Cross Validation Assessment of Clinical Performance Characteristics of a Test Table for the Risk of Maternal Autoantibody-Related Autism in High-Risk Populations

Background/Objectives: Specific combinations of maternal autoantibodies targeting fetal brain proteins are associated with Maternal Autoantibody-Related Autism (MARA), a biologically defined subset of autism. We previously translated a multi-ELISA assay to detect MARA-associated autoantibodies and identify mothers with transcendent levels of specific autoantibody combinations that exceed that of controls and are highly predictive of autism in their offspring. Our objective here was to derive a rule-based classifier, a test table of autoantibody pairwise combinations, each with pair-specific thresholds, and to validate its clinical performance characteristics by cross-validation. Methods: Plasma samples from mothers of 482 autism and 425 typically developing children were analyzed in a clinical lab for previously identified predictive combinations of autoantibodies. Stratified 20-fold cross-validation in all 907 participants assessed test table stability and predictive performance. The most stable test table was retrained on the full cohort to establish optimal thresholds and determine performance. Permutation testing assessed performance relative to chance. Results: Cross-validated performance showed a mean specificity across folds of 99.8%, an imputed mean PPV of 97.5%, a detection rate of 4.4%, and a positive likelihood ratio (LR+) of 18.5. Applied in-sample to the full cohort, the final test table detected 6.2% of cases with no false positives, giving an upper-range LR+ estimate of 26.5. Conclusion: The high specificity, PPV, and positive likelihood ratio of this assay support its potential clinical implementation for identifying a subset of mothers at very high risk of having a child with autism, and as a companion diagnostic for interventions targeting the MARA subtype of autism.