Development and validation of the NEOS2 score for prediction of long-term outcomes and improvement after first-line immunotherapy in patients with anti-NMDAR encephalitis: an international cohort study

Background: Anti-N-methyl-d-aspartate receptor (anti-NMDAR) encephalitis is a severe disease that primarily affects young people and can improve with adequate treatment. We aimed to refine the anti-NMDAR Encephalitis One-year functional Status (NEOS) score by developing NEOS2, an updated model using readily available data at the time of diagnosis. We assessed the predictive value of the NEOS2-score for (1) improvement following first-line treatment, (2) functional outcome at one-year follow-up, and (3) resumption of school or work within three years. Methods: In this international (France, Germany, Japan, the Netherlands and Spain) cohort study in patients with a definite anti-NMDAR encephalitis diagnosis (according to the clinical criteria plus antibody testing in CSF), we performed logistic regression analyses to develop and validate multivariable models to predict -based upon variables available at diagnosis- short (ΔmRS two weeks after first-line treatment), middle (modified Rankin Scale [mRS] at one year), and long-term (return to school or work within three years) outcomes. We included clinical variables and biomarkers available at diagnosis. Findings: We included 702 patients (mean age 23 years, 95%-CI 2-69; 79% female, 21% male) diagnosed between the discovery of the disease in 2007 and 2022. Most patients (96%; 672/702) had received first-line immunotherapy, and 38% (233/615) showed improvement within two weeks. One year after diagnosis, 80% (517/644) had a favourable functional outcome (mRS≤2). At three years, 73% (203/278) had resumed work/school. In multivariable analysis, higher age (odds ratio [OR] 0·35, 95%-CI 0·29-0·43, p < 0·0001), treatment delay (OR 0·49, 95%-CI 0·41-0·58, p < 0·0001), movement disorders (OR 0·32, 95%-CI 0·24-0·41, p < 0·0001), ICU-requirement (OR 0·34, 95%-CI 0·26-0·44, p < 0·0001) and increased CSF leucocyte count (OR 0·65, 95%-CI 0·60-0·71, p < 0·0001) independently predicted poorer outcomes (NEOS2, accuracy AUC 80%, 95%-CI 75-86%). The same variables, excluding age, were relevant in predicting improvement following first-line immunotherapy (NEOS2-T AUC 81-84%, 95%-CI 77-86%). Return-to-work or -school served as a useful measure of longer-term outcomes, predicted with equal accuracy as one-year functional outcome (NEOS2-W AUC 80%, 95%-CI 75-85%). The NEOS2-score, applied as an ordinal measure, enabled nuanced predictions of outcome probabilities across the score spectrum, ranging from a high (80%; n = 20/25) likelihood of improving after first-line immunotherapy and achieving a good outcome (100%; n = 32/32) to a high risk of first-line treatment failure (97%; n = 77/79) and no return to school/work (94%; n = 15/16). Interpretation: The NEOS2-score, readily available at diagnosis and easy to apply, can identify patients with either a favourable or poor prognosis, and those who may benefit from early intensified treatment. The value of the NEOS2-score for guiding treatment decisions and as a stratification tool in studies on optimal treatment regimens, should be confirmed in further prospective studies. Funding: This study was funded by Dioraphte (charity; project 2001 0403).