Refining developmental coordination disorder subtyping with multivariate statistical methods | JoVE Visualize

Area of Science:

Neuroscience
Developmental Psychology
Machine Learning in Healthcare

Background:

Penalization and ensemble learning methods show promise for predictive performance with numerous clinical indicators.
Their efficacy in clinical studies with limited cases, particularly for Developmental Coordination Disorder (DCD), remains underexplored.
This study investigates Random Forests and Partial Least Squares Discriminant Analysis for DCD feature selection and patient similarity assessment.

Purpose of the Study:

To identify salient neuropsychological and visuo-motor impairments for characterizing Developmental Coordination Disorder (DCD) subtypes.
To assess the performance of Random Forests and Partial Least Squares Discriminant Analysis in a clinical DCD sample.
To evaluate subtyping consistency and predictive accuracy using cluster analysis.

Main Methods:

Utilized a comprehensive testing battery for neuropsychological and visuo-motor impairments in 63 children with DCD.
Employed Random Forests and Partial Least Squares Discriminant Analysis to rank 49 items by variable importance.
Performed cluster analysis for subtyping consistency, with fitness and accuracy evaluated on a validation sample.

Main Results:

Both classifiers successfully identified key impairments discriminating three DCD subtypes: ideomotor, visual-spatial/constructional, and mixed dyspraxia.
Key discriminating impairments included digital perception, gesture imitation, digital praxia, block construction, spatial structuring, and motor integration.
Achieved classification accuracy exceeding 90% with satisfactory clustering fitness.

Conclusions:

Random Forests and Partial Least Squares Discriminant Analysis effectively extract salient features from correlated predictors and assess individual proximity.
A concise set of approximately 15 neuro-visual, psychomotor, and psychological tests may suffice for sensitive and specific DCD diagnosis.
Isolated markers identified can refine future understanding and diagnosis of DCD.