Overview
- Researchers tracked 88 adults aged 45 to 77 over 10 months using a smartphone app and smartwatch that passively collected 21 indicators including heart rate, activity, sleep, weather and air pollution.
- Participants completed questionnaires and cognitive tests every three months, providing benchmarks to compare against AI predictions from the passive data.
- Emotional states were predicted most accurately, with error rates typically between 5% and 10%, while cognitive states showed higher errors of roughly 10% to 20%.
- Air pollution, weather, daily heart rate and sleep variability emerged as the most informative signals for cognition, and weather, sleep variability and sleep heart rate were most informative for emotions.
- The work, published in npj Digital Medicine and conducted under the Providemus alz project, has moved into a 24‑month follow‑up to evaluate longer‑term performance and individual variability.