Polygenic scores and risk prediction for delay (Dale et al., 2020)
journal contributionposted on 2020-04-28, 21:32 authored by Philip S. Dale, Sophie von Stumm, Saskia Selzam, Marianna E. Hayiou-Thomas
Purpose: The ability to identify children early in development who are at substantial risk for language/literacy difficulties would have great benefit both for the children and for the educational and therapeutic institutions that serve them. Information that is relatively easily available prior to the age of 3 years, such as late talking, family history of language/literacy difficulties, and socioeconomic status, have some but very limited predictive power. Here, we examine whether the inclusion of a DNA-based genome-wide polygenic score that has been shown to capture children’s genetic propensity for educational attainment (EA3) adds enough prediction to yield a clinically useful score.
Method: Data are longitudinal scores of 1,420 children from the Twins Early Development Study, who were assessed at ages 2 and 3 years on language and nonverbal ability and at 12 years of age on oral language, word decoding, and reading comprehension. Five risk factors were examined: expressive vocabulary, nonverbal ability (these two from parent report), family history, mothers’ education, and EA3. Analyses were conducted both for continuous and categorically defined measures of risk and outcome.
Results: Language and literacy abilities at 12 years of age were significantly but modestly predicted by the risk factors, with a small but significant added prediction from EA3. Indices of diagnostic validity for poor outcomes, such as sensitivity and area under the curve statistics, were poor in all cases.
Conclusions: We conclude that, at present, clinically useful prediction from toddlerhood remains an unattained goal.
Methods S1. Genotyping and creating polygenic scores.
Table S1. Means and standard errors for Figures 1 and 2.
Table S2. Correlations among the five individual risk factors.
Figure S1. Distribution of number of risk factors at ages 2 and 3.
Figure S2. Phi coefficients for individual risk factors predicting outcomes at 12.
Figure S3. ROC curve plotting sensitivity versus specificity for cumulative risk measure at 2 and low receptive language at 12.
Figure S4. ROC curve plotting sensitivity versus specificity for cumulative risk measure at 2 and low word decoding at 12.
Figure S5. ROC curve plotting sensitivity versus specificity for cumulative risk measure at 2 and low reading comprehension at 12.
Figure S6. ROC curve plotting sensitivity versus specificity for cumulative risk measure at 3 and low receptive language at 12.
Figure S7. ROC curve plotting sensitivity versus specificity for cumulative risk measure at 3 and low word decoding at 12.
Figure S8. ROC curve plotting sensitivity versus specificity for cumulative risk measure at 3 and low reading comprehension at 12.
Dale, P. S., von Stumm, S., Seltsam, S., & Hayious-Thomas, M. E. (2020). Does the inclusion of a genome-wide polygenic score improve early risk prediction for later language and literacy delay? Journal of Speech, Language, and Hearing Research. Advance online publication. https://doi.org/10.1044/2020_JSLHR-19-00161
The Twins Early Development Study (TEDS) is supported by a program grant to Robert Plomin from the U.K. Medical Research Council (MR/M021475/1), with additional support from the U.S. National Institutes of Health (AG046938). The research leading to these results has also received funding from the European Research Council under the European Union’s Seventh Framework Programme (FP7/2007-2013)/Grant Agreement 602768. S. V. is supported by a Jacobs Fellowship (2017–2019).
languageliteracydelaychildrendevelopmentgenomepolygenicscoreriskpredictionpredictidentifydifficultylate talkingfamilyhistorysocioeconomic statusDNAgeneticpropensityeducational attainmentEA3clinicalusefulTwins Early Development Studyoralworddecodingreadingcomprehensionrisk factorsexpressivevocabularynonverbalparentreportmothereducationdiagnosticvaliditysensitivitytoddlerageimpairmentinterventionservicesclinicianvariabilitylong termlongitudinalLanguageGene Expression (incl. Microarray and other genome-wide approaches)