Neonatal AI-Based Early Warning System for Late-Onset Sepsis and Necrotizing Enterocolitis in the NICU: Assessment of Response and Effectiveness (NEO-AWARE)

Preterm infants born at less than 32 weeks of gestational age admitted to participating neonatal intensive care units (NICUs) after implementation of Intellicens Neo. The software generates visual alerts for potential late-onset sepsis (LOS) and necrotizing enterocolitis (NEC). Alerts are visible to the clinical care team, while clinical assessment and treatment decisions remain at the discretion of the treating clinicians. Study outcomes will be compared with a historical pre-implementation dataset.

Intellicens Neo is investigational medical device software designed to support early warning of late-onset sepsis (LOS) and necrotizing enterocolitis (NEC) in preterm infants admitted to the neonatal intensive care unit (NICU). The software calculates a NeoSep Index every 30 minutes and generates visual alerts when predefined alert criteria are met. The alerts are made visible to the clinical care team to support awareness of potential LOS or NEC. Clinical assessment and treatment decisions remain at the discretion of the treating clinicians. The intervention is evaluated in a prospective, multicenter, post-implementation clinical investigation, with outcomes compared with a pre-implementation historical dataset.

The time interval between the last Intellicens Neo visual alert preceding the diagnosis of late-onset sepsis (LOS) or necrotizing enterocolitis (NEC) and the administration of intravenous antibiotics for that episode. This outcome evaluates whether the availability of Intellicens Neo alerts is associated with earlier initiation of antibiotic treatment in preterm infants with LOS or NEC.

Primary outcome in the substudy, analysis done only in the first 150 patients enrolled. The proportion of Intellicens Neo visual alerts that are acknowledged by the clinical care team within 3 hours after the visual alert is generated.

The time interval between the start of the last Intellicens Neo visual alert preceding a LOS/NEC episode and the time of LOS/NEC diagnosis. The time of diagnosis is defined as the time of the blood culture order in the patient's electronic health record.

Occurrence of severe late-onset sepsis (LOS) or necrotizing enterocolitis (NEC) episodes in the study population. The number of severe LOS/NEC episodes will be compared between the prospective cohort and the pre-implementation historical cohort.

Total duration of the patient's stay in the neonatal intensive care unit (NICU), measured from NICU admission to NICU discharge.

Total number of days of intravenous antibiotic treatment during the patient's NICU stay.

Total number of suspected late-onset sepsis (LOS) episodes recorded during the patient's NICU stay.

Occurrence of mortality related to late-onset sepsis (LOS) or necrotizing enterocolitis (NEC), defined as mortality occurring within 7 days after LOS/NEC diagnosis.

Performance of the Intellicens Neo software device ML model on the prospective dataset, defined as the sensitivity/recall of the ML model: the percentage of detected LOS/NEC episodes by the Intellicens Neo software device

Performance of the Intellicens Neo software device ML model on the prospective dataset, defined as the specificity of the ML model

Performance of the Intellicens Neo software device ML model on the prospective dataset, defined as the precision of the ML model: the amount of (false positive) visual alerts generated by the Intellicens Neo software device represented by the total number of 24-hour time-windows in which a false positive visual alert occurred

Performance of the Intellicens Neo software device ML model on the prospective dataset, defined as the visual alert rate of the ML model: total number of alert days per patient per week.

Number of patients who need to be examined to detect one patient with the relevant disease (LOS or NEC).

The time interval between the time of LOS/NEC diagnosis, defined as the time of the blood culture order in the patient's electronic health record, and the time of intravenous antibiotic treatment, defined as the time of the order of the first dose of intravenous antibiotics in the patient's electronic health record.

The proportion of Intellicens Neo visual alerts that are acknowledged by the clinical care team within 1 hour after the visual alert is generated.

The proportion of Intellicens Neo visual alerts that are acknowledged by the clinical care team within 6 hours after the visual alert is generated.

The proportion of all Intellicens Neo visual alerts that are acknowledged by the clinical care team, irrespective of the time taken to acknowledge them.

The proportion of Intellicens Neo visual alerts that are followed by a clinical action related to suspected LOS or NEC.

The proportion of Intellicens Neo visual alerts that are followed by laboratory evaluation for suspected late-onset sepsis (LOS) or necrotizing enterocolitis (NEC).

The proportion of Intellicens Neo visual alerts that are followed by initiation of intravenous antibiotic treatment for suspected LOS or NEC.

The proportion of Intellicens Neo visual alerts that are dismissed by the clinical care team.

The time interval between an Intellicens Neo visual alert and the first clinical action related to that alert.

Evaluation of the effect of time of day on the time to acknowledgement of Intellicens Neo visual alerts. Time periods are defined as day shift (08:30-17:00), evening shift (17:00-23:00), and night shift (23:00-08:30).

Evaluation of the effect of alert burden on the time to acknowledgement of Intellicens Neo visual alerts. Alert burden is defined as the presence of at least 2 visual alerts in the same patient during the preceding week.