EEG-Derived Transdiagnostic Subtypes Map Onto Distinct Brain-Behavior Dimensions
J. Tabbal, A. Ebadi, G. Robert, A. Mheich, B. Rodríguez-Herreros, N. Chabane, A. Lefebvre, A. Iftimovici, S. Allouch, M. HassanIntroduction
Psychiatric diagnoses traditionally rely on symptom-based classifications, which, despite providing a shared clinical and research framework, show weak alignment with underlying neurobiology. In response, data-driven, biologically informed approaches have emerged, yet most still depend on group-level comparisons that assume homogeneity and obscure individual variability. Normative modeling (NM) offers a promising alternative, quantifying individual deviations from population-derived normative distributions and enabling the detection of neurophysiological alterations at the single-subject level (Marquand et al. Biol Psychiatry 2016; 80:552-561).
Objectives
This study aims to present a framework for clustering based on deviations from normative electrophysiological trajectories to identify transdiagnostic subtypes of neurodevelopmental and psychiatric disorders. The approach leverages a large, harmonized HD-EEG dataset to uncover biologically grounded brain–behavior profiles that transcend traditional diagnostic boundaries.
Methods
We analyzed a multisite EEG dataset comprising 1,701 participants, including healthy controls, and individuals with attention-deficit/hyperactivity disorder, autism spectrum disorder, anxiety, learning disorders, and their comorbidities. Normative models were built for a comprehensive set of EEG features (n = 1957), and individualized deviation scores were computed. These multivariate deviation profiles were then clustered using similarity network fusion to identify subgroups with distinct electrophysiological patterns. The resulting clusters were subsequently characterized in terms of diagnostic composition and behavioral profiles.
Results
We identified three neurophysiological subtypes that did not align neatly with traditional diagnostic categories (Fig. 1). Electrophysiologically, C2 (predominantly HC) showed deviation scores close to normative EEG trajectories, whereas C1 (internalizing symptoms) and C3 (externalizing diagnoses) exhibited more pronounced and often opposing deviations across key features. Behavioral comparisons revealed significant differences across behavioral domains (Fig. 2). C1 showed deficits in executive function and elevated emotional reactivity; C3 had higher hyperactivity-related scores; C2 showed the lowest scores across most domains, reflecting a broadly normative profile. Finally, EEG features significantly predicted cluster-specific behavioral traits (Fig. 3).
Image 1: Long description.