Predicting Cognitive Functioning in Mood Disorders Through Smartphone Typing Dynamics

Abstract

Mood disorders (MDs) such as major depressive disorder and bipolar disorder are associated with significant functional impairments, particularly in cognition, which can adversely affect daily functioning and social interactions. This study aims to predict cognitive functioning prospectively in individuals with MDs using passive data from smartphone typing dynamics. Over a period of approximately 28 days, participants (N= 127) utilized the BiAffect keyboard, which captured typing metadata such as keystroke timestamps and accelerometer data during typing sessions, while also undergoing in-lab neuropsychological assessments twice (at least 14 days apart). Principal component analysis was applied to keyboard features, and the component scores were subsequently used in structural equation modeling to predict performance on the NIH Toolbox cognitive tests and the Trail-Making Test, Part B. The results showed that slower typing speeds predicted worse NIH Toolbox performance only in healthy controls, suggesting a weaker or more variable relationship in MDs. However, for the Trail-Making Test, Part B, keystroke dynamics predicted performance equally across groups. These findings highlight the potential of keystroke dynamics as an ecologically valid, passive measure of cognitive function, while also underscoring its varying utility depending on the cognitive domain assessed and the population studied.