To test and compare the capability of 3 multimorbidity-based models to predict outcomes in early childhood among infants born with extremely low birth weight (<1000 g, ELBW).

Participants included 8332 surviving ELBW infants born 2010-2020 in North America who contributed follow-up data at 24-months corrected age to the Vermont Oxford Network. Neonatal morbidities included bronchopulmonary dysplasia, grade 3-4 intraventricular hemorrhage, periventricular leukomalacia, stage 3-4 retinopathy of prematurity, late infection, necrotizing enterocolitis, and spontaneous intestinal perforation. Outcomes included: 1) developmental delay (Bayley score <70 in ≥1 domain), 2) rehospitalization, and 3) therapeutic service use. We compared 3 gestational age-adjusted risk models with the following predictors: 1) morbidity count, 2) count of 3 morbidities (bronchopulmonary dysplasia, intraventricular hemorrhage, and retinopathy of prematurity), and 3) multimorbidity-based latent classes.

Thirty five percent of the study sample had ≥2 neonatal morbidities. Most (64%) received ≥2 therapeutic services, 36% were re-hospitalized, and 19% had developmental delay at 24-months. Morbidity counts and multimorbidity-based latent classes were associated with increased risk for poor 24-month outcomes compared with no morbidity. However, the predictive ability of all 3 models was modest (area under the receiver operating curve = 0.66).

Neonatal multimorbidity is common among ELBW infants and associated with later health and developmental outcomes. However, diagnosis-based multimorbidity risk models have poor prognostic ability. More robust characterization of multimorbidity symptom severity, physiologic impact, and environmental correlates may improve the clinical utility of future risk models.