AI Safety

Welltory Wearables Reveal Heart Rate Tracks State-Level Hardship

19M readings from 18,734 users show hardship raises resting heart rate by 1.33 bpm

Deep Dive

A new preprint by Maria Levchenko and colleagues uses passive wearable data from the Welltory app to show that resting heart rate—a key cardiovascular risk marker—tracks socioeconomic hardship at the state level. The study processed 19.1 million quality-filtered photoplethysmography readings from 18,734 opt-in users, computing cohort-adjusted mean daytime resting heart rate per US state. They then related this to a four-component material-hardship composite (uninsurance, food insecurity, utility shutoff, housing insecurity) across 41 states with sufficient coverage (12,497 contributing users). After adjusting for six state health indicators, latitude, median age, and population density, the partial Spearman correlation was +0.74 (bootstrap 95% CI [+0.31, +0.87]). Users in the five highest-hardship states averaged 1.33 bpm higher than those in the five lowest.

The gradient remained robust across numerous checks—leave-one-state-out (+0.67 to +0.79), demographic re-weighting, census-region fixed effects (+0.75 to +0.78), split-sample resampling (held-out median +0.51 to +0.56), and heavy-wear subsamples (+0.48). Notably, the association was absent in two other wearable metrics and did not match income or inequality under the same adjustment. The authors caution that the composite was selected after exploratory analysis (a broader precarity index gave only +0.29), and the results are cross-sectional and ecological. Still, the work suggests consumer wearables could passively track population-level health disparities without active surveys, opening new avenues for real-time public health monitoring.

Key Points
  • 19.1 million heart rate readings from 18,734 Welltory users across 41 US states showed a ρ=+0.74 correlation with a material-hardship index.
  • Users in the five highest-hardship states had average resting heart rates 1.33 bpm higher than those in the five lowest.
  • The gradient persisted under multiple robustness checks (leave-one-out, demographic re-weighting, split-sample) but was absent for income or inequality.

Why It Matters

Passive wearables could replace costly surveys for real-time, population-level health disparity surveillance.

📬 Get the top 10 AI stories daily