The Algorithm That Knows You're Sick Before You Do

Smartwatch algorithms are now detecting illness 48 hours before symptoms appear. Here is the specific data they use, what conditions they can flag, and the privacy cost of wearing a health monitor.

The Apple Watch's irregular heart rhythm detection has been available since Series 4 in 2018. In the years since, the range of health conditions that consumer wearable devices can detect or predict from continuous physiological monitoring has expanded to include COVID-19 onset (before symptomatic phase), flu illness, sleep apnoea, atrial fibrillation, and the specific hormonal patterns associated with menstrual cycle phase and fertility window in women.

The underlying data that enables these predictions consists of continuous streams of measurements that were not previously available for clinical purposes because clinical monitoring required clinical settings. Continuous heart rate data — not just resting heart rate but the second-by-second variation that constitutes heart rate variability. Continuous skin temperature. Blood oxygen saturation. Movement patterns. Sleep architecture inferred from movement and heart rate. The combination of these streams, analysed by machine learning models trained on data from populations whose outcomes are known, produces early warning signals that are genuinely informative.

The fever and illness detection capability is most validated: multiple studies have shown that smartwatch algorithms can detect the physiological changes preceding symptomatic illness — elevated resting heart rate, decreased heart rate variability, changes in sleep architecture — with sufficient accuracy to flag elevated illness risk 24-48 hours before symptoms appear. During COVID-19, this capability had specific public health value: infected individuals who were alerted to potential illness by their devices before becoming symptomatic could self-isolate earlier.

For the privacy consideration that receives less attention than the health benefit: these devices are generating continuous streams of intimate physiological data that reflects not just health status but emotional state, stress levels, and behavioural patterns. The data is stored on company servers, subject to privacy policies that most users haven't read, and potentially accessible to insurers, employers, and law enforcement under varying legal frameworks. The health monitoring value is genuine and growing. The privacy cost is also genuine and not yet adequately regulated.