Wearable and Digital Health Data in Real-World Research
Digital health real-world data is information about health and behavior collected outside the clinic by consumer and medical devices, such as wearables, smartphone apps, home monitors and connected sensors. Linked to clinical records, it fills the long gaps between appointments with continuous measures of activity, sleep, heart rate and symptoms, which makes it a growing source of evidence for clinical research.
This guide explains the main types of digital health data, what makes it hard to use for research, how to check its quality, and how to analyze it alongside clinical data without creating new privacy risks.
Why Digital Health Data Matters Now
Traditional real-world data, such as health records and claims, captures what happens when a patient meets the health system. It says little about the weeks in between. Digital health data does the opposite: it records daily life continuously but often lacks clinical context. Together, they give a fuller picture than either can alone.
Several trends make this practical. Wearable devices are common, smartphones carry sensors capable of measuring movement and speech, and remote monitoring has become a standard part of care for some chronic conditions. Regulators have also engaged. FDA guidance on digital health technologies for remote data acquisition in clinical investigations sets out how such tools can be used in trials, and European regulators have qualified some digital endpoints for specific uses.
For research teams, the result is a new kind of evidence: objective, frequent measures of how patients actually function, collected with less burden than clinic visits.
Types of Digital Health Real-World Data
Wearables
Wrist-worn and patch devices record step counts, activity intensity, heart rate, heart rhythm, sleep stages and, on some devices, blood oxygen. Consumer devices dominate by volume; medical-grade devices offer validated measurements for specific uses.
Smartphone Sensors and Apps
Phones collect movement, location patterns, typing and voice features, and patient-reported outcomes through app questionnaires. Symptom diaries in apps replace paper logs and capture data closer to when symptoms happen.
Home Monitoring Devices
Connected blood pressure cuffs, glucose monitors, scales, spirometers and pulse oximeters send readings to care teams. Continuous glucose monitors in particular produce dense, clinically meaningful data streams.
Remote Monitoring Programs
Health systems running remote monitoring for heart failure, diabetes or post-operative care generate digital data already linked to clinical records, which makes them especially valuable for research.
What Makes It Hard to Use
Volume and Granularity
A single wearable can produce readings every second. Raw sensor data for a cohort quickly reaches terabytes, while most research questions need summary measures, such as daily step count or nightly sleep duration. Deciding how to aggregate, and documenting it, is a core design choice.
Device and Algorithm Variation
Different devices, and different firmware versions of the same device, measure and calculate things differently. A step on one device is not quite a step on another. Without device and version metadata, comparisons across participants can mislead.
Missing Data and Wear Time
Participants take devices off, forget to charge them or stop using apps. A day with few steps may mean low activity or simply that the device was on the nightstand. Wear-time detection is essential before any analysis.
Representativeness
People who own and consistently use wearables are not a random sample of patients. Studies need to account for who is and is not represented.
Linkage and Standards
Digital data usually sits in vendor clouds, separate from clinical records. Linking it to the right patient, and mapping it to standards, takes deliberate work. FHIR offers resources for device observations, the IEEE 1752 family of standards covers some mobile health measures, and the OMOP CDM can hold derived measurements alongside clinical events. Consistent data harmonization is what makes cross-cohort analysis possible.
Quality Checks Before Analysis
- Device metadata: record make, model and firmware version for every data stream.
- Wear-time rules: define a valid day (for example a minimum number of hours worn) and apply it consistently.
- Plausibility checks: flag physiologically impossible values and sudden jumps caused by device errors.
- Time zones and clocks: normalize timestamps so events line up with clinical records.
- Derivation documentation: record how raw signals became summary measures, including algorithm versions.
- Validation evidence: for any measure used as an endpoint, keep evidence that it reflects what it claims to measure in the study population.
Building the Data Pipeline
A workable pipeline has four stages. Ingest raw or summarized data from vendor platforms or devices, with device metadata attached. Clean it by applying wear-time and plausibility rules. Derive the summary measures the study needs, with versioned algorithms. Then load the derived measures into the same analysis model as the clinical data, so a single query can draw on both. Keeping raw data and derived measures separate, and versioning every step, means results can be reproduced when an algorithm or device changes.
Comparing Digital Health Data With Traditional Real-World Data
| Dimension | Health records and claims | Digital health data |
|---|---|---|
| When data is collected | At care encounters | Continuously in daily life |
| What it captures best | Diagnoses, treatments, procedures, costs | Activity, sleep, vital signs, symptoms |
| Clinical context | Rich | Limited unless linked |
| Main quality issue | Coding variation and missing visits | Wear time and device variation |
| Typical volume | Moderate | Very high at raw level |
| Who holds it | Health systems, payers | Device vendors, app providers, remote monitoring programs |
The two are complements. The most useful studies link them, so a change in daily activity can be read against a diagnosis, a new prescription or a hospital admission. For the wider range of sources, see these real-world data examples.
An Illustrative Study Design
Consider a hypothetical study asking whether patients regain daily activity faster after one surgical approach than another. Health records identify the procedure, the date, complications and readmissions. Wearable data from consenting patients provides daily step counts before and after surgery. The analysis plan would define a valid wear day, set a pre-surgery baseline from the weeks before the operation, and compare the time each group takes to return to that baseline. Device model and firmware would be included as covariates, and patients with too few valid days would be handled with a pre-specified rule rather than dropped silently. Neither data source could answer the question alone: records lack daily function, and wearables lack the clinical events that give the activity curve its meaning.
Privacy and Governance
Digital health data can be surprisingly identifying. Location traces, daily routines and heart rhythm patterns can single out individuals even without names. When linked to clinical records, the combined dataset is highly sensitive and subject to health data protections such as GDPR special category rules.
Consent also needs care. Participants may have agreed to share data with a device vendor for one purpose, not for research. Research use should rest on clear, specific consent or another lawful basis, recorded in a form that software can enforce.
The safest model keeps linked data in a Trusted Research Environment (TRE). Researchers analyze the combined data inside a controlled workspace, data never leaves the source, and only aggregate results that pass output checks are released. When digital and clinical data sit with different custodians, a federated approach lets each one keep its data while a single approved analysis runs across both.
Research Uses
- Digital endpoints in trials: continuous measures such as walking activity used alongside or instead of clinic-based tests.
- Post-marketing studies: tracking how patients function after starting a treatment in routine care.
- Disease progression: spotting early changes in activity, sleep or speech that precede clinical events.
- Remote monitoring evaluation: measuring whether monitoring programs change outcomes.
- Recruitment and feasibility: finding participants whose daily data matches study criteria.
Practical Next Steps
- Start from the question. Decide which measure matters, such as daily steps or nightly sleep, before collecting raw streams.
- Fix the device policy. Limit studies to known devices, record versions and plan for updates.
- Define quality rules up front. Agree wear-time and plausibility rules in the protocol.
- Plan linkage and standards. Decide how digital data will be linked to clinical records and which models will hold it.
- Govern access in place. Keep linked data in a controlled environment and release only checked aggregate results.
Frequently asked questions
What is digital health real-world data?
It is health and behavior data collected outside the clinic by wearables, smartphone apps, home monitors and connected sensors, often linked to clinical records for research.
Can wearable data be used in clinical trials?
Yes, when the measure is fit for purpose. FDA guidance covers digital health technologies for remote data acquisition in clinical investigations, and some digital endpoints have been accepted for specific uses.
What are the main quality problems with wearable data?
Missing wear time, variation between devices and firmware versions, implausible values and inconsistent timestamps are the most common.
How is digital health data linked to health records?
Through a pseudonymized linkage process that matches device data to the right patient, then maps both to shared standards such as FHIR for exchange and OMOP for analysis.
Is wearable data sensitive?
Yes. Location, routines and physiological patterns can identify people, and linked to clinical records the data is highly sensitive health data.
How can researchers analyze it safely?
By working inside a Trusted Research Environment, where linked data stays at the source and only checked aggregate results are released.