Using the PDD Behavior Inventory as a Level 2 Screener: A Classification and Regression Trees Analysis

Area of Science:

• Neuroscience
• Developmental Psychology
• Data Science

Background:

• Accurate differential diagnosis between Autism Spectrum Disorder (ASD) and other neurodevelopmental disorders is challenging.
• Existing diagnostic tools may have limitations in discriminating between these conditions.

Purpose of the Study:

• To enhance diagnostic accuracy for Autism Spectrum Disorder (ASD) using a data mining approach.
• To identify distinct subtypes within ASD and non-ASD populations.

Main Methods:

• Applied Classification and Regression Trees (CART), a data mining technique, to a large, multi-site dataset of PDD Behavior Inventory (PDDBI) forms.
• Utilized data from children with and without ASD, including parent and teacher reports.
• Validated the model on an independent dataset and compared results with ADOS classifications.

Main Results:

• Achieved over 80% discrimination accuracy between ASD and similar disorders.
• Demonstrated generalization of accuracy across age groups, sites, and to an independent validation set.
• Identified three ASD subtypes (minimally verbal, verbal, atypical) and two non-ASD subtypes (social pragmatic problems, good social skills).

Conclusions:

• CART analysis significantly improves the accuracy of differentiating ASD from other neurodevelopmental disorders.
• The identified subtypes offer a more nuanced understanding of ASD heterogeneity and non-ASD developmental profiles.
• Parental PDDBI reports were more effective, with cross-informant agreement increasing diagnostic sensitivity.