Accurate classification of secondary progression in multiple sclerosis using a decision tree. (July 2021)

• Record Type: Journal Article
• Authors: Ramanujam, Ryan; Zhu, Feng; Fink, Katharina; Karrenbauer, Virginija Danylaitė; Lorscheider, Johannes; Benkert, Pascal; Kingwell, Elaine; Tremlett, Helen; Hillert, Jan; Manouchehrinia, Ali
• Is Part Of: Multiple sclerosis. Volume 27:Number 8(2021)
• Page Start: 1240
• Page End: 1249
• Publication Date: 2021-07
• Subjects: Multiple sclerosis -- classification -- secondary progressive -- decision tree; Central nervous system -- Diseases -- Periodicals; Myelin sheath -- Diseases -- Periodicals; Inflammation -- Periodicals; Multiple sclerosis -- Periodicals; Central Nervous System Diseases -- Periodicals; Demyelinating Diseases -- Periodicals; Inflammation -- Periodicals; Multiple Sclerosis -- Periodicals; Système nerveux central -- Maladies -- Périodiques; Gaine de myéline -- Maladies -- Périodiques; Inflammation (Pathologie) -- Périodiques; Sclérose en plaques -- Périodiques; Electronic journals; 616.834005
• DOI: 10.1177/1352458520975323
• Languages: English
• ISSNs: 1352-4585

Abstract

Background: The absence of reliable imaging or biological markers of phenotype transition in multiple sclerosis (MS) makes assignment of current phenotype status difficult. Objective: The authors sought to determine whether clinical information can be used to accurately assign current disease phenotypes. Methods: Data from the clinical visits of 14, 387 MS patients in Sweden were collected. Classifying algorithms based on several demographic and clinical factors were examined. Results obtained from the best classifier when predicting neurologist recorded disease classification were replicated in an independent cohort from British Columbia and were compared to a previously published algorithm and clinical judgment of three neurologists. Results: A decision tree (the classifier) containing only most recently available expanded disability scale status score and age obtained 89.3% (95% confidence intervals (CIs): 88.8–89.8) classification accuracy, defined as concordance with the latest reported status. Validation in the independent cohort resulted in 82.0% (95% CI: 81.0–83.1) accuracy. A previously published classification algorithm with slight modifications achieved 77.8% (95% CI: 77.1–78.4) accuracy. With complete patient history of 100 patients, three neurologists obtained 84.3% accuracy compared with 85% for the classifier using the same data. Conclusion: The classifier can be used to standardize definitions of disease phenotype across different cohorts. Clinically, this model could assist neurologists by providing additional information.