Early prediction of severe autoimmune encephalitis: development and validation of a model incorporating readily available lactate dehydrogenase.

Background: Autoimmune encephalitis (AE) is a severe neuroinflammatory disease with a substantial risk of progression to critical illness requiring intensive care. Early identification of patients at high risk of severe disease is essential but remains challenging because of heterogeneous presentations and the lack of objective, readily available prognostic tools. Lactate dehydrogenase (LDH), a ubiquitous enzyme associated with cellular injury and immune activation, has been linked to disease severity in systemic autoimmune disorders; however, its prognostic value in AE remains unexplored. This study aimed to develop a clinical prediction model for severe AE and to evaluate serum LDH as a core biomarker for prognosis and differential diagnosis. Methods: In this multicenter retrospective study, 299 adult patients with AE were analyzed. Severe AE was defined by intensive care unit admission or the presence of major disability (modified Rankin Scale ≥ 3). Independent predictors were identified using multivariable logistic regression and incorporated into a nomogram. The model underwent both internal and external validation. Serum LDH was additionally evaluated as a standalone biomarker by comparison with viral encephalitis (VE) controls (n = 243) and across AE antibody subtypes. Results: Elevated serum LDH, impaired consciousness at admission, and prodromal infection were identified as independent predictors of severe AE. The resulting nomogram demonstrated excellent discriminatory performance (area under the curve [AUC] 0.947 in the training cohort, 0.882 in the validation cohort) and good calibration. Serum LDH remained a robust predictor in the multivariable model. As a standalone predictor, LDH achieved an AUC of 0.887 for severe AE; an optimal cutoff value of 215 U/L yielded a sensitivity of 83.3% and a specificity of 84.1%. Notably, LDH levels were significantly higher in AE than in VE. Furthermore, elevated LDH demonstrated a significant positive correlation with the risk of severe disease across key AE subtypes, including anti-N-methyl-D-aspartate receptor, seronegative, anti-LGI1, anti-GAD65, and anti-GFAP encephalitis. Conclusion: This study presents a validated and readily applicable nomogram for early risk stratification in AE. Serum LDH emerges as a robust, accessible biomarker that supports both prognostic assessment and differential diagnosis, providing a simple objective threshold (215 U/L) to inform timely clinical decision-making.