Google-MIT AI Tool Identifies 66 Wearable Biomarkers Across 9,279 Observations

3 articles · Updated · Google Research · Aug 21

Across three cohorts totaling 9,279 participant-observations, the Biomarker Discovery Framework surfaced 66 candidate digital biomarkers from wearable and clinical data—41 for mental health and 25 for metabolic outcomes.
The Google-MIT system is designed to avoid spurious AI findings by separating feature construction from target signals and forcing candidates through an 11-test adversarial filtering stage under human supervision.
Among the signals it highlighted, sleep-duration variability tracked depression severity in one cohort with Spearman’s ρ of 0.252, while a steps-to-resting-heart-rate fitness index emerged as a non-invasive correlate of insulin resistance.
Adding the framework’s features to demographic data improved prediction modestly, with ΔR² of 0.040 for depression and 0.021 for insulin resistance, while the authors stressed the results are hypothesis-generating rather than clinically validated.
In a blinded review by 15 experts, the framework earned the highest mean scores across seven quality measures and was the only AI system to receive any Accept or Minor Revision recommendations.