Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neonatal Speech Auditory Brainstem Responses Predict Later Language Delay
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Shaoqi PAN1,2, Leyan Xie1,3, Fei Yuan1,4, Tak Fei Mau1, Hugh Simon Lam5, Tak Yeung Leung6, Nikolay Novitskiy1,4, Ting Fan Leung5,7, Patrick C.M. Wong1,4; 1Brain and Mind Institute, The Chinese University of Hong Kong, Hong Kong SAR, China, 2Department of Psychology, The Chinese University of Hong Kong, Hong Kong SAR, China, 3Department of Education, The Chinese University of Hong Kong, Hong Kong SAR, China, 4Department of Linguistics and Modern Languages, The Chinese University of Hong Kong, Hong Kong SAR, China, 5Department of Paediatrics, Prince of Wales Hospital, The Chinese University of Hong Kong, Hong Kong SAR, China, 6Department of Obstetrics and Gynaecology, The Chinese University of Hong Kong, Hong Kong SAR, China, 7Hong Kong Hub of Paediatric Excellence, The Chinese University of Hong Kong, Hong Kong SAR, China
Introduction: Early identification of risk for developmental language delay is essential for timely monitoring and intervention, yet formal diagnosis is typically not made until preschool age. Our previous studies suggest that infant speech auditory brainstem responses (speech-ABRs, also called Frequency-Following Responses) are associated with later language outcomes and can predict pre-school language delay. However, these studies have varied substantially in the age at which speech-ABRs were recorded. Because speech-ABR characteristics are age-dependent, this variability may confound the association between early neural responses and later language outcomes. Fortunately, newborns already show sensitivity to speech structure and language-relevant regularities, suggesting that neural speech-encoding mechanisms are active from the earliest stages of life. Moreover, for speech-ABR screening to provide prognostic information as early and efficiently as possible, it should ideally be implemented at birth, alongside standard click-based automated auditory brainstem response (AABR) screening. Therefore, we tried to record speech-ABRs from neonates here and asked whether these speech-ABRs can predict later language outcomes. Methods: We recorded EEG speech-ABRs from 109 neonates (mean: 1.56 (0–8) days) while they listened to three lexical-tone stimuli: Cantonese Tone 2, Tone 4, and Putonghua Tone 3. Language outcomes were assessed at follow-up using the Bayley-III language subscale (mean:17.33 (9.40–31.77) months). Infants scoring at or below the 16th percentile were classified as having language delay. We first examined group differences in neonatal speech-ABR features. To characterize the representational structure of neural responses, we applied CEBRA (Consistent EmBeddings of high-dimensional Recordings using Auxiliary variables), a deep-learning embedding method, to derive low-dimensional neural manifolds of tone encoding. We then constructed a speech-ABR index using a polygenic-style predictive pipeline and tested whether this index predicted later language delay after controlling for birth weight, sex, and gestational age. Finally, we evaluated the predictive performance of neonatal speech-ABR features using several machine-learning classifiers. Results: Neonatal speech-ABR features differed between infants with typical later language outcomes and those with later language delay. CEBRA embeddings revealed more organized neural representational structure in the typical-outcome group, with clearer separation among tone categories, whereas the delayed group showed weaker structure and greater overlap. Consistent with this pattern, decoding correlations derived from the CEBRA embeddings were significantly higher in the typical-outcome group than in the delayed group (r: 0.209 vs. 0.079, p < .001). The speech-ABR index significantly predicted later language delay after covariate adjustment, OR = 3.26, 95% CI [1.81, 5.89], p < .001. Machine-learning classifiers also predicted later language delay above chance from neonatal speech-ABR features, with the best-performing model achieving an AUC of 0.805. Conclusions: Speech-ABRs recorded within the first days of life can predict later language delay. These findings suggest that clinically useful information about future language risk may already be present in neonatal neural speech-encoding responses. Integrating speech-ABR measures with standard newborn hearing screening could provide an early and efficient approach for obtaining both hearing- and speech-related prognostic information, thereby helping to identify infants who may benefit from enhanced monitoring and pre-emptive language support.
Topic Areas: Language Development/Acquisition, Disorders: Developmental