Machine Learning to Improve the Access to the Clinical Pathway for Children with Specific Learning Disabilities.

Area of Science:

• Developmental Psychology
• Educational Psychology
• Machine Learning in Healthcare

Background:

• Overwhelmed healthcare systems hinder timely diagnosis of specific learning disabilities (SLDs).
• Teachers require improved school-based screening tools due to limited clinical preparation.
• Early identification of SLDs is crucial for effective intervention and academic support.

Purpose of the Study:

• To develop methods for identifying children's specific learning disability profiles.
• To create a system for selecting children who require clinical consultation.
• To enhance understanding of factors associated with initiating the clinical pathway for SLDs.

Main Methods:

• Analyzed data from 364 children referred for clinical consultation.
• Utilized a 96-item teacher-completed screening questionnaire.
• Applied item response theory for severity scores, K-means clustering for profile segmentation, and machine learning models (SVC, Naive Bayes) with Shapley values for classification and interpretation.

Main Results:

• Cluster analysis identified two distinct children's profiles based on learning difficulty severity.
• Machine learning models achieved high classification performance (AUC 0.96 with SVC, 0.69 with Naive Bayes), indicating effectiveness in differentiating profiles.
• Shapley values revealed that classification models identified severity and complex interactions among learning difficulties.

Conclusions:

• The proposed methods offer a step forward in managing SLDs from an early, preclinical stage.
• The findings support the consideration of children with less apparent difficulties, who may still have undiagnosed SLDs.
• This work provides valuable insights for referring children with specific learning disabilities to clinical settings.