Prediction of Language Development in Neonates Born at Less than 32 Weeks of Gestation

OBJECTIVE: To evaluate neonatal risk factors and the predictive value of prior assessments for language development at age 3 in toddlers born before 32 weeks of gestation. STUDY DESIGN: Data were acquired retrospectively from infants' hospitalizations (level 3 perinatal center, Medical University of Vienna) and annual assessments (Bayley Scales of Infant and Toddler Development). We included children born in 2012-2020 at < 32 weeks of gestation, with annual follow-ups, at least one German-speaking parent, and unimpaired hearing. Predictors of 3-year language development were analyzed using Gradient Boost Decision Trees and interpreted through SHapely Additive exPlanations (SHAP) values. RESULTS: The cohort comprised 476 children (202 [42.4%] females; mean [standard deviation] gestational age, 27.4 [2.0] weeks). Risk factors were multilingualism (mean|SHAP| = 3.93, P ≤ .001), lower maternal education (mean|SHAP| = 2.51, P ≤ .001), reduced birth weight (mean|SHAP| = 1.92, P = .003), shorter gestational age (mean|SHAP| = 1.64, P = .009), grade of intraventricular hemorrhage in the left hemisphere (mean|SHAP| = 1.07, P = .005), severe intraventricular hemorrhage (mean|SHAP| = 0.80, P = .011, surgical necrotizing enterocolitis (mean|SHAP| = 0.80, P = .018), and surgical retinopathy of prematurity (mean|SHAP| = 0.59, P = .045). CONCLUSIONS: The identification of SHAP values allowed us to evaluate the additive effect of neonatal and sociodemographic risk factors predicting language impairments in this cohort of children born preterm. Replication in other cohorts will be important, but SHAP-based analyses may be useful for tailored monitoring and implementation of early language support.