Smartphone-derived keystroke dynamics are sensitive to relevant changes in multiple sclerosis

Background To investigate smartphone keystroke dynamics (KD), derived from regular typing, on sensitivity to relevant change in disease activity, fatigue, and clinical disability in multiple sclerosis (MS). Methods Preplanned interim analysis of a cohort study with 102 MS patients assessed at baseline and 3-month follow-up for gadolinium-enhancing lesions on magnetic resonance imaging, relapses, fatigue and clinical disability outcomes. Keyboard interactions were unobtrusively collected during typing using the Neurokeys App. From these interactions 15 keystroke features were derived and aggregated using 16 summary and time series statistics. Responsiveness of KD to clinical anchor-based change was assessed by calculating the area under the receiver operating characteristic curve (AUC). The optimal cut-point was used to determine the minimal clinically important difference (MCID) and compared to the smallest real change (SRC). Commonly used clinical measures were analyzed for comparison. Results A total of 94 patients completed the follow-up. The five best performing keystroke features had AUC-values in the range 0.72-0.78 for change in gadolinium-enhancing lesions, 0.67-0.70 for the Checklist Individual Strength Fatigue subscale, 0.66-0.79 for the Expanded Disability Status Scale, 0.69-0.73 for the Ambulation Functional System, and 0.72-0.75 for Arm function in MS Questionnaire. The MCID of these features exceeded the SRC on group level. KD had higher AUC-values than comparative clinical measures for the study outcomes, aside from ambulatory function. Conclusions Keystroke dynamics demonstrated good responsiveness to changes in disease activity, fatigue, and clinical disability in MS, and detected important change beyond measurement error on group level. Responsiveness of KD was better than commonly used clinical measures.

Automated summary

The study found that smartphone keystroke dynamics are sensitive to changes in disease activity, fatigue, and clinical disability in multiple sclerosis patients, detecting important changes beyond measurement error on a group level. Keystroke dynamics showed better responsiveness than commonly used clinical measures.