Harnessing Smartphone Typing Behaviors for Passive Assessment of Processing Speed and Executive Function

Alongside mood fluctuations, cognitive deficits affect daily functioning in individuals with mood disorders. Current methods to assess these symptoms lack the granularity and ecological validity needed to provide longitudinal but unobtrusive monitoring. As such, improved modalities to accurately monitor the extent and fluctuations of the impairments are needed. The reliance of smartphones in everyday life has facilitated passive monitoring of behaviors through these devices without additional burden on the individuals. Nuances in smartphone typing behaviors may reveal aspects relating to individuals’ cognitive function. The objective of this study was to develop a model entirely based on smartphone typing behaviors collected in-the-wild to predict processing speed and executive functioning in a sample of individuals with and without a mood disorder. Using explainable machine learning models, we were able to predict first whether participants were experiencing depressive symptoms with an accuracy of 82.4% by incorporating novel features developed around phone orientation. We combine selected features relating to mood into entirely passive models predicting completion times of a digital trail-making test part B, a well-validated assessment of processing speed and executive function, finding that task performance could be predicted within less than one standard deviation of the mean. Improvement of the model was conducted through feature engineering of behaviors thought to relate to diurnal patterns and sleep. Overall, our findings suggest that smartphone typing behaviors can be leveraged to passively assess processing speed and executive functioning and support the use of smartphones in unobtrusive monitoring of mood disorder symptomatology.