Context-Specific Decoupling and Competing Phenotypes: Transdiagnostic Eye-Tracking Biomarkers of ASD and ADHD During Naturalistic Viewing in a Large Pediatric Cohort

Background: Autism Spectrum Disorder (ASD) and Attention-Deficit/Hyperactivity Disorder (ADHD) exhibit high clinical overlap, but categorical diagnostic boundaries obscure their shared, dynamic physiological vulnerabilities during real-world sensory processing. Methods: We analyzed multimodal eye-tracking synchrony in a large transdiagnostic pediatric cohort (N = 2,026) during naturalistic viewing of four distinct media paradigms. A novel 2D complex correlation framework captured gaze inter-subject correlation (ISC) magnitude and spatiotemporal phase divergence, while 1D pupil ISC measured autonomic arousal synchrony. Linear models evaluated dimensional (RDoC) and categorical (2x2 ANCOVA) diagnostic frameworks alongside rigorous medication and severity controls. Results: Dimensional models revealed a domain-general vulnerability: autistic traits independently predicted widespread reductions across gaze synchrony in all media contexts, and pupillary synchrony in narrative-driven contexts, whereas continuous ADHD traits showed minimal independent effects. In contrast, severe spatiotemporal misalignment (phase divergence) did not scale dimensionally but emerged strictly at clinical boundaries, reflecting highly idiosyncratic spatial locking in isolated ASD. Furthermore, categorical models demonstrated a robust, non-additive interaction: the clinical co-occurrence of ADHD paradoxically buffered against this severe spatiotemporal decoupling. Crucially, this protective phenotype was localized strictly to character-driven social narratives and remained highly significant after rigorously adjusting for daily stimulant medication, outlier instability, and baseline autism trait severity. Conclusions: These findings validate model-free physiological synchrony as a candidate transdiagnostic biomarker. Rather than compounding impairment, comorbid ASD and ADHD reflect competing, non-additive neurocognitive strategies that yield distinct, context-dependent visual phenotypes.