Passive logging of smartphone interactions as a feasible and reliable measure to monitor everyday cognition in Alzheimer’s Disease

Abstract

AbstractBackgroundAs cognitive assessments can be burdensome to patients, new approaches are emerging to assess symptomatic changes in Alzheimer’s Disease (AD). Since typing on smartphones requires a variety of cognitive functions, passively and remotely collected typing behaviour has the potential to provide day‐to‐day insights into cognitive decline in AD. We investigated therefore the feasibility and reliability of a keyboard application for smartphones (Neurokeys) that passively collects keystroke dynamics (KD) to monitor cognition in people with preclinical and clinical AD.MethodParticipants with preclinical AD (N = 8) and patients with a clinical AD diagnosis (N = 8, Table 1) were included from the Amsterdam Dementia Cohort. Participants used Neurokeys for 28 consecutive days in the context of the ‘A Personalized Medicine Approach for Alzheimer’s Disease’ (ABOARD)‐project. The retention rate (active users on day 28) and the average number of daily keystrokes were calculated to determine feasibility. KD were aggregated into two commonly investigated features: time between keystrokes (flight‐time) and duration of a keystroke (hold‐time), which were standardized into z‐scores. Intra‐class correlations (ICC) between first and last 14 days of data collection for both features were determined to assess reliability. Finally, KD features were correlated to scores on the Mini‐Mental State Examination (MMSE) to explore the association between KD and global cognition.ResultsThe retention rate was 93.75% for all participants (Figure 1). Participants with preclinical AD used the keyboard for 26.25±2.12 days with 653.67±305.61 keystrokes each day, and clinical AD patients used the keyboard for 22.63±4.10 days with 409.44±288.73 daily keystrokes. For all participants, the ICC for flight‐time and hold‐time are 0.85 and 0.89 respectively. MMSE is negatively correlated with flight‐time (r = ‐0.68, p = 0.003), but not with hold‐time (r = ‐0.20, p = 0.45, Figure 2).ConclusionThe number of daily keystrokes is sufficient to obtain reliable KD features in preclinical and clinical AD. The results show that people with better global cognition type faster, indicating that passive logging of keystrokes can be a feasible measure for cognition in AD. To elaborate on these first findings, a larger validation study can further explore the utility of smartphone interactions and aid the development of digital biomarkers in AD.