Doshi-Velez lab's wearable stress tool: social time cuts heart rate 5 bpm

A team of researchers including Esther Brown, Karis Moon, Victoria Dean, and Harvard's Finale Doshi-Velez has released a paper on arXiv proposing a framework that turns noisy wearable data into personalized, actionable stress insights. The core problem: raw streams like heart rate and respiration are context-dependent—a spike could come from sprinting, a tense presentation, or laughing. To disambiguate, the framework layers user annotations of activities, stress events, and interventions directly onto physiological data in an interactive web visualization, helping users spot patterns they'd otherwise miss.

In a four-week pilot with seven university students who logged 269 events, the system quantified the impact of different coping strategies. Social interaction reduced average heart rate by 4.35 to 5.0 beats per minute, deliberate rest lowered Garmin stress scores by 10.03 to 13.83 points, and mindfulness activities decreased HRV by 6.61 to 13.22 milliseconds. These effect sizes give users concrete feedback on what actually works for their bodies. While the sample is small, the approach demonstrates a replicable method for transforming noisy wearable streams into reflective, behavior-changing tools—paving the way for larger studies and consumer health products that don't just track data, but tell you what to do with it.