Passively detecting anxiety in clinical staff using
consumer-grade wearables and real-time self-report.
Passive sensing
How are you feeling right now?
ESM self-report
The Problem
Anxiety among healthcare workers directly affects clinical performance, decision-making, and long-term wellbeing. Yet the most anxious moments — during critical incidents, handovers, and complex procedures — are exactly when monitoring is hardest.
Investigating the use of consumer-grade wearable devices (Apple Watch, Fitbit) to passively detect anxiety in clinicians and patients. By leveraging physiological signals such as heart rate variability (HRV), skin conductance, and movement data, this project aims to provide real-time mental health insights without disrupting clinical workflows.
Continuous physiological signal monitoring. HR, HRV, and EDA streams, analysed in real time.
Our Approach
Off-the-shelf consumer wearables worn by clinical staff throughout their shifts. No additional hardware, no patient burden, no workflow interruptions. The sensors are already there — collecting HRV, EDA, accelerometry, and skin temperature around the clock.
Experience Sampling Method (ESM) prompts are sent to participants' phones at random intervals throughout each shift. These brief in-the-moment check-ins ask participants to rate their current anxiety or stress level, providing time-stamped ground-truth labels that are synchronised with the continuous wearable stream — enabling supervised model training and validation.
Apple Watch and Fitbit devices provide access to heart rate, HRV, accelerometry, and skin temperature. We work within the constraints of consumer hardware — noisy, intermittent data that must be modelled robustly without reliance on clinical-grade equipment.
Machine learning models learn to distinguish anxiety-related physiological patterns from physical activity, arousal, and normal variation — using ESM labels as the training signal and physiological time-series features as inputs. The result: passive detection that runs silently in a demanding real-world environment.
Methodology
ESM bridges the gap between passive physiological signals and subjective experience — turning self-reports into precision training labels.
Experience Sampling Method (ESM), also known as Ecological Momentary Assessment (EMA), is a structured data-collection approach in which participants receive periodic prompts throughout their day and respond with a brief self-report of their current psychological or emotional state. Unlike retrospective questionnaires, ESM captures in-the-moment experience — reducing recall bias and anchoring responses to the same temporal window as the physiological data stream.
In this project, participants receive short smartphone prompts at pseudo-random intervals during their clinical shift. Each prompt asks them to rate their current anxiety or stress level on a validated scale. Responses are time-stamped and automatically linked to the concurrent wearable signal epoch, creating aligned pairs of (physiological features, self-reported label) for model training.
Notifications are sent at variable intervals within fixed time windows, preventing habituation and ensuring coverage across a full shift.
Prompts use validated single-item or short-form anxiety scales designed for ESM use — minimising participant burden while maintaining psychometric reliability.
Each ESM response is matched to the wearable signal window immediately preceding the prompt — creating labelled training examples that reflect real physiological correlates of anxiety.
Repeated ESM responses per participant support mixed-effects and person-specific model approaches, accounting for individual differences in baseline physiology.
Continuous HRV, EDA, movement, and temperature — collected passively throughout the shift with no participant action required.
Combined into
Periodic in-the-moment self-reports that anchor physiological patterns to subjective anxiety state — the ground truth that makes supervised learning possible.
Labelled (physiology, ESM) pairs train and validate detection models in ecologically valid, real-world clinical settings.
Physiological Signals
Heart Rate Variability — autonomic nervous system tone
Electrodermal Activity — sympathetic arousal and stress response
Movement and activity patterns as contextual covariates
Peripheral temperature as an indicator of autonomic state
Passive
Data collection
ESM
Ground-truth labels
Real-time
Inference
Active
Project status