A Novel Approach to Clustering Accelerometer Data for Application in Passive Predictions of Changes in Depression Severity

Abstract

The treatment of mood disorders, which can become a lifelong process, varies widely in efficacy between individuals. Most options to monitor mood rely on subjective self-reports and clinical visits, which can be burdensome and may not portray an accurate representation of what the individual is experiencing. A passive method to monitor mood could be a useful tool for those with these disorders. Some previously proposed models utilized sensors from smartphones and wearables, such as the accelerometer. This study examined a novel approach of processing accelerometer data collected from smartphones only while participants of the open-science branch of the BiAffect study were typing. The data were modeled by von Mises-Fisher distributions and weighted networks to identify clusters relating to different typing positions unique for each participant. Longitudinal features were derived from the clustered data and used in machine learning models to predict clinically relevant changes in depression from clinical and typing measures. Model accuracy was approximately 95%, with 97% area under the ROC curve (AUC). The accelerometer features outperformed the vast majority of clinical and typing features, which suggested that this new approach to analyzing accelerometer data could contribute towards unobtrusive detection of changes in depression severity without the need for clinical input.

Improving Depression Severity Prediction from Passive Sensing: Symptom-Profiling Approach

“…We summarize the previous studies in Table 1, including information on the number and type of participants, duration of data collection, questionnaires used for EMAs, assessments for pre-/post-tests, and devices utilized for collecting the sensor data. Recent studies have used digital phenotyping to predict mental well-being and study the association between passively collected data and depression [7,8,[18][19][20][21][23][24][25][26]. Wang et al [18] conducted a study to evaluate mental health, including depression, stress, sleep, activity level, mood, sociability, and academic performance, in undergraduate and graduate students.…”

“…The classification model achieved an F 1 score of 0.88 to 0.94. Ross et al [26] introduced an innovative method for predicting fluctuations in depression severity, focusing on clustering accelerometer data specifically during participants' typing activities. The model achieved an accuracy of around 95%, accompanied by an area under the ROC curve of 97%.…”

Towards Personalised Depression Modelling and Explanation from Wearable Data

“…Given the prevalence of devices with sensors that can be used to monitor lifestyle activities, such as smartphones and smartwatches, researchers are proposing using such devices for detecting, monitoring and managing depression [11]. The use of wearable technology to supplement clinical approaches is particularly appealing as it is unobtrusive, real-time, often passive (requiring little or no active participation by the patient), of finer granularity (more data in the same time period) and allows assessments to occur in the person's usual environment [12].…”

Towards Personalised Mood Prediction and Explanation for Depression from Biophysical Data

“…Given the prevalence of devices with sensors that can be used to monitor lifestyle activities, such as smartphones and smartwatches, researchers are proposing using such devices to detect, monitor and manage depression [ 12 ]. The use of wearable technology to supplement clinical approaches is particularly appealing as it is unobtrusive, real time, often passive (requiring little or no active input by a depressed individual/patient), of finer granularity (more data in the same time period) and allows assessments to occur in the person’s usual environment [ 13 ].…”