[Paper Review] Identifying differences in physical activity and autonomic function patterns between psychotic patients and controls over a long period of continuous monitoring using wearable sensors

Panagiotis P. Filntisis, Athanasia Zlatintsi+6 more|arXiv (Cornell University)|Oct 30, 2020

Mental Health Research Topics29 references4 citations

TL;DR

This study uses long-term wearable sensor data from smartwatches to identify significant differences in physical activity and autonomic function between psychotic patients and healthy controls. It applies short-time signal analysis with both standard and novel nonlinear features—particularly MFD, sample entropy, and HRV metrics—revealing that patients exhibit higher variability in movement and autonomic markers during wakefulness and reduced activity during sleep, suggesting potential biomarkers for relapse prediction.

ABSTRACT

Digital phenotyping is a nascent multidisciplinary field that has the potential to revolutionize psychiatry and its clinical practice. In this paper, we present a rigorous statistical analysis of short-time features extracted from wearable data, during long-term continuous monitoring of patients with psychotic disorders and healthy control counterparts. Our novel analysis identifies features that fluctuate significantly between the two groups, and offers insights on several factors that differentiate them, which could be leveraged in the future for relapse prevention and individualized assistance.

Motivation & Objective

• To detect reliable, continuous biomarkers of psychotic disorders using passive, unobtrusive wearable sensor data.
• To compare long-term physical activity and autonomic nervous system function between psychotic patients and healthy controls.
• To evaluate the discriminative power of both conventional and novel nonlinear signal features in distinguishing patient and control groups.
• To explore the feasibility of using consumer smartwatches for long-term, real-world monitoring in psychiatric populations.
• To lay the groundwork for future relapse prediction and personalized clinical interventions using digital phenotyping.

Proposed method

• Employed a commercial off-the-shelf smartwatch for 24/7, long-term (over 1 year) continuous monitoring of 23 controls and 22 psychotic patients.
• Extracted short-time features from accelerometer (acc), gyroscope (gyr), and heart rate variability (HRV) signals using time-domain, nonlinear, and fractal analysis techniques.
• Applied standard deviation, sample entropy (sampen), Poincaré plot parameters (SD1, SD2), and Multiscale Fractal Dimension (MFD) features to detect group differences.
• Used Mann-Whitney U tests to assess statistical significance of feature distributions between patients and controls during wake and sleep states.
• Conducted descriptive statistics and boxplot visualization to compare mean, standard deviation, and variability of movement and autonomic markers.
• Focused on features derived from short-time energy, HRV, and fractal geometry to identify subtle, persistent physiological differences.

Experimental results

Research questions

• RQ1 Which physical activity and autonomic function features significantly differ between psychotic patients and healthy controls during prolonged monitoring?
• RQ2 How do short-time signal features—especially nonlinear and fractal measures—discriminate between patients and controls?
• RQ3 To what extent do movement variability and autonomic regulation patterns differ between patients and controls during wakefulness and sleep?
• RQ4 Can wearable sensor data reveal persistent, clinically relevant biomarkers for psychotic disorders beyond traditional clinical assessments?
• RQ5 How do medication effects and lifestyle factors (e.g., smoking) potentially influence the observed physiological differences?

Key findings

• Patients showed significantly higher standard deviation in short-time energy of accelerometer and gyroscope signals during wakefulness, indicating greater movement variability.
• Significant differences were found in the standard deviation of Poincaré plot parameters (SD1 and SD2), suggesting altered short-term heart rate variability in patients.
• Sample entropy mean and standard deviation of HRV showed significant group differences during both wake and sleep states, particularly in sleep.
• Multiscale Fractal Dimension (MFD) features—especially mean, min, max, and standard deviation of MFD—differed significantly between groups across both wake and sleep periods.
• The sleep-wake ratio and daily total steps per day showed highly significant differences (p < 0.001) between patients and controls, with patients exhibiting lower and more variable step counts.
• Spectral analysis using the Lomb-Scargle periodogram did not reveal significant differences, indicating that nonlinear and time-domain features were more discriminative than frequency-domain features.