App Advisor [uncertain: not verified].
2. Ask: What data are collected? Where are they stored and for how long? Who can access them (company, clinic, researchers, third parties)? Are they encrypted? Can the client delete or export them? What evidence supports benefit? [source: APA Ethics Committee 2021; Vaidyam et al. 2022].
3. Apply the same review whether the client, you, an employer or an insurer proposed the app [source: APA Ethics Committee 2021].
4. Check regulatory claims (FDA, NICE, DiGA, CE/UKCA) at the regulator's own site on the date of use, and note the date in the record [synthesis].
5. Ask what happens to data if the company closes or is sold; Mindstrong announced closure in 2023 [source: STAT 2023]; reports that its technology was sold on are [uncertain: not verified].
Example language (illustrative): "Let's read the privacy page together. If we can't tell who sees your location, that's a reason not to use it."
Watch for: vague privacy policies, data sharing for marketing, no deletion route, health claims without evidence. Working if: you and the client can answer each question in writing.

4.5 Staged informed consent

Purpose: make consent real, specific and revocable.
Steps:
1. Go stream by stream (location, movement, screen, call/text logs, keyboard, microphone, surveys). For each, say what it records, what it cannot show, and why you suggest it [source: Martinez-Martin et al. 2018; Nestor et al. 2024].
2. Explain how often data are sent and who sees them; mindLAMP's developers state that clinicians should explain data types and upload frequency and obtain consent [source: Vaidyam et al. 2022].
3. Explain limits: probabilities, false alarms, missed signals; not monitored in real time unless agreed; not a crisis service [source: Martinez-Martin et al. 2018] [synthesis].
4. Explain legal exposure where relevant: records could be requested by courts or law enforcement [source: Resnick and Appelbaum 2019].
5. State clearly that the client can pause or stop at any time and keep their therapy [source: Martinez-Martin et al. 2018].
6. Ask for a teach-back, then record consent and the agreed streams in the notes.
Example language (illustrative): "You can turn this off any time, for a day or for good, and our sessions carry on exactly the same. Can you tell me in your own words what the app will record?"
Watch for: agreement without understanding, consent given in a crisis, or consent driven by wanting to please you. Working if: teach-back is accurate and the client names how to stop.

4.6 Choosing the minimum necessary data streams

Purpose: reduce privacy intrusion while keeping clinical value.
Steps:
1. Tie each stream to the agreed question (for example, screen-on times and movement for sleep and activity).
2. Leave off high-sensitivity streams (keyboard, microphone, social-communication logs) unless there is a strong reason and explicit agreement; these drew the most concern from adolescents and parents [source: Nestor et al. 2024].
3. Use week-by-week switching where the platform allows it [source: Vaidyam et al. 2022].
4. Keep active self-report alongside, since passive measures are still being validated against it [source: Bufano et al. 2023].
Example language (illustrative): "We only need to know roughly when your phone goes quiet at night. We don't need location or messages for that, so we'll leave them off."
Watch for: collecting "just in case". Working if: every stream switched on has a named use.

4.7 Onboarding and working with a digital navigator

Purpose: remove technical burden so sessions stay clinical.
Steps:
1. Where a navigator exists, refer for setup, troubleshooting and weekly summaries [source: Wisniewski et al. 2020; CIBHS handout].
2. Agree who checks data, how often, and what the navigator does with a worrying entry (route to you, never act alone) [synthesis].
3. Without a navigator, do a 10-minute setup in session and test that data appear on the dashboard [synthesis].
Example language (illustrative): "Sam, our navigator, will help you install it and check it's working. Sam won't make decisions about your care; anything important comes to me and we talk about it."
Watch for: battery drain, missing data from permissions or phone type, frustration that becomes a reason to drop out. Sensor availability depends on the device [source: Vaidyam et al. 2022]. Working if: data are arriving and the client feels the burden is small.

4.8 Collaborative data review in session

Purpose: turn data into shared reflection, with the client adding context. This is the main therapeutic use.
Steps:
1. Open the summary or chart together at the start of the session (mindLAMP's "Prevent" view shows charts to both; a navigator may prepare a weekly report) [source: Vaidyam et al. 2022; CIBHS handout].
2. Ask the client to read it first: "What do you notice?"
3. Ask for context for any change before offering any interpretation; clinicians reviewing data with patients gives them a chance to add context [source: Nestor et al. 2024].
4. Compare with active self-report and with what the client tells you; treat disagreement as information, not as the client being wrong [source: Bufano et al. 2023] [synthesis].
5. Link one pattern to a therapy goal (for example, behavioural activation, sleep routine, social contact) and agree one small experiment [synthesis].
6. Where a structured protocol uses diary cards (for example DBT), the app's diary card can replace the paper card and be reviewed in the usual way [source: Vaidyam et al. 2022].
7. Keep it short (about 5 to 10 minutes) so data do not crowd out the therapy [synthesis].
Example language (illustrative): "The chart shows your phone moved much less from Wednesday. Before I guess, what was going on for you then?" / "You rated your mood as fine, and the phone shows late nights. Which feels closer to the truth?"
Watch for: the client feeling judged or caught out; hiding distress to avoid "bad" data (a concern adolescents raised); the session turning into a data audit. Working if: the client starts to bring their own hypotheses about the charts, and the review leads to concrete plans [source: Nestor et al. 2024] [synthesis].

4.9 Handling alerts and worrying signals: no decisions from signals alone

Purpose: respond safely to a concerning pattern without over-reacting or under-reacting.
Steps:
1. Before starting, agree in writing whether anyone watches data between sessions, how often, and what happens if an alert fires. Duty-to-act rules for apps are unsettled, so make them explicit [source: Martinez-Martin et al. 2018] [synthesis].
2. Tell the client that data are not watched in real time unless that is agreed, and give the usual crisis pathway (crisis line, emergency services, out-of-hours team) in writing [synthesis].
3. When a signal looks worrying, contact the client and assess as you would for any concern: ask directly about mood, safety and suicidal thoughts; follow your service's standard risk-assessment and safety-planning procedure [synthesis].
4. Do not escalate (hospitalisation, report to a third party, change of risk level) on the signal alone. Fine-grained data can trigger over-reaction to minor fluctuations, and a gap in data may be technical rather than clinical [source: Resnick and Appelbaum 2019; Martinez-Martin et al. 2018].
5. Document what the signal was, how you checked it, what the client said, and your clinical reasoning [synthesis].
Example language (illustrative): "I saw your app hasn't sent anything for four days. That might just be the phone, but I wanted to check in. How have you been, honestly?"
Watch for: missing data read as relapse; alerts creating client anxiety; your own anxiety pushing you to check data too often. Working if: concerns are checked with the person quickly and proportionately.

4.10 Re-consent and review of the monitoring itself

Purpose: keep consent current as the client's state and circumstances change.
Steps:
1. Review every 4 to 6 weeks and at any change in symptoms, living situation, relationship or legal status [synthesis].
2. Ask whether the data are still useful, whether any stream feels intrusive, and whether they want to pause [synthesis].
3. Revisit consent when symptoms return, since people who agreed while well may object later [source: Martinez-Martin et al. 2018].
4. Plan in advance, while the client is well, what they want to happen with monitoring if they become unwell (an advance-statement-style agreement) [synthesis].
Example language (illustrative): "When we started, you said the sleep data helped. Is that still true, or has it started to feel like being watched?"
Watch for: drift into routine collection with no use; withdrawal of consent expressed indirectly (phone left at home, app deleted). Working if: the client actively chooses each renewal.

4.11 Young people: confidentiality and family agreements

Purpose: protect a young person's privacy and the therapeutic alliance when passive data are used with minors.
Steps:
1. Follow local law and service policy on who can consent for a minor; parental consent and teen assent both matter [source: Martinez-Martin et al. 2018]. Age thresholds vary by jurisdiction and platform [uncertain: not verified].
2. Agree with the young person, then with parents, what parents will and will not see. In one survey, adolescents were least willing to share social-activity data with parents and most willing to share activity level [source: Nestor et al. 2024].
3. Use the lowest-sensitivity streams; keyboard and microphone had the lowest endorsement from adolescents for clinical use [source: Nestor et al. 2024].
4. Explain the usual limits of confidentiality (risk of serious harm) in the same way you do for the therapy itself [synthesis].
5. Make clear that the phone data never replace talking: one adolescent suggested that just having a conversation would be better [source: Nestor et al. 2024].
6. Do not agree to a parent's or school's request to monitor a young person covertly or as a condition of support [source: Martinez-Martin et al. 2018] [synthesis].
Example language (illustrative): "Your parents will see whether you're doing your check-ins, not what you write. If that ever needs to change, we'll talk about it with you first."
Watch for: teens hiding distress, family conflict over monitoring, both raised by respondents as possible harms [source: Nestor et al. 2024]. Working if: the young person can say who sees what and still feels the space is theirs.

4.12 Forensic, mandated and other coercive settings: declining coercive use

Purpose: protect clients from monitoring used as control, and protect the therapy from becoming surveillance.
Steps:
1. Identify when monitoring is linked to a court order, probation or parole condition, mental health court, conditional release, employment or insurance [source: Resnick and Appelbaum 2019; Martinez-Martin et al. 2018].
2. Tell the client plainly that data in these settings may be sought by police, prosecutors or courts, and that therapist-patient privilege may not protect them; HIPAA has law-enforcement exceptions [source: Resnick and Appelbaum 2019].
3. Keep your therapeutic role separate from any monitoring role. Do not collect, review or report passive data for a court or supervising agency within therapy [synthesis].
4. If asked to use passive data as a compliance tool, decline and explain why: the benefit of passive monitoring has not been shown in forensic populations, refusal is rarely a free choice, and over-reaction to small changes can lead to rehospitalisation or reincarceration [source: Resnick and Appelbaum 2019].
5. Where monitoring is already imposed by others, help the client understand it, cope with it, and seek legal advice about its scope [synthesis]. Resnick and Appelbaum recommend calibrated decision rules, training for clinicians and courts, device functions limited to what is needed, and safeguards against misuse [source: Resnick and Appelbaum 2019].
6. Consult your professional body's ethics guidance and your service's legal team [source: Resnick and Appelbaum 2019; APA Ethics Committee 2021].
Example language (illustrative): "My job is to help you, not to monitor you for the court. I won't be collecting phone data in our work, and if anyone asks me to, I'll tell you first." / to a referrer: "I can't use passive phone data as a compliance measure. It hasn't been validated for that, and it would undermine the therapy."
Watch for: pressure from referrers, clients agreeing out of fear, guarded or performative sessions. Working if: the client trusts that sessions are not a monitoring channel.

4.13 Clients who self-track with consumer apps and wearables

Purpose: use data the client already collects, without endorsing unproven claims. [synthesis throughout, drawing on the principles in section 3]
Steps:
1. Ask what they track, why, and how it affects them.
2. Invite them to bring summaries if they wish; never ask for account access or raw data.
3. Review collaboratively as in 4.8, treating device "stress", "readiness" or "mood" scores as company estimates, not clinical measures.
4. Watch for obsessive checking, sleep or body-related preoccupation, or self-criticism driven by scores, and reduce tracking if so.
Example language (illustrative): "Your watch says your stress score was high. What does your body tell you about that day?"
Watch for: health anxiety, orthosomnia-like worry about sleep scores [uncertain: term not verified in sources here], eating-disorder or exercise preoccupation. Working if: the data serve the client's goals rather than driving their mood.

4.14 Escalating medical questions to the prescriber or physician

Purpose: keep medical and medication questions with the right professional.
Steps:
1. If data or discussion raise questions about medication effects, sleep disorders or physical health, encourage the client to take them to their prescriber or GP, with a short summary if they wish [synthesis].
2. With consent, share relevant observations with the prescribing team; never advise starting, stopping or changing medication [synthesis].
Example language (illustrative): "This sleep change might be worth raising with your psychiatrist. Would you like me to write a short note you can take?"
Watch for: clients using data to self-adjust medication. Working if: medical questions reach the prescriber.

4.15 Ending monitoring and handling the data

Purpose: close the monitoring cleanly and carry forward what was learned.
Steps:
1. Agree a stop date; switch off all streams and confirm on the dashboard that collection has stopped [synthesis].
2. Explain what happens to stored data (retention period, deletion or export options) according to the platform's policy and your records policy [source: Vaidyam et al. 2022; APA Ethics Committee 2021] [synthesis].
3. Summarise, in words, the patterns the client found useful (early warning signs, helpful routines) and add them to the relapse-prevention plan [synthesis].
4. Remind the client how to remove the app and revoke phone permissions [synthesis].
Example language (illustrative): "Let's write down the two things the data taught you, like late nights before low weeks, so you keep them without needing the app."
Watch for: anxiety about losing a "safety net"; data retained longer than needed. Working if: the client keeps the insight and the collection has fully stopped.

4.16 Documentation and supervision

Purpose: make data use transparent and accountable. [synthesis]
Steps:
1. Record the clinical question, platform, version, streams, consent date, review schedule and duty-to-act agreement.
2. Record the regulatory and evidence status you checked and the date ("as of <month year> per <source>").
3. Note in each session record whether data were reviewed and what the client added.
4. Bring data-related decisions to supervision, especially any escalation based partly on data.
Example language (illustrative, to a supervisor): "The phone data showed a drop, but the client explained it as a holiday. I didn't change the risk plan. Does that reasoning hold?"
Watch for: notes that quote data as fact without the client's context. Working if: anyone reading the record can see how data and clinical judgement were combined.

5. Handling problems

Most guidance below is [synthesis] unless a source is given; the developer sources do not offer a problem-handling protocol.

Client feels watched or becomes more suspicious. Pause collection at once, validate the reaction, and discuss whether to stop. For clients with persecutory beliefs, avoid passive sensing; active self-report is a less intrusive option [synthesis]. Illustrative: "Let's switch it off now. Your sense of safety matters more than the data."
Data and the client's account disagree. Treat the client's account as primary and explore the gap with curiosity; phones left in bags or apps running in the background can give misleading data [source: Nestor et al. 2024]. Do not confront the client with data as proof.
Non-adherence: missing data, app deleted, permissions off. Check technical causes first (device, battery, permissions; sensor availability depends on the device) [source: Vaidyam et al. 2022]. Then ask openly whether the client still wants to do it; disengagement may be a form of withdrawing consent [synthesis]. Never treat missing data as non-compliance with treatment [source: Resnick and Appelbaum 2019].
Overwhelm or obsessive checking. Reduce what the client sees (weekly summary instead of live charts), reduce streams, or stop [synthesis].
Rupture: client feels judged or "caught out" by data. Acknowledge it, own any part you played (for example leading with data rather than with the client), and return to the client-led review order: they read first, they give context first [synthesis] [source: Nestor et al. 2024].
Stuck: data reviewed but nothing changes. Check that the data answer the original question; if not, change streams or stop. Link one pattern to one behavioural experiment per review [synthesis].
Third-party pressure (parent, partner, court, employer, school, insurer). Do not share data without the client's informed consent, except where law or the usual limits of confidentiality require. Refuse covert or coercive monitoring and explain why [source: Martinez-Martin et al. 2018; Resnick and Appelbaum 2019].
Intimate-partner or family control. If a partner or family member has access to the client's phone or accounts, passive sensing could expose location or contacts. Do not start monitoring; consider safety planning and specialist domestic-abuse advice [synthesis].
Request for data by court, police or lawyer. Do not release without legal advice and the usual procedures; inform the client. Data may fall outside privilege [source: Resnick and Appelbaum 2019].
Worrying signal outside session. Follow the pre-agreed pathway in 4.9; assess directly; do not act on the signal alone [source: Martinez-Martin et al. 2018] [synthesis].
Platform closes, is sold or changes policy. Tell the client, stop collection, and check what happens to their stored data; commercial digital phenotyping services have closed (for example Mindstrong in 2023) [source: STAT 2023].
Bias or poor fit for the client's group. Be cautious when the client's age, culture, disability or device use differs from typical study samples, which are often small, short and from the US and Europe [source: Bufano et al. 2023; Martinez-Martin et al. 2018].
Therapist over-reliance on data. If you notice you check dashboards instead of asking, or feel reassured by data more than by the client's report, bring it to supervision [synthesis].

Sources

Web sources:

Further foundational paper (details [uncertain: not verified]): Onnela J-P, Rauch SL (2016). Harnessing smartphone-based digital phenotyping to enhance behavioral and mental health. Neuropsychopharmacology.

← Previous · IX-099Ecological momentary intervention Next · IX-101 →Online peer support and crisis text services