The Apple Watch is already technically capable of predicting stressful situations in people. Such a function could theoretically be supplied later via software update. This is the conclusion reached by researchers at the Canadian University of Waterloo after a pilot study with 36 participants. For their research project, they evaluated ECG data from the Apple Watch in order to use machine learning (ML) to discover signs typical of stress.
Despite various health sensors such as pulse measurement, ECG and blood oxygen measurement, Apple itself has not yet offered any software function that allows conclusions to be drawn about the stress level of the wearer. Competitors like Fitbit and Garmin are already using their watch sensors on some of their devices to assess user stress levels.
Stress detection: Benefits for the entire population
The Canadian researchers see after their pilot study the benefits of such a stress function not only for the wearer of the watch. Important information about the exposure of the population could also be gained in anonymous form for the health authorities. In this way, suitable measures and initiatives can be taken in good time to counteract this. High levels of stress are linked to health problems such as depression, obesity and cardiovascular disease. In Canada, every fifth citizen complains of high levels of stress. Digital watches have the advantage that they are already widespread and socially accepted. So they are well suited to collect such data in real time.
Apple’s ECG sensor is similar to a single-lead ECG. The software function allows users to record a 30-second ECG, the data from which can also be exported. What Apple advertises for checking sinus rhythm and detecting atrial fibrillation can also be used in stress detection, the researchers report. They used a homegrown app based on the developer interface HealthKit, which the study participants used to record an EKG every three hours for two weeks. They then evaluate the data and use it to train a machine learning model.
The stress detection worked so well
Among other things, the researchers were able to determine when a stressful situation began using characteristics such as heart acceleration and heartbeat delays. They found that detecting non-stress conditions was more accurate than actually detecting high levels of stress. If the stress detection was correct at 52 to 64 percent, the stress-free times were correctly detected at over 60 percent. For comparison: the devices already available for stress detection are 60 to 80 percent correct.
Among other things, a longer duration of the ECG recording could further improve the detection, it is said. But the combination with other data, such as sleep tracking, could improve the function even further, so that real-time detection is possible.
(mki)
