Abstract
BACKGROUND: Considering the central role of the Diagnostic and Statistical Manual of Mental Disorders (DSM) in psychiatric classification, multiple studies have examined how it describes Attention-Deficit/Hyperactivity Disorder (ADHD) - one of the most common psychiatric diagnoses. However, despite analyzing the same DSM texts, these studies yielded conflicting conclusions, likely influenced by the subjectivity of qualitative research and the challenge of systematically tracking subtle changes in large textual corpora. This study addresses these limitations by providing the first systematic, Artificial Intelligence (AI)-assisted analysis of all ADHD-related texts across six DSM editions (DSM-III to DSM-5-TR). METHODS: The analysis employed two AI models (GPT-4o and Claude 3.5 Sonnet) and followed five structured steps: (A) preliminary human review, (B) AI-assisted comparative analysis, (C) refinement through AI self-prompting to detect subtle linguistic changes, such as tone and diagnostic uncertainties, (D) thematic synthesis by each model, and (E) cross-model validation. Strict adherence to DSM texts ensured all findings were grounded in verifiable textual evidence. RESULTS: The analysis identified six overarching trends (1): a shift from a behavioral disorder to a neurodevelopmental framework (2), expansion to a lifespan condition across genders, (3) a broadening concept of impairment, (4) increasing diagnostic flexibility, (5) an expanding scope of comorbidities and differential diagnoses, and (6) growing acknowledgment of cultural and contextual influences. CONCLUSIONS: The six overarching shifts alongside the detailed systematic analysis results (Supplementary Materials) provide a transparent and replicable reference point for how ADHD has been described and classified in the DSM over four decades. Additionally, the innovative methodology can improve reliability of future research into complex psychiatric discourse.