Clinical characteristics, risk factors and machine‑learning prediction models for ultra‑early‑onset neonatal sepsis: a retrospective cohort study

This classification paradigm of neonatal sepsis is derived from high-income country datasets, but clinical microbiology evidence suggests that this classification may not be suitable for low- and middle-income countries. Therefore, we proposed the concept of ultra-early onset of neonatal sepsis (ultra EOS) to represent neonatal sepsis patients diagnosed within 24 h after birth, and aimed to describe its clinical characteristics and identify relevant risk factors. All the patients were diagnosed with early onset of neonatal sepsis (EOS) and treated in Nanyang Central Hospital between July 2023 and June 2025. The data of newborns with EOS and maternal information were retrospectively collected from medical records and patient registration information. The ultra EOS means EOS diagnosed within 24 h after birth. The Kaplan–Meier curve analysis was used to compare treatment outcomes. The logistic regression analysis was used to explore independent risk factors. Multiple machine learning models were used to establish predictive models and interpret them by SHapley Additive exPlanations (SHAP) algorithm. The study included 583 EOS patients, with 447 ultra EOS and 136 non-ultra EOS cases. Compared with the non-ultra EOS group, the ultra EOS group showed significant differences in demographic information, clinical manifestations and laboratory tests. Overall, the condition of the ultra EOS group was more severe, and patients in ultra EOS group require more time of anti-infection treatment and had a longer length of stay. We identified 7 factors independently associated with ultra EOS including blood Ca concentration, 1-min Apgar score, low response symptoms, gestational age, fever, jaundice and birth place. On this basis, we established a prediction model (AUC 0.864, 95% confidence interval: 0.804–0.924) using neonate birth information and the maternal information during pregnancy to assist in the early diagnosis of ultra EOS patients. In conclusion, our study explored the possibility of a new EOS classification, hoping to contribute to the diagnosis and treatment of neonatal sepsis, especially in low- and middle-income countries (LMICs). However, some clinically valuable factors have not been fully collected. In the future, prospective studies may solve these problems.