Where to start?

Do you ever wonder how to decide if a bilingual child has a developmental language disorder (DLD)? We have a new paper here where we outline one approach we developed to identify DLD in bilinguals when there isn’t a gold standard.

Usually, when you develop a new test or approach you compare how well it works against a gold standard. Sensitivity and specificity percentages are how well the new test classifies compared to the gold standard. If the new test is at least 80% accurate in its classification (both sensitivity and specificity), then we say it’s a good test. But, what do we do when there is no gold standard as is often the case when we are testing bilingual kids? Can we use this approach for different language pairs? WHERE DO WE START?

1. We started with expert clinicians. Clinicians who were Spanish-English bilinguals, who had several years of clinical experience, who had worked a lot with Spanish-English bilingual kids.
2. We looked at the literature. How does English DLD present? How does Spanish DLD present? Where would we see this?
3. Our expert clinicians reviewed test, interview and language sample data for 167 kids and made judgements about each of their languages in three domains: narratives, grammar, & semantics.

We made a form, where we asked clinicians to review each case, and to make notes. The included the three domains across the top, and the sources of information (here language samples) along the side.

Clinicians made their notes as they went through the files. We asked them to look at patterns of performance rather than at test scores (see example):

Then, we asked them to provide a rating for each language x domain, and an overall rating, based on their observations of children’s patterns of performance and their knowledge of ESL.

Every 10 cases, the three clinicians discussed the case together and their reasons for their ratings. They re-rated those cases after the discussion. Our findings are that there was high agreement overall in the domain ratings and in the decisions about who had impairment. When there were disagreements, they were usually a 1-point difference. What’s interesting is that even when they paid attention to different things, they were able to make judgements that were consistent with each other.

We then went back to look at the test scores of the kids that were judged to have and not have DLD. And we found large, significant differences on almost all measures of interest. We propose that this might be a good way to structure the data we collect when conducting bilingual assessment. We can then systematically rate each domain to then compare across languages and come up with a rating. We hope to expand this work, and we hope to see what kinds of measures and observations are informative for clinical decision making.