Pediatr Res. 2026 Nov 14. doi: 10.1038/s41390-026-05665-2. Online ahead of print.
ABSTRACT
BACKGROUND: Late-onset sepsis (LOS) causes morbidity and mortality in very-low birthweight (VLBW) neonates. We previously developed a machine learning model, the Pulse Oximetry Warning System (POWS), to predict LOS in the next 24 h. The neonatal Sequential Organ Failure Assessment (nSOFA) score predicts sepsis mortality. We hypothesized that POWS scores would differentiate severe LOS events, defined by nSOFA >4, from milder events.
METHODS: We conducted a multicenter, retrospective cohort study including VLBW infants diagnosed with culture-confirmed LOS with continuous vital sign data available. Mixed-effects regression was used to examine the association between POWS and nSOFA.
RESULTS: We analyzed 89 LOS episodes in 84 VLBW infants. Severe sepsis (nSOFA >4 within 48 h of blood culture) occurred in 27 (30%) episodes. Mean POWS scores were higher in the 24 h before severe LOS than non-severe LOS. POWS rose more steeply in the hours before sepsis diagnosis among infants with severe LOS.
CONCLUSION: POWS scores, predicting risk of LOS in the next 24 h, have a significantly higher rate of rise in severe episodes as compared to less severe episodes as measured by nSOFA. This result adds to evidence that POWS could prompt earlier intervention that improves sepsis outcomes.
IMPACT: Late-onset sepsis (LOS) is a leading cause of death and disability in premature infants. The Pulse Oximetry Warning System (POWS) uses continuous heart rate and oxygen saturation data to predict sepsis up to 24 h before clinical diagnosis. This study demonstrates that POWS rises more steeply before severe sepsis episodes, as defined by the neonatal Sequential Organ Failure Assessment (nSOFA) score, than before milder episodes. POWS thus may provide earlier warning of the most severe sepsis events, potentially enabling earlier intervention in this vulnerable population.
PMID:43034165 | DOI:10.1038/s41390-026-05665-2

