Enhancing the Effectiveness of Attention-Deficit/Hyperactivity Disorder Screening Using the SNAP-IV: A Deep Learning...

Area of Science:

• Neurodevelopmental disorders
• Psychometrics
• Machine learning applications in healthcare

Background:

• Attention-deficit/hyperactivity disorder (ADHD) is a prevalent neurodevelopmental condition necessitating early identification for improved outcomes.
• The Swanson, Nolan, and Pelham Rating Scale (SNAP-IV) is a comprehensive tool for assessing ADHD symptoms.
• The full SNAP-IV's length can impede its use in large-scale screening initiatives.

Purpose of the Study:

• To refine the 18 core ADHD items of the SNAP-IV using a machine learning framework.
• To identify the most predictive items for ADHD screening while maintaining balanced symptom representation.
• To develop a shorter, efficient, and psychometrically sound SNAP-IV scale for ADHD assessment.

Main Methods:

• A multi-algorithm machine-learning framework was employed to analyze SNAP-IV data.
• Cross-model consensus ranking was used to identify the most predictive ADHD items.
• Data from two Taiwanese cohorts (Taiwan National Epidemiological Study of Child Mental Disorders and National Taiwan University Hospital) were utilized.
• Ten classifiers were optimized for screening, prioritizing sensitivity.

Main Results:

• Reduced subsets of 4 parent-reported and 6 teacher-reported items demonstrated robust predictive performance for ADHD across cohorts.
• Confirmatory factor analysis supported the structural validity of the shortened scales, aligning with Inattention and Hyperactivity-Impulsivity factors.
• High latent reliability (McDonald's omega) was observed for the shortened scales.

Conclusions:

• A machine learning-derived, construct-balanced short form of the SNAP-IV offers an efficient screening tool for ADHD.
• The shortened scale maintains psychometric soundness and predictive accuracy.
• This refined tool can facilitate large-scale ADHD screening and early identification.