The Pulse

Google Research tests wearable models for CKM health coaching

Google Research has presented an investigational framework that combines wearable signals with simulated habit changes for health coaching. A companion model also uses blood biomarkers to examine cardiovascular, kidney and metabolic health.

AI.info Team · · fact-checked · against 4 sources ↓ · Published 12 Nov 2026

Google Research has presented, in an arXiv preprint and a blog post, an investigational framework that uses wearable data to estimate health age and simulate how changes in habits might affect it. The work combines a wearable-based metric with a model that generates 12-week coaching scenarios, alongside a companion model that adds blood biomarkers to assess cardiovascular, kidney and metabolic health.

The models are research prototypes, not clinical diagnostic tools. Google said its evaluations were conducted offline using retrospective research cohorts, and described prospective trials as a future step.

A health-age measure

The measure draws on four signals captured by consumer wearables: resting heart rate, daily steps, sleep duration and weekly Active Zone Minutes. Optional inputs include measures of aerobic fitness, sleep-schedule regularity and insulin resistance when available.

It compares an estimated health age with chronological age and adjusts for overlap among the signals. A negative difference indicates a physiological profile that appears younger than the person’s chronological age; a positive difference indicates accelerated aging, according to the researchers.

Across cohorts including 22,748 participants in the All of Us research program, the researchers reported that adding the measure to age, sex and body mass index improved prognostic discrimination for incident cardiovascular and metabolic disease, according to the preprint's abstract. They also examined wearable records over time and compared the measure with clinical scores and self-reported health.

Adding blood markers

In the companion study, the researchers evaluated a cardiovascular-kidney-metabolic model in 819 participants from the WEAR-ME study. It combined wearable signals with cholesterol-to-HDL ratio, a kidney-filtration measure called eGFR and HOMA-IR, a measure of insulin resistance.

The researchers reported an association between cardiovascular and metabolic disruptions, while early kidney impairment appeared independently. Using a normalized deviance score, they identified significant health impairments in 29.0% of the cohort.

The model measures how far individual markers depart from healthy clinical thresholds. The resulting health-age gap also aligned with clinical scores and participants’ reported health, the researchers said.

Testing coaching scenarios

A separate component, SensorWM, was pretrained on over 2.2 million person-weeks of longitudinal sensor data from approximately 100,000 participants. It is designed to forecast physiological responses under different activity and sleep routines, with a language-model agent turning selected scenarios into coaching plans.

Given two weeks of baseline wearable data, the system forecasts 20 physiological markers over the next 12 weeks; it compares candidate habit combinations with a projection in which habits remain unchanged. An optimizer weighs projected changes against a person’s historical flexibility and clinical bounds before selecting pathways for the coaching agent.

In blinded paired reviews, lifestyle coaches preferred the simulation-grounded plans over plans based on wearable history alone for overall quality in 71.4% of evaluations and for anticipated adherence in 71.0%. Board-certified clinicians judged 99.2% of the simulation-grounded plans to meet clinical safety criteria, compared with 97.6% of comparison plans.