Differential temporal utility of passively sensed smartphone features for depression and anxiety symptom prediction: a longitudinal cohort study

Passive sensing data predicts stress in university students: a supervised machine learning method for digital phenotyping

Introduction

University students are particularly susceptible to developing high levels of stress, which occur when environmental demands outweigh an individual's ability to cope. The growing advent of mental health smartphone apps has led to a surge in use by university students seeking ways to help them cope with stress. Use of these apps has afforded researchers the unique ability to collect extensive amounts of passive sensing data including GPS and step detection. Despite this, little is known about the relationship between passive sensing data and stress. Further, there are no established methodologies or tools to predict stress from passive sensing data in this group.

Methods

In this study, we establish a clear machine learning-based methodological pipeline for processing passive sensing data and extracting features that may be relevant in the context of mental health.

Results

We then use this methodology to determine the relationship between passive sensing data and stress in university students.

Discussion

In doing so, we offer the first proof-of-principle data for the utility of our methodological pipeline and highlight that passive sensing data can indeed digitally phenotype stress in university students.

Clinical trial registration

Australia New Zealand Clinical Trials Registry (ANZCTR), identifier ACTRN12621001223820.

Longitudinal associations of daily affective dynamics with depression, generalized anxiety, and social anxiety symptoms

BACKGROUND

Low average affect, measured using ecological momentary assessment (EMA), has been consistently linked with depression, generalized anxiety, and social anxiety, supporting trait-like negative affect as a shared underlying feature. However, while theoretical models of emotion regulation would also implicate greater variability in daily affect in these conditions, empirical evidence linking EMA of mood variability with affective disorders is mixed. We used multilevel modeling to test relationships of daily mood and mood variability with depression, generalized anxiety, and social anxiety symptoms.

METHODS

Participants (N = 1004; 72.31 % female; Mage = 40.85) responded to EMA of mood 2-3×/day and completed measures of depression (PHQ-8), generalized anxiety (GAD-7), and social anxiety (SPIN) every three weeks.

RESULTS

Lower mean affect predicted all symptoms at both the between-person (PHQ-8: β = -0.486, p < 0.001; GAD-7: β = -0.429, p < 0.001; SPIN: β = -0.284, p < 0.001) and within-person (PHQ-8: β = -0.219, p < 0.001; GAD-7: β = -0.196, p < 0.001; SPIN: β = -0.049, p < 0.001) levels. Similarly, at the between-person level, greater affective variability was linked with all three clinical symptoms (PHQ-8: β = 0.617, p < 0.001; GAD-7: β = 0.703, p < 0.001; SPIN: β = 0.449, p < 0.001). However, within-person, affective variability related to depression (β = 0.144, p < 0.001) and generalized anxiety (β = 0.150, p < 0.001), but not social anxiety (β = 0.006, p = 0.712).

LIMITATIONS

The COVID-19 pandemic lockdown period occurred midway through the study.

CONCLUSION

Findings point to common and specific emotion dynamics that characterize affective symptoms severity, with implications for affective monitoring in a clinical context.

Multiclassification of the symptom severity of social anxiety disorder using digital phenotypes and feature representation learning

Objective

Social anxiety disorder (SAD) is characterized by heightened sensitivity to social interactions or settings, which disrupts daily activities and social relationships. This study aimed to explore the feasibility of utilizing digital phenotypes for predicting the severity of these symptoms and to elucidate how the main predictive digital phenotypes differed depending on the symptom severity.

Method

We collected 511 behavioral and physiological data over 7 to 13 weeks from 27 SAD and 31 healthy individuals using smartphones and smartbands, from which we extracted 76 digital phenotype features. To reduce data dimensionality, we employed an autoencoder, an unsupervised machine learning model that transformed these features into low-dimensional latent representations. Symptom severity was assessed with three social anxiety-specific and nine additional psychological scales. For each symptom, we developed individual classifiers to predict the severity and applied integrated gradients to identify critical predictive features.

Results

Classifiers targeting social anxiety symptoms outperformed baseline accuracy, achieving mean accuracy and F1 scores of 87% (with both metrics in the range 84-90%). For secondary psychological symptoms, classifiers demonstrated mean accuracy and F1 scores of 85%. Application of integrated gradients revealed key digital phenotypes with substantial influence on the predictive models, differentiated by symptom types and levels of severity.

Conclusions

Leveraging digital phenotypes through feature representation learning could effectively classify symptom severities in SAD. It identifies distinct digital phenotypes associated with the cognitive, emotional, and behavioral dimensions of SAD, thereby advancing the understanding of SAD. These findings underscore the potential utility of digital phenotypes in informing clinical management.