Integrating explainable AI with clinical features to enhance ADHD diagnostic understanding

Comprehensive Summary

Due to the vastness and challenges that come with accurately diagnosing Attention Deficit Hyperactivity Disorder (ADHD) in adults, over- and under- diagnosis remain common as providers rely on subjective clinical judgement, and machine learning (ML) tools may be beneficial if clinically adopted. Through 786 adult assessments from a UK specialist mental health service, model interpretability was examined using SHapley Additive exPlanations (SHAP), which were used with traditional exploratory analyses. This model proved 77% accuracy and revealed important interaction patterns, including depressive symptom severity heavily correlating with predictive contribution of ADHD symptom scales. Integrating multimodal clinical features with clear ML methods allows for the analysis of interpretable, clinically aligned insights into adult ADHD diagnosis.

Outcomes and Implications

The combination of the SHAP-EDA approach allows for identifying actionable thresholds, clarified differential feature contributions, and highlights the importance of comorbidity and demographic context in diagnostic evaluation. This allows for a patient-centered and data-driven approach to improving the consistency of accurate diagnosis in clinical practice. Future research may allow for multi-site validation and temporal analyses to assess generalizability and stability of feature influences over time.