To determine whether conventional brain imaging features at onset predict subsequent disease activity in children with multiple sclerosis (MS).

Early identification of patients at risk of a poor disease course would aid treatment selection.

Participants with a diagnosis of MS and seronegative for AQP4 and MOG antibodies, were selected from a prospective cohort of children with incident demyelinating syndromes. Baseline (within 30 days of presentation) MRI features, and 2001, 2010 and 2017 international diagnostic criteria were evaluated using a validated scoring instrument. Relapses, normalized brain volume, and new lesions were recorded on follow-up (imaging at 0, 3, 6, 12 months and annually thereafter). Cox hazard models assessed baseline features predicting time to second attack and gaussian or binomial mixed effects models were applied for predicting normalized brain volume or new lesions; sensitivity, specificity and predictive values were computed for identifying patients with high frequency of attacks(≥3 in first two years).

32 potential predictors were scored in 56 patients (42 F; median (IQR) age at onset 14.25 (12.92-15.40)y; 39 with clinical relapses; median (IQR) EDSS at two years 1 (0-2); 483 scans; median (IQR) follow-up 6.6 (4.8)y). Black holes at baseline (93% positive) were the best predictor of time to second attack (hazard ratio [95%CI]: 6 [0.8-46]), and predicted new lesions, either enhancing (HR [95%CI]: 12 [1.4-104]) or T2 hyperintense (HR [95% CI]: 6 [2-20]). Two year change in brain volume z-score was -0.15±0.09 and was not predicted significantly by any MRI feature, including baseline brain volume. Baseline 2010 criterion of dissemination in time best predicted the high relapse group (sensitivity/specificity 1/0.4; NPV/PPV 1/0.3). Baseline lesion count and T2 lesion volume were poor predictors.

Conventional baseline imaging features poorly predict prognosis in children, motivating efforts to identify novel imaging or biological biomarkers of aggressive MS.