知Yizhi Daily
← Back to this edition

AI

Google Research Proposes Wearable-Based CKM Health-Risk and Personalized Coaching Framework

Edition 2026-11-12 · 08:00 Beijing time

Google Research presented an experimental machine-learning framework that converts passively collected smartwatch signals—resting heart rate, steps, sleep, and activity duration—into an actionable “coachable health age,” uses an action-conditioned world model to simulate 12 weeks of habit changes, and incorporates blood biomarkers to track cardiovascular-kidney-metabolic risk.

Google Research research scientists Zeinab Esmaeilpour and Ahmed A. Metwally presented a research framework on November 11. Metabolic diseases such as cardiovascular disease, kidney problems, and type 2 diabetes are interconnected and are known as cardiovascular-kidney-metabolic syndrome. Its progression often occurs silently for years before symptoms appear and is influenced by daily habits such as exercise and sleep.

The researchers noted that many existing biological-age models primarily estimate age based on blood or wearable data, often relying on characteristics that individuals cannot easily change, and making it difficult to explain how adjusting a particular habit would alter future health. Rather than training a model to guess age, the new framework draws on published epidemiological research to construct a “coachable health age” that can be influenced through everyday choices.

The model integrates four core signals passively measured by consumer wearables: resting heart rate, daily steps, sleep duration, and weekly Active Zone Minutes; when available, it also incorporates optional indicators such as VO2max, sleep regularity, and insulin resistance. The model mathematically adjusts for overlapping signals to distinguish the contributions of different habits to the health-age difference: a negative difference means physiological state is younger than chronological age, while a positive one means accelerated aging.

The framework also uses an action-conditioned world model to answer counterfactual questions such as “What would happen over the next 12 weeks if I changed a certain habit?” A companion study incorporates routine blood biomarkers into the CKM health-age assessment. The materials say the method has been validated in three independent cohorts, but the provided portion of the original paper does not list specific results.

Original sources

Google Research ↗
Back to this editionOpen interactive reader