Combining genetic and behavioral predictors of 11-year language outcome

Background: Rapid population-level identification of language disorders could help provide care to young children to improve their outcomes. Two previous studies identified and replicated up to six parent-reported items that predicted 11-year language outcome with ≥71% sensitivity and specificity. Here, we assess whether including genetic propensity for toddlerhood vocabulary improves predictive accuracy. Method: The Early Language in Victoria Study (ELVS) recruited 1,910 8-month-olds in Melbourne in 2003-2004. The Longitudinal Study of Australian Children (LSAC) recruited 5,107 0-1-year-olds across Australia in 2004. Both collected parent-reported items at 2-3 years, a comparable 11-year language outcome: the Clinical Evaluation of Language Fundamentals (CELF-4) Core Language score or Recalling Sentences subtest, and biospecimens for genotyping. We derived polygenic scores capturing participants’ genetic propensity for parent-reported 24-38-month vocabulary. We calculated univariate associations with continuous language outcomes. We used ensemble method SuperLearner to estimate how accurately the parent-reported predictors and polygenic scores predict low 11-year language outcome (>1.5 standard deviations below the mean) in each cohort. Results: Language outcome was available for 839 ELVS and 1,441 LSAC participants. Polygenic scores accounted for little variance in continuous language outcomes (R-squared < 1.5%). Adding polygenic scores to the predictor sets increased accuracy of predicting language outcome by up to 7%, but inconsistently between analyses. Conclusions: Polygenic scores derived for toddlerhood vocabulary did not meaningfully predict late childhood language. Polygenic scores derived for later global language measures in larger samples may be better predictors. Presently, parent-reported measures or clinician observation appear best for predicting language outcome at this age.