An Algorithm to Predict the Need for Magnetic Resonance Imaging to Support Disease Monitoring in Multiple Sclerosis

Michael Robinette,Karla Gray-Roncal,Kathryn Fitzgerald,Kadija Ferryman,Casey Overby Taylor,John Scott,Elias Sotirchos,Peter Calabresi,Ellen M Mowry,William Gray-Roncal

DOI: https://doi.org/10.1101/2024.06.20.24309267

2024-07-05

Abstract:People with multiple sclerosis (MS), a chronic neurologic disease, typically undergo brain magnetic resonance imaging (MRI) to assess for new disease activity or to evaluate response to disease-modifying therapy. The required frequency of these imaging studies has not yet been determined, and clinicians rely on expert panel recommendations or intuition to select the interval length for their patients. These former recommendations are often broad and lack the incorporation of individualized information (i.e., age, presence of recent disease activity, type of disease modifying therapy) into the synthesis of their assessment. We developed an algorithm that can predict new disease activity (i.e., lesions) on the next MRI scan. Multiple data sources from the electronic health records of 1045 patients were used to train and test the algorithm which resulted in an accurate solution with an area under the curve (AUC) of 0.8 that minimizes missed lesions while accurately identifying two-thirds of patients as not having a new lesion in subsequent MRI. We believe that this algorithm can be developed into a clinical decision support tool as input into a clinician-patient assessment of the appropriate interval length between MRIs for a specific patient.