To explore the effects of clinical heterogeneity on classification performance, we selected subgroups with particular demographic and clinical characteristics: medication use, OCD severity, age of onset (AO) and duration of illness. Classifications performed were HC vs. low (YBOCS <= 24; mild-moderate37) and high severity (YBOCS > 24; moderate-severe) OCD; HC vs. early (<18 yrs) and late AO (>= 18 yrs) OCD; HC vs. short (<=7 yrs) and long duration (>7 yrs) OCD; and HC vs. unmedicated and medicated OCD. For disease duration and severity, median splits were used to define groups; the 18 year threshold for AO was chosen in line with prior ENIGMA-OCD mega-analyses6,7. Finally, as particular clinical variables can co-occur, we performed a post-hoc sensitivity analysis to investigate the effects of potential clinical covariance for results with AUC ≥ 0.8. First, correlations between all clinical features were computed using point-biserial correlations between dichotomous and continuous variables, phi correlation for dichotomous variables and Pearson correlation for continuous variables. Only those clinical features that were significantly correlated (Bonferroni-corrected) were investigated further by rerunning previously described classifications, but now using samples
