Development and validation of a model predicting preterm infant discharge in level 2 care.
retrospective_cohort · Level III
Where this comes from
- Record sourced from PubMed, PMID 41820625.
- Also identified by DOI 10.1038/s41390-026-04782-2.
- No licence information is recorded for this record.
- Because redistribution is not established, this page shows the abstract only. Follow the links below for the full text.
Abstract
To develop and validate a model for predicting upcoming discharge home of preterm infants in a level 2 neonatal ward. This retrospective cohort study included preterm infants admitted to the two-location study site between January 2016 and December 2023. A multivariable logistic regression model was developed using backward selection, with day 7 of admission selected as the prediction time. Primary outcome was discharge within one week (i.e. between admission day 7 and 14). On our wards, discharge required a minimum postconceptional age (PCA) of 35 weeks. Thus, infants with a PCA < 33 weeks at admission were excluded. The 1083 infants included were allocated to the development (n = 614) or validation (n = 469) set. Nine predictors were identified: mode of delivery, syndromal diagnoses, gestational and postconceptional age, tube feeding, provision of mother's own milk, weight, monitor surveillance, and caffeine administration. Internal and external validation showed excellent discrimination (AUC 0.93, CI 0.90-0.95) and acceptable calibration (slope 1.13, CI 0.91-1.35; intercept -0.14, CI -0.45 to 0.16). A probability threshold of 0.60 provided a sensitivity of 88% and specificity of 89%. A combination of perinatal and neonatal characteristics can adequately predict upcoming discharge home of preterm infants in a level 2 neonatal setting. Although models estimating total length of hospital stay in preterm infants have been reported, no models predict upcoming discharge, and the level 2 neonatal population remains underreported. We developed a tool to estimate the odds of discharge home within one week from the time of prediction, identifying nine (mainly clinical neonatal) predictors. The tool showed excellent discrimination and acceptable calibration, providing high sensitivity and specificity. The tool could optimize parent-provider communication and hospital capacity management, and should be validated further in prospective studies.