Author Of 1 Presentation
P0163 - Smartphone keystroke dynamics are sensitive to changes in disease activity and clinical disability measures in multiple sclerosis (ID 748)
Abstract
Background
Typing behavior on a smartphone may be used as a biomarker in patients with multiple sclerosis (MS) by analyzing their keystroke dynamics (KD). The continuous acquisition of high sample rate data may provide unprecedented insights in short-term changes in important health outcomes in MS.
Objectives
To investigate the sensitivity of KD to clinically relevant change (i.e. responsiveness) in disease activity, fatigue, and clinical disability outcomes in patients with MS.
Methods
Patients with MS were recruited in this cohort study. Clinical outcomes were assessed at baseline and 3 months follow-up, including: MRI gadolinium-enhancing lesions (Gd-EL), patient-perceived fatigue, and clinical disability measures (Expanded Disability Status Scale, EDSS; Timed 25-foot Walk Test, TWT; Nine-Hole Peg Test, NHPT; Arm function in MS Questionnaire, AMSQ). Throughout the study, patients used the Neurokeys App which replaces the native keyboard with a smart-keyboard and unobtrusively collects time-stamped key press and release events in the real-world setting. Keystroke data of 14 days surrounding the clinical visits were aggregated for the analyses. The area under the receiver operating characteristics curve (AUROC) was calculated to assess responsiveness of KD in classifying anchor-based change within clinical outcomes. The minimally important change (MIC) was calculated as the mean change in KD in the lower +2 SD portion (to approximate minimal change) of patients with clinically relevant change for each clinical outcome. The MIC was compared to the smallest detectable change (SDC) to assess the capability of KD to distinguish important change from measurement error.
Results
102 patients with MS were included, of whom 94 completed follow-up. Responsiveness of KD were acceptable for change in number of MRI Gd-EL (highest AUROC = 0.73) and arm function based on the AMSQ score (highest AUROCs = 0.75). KD had excellent responsiveness to change in ambulatory function measured with TWT (highest AUROC = 0.84). EDSS and NHPT had lower AUC values than KD in classifying change in Gd-EL and AMSQ, respectively. For all keystroke features the MIC exceeded the SDC with differences ranging from 3.6 to 92.4%.
Conclusions
KD collected in patients with MS using the Neurokeys App demonstrated responsiveness to clinically relevant changes in gadolinium-enhancing lesions on MRI and clinical disability measures for arm and ambulatory function. Responsiveness of KD was higher than commonly used clinical measures in MS and sensitive enough to discriminate important change from measurement error.