AI News

Trending story

Google Research proposes a coachable health age and 12-week habit simulation framework

1 reports1 story updatesUpdated Nov 12 04:46

Understand this story

AI overview

Google Research has proposed a research machine learning framework that converts passive signals from wearable devices into an actionable "Coachable Health Age" (CHA), and uses an action-conditioned world model, SensorWM, to simulate health trajectories after personalized 12-week habit changes. The framework's SensorWM is pretrained on de-identified data from approximately 100,000 Fitbit users covering 2.28 million person-weeks, and can predict changes in 20 physiological indicators over 12 weeks. In a blinded evaluation by 5 lifestyle coaches and 5 physicians, coaching plans based on counterfactual simulation were rated as the best plan in 71.4% of assessments, and physicians judged 99.2% of the plans to be within a safe exercise intensity. The framework can also be extended to cardio-kidney-metabolic (CKM) multi-organ health assessment by incorporating three routine blood indicators: the cholesterol-to-HDL ratio, eGFR, and HOMA-IR. There are currently no further developments.

AI-generated from public reports

Latest development · Report

Google Research uses wearable data to build an actionable health age and a 12-week habit simulation framework

Developments

1 confirmed developments

  1. Google Research proposes a coachable health age and counterfactual wearable trajectory framework

    Source:Google Research

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 12

  1. Google ResearchFirst-party

    Google Research uses wearable data to build an actionable health age and a 12-week habit simulation framework

    Google Research proposes a research-oriented machine learning framework that converts passive signals from wearable devices into an actionable "Coachable Health Age" (CHA), and uses an action-conditioned world model, SensorWM, to simulate health trajectories after personalized 12-week habit changes. SensorWM is pretrained on de-identified data from approximately 100,000 Fitbit users and 2.28 million person-weeks, and can predict changes in 20 physiological metrics over 12 weeks; in a blinded evaluation by 5 lifestyle coaches and 5 physicians, coaching plans based on counterfactual simulation were rated as the best plan in 71.4% of assessments, and physicians judged 99.2% of plans to be within safe exercise intensity. The framework can also incorporate three routine blood metrics—cholesterol-to-HDL ratio, eGFR, and HOMA-IR—to extend to cardio-kidney-metabolic (CKM) multi-organ health assessment.

Related stories

  1. Related storiesCSIRO releases two high-quality reference genomes of Australian giant kelp
  2. Related storiesGoogle Research proposes a cuffless, calibration-free AI system for smartwatch hypertension detection
  3. Related storiesGoogle Research proposes a sleep framework to help LLMs retain knowledge in continual learning
  4. Related storiesGoogle DeepMind and others release the global forest type map ForTy 2020