Aim: To determine whether trends in routinely collected physiological variables can be used retrospectively to classify infants according to the presence or absence of late-onset neonatal sepsis.
Methods: Case control study. Thirty infants born < or =32 weeks of gestation who developed late-onset sepsis were matched with 30 controls for gestational and postnatal age but remained sepsis free. For each infant, 25 clinicians inspected 48 h of routine monitoring of heart rate, respiratory rate and oxygen saturation. Clinicians were asked to determine whether the recording was obtained from an infant who did or did not develop sepsis and also indicate how confident they were in their judgement. Clinicians were stratified into three groups by professional role.
Results: The median correct assignment of infant's recordings was 67% (IQR 62-72). When very confident, this improved to 82% (IQR 67-88). Overall sensitivity was 53% (IQR 43-63) and specificity 80% (IQR 67-87). Advanced neonatal nurse practitioners consistently assigned babies to the correct group more often than other professional groups.
Conclusion: The simple observation physiological trend graphs can classify infants according to the presence or absence of late-onset neonatal sepsis. The accuracy of this method is good to strong but varies with experience of neonatal intensive care.