AI in Neonatal Intensive Care

Some neonatal monitoring systems analyze patterns in heart rate that may change before a very-low-birth-weight infant shows obvious signs of illness. Heart-rate-characteristics (HRC) monitoring uses features such as reduced variability and transient decelerations to calculate a risk index associated with late-onset sepsis. This is an early-warning signal, not a blood-culture result or a diagnosis. A high score may prompt clinicians to reassess the infant and consider whether more testing is needed; it does not itself establish infection or require a particular treatment. A multicenter randomized trial studied HRC monitoring in 3,003 very-low-birth-weight infants across nine neonatal intensive care units. When the HRC display was available to clinicians, inpatient mortality was 8.1%, compared with 10.2% in the masked group; the estimated relative hazard was 0.78. However, the trial’s primary outcome of days alive and ventilator-free showed only a non-significant trend, and there were no significant differences in several other measures such as ventilator days or NICU stay. The findings concern a defined high-risk population and a specific monitored workflow. They do not prove that all neonatal AI alerts reduce mortality. The FDA-cleared HeRO system measures heart-rate variability for use by trained operators under licensed-practitioner supervision in hospital neonatal or pediatric ICUs; its FDA decision document says those measurements are not approved for a specific clinical diagnosis. NICU teams should monitor alert burden, false alarms and response protocols. Premature infants need continuous bedside care, laboratory confirmation where indicated and individualized decisions. AI can help surface a pattern; clinicians decide how to evaluate it.