Prediction of vasculitic neuropathy using supervised machine learning approaches

Abstract

Vasculitic neuropathy is an inflammation-driven nerve condition that often goes undiagnosed until irreversible damage occurs. This study developed and validated a supervised machine learning model to predict future onset of vasculitic neuropathy using electronic health record data from 450 cases and 1,800 matched controls. The predictive algorithm analyzed 134 structured features related to diagnoses, medications, lab tests and clinical notes. Selected logistic regression model with L2 regularization achieved an AUC of 0.92 (0.89-0.94 CI) internally, and maintained an AUC of 0.90 (0.84-0.93 CI) in the temporal validation cohort. At peak operating threshold, external sensitivity was 0.81 and specificity 0.79. Among highest risk decile, positive predictive value reached 47%. Key features driving predictions included inflammatory markers, neuropathic symptoms and vascular imaging patterns. This methodology demonstrates feasibility of leveraging machine learning for early detection of impending vasculitic neuropathy prior to confirmatory biopsy to enable prompt treatment and improved outcomes.