A Deep Learning Approach for Missing Data Imputation of Rating Scales Assessing Attention-Deficit Hyperactivity...

Area of Science:

• Neuroscience
• Psychiatry
• Data Science

Background:

• Missing data is a significant challenge in attention-deficit/hyperactivity disorder (ADHD) behavioral studies.
• Accurate data is crucial for diagnosing and understanding ADHD.
• Existing methods for handling missing data may introduce bias.

Purpose of the Study:

• To apply a deep learning method for imputing missing data in ADHD rating scales.
• To evaluate the effectiveness of the imputed dataset in distinguishing youths with ADHD from typically developing (TD) youths.
• To assess whether deep learning imputation introduces bias compared to complete datasets.

Main Methods:

• Utilized a deep learning model trained on a complete dataset of 1220 youths (799 with ADHD, 421 TD) from Northern Taiwan.
• Imputed missing values in ADHD rating scales and generated an imputation order based on accuracy.
• Employed support vector machine (SVM) to classify ADHD vs. TD groups using both the imputed and original datasets.

Main Results:

• The deep learning imputed dataset achieved 89% accuracy in classifying ADHD vs. TD groups.
• This accuracy was comparable to the 89% accuracy obtained using the original, complete reference dataset.
• High discriminatory accuracy was observed for oppositional behaviors (teachers) and hyperactivity/impulsivity (parents and teachers).

Conclusions:

• Deep learning provides a viable solution for imputing missing data in ADHD behavioral studies.
• The imputation method does not appear to introduce bias, maintaining classification accuracy.
• Specific behavioral symptoms like oppositional behavior and hyperactivity/impulsivity are key discriminators for ADHD.