Longitudinal transitions, remission predictors, symptom network structure, and classification accuracy in childhood ADHD: a four-year prospective study.

Abstract

Objective: To examine four-year longitudinal transitions in ADHD diagnostic status and DSM-IV subtypes, identify predictors of remission, characterize changes in symptom network structure, and evaluate the classification performance of network-derived indices.

Method: This secondary analysis included children with ADHD (n = 89) and matched controls (n = 82) from a prospective four-wave school-based cohort (2008-2011). Participants were assessed annually using the Turgay DSM-IV-Based Scale and structured clinical interviews. Multivariable logistic regression with bootstrap resampling identified predictors of remission. Regularized partial correlation networks were estimated at year 1 and year 4 and compared using the network comparison test. Receiver operating characteristic analyses evaluated the classification accuracy of network-derived indices.

Results: Among participants reassessed at year 4 (94.7% retention), 32.1% achieved remission, and 67.9% showed persistent ADHD. Greater early reduction in teacher-rated inattention and lower baseline inattention independently predicted remission. Symptom network structure remained highly stable over time, whereas global connectivity increased significantly (p = 0.024). An inattention-dominant composite AI index showed high apparent classification accuracy (AUC = 0.94), although external validation is required.

Conclusion: Early inattention trajectory predicts ADHD remission better than baseline severity. Stable but increasingly connected symptom networks and network-derived indices may improve clinical stratification pending independent replication.