Validated, Quantitative, Machine Learning-Generated Neurologic Assessment of Multiple Sclerosis Using a Mobile Application

Abstract

BACKGROUND: The BeCare MS Link mobile app collects data as users complete different in-app assessments. It was specifically developed to evaluate the symptomatology and neurologic function of patients with multiple sclerosis (MS) and to become a digital equivalent of the Expanded Disability Status Scale (EDSS) and other standard clinical metrics of MS progression. METHODS: Our research compared EDSS scores derived from the BeCare MS link app to EDSS scores derived from neurologist assessment for the same cohort of 35 patients diagnosed with MS. App-derived data were supplied to 4 different machine learning algorithms (MLAs) with an independent EDSS score prediction generated from each. These scores were compared with the clinically derived EDSS score to assess the similarity of the scores and to determine an accuracy estimate for each. RESULTS: Of the 4 MLAs employed, the most accurate MLA produced 19 EDSS score predictions that exactly matched the clinically derived scores, 21 score predictions within 0.5 EDSS points, and 32 score predictions within 1 EDSS point. The remaining MLAs also provided a relatively high level of accuracy in predicting EDSS scores when compared with clinically derived EDSS, with over 80% of scores predicted within 1 point and a mean squared error with a range of 1.05 to 1.37. CONCLUSIONS: The BeCare MS Link app can replicate the clinically derived EDSS assessment of a patient with MS. The app may also offer a more complete evaluation of disability in patients with MS.