Differential diagnosis and comparison of diagnostic algorithms in children and adolescents with autoimmune encephalitis in Spain: a prospective cohort study and retrospective analysis

Background: The usefulness of current diagnostic approaches in children with suspected autoimmune encephalitis is unknown. We aimed to assess the diagnosis of autoimmune encephalitis in clinical practice and to compare the performance of two international diagnostic algorithms (one intended for patients of any age [general], the other intended for paediatric patients), with particular emphasis on the evaluation of patients with probable antibody-negative autoimmune encephalitis because this diagnosis suggests that immunotherapy should be continued or escalated but is difficult to establish. Methods: We did a prospective cohort study that included all patients (<18 years of age) with suspected autoimmune encephalitis recruited at 40 hospitals in Spain whose physicians provided clinical information every 6 months for 2 years or more. Neural antibody testing to confirm diagnosis of antibody-positive autoimmune encephalitis was done at Institut d'Investigacions Biomèdiques August Pi i Sunyer-Hospital Clínic, Barcelona. Patients were classified according to the most probable diagnosis at last follow-up into four prespecified categories. We used multivariable logistic analysis to assess a potential association between immunotherapy and outcome in individuals with probable antibody-negative autoimmune encephalitis. We also did a retrospective analysis of agreement, assessed with the kappa index, between diagnoses made according to the general and paediatric diagnostic algorithms. Findings: Between June 1, 2013, and May 31, 2021, 729 children (mean age 7·1 years [SD 4·9]; 383 boys [53%], 346 girls [47%]) with suspected autoimmune encephalitis were recruited. After a median follow-up of 36 months (IQR 26–60), patients were classified according to their most probable diagnosis: definite autoimmune encephalitis or well defined inflammatory or autoimmune disorders (n=230 [32%]); CNS infections (n=112 [15%]); inflammatory CNS disorders of unknown cause (n=81 [11%], including three (4%) with a novel Klüver-Bucy-like syndrome; and non-inflammatory disorders (n=306 [42%]), which were predominantly epileptic or psychiatric disorders (177 [58%] of 306). Neural antibodies were detected in 150 (65%) of 230 patients who had definite autoimmune encephalitis; 127 (85%) of these 150 individuals had antibodies to the NMDA receptor or myelin oligodendrocyte glycoprotein (MOG). Agreement between algorithms was excellent (kappa index 0·99, 95% CI 0·97–1·00) for the diagnosis of children with antibody-positive autoimmune encephalitis, good (0·59, 0·54–0·65) for recommendations of empiric immunotherapy, and poor (0·29, 0·21–0·37) for the diagnosis of probable antibody-negative autoimmune encephalitis. Compared with the general algorithm, the paediatric algorithm included more patients in the probable antibody-negative autoimmune encephalitis category (173 vs 41). These patients included some of those who had a diagnosis of CNS inflammatory disorder of unknown cause at the last follow-up (80 of 81 with the paediatric algorithm vs 31 of 81 with the general algorithm), who might have benefitted from immunotherapy, and some of those diagnosed with a non-inflammatory disorder at the last follow-up (47 of 306 with the paediatric algorithm vs six of 306 with the general algorithm), who did not need immunotherapy. Interpretation: About a third of children with suspected autoimmune encephalitis eventually had confirmation of this diagnosis, or diagnosis of another well defined inflammatory disorder. Frequent mimics of autoimmune encephalitis were infectious, epileptic, and psychiatric disorders. Both algorithms performed well in the diagnosis of antibody-positive autoimmune encephalitis, but the paediatric algorithm under-recognised definite autoimmune encephalitis that can occur without autoantibodies and might have overdiagnosed patients with probable antibody-negative autoimmune encephalitis. By contrast, the general algorithm might have underdiagnosed patients with probable antibody-negative autoimmune encephalitis. Given that the diagnosis of probable antibody-negative autoimmune encephalitis has treatment implications, inaccuracies on this diagnostic category leads to overuse or underuse of immunotherapy. Funding: Instituto de Salud Carlos III, Fundació Clínic per la Recerca Biomèdica, The Edmond J Safra Foundation, and la Caixa Foundation. Translation: For the Spanish translation of the abstract see Supplementary Materials section.