Retrospective Pediatric Cohort Study Validates NEOS Score and Demonstrates Applicability in Children With Anti-NMDAR Encephalitis

Abstract

Background and objectives: Anti-N-methyl-D-aspartate receptor encephalitis (NMDARE) is the most common form of autoimmune encephalitis in children and adults. Although our understanding of the disease mechanisms has progressed, little is known about estimating patient outcomes. Therefore, the NEOS (anti-NMDAR Encephalitis One-Year Functional Status) score was introduced as a tool to predict disease progression in NMDARE. Developed in a mixed-age cohort, it currently remains unclear whether NEOS can be optimized for pediatric NMDARE.

Methods: This retrospective observational study aimed to validate NEOS in a large pediatric-only cohort of 59 patients (median age of 8 years). We reconstructed the original score, adapted it, evaluated additional variables, and assessed its predictive power (median follow-up of 20 months). Generalized linear regression models were used to examine predictability of binary outcomes based on the modified Rankin Scale (mRS). In addition, neuropsychological test results were investigated as alternative cognitive outcome.

Results: The NEOS score reliably predicted poor clinical outcome (mRS ≥3) in children in the first year after diagnosis (p = 0.0014) and beyond (p = 0.036, 16 months after diagnosis). A score adapted to the pediatric cohort by adjusting the cutoffs of the 5 NEOS components did not improve predictive power. In addition to these 5 variables, further patient characteristics such as the "Herpes simplex virus encephalitis (HSE) status" and "age at disease onset" influenced predictability and could potentially be useful to define risk groups. NEOS also predicted cognitive outcome with higher scores associated with deficits of executive function (p = 0.048) and memory (p = 0.043).

Discussion: Our data support the applicability of the NEOS score in children with NMDARE. Although not yet validated in prospective studies, NEOS also predicted cognitive impairment in our cohort. Consequently, the score could help identify patients at risk of poor overall clinical outcome and poor cognitive outcome and thus aid in selecting not only optimized initial therapies for these patients but also cognitive rehabilitation to improve long-term outcomes.

Figure 1. Validation of the NEOS Score in Children

(A) Association of the original NEOS score with mRS-based outcomes (good outcome mRS ≤ 2, poor outcome mRS ≥ 3) at 1 year after diagnosis. Box plots represent IQR, solid lines mark the median, whiskers display range (upper/lower quartile ± 1.5*IQR), and circles show outliers. “n” indicates the number of subjects included at each time point. (B) Predictability analysis of mRS-based clinical outcomes by the NEOS score with binomial generalized linear models (GLMs). Line plots show association of the original (red curve) and adapted (blue curve) NEOS scores with poor clinical outcome (mRS ≥ 3) at 1 year after diagnosis. Solid lines represent best fit and shadows indicate confidence intervals. tick marks on the upper and lower x axes indicate the number of subjects (also written next to every graph) with each score with a good or poor mRS-based clinical outcome. A small random jitter was added to spread ticks around the discrete NEOS score values, to discriminate single data points. The p values were adjusted for multiple testing. NEOS 0: n = 5, NEOS 1: n = 12, NEOS 2: n = 16, NEOS 3: n = 9, NEOS 4: n = 4, NEOS 5: n = 0. For further results (comparison of original and adapted NEOS score, analysis over time), see eFigure 2 (links.lww.com/NXI/A812) and eTable 2 (links.lww.com/NXI/A813).

Figure 2. Association of the NEOS Score With Cognitive Outcome

(A) Bar plots display the frequency of pathologic test scores from neuropsychologic assessments early (2 months) in follow-up and 1 year after diagnosis. The categories intelligence, memory, language, executive function, visuospatial function, and behavior contain raw scores of various test batteries grouped into a binary measure—normal (gray bars) vs pathologic (black bars). White bars (N/A) indicate cases with incomplete data. (B) Predictability analysis using binomial generalized linear models (GLMs) with a now cognition-based outcome reference. After assigning cognitive test scores instead of mRS values to each patient's NEOS score, models revealed associations of a poor outcome (mRS ≥ 3) with deficits in executive function and in memory at 1 year after diagnosis, here shown for the adapted NEOS score. No associations were found with intelligence, behavior, language, and visuospatial function. Solid lines represent best fit and shadows indicate confidence intervals. Tick marks on the upper and lower X axes indicate the number of subjects included, also written next to each line plot. A small random jitter was added to spread ticks around the discrete NEOS score values, to discriminate single data points. NEOS 0: n = 5, NEOS 1: n = 12, NEOS 2: n = 16, NEOS 3: n = 9, NEOS 4: n = 4, NEOS 5: n = 0. For further results (original score, analysis over time), see eTable 5 (links.lww.com/NXI/A813).