We conducted a retrospective study of consecutive patients referred for suspected AE (January 2017–May 2023). Final diagnosis was determined by expert consensus based on clinical features, exclusion of alternative diagnoses, and response to immunotherapy. APE², Graus criteria, and APE2SM scores were assigned retrospectively. The APE2SM score weights movement disorders and seizure presentations more heavily compared to the APE² score. Diagnostic performance was assessed using receiver operating characteristic analysis, contingency tables, and logistic regression to model probability of AE across score thresholds.

Eighty-five patients were included (56.5% female; mean age 53.8 years), of whom 59 (69.4%) had AE, including 34 antibody-positive cases. In exploratory analyses, APE2SM demonstrated the strongest overall diagnostic performance, with comparable discrimination to APE² but improved sensitivity and negative predictive value, particularly at clinically relevant thresholds and in antibody-negative patients. APE² outperformed Graus criteria in the full cohort (AUC 0.894 vs 0.804) and in antibody-negative patients (AUC 0.905 vs 0.850). An APE² cutoff ≥4 achieved balanced sensitivity and specificity near or above 80%, whereas no Graus threshold achieved similar balance.