To describe first week electrolyte trajectories and develop machine learning models to estimate serum phosphate levels during the first postnatal week in very preterm infants.
Approach:
Machine Learning Models: Defined three regression tasks for phosphate prediction at blood gas analysis and forecasts 12 h and 24 h ahead, using 22 candidate predictors selected through RENT-based variable selection within leave-one-group-out cross-validation.
Key Findings:
33.6% of infants developed hypophosphatemia in the first week.
SGA infants had a significantly higher risk of hypophosphatemia compared to appropriate-for-gestational-age infants.
Six predictors were consistently retained across all prediction tasks.
Elastic net models performed best on all prediction tasks.
Interpretation:
SGA status is a major determinant of early hypophosphatemia risk in very preterm infants. ML models can moderately estimate serum phosphate levels using routine clinical data.
Limitations:
Prospective validation is warranted before clinical implementation.
The study is limited to a specific cohort of infants born <29 weeks' gestation and may not generalize to all very preterm infants.
Conclusion:
ML models may serve as screening tools to guide nutritional management and reduce unnecessary blood sampling.
