Apple Watch Aging Clock Predicts Heart Disease Risk from Your Wrist
Researchers built an aging clock from wrist PPG wearable data of 213,000+ people, linking biological age gap to heart disease, diabetes, and behavior.
Summary
Scientists at Apple developed PpgAge, an aging clock built from photoplethysmography (PPG) waveforms captured passively by Apple Watch. Using data from over 213,000 participants in the Apple Heart & Movement Study across 149 million participant-days, the model predicts chronological age with a mean absolute error of just 2.43 years in healthy individuals. Critically, the 'PpgAge gap'—the difference between predicted and actual age—strongly associates with heart disease, heart failure, and diabetes diagnoses, and predicts new cardiac events even after controlling for standard risk factors. The clock also reflects behavioral factors like sleep, exercise, and smoking, and captures physiological changes during pregnancy and cardiac events longitudinally.
Detailed Summary
Biological aging clocks have traditionally relied on invasive or expensive measurements—DNA methylation from blood, clinical ECGs, or brain MRI. These approaches are difficult to scale and rarely collected longitudinally. PpgAge represents a paradigm shift: a passive, non-invasive aging biomarker derived continuously from wrist-worn consumer wearables.
The research team, based at Apple, trained a deep learning model on approximately 20 million 60-second PPG segments from 172,318 participants using self-supervised contrastive learning. This produced a 256-dimensional feature vector for each PPG segment capturing waveform morphology reflecting cardiac, vascular, respiratory, and autonomic nervous system function. A linear regression model then mapped these averaged representations to chronological age using a curated healthy subcohort (n=6,728; 80% training, 20% test). The 'PpgAge gap'—predicted minus chronological age—was the primary biomarker of interest.
PpgAge achieved a MAE of 2.43 years in the healthy test cohort, with consistent accuracy across biological sex, race/ethnicity, and BMI subgroups. In the broader general population (n=120,235), MAE rose modestly to ~3.2 years, expected given that non-healthy participants were excluded from training. Critically, the PpgAge gap showed strong cross-sectional associations with chronic disease. Among 35–45 year old women, diabetes prevalence was 6.3% on average but jumped to 14.9% (2.38x) among those with a >6-year PpgAge gap, and dropped to 3.7% among those with a < −2-year gap. Similar patterns held for heart disease and heart failure. Prospectively, elevated PpgAge gap significantly predicted incident heart disease events independent of traditional cardiovascular risk factors.
Beyond disease, PpgAge gap tracked behavioral health: higher gaps were associated with smoking, poor sleep, and lower physical activity levels. Longitudinally, PpgAge exhibited sharp increases during pregnancy and around the time of cardiac events—suggesting real-time sensitivity to acute physiological changes. This longitudinal responsiveness distinguishes PpgAge from static biomarkers and opens potential applications in monitoring interventions or disease progression over time.
The study's scale and passivity are its greatest strengths—continuous, real-world data from over 213,000 participants spanning years. However, the population skews toward Apple Watch users, who may be healthier and more affluent than the general public. Self-reported disease diagnoses introduce classification noise, and causal interpretation of the age gap remains limited by the observational design. Nonetheless, PpgAge represents a compelling proof-of-concept that commercially available wearables can yield clinically meaningful aging biomarkers.
Key Findings
- PpgAge predicted chronological age with MAE of 2.43 years in healthy participants using wrist PPG waveforms.
- A >6-year PpgAge gap was associated with 2.38x higher diabetes diagnosis rates in 35–45-year-old women.
- Elevated PpgAge gap significantly predicted incident heart disease events, independent of standard risk factors.
- PpgAge gap correlated with smoking, sleep quality, and exercise levels, reflecting behavioral aging.
- Longitudinally, PpgAge spiked during pregnancy and around cardiac events, showing real-time physiological sensitivity.
Methodology
Self-supervised contrastive deep learning was used to extract 256-dimensional PPG waveform features from ~20 million 60-second Apple Watch segments. A linear regression model trained on a curated healthy subcohort (n=5,355) predicted chronological age; associations with disease, behavior, and longitudinal events were assessed in held-out healthy (n=1,373) and general (n=120,235) cohorts from the Apple Heart & Movement Study (NCT04198194).
Study Limitations
The AHMS cohort skews toward Apple Watch owners, who may be healthier and more affluent than the general population, limiting generalizability. Disease labels rely on self-report, introducing misclassification risk. The observational design precludes causal inference about the PpgAge gap's role in disease development.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.