Development and validation of an early prediction model for post-discharge home oxygen therapy requirement in preterm infants

Background: Home oxygen therapy (HOT) has become an important supportive measure for preterm infants with chronic lung disease. This study aimed to develop and validate a prediction model for identifying preterm infants who will require HOT after discharge. Methods: This single-center retrospective cohort study included preterm infants with birth weight ≤1500g admitted to the neonatal intensive care unit (NICU) between June 2017 and December 2023. The whole cohort was randomly split into training and validation sets. Predictors were screened using a combined strategy of least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm. The final predictors was entered into a multivariable logistic regression model. Results: A total of 810 infants were included, and 129 (15.9%) required HOT. LASSO regression and Boruta algorithm identified four variables: antenatal antibiotic exposure, gestational age, the highest FiO2 during the first 24h after birth, and days of invasive mechanical ventilation during the first postnatal week. The model achieved area under the curves (AUCs) of 0.794 (95% confidence interval (CI): 0.745–0.843) in the training set and 0.820 (95% CI: 0.755–0.885) in the validation set, with Hosmer-Lemeshow P values of 0.709 and 0.146. Calibration curves demonstrated good agreement between predicted probabilities and observed outcomes. After 1000 bootstrap resamples in which the entire variable selection procedure was repeated from scratch, the optimism-corrected AUC was 0.764 (95% CI: 0.731–0.797). Conclusion: The model exhibits good discrimination and calibration and may provide an objective tool to support early risk stratification and discharge planning in the NICU.