Predicting Developmental Language Disorder in Bilinguals From Multiple Sources of Evidence: A Decision Tree Approach

PURPOSE: Speech-language pathologists consider multiple pieces of assessment data when making diagnostic decisions for bilingual children. The current study explored the converging evidence framework by applying a decision tree model, a user-friendly prediction tool, to identify bilingual children with developmental language disorder (DLD) using multiple assessment data. METHOD: One hundred eighty-six Spanish-English bilingual children in kindergarten completed a bilingual language assessment battery, consisting of parents' and teachers' questionnaires, standardized language assessment, narrative sample analysis, and dynamic assessment. Decision tree models were constructed to identify DLD. RESULTS: The decision tree models using (a) bilingual assessment data and (b) bilingual assessment data with parents' and teachers' data as fixed root achieved acceptable sensitivity and specificity. Decision tree models using English-only assessment achieved minimum sensitivity but not minimum specificity. Those using English-only assessment with parents' and teachers' data as fixed root did not achieve acceptable sensitivity or specificity. CONCLUSIONS: This study suggests that decision tree models can help speech-language pathologists identify bilingual children with DLD using different evidence, aligning with the converging evidence framework. The feasibility of English-only assessment and the clinical use of decision tree models are discussed. SUPPLEMENTAL MATERIAL: https://doi.org/10.23641/asha.34024689.