Date of Award
Fall 2013
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Speech, Language, and Hearing Sciences
First Advisor
Laurence B. Leonard
Committee Chair
Laurence B. Leonard
Committee Member 1
Alexander Francis
Committee Member 2
Anu Subramanian
Committee Member 3
Lisa Goffman
Committee Member 4
George Hollich
Abstract
The etiology of specific language impairment (SLI) is multifactorial. Research has shown that genetic, environmental, and developmental factors may influence the course of its development. Because many of these factors are present even before a child is born, it is possible that a child's risk of developing the disorder can be identified long before grammatical deficits are observed. The goal of this study was to develop and validate a screening tool to discriminate between children with SLI and typically developing (TD) children using risk factor information including gender, family history of communication or reading disorders, socioeconomic status, maternal and paternal education level, birth order, premature birth, and birth weight. Participants included 211 children between the ages of 4;0 and 7;0 years. Two diagnostic classification schemes were used to examine the effects of clinical sample heterogeneity on risk factor model accuracy. In the first scheme, children were classified as SLI or TD based on their performance on a standardized expressive language test (SPELT classification). In the second scheme, children were classified based on their combined performance on both the standardized test and a measure of tense and agreement morpheme use (SPELT-FVM classification), thereby limiting the clinical sample to children with specific deficits in grammatical morphology. Step-wise logistic regression and five-fold cross-validation were used to derive risk factor models, probability equations, and cut-off scores for the two classification groups. Sensitivity, specificity, likelihood ratios, and diagnostic odds ratios were calculated for each cut-off value. Family history and maternal education arose as highly significant variables in both models. Gender was also included in the SPELT-FVM classification model. By design, the sensitivity of these models was high, but their specificity levels were inadequate for clinical use. A behavioral measure, late word combiner status at 24 months, was added to both models resulting in a substantial increase in diagnostic accuracy. With the behavioral measure, the SPELT-FVM classification model yielded a sensitivity of 91%, a specificity of 72%, a positive likelihood ratio of 3.24, and a negative likelihood ratio of 0.12. Based on diagnostic odds ratio values, this model was more accurate than all previous risk factor models and more accurate than current early identification screening methods. Although further verification of this model is needed, it may become an efficient and effective tool for identifying toddlers at risk for developing SLI, thereby leading to earlier diagnoses and better-targeted early intervention efforts.
Recommended Citation
Rudolph, Johanna Maria, "Predicting Language Impairment Status: A Risk Factor Model" (2013). Open Access Dissertations. 71.
https://docs.lib.purdue.edu/open_access_dissertations/71