Wearable Stress Score as a Digital Biomarker

A wearable stress score is a composite number produced by a consumer device, typically derived from heart rate variability, heart rate, movement and sometimes electrodermal activity or skin temperature, and presented to the user as a measure of stress.

This directory lists it because researchers are increasingly offered these scores as study data, and because someone should say plainly what they are. Every component is a real physiological signal. The composite is a proprietary algorithm that differs between manufacturers, is generally unpublished, and can change with a firmware update. The label attached to it asserts a psychological construct that the underlying signals cannot identify.

Physiological arousal rises with exercise, caffeine, heat, illness, excitement and effort as readily as with psychological distress. A measure that cannot separate a hard workout from a difficult conversation is measuring activation, not stress.

The inputs are standard: beat interval variability, heart rate relative to a personal baseline, movement to identify rest, and in some devices skin conductance or temperature. The combination is where the difference lies, and it is usually not disclosed. Some devices report a continuous score across the day, others a daily summary, others a categorical state.

Systematic review of machine learning for stress monitoring from wearables shows the pattern clearly: models trained in laboratory conditions, where a stressor is applied deliberately and confounders are controlled, perform far better than the same approaches applied to free living data. Work specifically addressing the move from laboratory to ambulatory settings documents how much accuracy is lost when the context is no longer known.

Because the vendor score is not interpretable, the defensible research approach is to collect the underlying signals rather than the composite, and to pair them with a validated psychological instrument. The Perceived Stress Scale is the direct counterpart and the standard reference for the construct; the WHO-5 Well-Being Index and burnout and wellbeing scales cover related dimensions.

Used that way, the physiological data can genuinely add value: it shows when arousal occurred, how long it lasted and how quickly it resolved, which self report at the end of a week cannot. What it cannot do is replace the questionnaire, and studies that treat a vendor stress score as an outcome measure have substituted a marketing artefact for a measurement.

Ecological momentary assessment, which prompts participants to report state in the moment, is the practical way to attach meaning to physiological arousal in daily life.

The central limitation is construct validity, not accuracy. No consumer stress score in common use has published evidence that it measures psychological stress, and the algorithms are proprietary, undocumented and mutable.

Scores are not comparable between devices or between people, and firmware changes can shift values mid study without notice. Baseline calibration periods differ, so early data in a study may not be equivalent to later data from the same participant.

There is also a duty of care. Presenting a person with a daily stress number can itself cause distress, particularly in mental health populations, and studies should consider whether participants see the score at all.

• Cohen S, Kamarck T, Mermelstein R. A global measure of perceived stress. J Health Soc Behav. 1983. pubmed.ncbi.nlm.nih.gov
• Vos G, et al. Generalizable machine learning for stress monitoring from wearable devices: a systematic literature review. Int J Med Inform. 2023. pubmed.ncbi.nlm.nih.gov
• Smets E, et al. Into the wild: the challenges of physiological stress detection in laboratory and ambulatory settings. IEEE J Biomed Health Inform. 2019. pubmed.ncbi.nlm.nih.gov
• Smets E, et al. Large-scale wearable data reveal digital phenotypes for daily-life stress detection. NPJ Digit Med. 2018. pubmed.ncbi.nlm.nih.gov