A call for open data to develop mental health digital biomarkers

Area of Science:

• Digital health
• Mental health technology
• Machine learning in healthcare

Background:

• Digital biomarkers, derived from everyday technology use (smartphones, wearables, social media), offer continuous, remote monitoring of behaviors linked to mental health symptoms.
• These digital biomarkers have the potential to transform mental health diagnosis and treatment through unobtrusive data collection.

Purpose of the Study:

• To caution clinicians about the current limitations of digital biomarkers in mental health.
• To emphasize the critical need for assessing the equitable performance of machine learning models across diverse populations and data types.
• To advocate for a centralized repository of open, de-identified data for robust digital biomarker development.

Main Methods:

• Review of current practices and challenges in digital biomarker development for mental health.
• Discussion of machine learning model applications in analyzing data traces from digital devices.
• Conceptual framework for assessing 'model equity' in digital biomarker predictions.

Main Results:

• Digital biomarkers can provide near-continuous, remote behavioral insights for mental health.
• Current machine learning models require rigorous assessment for equitable predictions across diverse demographics, behaviors, and data sources.
• Ensuring model equity is challenging for individual clinics or large studies.

Conclusions:

• Widespread clinical adoption of digital biomarkers for mental health should be deferred until model equity is established.
• A collaborative approach, including a public repository of de-identified data, is necessary for developing and validating equitable digital biomarkers.
• Future research must prioritize fairness and generalizability in AI-driven mental health tools.