Abstract
Attention-Deficit/Hyperactivity Disorder (ADHD) diagnosis has historically been shaped by symptom presentations more common in males, leading to under-identification in females and limited recognition of broader symptom domains. To address these gaps, this study evaluates the Neuro-Emotional Affect Regulation Dysregulation Scale (NEARD), a newly developed ADHD screening tool designed to capture both traditional and emerging symptom domains with sensitivity to gender differences. The study aimed to evaluate the scale’s construct and criterion validity while assessing potential gender bias. Data were analysed using exploratory and confirmatory factor analyses to examine latent structure, and machine learning models were employed to evaluate predictive validity against clinical ADHD diagnoses. The scale showed promise for improved screening and preliminary evidence for similar predictive performance across genders within a small-scale clinical sample (N = 142). Results indicated satisfactory construct validity with a four-factor solution (RMSEA = .062; SRMR = .058; CFI = .962; TLI = .959; GFI = .988) and high internal reliability (Factor CR range [0.826, 0.927]), but raised some concerns regarding factor discriminant validity, indicating a strong general factor. The Ensemble (tuned logistic regression + XGBoost) machine learning model achieved predictive accuracy of 75.3% on a holdout sample (17.5% above chance classification) and displayed statistically insignificant differences between male and female sample performance.
1. Introduction
Given the high prevalence and adverse consequences of underdiagnosed and misdiagnosed ADHD, it is essential to examine the shortcomings of current screening and assessment practices employed for diagnostics. Current academic literature underpins persistent gender biases associated with diagnostic procedures, as well as the need for revision and extension of diagnostic criteria. Recent data show a reversal in male-to-female ADHD diagnosis ratios from childhood (higher in males) to adulthood (near parity), likely due to under-recognition of internalized symptoms in women. Women with ADHD are often diagnosed later, experiencing adverse impacts from delays, such as higher comorbidity rates.
Advances in Machine Learning and statistical methodology show promise in enhancing the process of psychometric scale development and validation. Analysis of current research reveals an emphasis on the biological underpinnings of ADHD in academic and professional literature and highlights the benefits of using machine learning in diagnostics and scale development. Additionally, a gap in the use of machine learning for investigating gender differences in scale development was noted, underlining the necessity for further research. This study examines the...