Predicting Mood Disturbance Severity with Mobile Phone Keystroke Metadata: A BiAffect Digital Phenotyping Study

Abstract

BackgroundMood disorders are common and associated with significant morbidity and mortality. Better tools are needed for their diagnosis and treatment. Deeper phenotypic understanding of these disorders is integral to the development of such tools. This study is the first effort to use passively collected mobile phone keyboard activity to build deep digital phenotypes of depression and mania.ObjectiveThe objective of our study was to investigate the relationship between mobile phone keyboard activity and mood disturbance in subjects with bipolar disorders and to demonstrate the feasibility of using passively collected mobile phone keyboard metadata features to predict manic and depressive signs and symptoms as measured via clinician-administered rating scales.MethodsUsing a within-subject design of 8 weeks, subjects were provided a mobile phone loaded with a customized keyboard that passively collected keystroke metadata. Subjects were administered the Hamilton Depression Rating Scale (HDRS) and Young Mania Rating Scale (YMRS) weekly. Linear mixed-effects models were created to predict HDRS and YMRS scores. The total number of keystrokes was 626,641, with a weekly average of 9791 (7861), and that of accelerometer readings was 6,660,890, with a weekly average 104,076 (68,912).ResultsA statistically significant mixed-effects regression model for the prediction of HDRS-17 item scores was created: conditional R2=.63, P=.01. A mixed-effects regression model for YMRS scores showed the variance accounted for by random effect was zero, and so an ordinary least squares linear regression model was created: R2=.34, P=.001. Multiple significant variables were demonstrated for each measure.ConclusionsMood states in bipolar disorder appear to correlate with specific changes in mobile phone usage. The creation of these models provides evidence for the feasibility of using passively collected keyboard metadata to detect and monitor mood disturbances.

A comparative study of engagement in mobile and wearable health monitoring for bipolar disorder

“…The study required participants to utilize mobile and wearable health technology; thus, there was likely a selection bias based on participants’ willingness or interest in using smartphone and wearable technology. Another limitation is that an activity tracker was used for passive monitoring, but there are other forms of passive monitoring such as keystrokes, GPS, acoustics, call patterns, which may achieve better adherence rates than an activity tracker because they are collected directly from a smartphone. Another limitation is all participants were assessed weekly, which may dampen the effect of reviewing recorded symptoms on engagement compared to a situation in which only those who reviewed recorded symptoms interacted with an interviewer.…”

“…Sensors and wearable devices allow symptoms to monitored passively and objectively . For example, MONARCA, PRIORI, and Bi‐Affect use patterns of speech and behavior from recorded calls, keystrokes, number of phone calls, and duration of phone calls to predict mood. Wearable devices, such as activity trackers, also offer direct and indirect measurements of a wide range of clinically important variables of BP such as mood, physical activity, heart rate variability, sleep patterns, and circadian rhythms .…”

Circadian rhythm disruption and mental health

“…Indeed, passively collected data from mobile phones can be used to monitor and predict real time behavioral indicators of depression and PTSD 175 . Furthermore, passively collected keystroke metadata may also be used to predict state changes found in BD 176 . In future studies, it might be informative to consider leveraging data from the devices' sensors to monitor ambient light, at eye level, in association with the other data collected to investigate whether exposure to light at night is a predictive cofactor in affective disorder state.…”

Smartphone-based Mental State Estimation: A Survey from a Machine Learning Perspective

“…For feature PHO-9, number of calendar events, Wahle et al assumed that too many calendar events could influence depression levels and tracked the number of stored calendar events [67]. To define feature PHO-10, keypress features, Zulueta et al considered various metrics, such as average time between keystrokes, number of backspace keypresses, number of autocorrect events, average accelerometer amplitude while typing, number of keypress sessions, and average keypress session length [75]. Mehrotra et al defined the number of normal clicks and long clicks on the phone screen as a feature [74].…”