Search

Background: Anti-N-methyl-D-aspartate receptor (NMDAR) encephalitis, a recently reported autoimmune disorder, can be mistakenly diagnosed as a psychotic disorder, especially schizophrenia, as patients can present with prominent psychotic symptoms, in particular persecutory ideation, hallucinations and disturbed speech. In this study we used machine learning of the clinical data in a large cohort of persons with a positive psychosis history to ascertain whether we could predict NMDAR-positive cases, and which variables most accurately distinguished between NMDAR-positive and -negative cases. Methods: SHIP collected nationally representative data from 1825 individuals with a psychotic illness. Plasma samples were available for n=472. To investigate the prevalence of NMDAR autoantibodies a recombinant indirect immunofluorescence test was performed (EuroImmun AG, Lübeck, Germany), with NMDAR transfected human embryonic kidney (HEK) 293 cells quantified using NIS Elements software. NMDAR-positive cases were estimated. Gradient boosting machine learning (the data were randomly split: 60% for initial ascertainment and 40% for validation) was subsequently performed using the clinical data available: 120 variables in total across various domains of sociodemographic, medical history, psychiatric diagnosis and current psychiatric symptoms. Only the variables found to have significant (or near significant) association with being NMDAR-positive were used to develop rules for identifying cases. Results: There were 38 NMDAR-positive cases. They were more likely to be associated with a schizophrenia /schizoaffective and a depressive psychosis diagnosis, and less likely to be associated with a bipolar diagnosis, than antibody-negative cases. They were also more likely to be associated with a single episode with good recovery, and with anxiety symptoms and dizziness in the prior 12 months (which included light headedness, feeling faint and unsteady). For the present state symptoms, restricted affect was more likely to be present whereas poverty of speech was rare. Initial insomnia and a medical history that included epilepsy were not present for any of the NMDAR-positive cases. The machine learning algorithm was able to successfully classify 94% of cases to the correct antibody group. Discussion: In this significant Australian epidemiological cohort, we have identified key clinical features associated with anti-NMDAR encephalitis, including diagnosis, and symptoms and clinical course. The novel and insightful analyses afforded by using machine learning should be replicated in other samples to confirm the important clinical findings reported in the current work. [ABSTRACT FROM AUTHOR]