Disease severity classification using passively collected smartphone-based keystroke dynamics within multiple...

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Multiple Sclerosis (MS) is a central nervous system disorder causing varied motor and non-motor symptoms.
• Current disability assessments for people with MS (pwMS) are limited by infrequent evaluations and potential inaccuracies.
• Remote health monitoring offers a cost-effective approach for continuous, objective data collection.

Purpose of the Study:

• To investigate if smartphone keyboard interactions can provide insights into the clinical status of pwMS.
• To leverage machine learning to analyze keystroke dynamics for assessing MS-related disability.
• To explore an unobtrusive method for frequent monitoring of pwMS.

Main Methods:

• Collected one year of smartphone interaction data from 24 Healthy Controls (HC) and 102 pwMS.
• Utilized machine learning to train four classification pipelines.
• Used clinical outcome measures as targets for model training and validation.

Main Results:

• Models achieved an AUC-ROC above 0.7 in the hold-out set.
• The highest performance was in estimating fine motor skills (AUC-ROC=0.753).
• Keyboard interaction data successfully discerned pwMS from HC and estimated clinical status.

Conclusions:

• Smartphone keyboard interactions, analyzed with machine learning, can serve as an effective tool for monitoring MS clinical disability.
• This method allows for high-frequency, unobtrusive data collection without increasing patient burden.
• Findings support the use of digital biomarkers for remote health monitoring in neurological conditions.