Your Smartwatch May Be Able to Predict Cognitive Decline — Before You Notice It
BiohackingThe case for wearing a smartwatch just got significantly stronger — not for fitness tracking, but for brain health monitoring. A new study from the University of Geneva demonstrates that consumer-grade wearables can predict fluctuations in cognitive and emotional health with meaningful accuracy, using only passive sensor data collected during normal daily life.
Context
Brain health has historically been measured in clinical snapshots: a cognitive test every few months, a questionnaire at an annual check-up. The gaps between those snapshots are exactly where early signals tend to get missed. The premise of continuous monitoring is that if you can capture what’s happening to the brain day-to-day — not just at the moments when someone enters a clinic — you can identify deviations from a personal baseline long before they become clinically significant.
Wearables already collect data continuously. The question is whether that data contains enough signal to be clinically useful. This study is the first to answer that question at scale, in a healthy aging population, over a long enough timeframe to be meaningful.
The Detail
The study, published in npj Digital Medicine (Matias et al., 2026), enrolled 88 volunteers ages 45–77 and tracked them for 10 months using a smartphone app and a consumer smartwatch. Participants made no behavioral changes — the devices passively collected heart rate, physical activity, sleep patterns, and environmental data including weather conditions and air pollution exposure.
Every three months, participants completed emotional state questionnaires and cognitive performance tests. The AI model was trained to predict those active assessments from the continuous passive data — essentially asking: does what the watch sees on a Tuesday predict how well someone will perform on a cognitive test that Friday?
The results were sharper than expected:
- Average AI prediction error: 12.5% across all 21 tracked indicators
- Emotional states: 5–10% error — the model predicted emotional questionnaire responses with high accuracy, with sleep variability, heart rate during sleep, and weather conditions as the most predictive passive signals
- Cognitive states: 10–20% error — harder to predict, but still well above chance; air pollution, weather, daily heart rate, and sleep variability were the strongest predictors
Lead author Igor Matias: “On average, the error rate was just 12.5%, opening up new possibilities for the use of connected devices in the early detection of abnormalities or changes in brain health.”
The study was conducted as part of the Providemus alz research project, supervised by professors Katarzyna Wac and Matthias Kliegel at UNIGE’s Research Institute for Statistics and Information Science.
The So What
The WHO estimates more than 1 in 3 people worldwide live with a neurological disorder, and more than 1 in 2 individuals will experience a mental health disorder in their lifetime. Both categories involve gradual changes that are currently invisible between clinical check-ins. This study doesn’t prove that consumer wearables can predict who is at clinical risk — that’s a next step, and a follow-up phase is already underway. What it demonstrates is that the passive signal is there: the data your watch already collects can track meaningful variability in cognitive and emotional state with reasonable accuracy in healthy adults over 45.
The practical implication for biohackers and self-trackers is to take sleep quality data and HRV-during-sleep metrics seriously — not just as fitness indicators, but as potential cognitive health proxies. The most predictive inputs in the model weren’t exotic biomarkers; they were sleep variability and resting heart rate during sleep, both measurable by any Garmin, Apple Watch, or Fitbit you’re likely already wearing.
The larger signal here is directional: continuous, ambient monitoring is coming to brain health the same way it came to cardiovascular health. The question is no longer whether the data can capture something useful. It’s how long before it’s clinical-grade.
Photo by Andrey Matveev on Pexels
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →