mindLAMP: Phone and Wearable Data in Psychiatric Care

LAMP, short for Learn, Assess, Manage, Prevent, is an open source smartphone platform from the Division of Digital Psychiatry at Beth Israel Deaconess Medical Center. Its app, mindLAMP, combines passive phone and wearable data with surveys and brief cognitive tests, so researchers and clinicians can see how a person is doing between appointments, not only during them.

This guide is for health IT leaders, researchers and clinical teams weighing digital phenotyping tools. It explains what the platform collects, where machine learning fits, and the privacy and equity questions you need to settle before a pilot.

Key takeaways

  • mindLAMP is organized around four tabs: Learn (tips and psychoeducation), Assess (surveys and cognitive tests), Manage (activities such as mindfulness and journaling) and Prevent (personal data trends).
  • Passive data such as step count, sleep, screen use and GPS-derived mobility summaries can be collected with no extra effort from the patient.
  • The code is open source. Sites host their own data, and the LAMP Consortium has supported dozens of research and clinical sites worldwide.
  • Prediction models built on this kind of data are still research tools, not validated diagnostic tests.

What is the LAMP platform and who built it?

LAMP was designed by the Division of Digital Psychiatry at BIDMC, a Harvard Medical School teaching hospital, in partnership with a software development firm. The problem it targets is simple to state. A psychiatrist might see a patient for an hour a month. Everything that happens in the other weeks, sleep changes, withdrawal from friends, a drop in activity, reaches the clinician only through the patient's memory.

mindLAMP fills that gap in two ways. Active data comes from short surveys about mood, anxiety, sleep, medication and psychosis symptoms, plus tap-based cognitive games that measure attention, memory and processing speed. Passive data comes from phone sensors and connected health platforms such as Apple HealthKit and Google Fit. A clinic or study can switch individual features on or off, so a depression study and a schizophrenia relapse study can run very different versions of the same app.

Where does machine learning fit?

The data collection layer comes first. Machine learning sits on top and looks for patterns across many weeks of signals. Researchers in digital psychiatry use these "digital phenotyping" methods to ask questions like: does a drop in mobility and a shift in sleep timing come before a relapse? Do changes in phone use track with negative symptoms in schizophrenia?

The LAMP team has described an analysis pipeline called Cortex, intended to help sites turn raw sensor streams into usable features without needing an in-house data scientist. That matters because raw GPS points and accelerometer readings are noisy and full of gaps. A lot of the real work is cleaning data, handling missing days and building features such as time spent at home or number of places visited.

Be careful with claims that go beyond the evidence. Early studies are promising, but many are small, and models trained in one population often perform worse in another. A model flag should start a conversation with the patient, not trigger an automatic clinical decision. Our piece on the ethics of AI in medical diagnosis covers why human review matters here.

What data does mindLAMP collect?

Data typeExamplesWhy researchers use it
SurveysMood, anxiety, sleep, social behavior, psychosis symptoms, medication useCaptures how the person says they feel, close to real time
Cognitive testsTap-based games such as a Trail Making Test variantTracks attention, memory and executive function over time
MobilityGPS-derived summaries such as time at home (actual locations need not be stored)Withdrawal and reduced movement can signal worsening mood
Activity and sleepStep count, sleep duration, heart rate variability from phone or watchSleep disruption is closely watched in mood disorders
Phone use metadataScreen on and off eventsChanges in routine can show up in how often the phone is used

How wearables add to the picture

Phones capture behavior well, but they spend time on desks and in bags. A smartwatch or fitness tracker fills in sleep and heart rate data more consistently. Because LAMP pulls from common health platforms rather than one brand, participants can often use a device they already own. If you are comparing device options, our overview of wearable technology in healthcare covers accuracy and adoption issues.

Two limits are worth knowing. Consumer devices estimate sleep and heart rate with different algorithms, so values from two brands are not directly comparable. And a person who stops wearing the device creates gaps that can look like a behavior change. Good studies record wear time and treat missing data as missing, not as zero.

Privacy, ethics and equity: what to settle first

Psychiatric data is among the most sensitive information a health system holds. Location history and phone use patterns add to that sensitivity. Published LAMP study protocols describe randomly generated user IDs, encrypted transmission, password-protected accounts and storage on a secure server controlled by the site. Because the platform is self-hosted, your organization is responsible for the server, its access controls and your HIPAA compliance.

Consent has to be specific. Patients should know which sensors are on, who sees the data, how often it is reviewed and what happens if a pattern looks worrying. Some people find monitoring reassuring. Others find it intrusive, and they should be able to opt out of passive sensing while still using surveys or activities.

Equity is the third issue. People with serious mental illness are less likely to have a recent smartphone or a wearable, and data plans can run out. Research sites often provide devices. A clinical rollout needs a plan for that too.

Pilot readiness checklist

  • A named clinical question the data will help answer
  • IRB or clinical governance approval, with a clear consent script
  • Hosting, security review and data retention rules agreed with IT
  • A person responsible for reviewing data and following up
  • A device and data plan for participants who need one
  • A plan for what happens if a survey shows risk of harm

Why open source matters here

Commercial digital health tools can change features, pricing or data terms with little notice. Because mindLAMP's code is public, a site can inspect it, adapt it and keep its own data. The LAMP Consortium, described in a 2022 paper, grew to more than 50 sites, most of them research sites, sharing setup support, webinars and a user forum. Surveys, data viewing and GPS-based passive collection were the most used features. The trade-off is that open source shifts responsibility for hosting, updates and support onto your own team. Organizations that already run AI and ML programs in healthcare will recognize the governance questions.

For a look at how the same platform is used to deliver care rather than research, see LAMP and the digital clinic model.

FAQ

What does LAMP stand for?

LAMP stands for Learn, Assess, Manage, Prevent. Those are the four sections of the mindLAMP app: Learn offers tips and psychoeducation, Assess holds surveys and cognitive tests, Manage includes activities such as mindfulness and journaling, and Prevent shows the person's own data trends over time.

Is mindLAMP free to use?

The software is open source and shared publicly, so there is no license fee. Sites still pay for hosting their own data, security, staff time and any devices they provide. Joining the LAMP Consortium has been free but requires ethics approval and agreement to data-sharing and compliance terms.

Can mindLAMP predict a relapse?

Not reliably on its own. Research using mindLAMP and similar tools is testing whether changes in sleep, mobility and survey answers come before relapse in conditions such as schizophrenia and bipolar disorder. Results are promising but early. Any alert should prompt a clinician to check in with the patient, not replace clinical judgment.

What to do next

  • Write down one clinical or research question you want passive data to answer.
  • Review the published LAMP papers and the consortium's documentation with your IT security team.
  • Run a small pilot with clear consent and a named person who reviews the data weekly.
  • Decide in advance how you will measure whether the pilot helped patients.

Sources

Last reviewed: October 2026.

Comments

Popular Posts

Why Workday Onboarding Breaks Down for Frontline Employees

How to Improve the Customer Experience (CX)

Interview Tips for Beginners: A Step-by-Step Prep Plan

How Do I Log In and Sign In to Workday HCM

Dollar General Store Walkout: Lessons for Retail HR

Apple Watch Health Features: Blood Pressure and Glucose

UKG Union Contract and Overtime Rules for Multi-Site Food Plants

The End of Remote Work: Why Companies Want You Back

ERP Solution Guide: How to Choose the Best ERP for You

How to Search in Workday: Find Reqs, Reports, People and Tasks