Optimising enteral feeding prevents postnatal growth faltering in preterm infants: a retrospective cohort study.

Background: Despite advances in neonatal care, many preterm infants continue to experience growth faltering, with long-term consequences. We aimed to identify early, modifiable feeding-related risk factors and develop machine learning models to predict growth faltering at discharge. Methods: We retrospectively analysed 700 preterm infants (≤34 weeks gestational age (GA), ≤1800 g birth weight (BW)) admitted to a single level IV neonatal intensive care unit between 2014 and 2022, representing a relatively homogeneous population with standardised feeding practices. Over 100 demographic, clinical and nutritional variables, including parenteral and enteral intake during the first 28 days, were evaluated. Growth faltering was defined as a longitudinal decline in weight z-score of ≥1.2 SD from birth to 36 weeks postmenstrual age, rather than an absolute cross-sectional threshold. Supervised machine-learning models were trained using a 70% training and 30% testing split. Results: Growth faltering occurred in 17.7% (n=124). Affected infants had lower GA, lower BW and greater oxygen dependence at day of life 28 (all p<0.001). Although both enteral feeding volume and caloric intake were lower among infants with growth faltering, differences in enteral volume emerged earlier and persisted more strongly over time. The final predictive model demonstrated good discrimination (area under the curve=0.85). In adjusted models evaluating individual nutritional exposures, both enteral feeding volume and caloric intake were significantly associated with reduced odds of growth faltering. However, enteral volume demonstrated earlier divergence and a greater consistency across time points. Conclusions: Although both volume and caloric intake were significant when modelled independently, enteral feeding volume emerged as the more robust and clinically informative predictor, likely reflecting feeding tolerance and advancement decisions. Combining dynamic nutritional monitoring with artificial intelligence-based prediction tools may enable earlier identification of at-risk infants and guide individualised nutrition strategies.