Improved prediction of early cognitive impairment in multiple sclerosis combining blood and imaging biomarkers. Issue 4 (8th July 2022)

• Record Type: Journal Article
• Authors: Brummer, Tobias; Muthuraman, Muthuraman; Steffen, Falk; Uphaus, Timo; Minch, Lena; Person, Maren; Zipp, Frauke; Groppa, Sergiu; Bittner, Stefan; Fleischer, Vinzenz
• Is Part Of: Brain communications. Volume 4:Issue 4(2022)
• Publication Date: 2022-07-08
• Subjects: multiple sclerosis -- serum neurofilament -- cognition -- lesion volume -- grey matter; 616
• DOI: 10.1093/braincomms/fcac153
• Languages: English
• ISSNs: 2632-1297

Abstract

Abstract: Disability in multiple sclerosis is generally classified by sensory and motor symptoms, yet cognitive impairment has been identified as a frequent manifestation already in the early disease stages. Imaging- and more recently blood-based biomarkers have become increasingly important for understanding cognitive decline associated with multiple sclerosis. Thus, we sought to determine the prognostic utility of serum neurofilament light chain levels alone and in combination with MRI markers by examining their ability to predict cognitive impairment in early multiple sclerosis. A comprehensive and detailed assessment of 152 early multiple sclerosis patients (Expanded Disability Status Scale: 1.3 ± 1.2, mean age: 33.0 ± 10.0 years) was performed, which included serum neurofilament light chain measurement, MRI markers (i.e. T2 -hyperintense lesion volume and grey matter volume) acquisition and completion of a set of cognitive tests (Symbol Digits Modalities Test, Paced Auditory Serial Addition Test, Verbal Learning and Memory Test) and mood questionnaires (Hospital Anxiety and Depression scale, Fatigue Scale for Motor and Cognitive Functions). Support vector regression, a branch of unsupervised machine learning, was applied to test serum neurofilament light chain and combination models of biomarkers for the prediction of neuropsychological test performance. The support vector regression results were validated in a replication cohort of 101 early multiple sclerosis patients (Expanded Disability Status Scale: 1.1 ± 1.2, mean age: 34.4 ± 10.6 years). Higher serum neurofilament light chain levels were associated with worse Symbol Digits Modalities Test scores after adjusting for age, sex Expanded Disability Status Scale, disease duration and disease-modifying therapy (B = −0.561; SE = 0.192; P = 0.004; 95% CI = −0.940 to −0.182). Besides this association, serum neurofilament light chain levels were not linked to any other cognitive or mood measures (all P -values > 0.05). The tripartite combination of serum neurofilament light chain levels, lesion volume and grey matter volume showed a cross-validated accuracy of 88.7% (90.8% in the replication cohort) in predicting Symbol Digits Modalities Test performance in the support vector regression approach, and outperformed each single biomarker (accuracy range: 68.6–75.6% and 68.9–77.8% in the replication cohort), as well as the dual biomarker combinations (accuracy range: 71.8–82.3% and 72.6–85.6% in the replication cohort). Taken together, early neuro-axonal loss reflects worse information processing speed, the key deficit underlying cognitive dysfunction in multiple sclerosis. Our findings demonstrate that combining blood and imaging measures improves the accuracy of predicting cognitive impairment, highlighting the clinical utility of cross-modal biomarkers in multiple sclerosis. Abstract : Brummer et al. report that a machine learning approach using a combination of blood and imaging measures improves the predictive accuracy of cognitive impairment in early multiple sclerosis. Their findings highlight the clinical utility of cross-modal biomarkers in multiple sclerosis patients who are likely to experience cognitive decline. Graphical Abstract: Graphical abstract