Protecting Participant Privacy in the Age of Wearable Digital Health Research
In the rapidly evolving world of digital health, wearable devices like smartwatches and activity trackers provide researchers with a continuous stream of valuable information, such as accelerometer data, to study sleep, stress, and physical activity. However, a new self-paced tutorial supported by the mDOT Center warns that with this power comes a significant responsibility: navigating the inherent privacy risks of sharing individual-level high-frequency sensor data.
The “WristPrint”: Why Motion Data is Unique
A primary concern highlighted in the sources is re-identification, the process by which anonymized data is linked back to a specific individual. This re-identification potential can risk serious harms, including discrimination, stigma, or financial, legal, and reputational damage.
Research from the “WristPrint” study has revealed that human motion patterns captured by wrist-worn sensors are highly distinctive—almost like a fingerprint. This unique “signature” is influenced by several key factors that researchers must consider:
- Activity Type: Dynamic movements like running are more individually distinctive than stationary activities like sitting.
- Intensity: Higher intensity activities generate more unique biomechanical signatures.
- Duration: Sharing data over longer periods (e.g., days vs. hours vs. minutes) significantly increases the chance that unique, identifiable patterns will emerge.
Strategies for Ethical Research
The tutorial emphasizes that the goal is to strike a thoughtful balance between research utility and participant privacy. To achieve this, researchers are encouraged to consider mitigation strategies to reduce re-identification risks, within the context of the data they need to collect or share:
- Data Aggregation: Summarizing raw readings into metrics like “steps per minute” to mask unique patterns.
- Differential Privacy: Adding mathematically calibrated statistical noise to the data set to protect individual identities while preserving overall patterns.
- Data Reduction: Following the principle of data minimization by sharing only the specific segments or duration necessary for the research question.
- Access Control: Limiting data release to trusted collaborators through secure repositories and formal data use agreements.
- Synthetic Data: Using machine learning to create entirely artificial data sets that mimic real statistical patterns without using actual individual records.
Take Action
The mHealthHub now hosts the “WristPrint: Guided Tutorial“ to equip clinicians, behavioral scientists, and computer scientists with the tools needed to conduct ethically sound research.
Researchers are encouraged to visit the mHealthHub website to complete the tutorial and download a Mitigation Checklist to use in the design phase of future projects. By taking a proactive approach to privacy, the research community can continue to unlock the potential of wearable data while rigorously protecting the individuals who make that science possible.
This tutorial and research project were supported by the National Institutes of Health (NIH) under administrative supplement 3P41EB028242-05: Administrative Supplement to P41EB028242 award (The mDOT Center): WristPrint: Bioethical Policy Implications of the Emerging Re-identifiability Risks from Wrist-worn Activity Data.