A Clinical Decision Support Tool to Reduce the Need for Magnetic Resonance Imaging for Disease Monitoring in Multiple Sclerosis

BackgroundPeople with multiple sclerosis (MS) undergo magnetic resonance imaging (MRI) to monitor disease activity and treatment response. Current scan frequency recommendations are non-individualized, potentially increasing unnecessary imaging and costs. Risk-based tools could enable more personalized surveillance strategies. ObjectiveTo develop an algorithm that predicts new lesions in subsequent brain MRI. MethodsUsing longitudinal data from adults with MS at the Johns Hopkins MS Center (2017-2025) with [≥]2 visits and [≥]1 MRI scan, a logistic regression model with 5-fold stratified cross-validation predicted new lesions, using a sensitivity-prioritized threshold. Features included disease activity history, therapy class, and patient-reported outcomes. ResultsAmong 1,131 participants (3:1 female-to-male; mean age 48, SD 12.3), 8.8% developed new MRI lesions. At a 0.08 threshold, sensitivity was 0.72 and specificity 0.75, with AUC 0.80. The model identified 72 patients with new lesions and 772 without (low risk). Prior MRI activity and recent relapse predicted new lesions, while older age and high-efficacy DMT use were associated with lower risk. ConclusionThe algorithm accurately stratified patients by risk of MRI lesion activity and identified who could undergo longer surveillance intervals with low risk of missed inflammation. With validation and integration, it may enable personalized monitoring in MS care.