OBJECTIVE Although preterm birth is among the best predictors of language delay, it is not precise enough to make child-level prediction that will enable the prescription of the highly effective early intervention (EI). This study aims to develop and validate predictive models of language delay using neural data collected from as early as infancy to forecast language delay as to indicate EI before preschool years. METHOD Electroencephalography (EEG) neural speech encoding (ie, “speech auditory brainstem response ABR”) was recorded from 423 Chinese-learning infants between 1 and 24 months, and language outcomes were collected from the same children between 7 and 32 months in this cohort study. Data were collected from 2016 to 2024, with an analysis cutoff on October 3, 2024. Early-latency and long-latency EEG responses to 3 speech stimuli (2 native, 1 non-native) were collected. Model outcome was the language subscale of the Bayley Scales of Infant and Toddler Development, third edition. RESULTS Random forest was used to classify children into binary groups based on outcome measure: below/at the 16th percentile vs above. Different predictive models were constructed and compared, including those with and without EEG and clinical measures. Models with non-neural measures (eg, gestational age and birth weight) predicted language outcomes above chance. Models with EEG measures alone outperformed any non-neural models, achieving sensitivity and area under the receiver operating characteristic curve (AUC) well above 90% for the best models. When EEG models were externally validated, sensitivity and AUC remained above 80% and 90%, respectively. CONCLUSION Speech ABR can be a novel screening tool for language delay, allowing families of screened children to adopt EI preemptively for enhanced language development.
Wong et al. (2026) studied this question.