A Machine-Based Prediction Model of ADHD Using CPT Data

Area of Science:

• Neuroscience
• Computational Psychiatry
• Developmental Psychology

Background:

• The Continuous Performance Test (CPT) is widely used for diagnosing attention-deficit/hyperactivity disorder (ADHD), but its diagnostic accuracy is debated.
• Traditional analysis of CPT data has limitations in specificity, sensitivity, and ecological validity.
• Machine learning (ML) offers potential for improved analysis of complex behavioral data.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting ADHD using CPT indices.
• To compare the performance of the ML model against traditional clinical data benchmarks.
• To identify key CPT indices and control variables crucial for ADHD discrimination.

Main Methods:

• A retrospective factorial fitting and bootstrap technique were used to train, cross-validate, and test ML models.
• CPT performance data from 458 children (6-12 years) were analyzed, including 213 with ADHD and 245 typically developing controls.
• The MOXO-CPT, incorporating visual and auditory distractors, was utilized. Models incorporated CPT total scores and control variables (age, gender, day/time of day).

Main Results:

• The ML model achieved significantly higher accuracy (87%) compared to benchmark models using only clinical data.
• The ML model demonstrated high sensitivity (89%) and specificity (84%) in classifying children with ADHD.
• CPT total scores and control variables (age, gender, DoW, ToD) were the most salient predictors for ADHD discrimination.

Conclusions:

• Machine learning models can accurately classify children with ADHD based on CPT performance data.
• The proposed ML model shows promise for enhancing and potentially complementing traditional behavioral assessments for ADHD.
• This approach may serve as a valuable supportive tool in the comprehensive evaluation of ADHD.