Identification of risk factors for hyperglycemia during the neonatal period in preterm infants: a machine learning-based study

Preterm infants are susceptible to hyperglycemia within the neonatal period, leading to increased mortality and morbidity. Nevertheless, the risk factors contributing to hyperglycemia in this population remain incompletely elucidated. To address this issue, we employed machine learning algorithms to identify the risk factors for hyperglycemia detected by continuous glucose monitoring (CGM) in preterm neonates. In this retrospective observational study, we included 66 preterm infants (< 37 weeks’ gestation) who were hospitalized at the Children’s Hospital of Fudan University from June 2015 to May 2023 and underwent CGM initiated within 7 days after birth during the neonatal period. We evaluated maternal, neonatal, and postnatal factors associated with hyperglycemia frequency, defined as CGM glucose > 10 mmol/L and quantified by the proportion of CGM time above this threshold. Pairwise correlation, structural equation modeling (SEM), and a histogram-based gradient boosting regression tree (HGBRT) with SHAP-based feature ranking were used for analysis. Hyperglycemia (> 10 mmol/L) was detected in 17 of 66 CGM-monitored preterm neonates (25.76%). Neonatal respiratory distress syndrome (NRDS), metabolic acidosis, Apgar score at 5 min after birth, maternal anemia, Apgar score at 1 min after birth and maternal age were significantly correlated with hyperglycemia frequency. SEM analysis demonstrated that postnatal factors, rather than maternal factors or neonatal factors are causally associated with higher hyperglycemia frequency. Machine learning-based modeling using HGBRT identified lower Apgar score at 1 min after birth as the top risk factor for hyperglycemia, and other risk factors including infection, lower Apgar score at 5 min after birth, lower gestational age, higher parity and higher maternal age. Integrating analyses from pairwise correlation, SEM and machine learning, we found that postnatal factors, including lower Apgar score at 1 min after birth, lower Apgar score at 5 min after birth, and higher maternal age, significantly contributed to higher hyperglycemia frequency within the neonatal period in preterm infants. Our results indicate that preterm infants with postnatal risk factors should be prioritized for CGM use to enhance hyperglycemia surveillance.