Association between Serum Biomarker Profile and Real-World Evidence of Disability in Multiple Sclerosis

Abstract

AbstractObjectiveBiomarkers could inform disease worsening and severity in people with MS (pwMS). Few studies have examined blood biomarkers informative of patient-reported outcome (PRO) of disability in pwMS. In this study we examine the associations between serum protein biomarker profiles and patient-reported disability in pwMS.MethodsThis cross-sectional study included adults with a neurologist-confirmed diagnosis of MS from the University of Pittsburgh Medical Center (Pittsburgh, PA) and the Rocky Mountain MS Clinic (Salt Lake City, Utah) between 2017 and 2020. For exposure, we included 19 serum protein biomarkers potentially associated with MS inflammatory disease activity and 7 key clinical factors (age at sample collection, sex, race/ethnicity, disease subtype, disease duration, disease-modifying treatment, and time interval between sample collection and closest PRO assessment). Using 6 machine learning approaches (Least Absolute Shrinkage and Selection Operator [LASSO] regression, Random Forest [RF], XGBoost, Support-Vector Machines [SVM], stacking ensemble learning, and stacking classification algorithm), we examined model performance in predicting Patient Determined Disease Steps (PDDS) as the primary outcome. We assessed model prediction of Patient-Reported Outcomes Measurement Information System (PROMIS) physical function in a subgroup. We reported model performance using the held-out testing set.ResultsWe included 431 unique participants (mean age 49 years, 81% women, 94% non-Hispanic White). Using binary outcomes, models comprising both routine clinical factors and the 19 proteins as features consistently outperformed base models (containing clinical features alone or clinical features plus single protein) in predicting severe (PDDS≥4, PROMIS<35) versus mild/moderate (PDDS<4, PROMIS≥35) disability for all machine learning approaches, with LASSO achieving the best area under the curve (AUCPDDS=0.91, AUCPROMIS=0.90). Using ordinal/continuous outcomes, LASSO models with combined clinical factors and 19 proteins as features (R2PDDS=0.31, R2PROMIS=0.35) again outperformed base models. The four LASSO models (PDDS, PROMIS; both binary and ordinal/continuous) with combined clinical and protein features shared 2 clinical features (disease subtype, disease duration) and 4 protein biomarkers (CDCP1, IL-12B, NEFL, PRTG).ConclusionsSerum protein biomarker profiles improve the prediction of real-world MS disability status beyond clinical profile alone or clinical profile plus individual protein biomarker, reaching clinically actionable performance.