ONE MODEL DOES NOT FIT ALL: HETEROGENEITY IN POLYGENIC PROFILES OF ADHD ACROSS FOUR DEVELOPMENTAL STAGES

Although diagnostic manuals define ADHD as a disorder with onset before age 12, many individuals are diagnosed later. This raises the question of whether late-onset ADHD reflects the same causes as early-onset ADHD. Using the iPSYCH cohort linked to Danish national registers, we examined the polygenic architectures of ADHD across preschool- (n = 3,297), childhood- (n = 10,260), adolescent- (n = 3,582), and adult-diagnosed (n = 7,098) groups, each compared with birth-year–matched controls at the corresponding developmental stage. For each group, we constructed prediction models incorporating 955 polygenic scores (PGSs). Within-group prediction yielded area under the curve (AUC) values ranging from 0.67 to 0.72. In contrast, cross–group prediction showed reduced performance, with AUC decreases of 10%, most pronounced when the model derived from preschool-diagnosed cases (AUC=0.71) was applied to adult-diagnosed (0.57) and adolescent-diagnosed (0.58) groups. To further characterize shared and distinct genetic characteristics across the four ADHD groups, we conducted Shapley Additive Explanations (SHAP) value–based feature ranking for each model, principal component analysis of SHAP value patterns, and conditional time-to-event analyses. Several PGS – including the ADHD-PGS – demonstrated similar effects across the four ADHD groups. However, preschool and childhood ADHD appeared to be mainly driven by genetic liability for behaviour-related traits (e.g., PGS for smoking and other risk-taking behaviour), whereas adolescent and adult ADHD showed a more complex polygenic architecture, with stronger and broader associations across PGSs. These findings suggest that ADHD may be conceptualized as two main subtypes (e.g., ADHD type I and ADHD type II) that share the same clinical diagnosis but differ in their underlying genetic architecture. If replicated, these results may warrant reconsideration of ADHD’s placement within current diagnostic classification systems.