Reconceptualizing developmental language disorder as a spectrum disorder: issues and evidence

Abstract

Background: There is considerable variability in the presentation of developmental language disorder (DLD). Disagreement amongst professionals about how to characterize and interpret the variability complicates both the research on understanding the nature of DLD and the best clinical framework for diagnosing and treating children with DLD. We describe and statistically examine three primary possible models for characterizing the variability in presentation in DLD: predictable subtypes; individual differences; and continuum/spectrum.

Aims: To test these three models of DLD in a population-based sample using two distinct types of cluster analyses.

Methods & procedures: This study included children with DLD (n = 505) from the US Epidemiological Study of Language Impairment database. All available language and cognitive measures were included. Two cluster methods were used: Ward's method and K-means. Optimal cluster sizes were selected using Bayesian information criteria (BIC). Bootstrapping and permutation methods were used to evaluate randomness of clustering.

Outcomes & results: Both clustering analyses yielded more than 10 clusters, and the clusters did not have spatial distinction: many of these clusters were not clinically interpretable. However, tests of random clustering revealed that the cluster solutions obtained did not arise from random aggregation.

Conclusions & implications: Non-random clustering coupled with a large number of non-interpretable subtypes provides empirical support for the continuum/spectrum and individual differences models. Although there was substantial support for the continuum/spectrum model and weaker support for the individual differences model, additional research testing these models should be completed. Based on these results, clinicians working with children with DLD should focus on creating treatment plans that address the severity of functioning rather than seeking to identify and treat distinct subtypes. Additional consideration should be given to reconceptualizing DLD as a spectrum condition.

Keywords: developmental language impairment; quantitative; specific language impairment.

Figure 1.

Flow chart summary of the analysis.

Figure 2.

Bayesian information criteria (BIC) curves for K-means clustering for the data set. BIC curves are for the original data, permutated data and 95% confidence interval envelope for the permutated curve.

Figure 3.

Ward’s hierarchical clustering results for the data set. (A) The dendrogram for the hierarchical, where the y-axis is height, which is the criterion value for a merge and represents the total distance accounted for by that merge. The black box on the dendrogram represents the identified minimum Bayesian information criteria (BIC) value. (B) Associated BIC curve for clusters 1–100 for the data set. The vertical line represents the identified minimum BIC value.

Figure 4.

Ward’s method cluster scatter plot (A) and K-means cluster scatterplot (B) for the developmental language disorder (DLD) data set. Circles represent clusters. [Colour figure can be viewed at wileyonlinelibrary.com]