This paper presents an approach to enhancing neonatal care through the application of artificial intelligence (AI). Utilizing network-oriented modeling methodologies, the study aims to develop a network model to improve outcomes in neonatal respiratory support. The introduction sets the stage by outlining the significance of neonatal respiratory support and the challenges faced in this domain. The literature review delves into the existing body of work, highlighting the gaps and the need for a network modeling approach. The network-oriented modeling approach provides a robust framework that captures various states, such as world states, doctors’ mental states, and AI coach states, facilitating a comprehensive understanding of the complex interactions in neonatal respiratory support. Through Matlab simulations, the study investigates multiple scenarios, from optimal conditions to deviations from standard protocol. The main contribution focuses on the introduction of an AI coach, which serves as a real-time intervention mechanism to fill in the doctor's knowledge gaps. The research serves as a seminal work in the intersection of artificial intelligence and healthcare, demonstrating the potential of network-oriented modeling in improving patient outcomes and streamlining healthcare protocols.

Background: Preterm neonates are extensively monitored to require strict oxygen target attainment for optimal outcomes. In daily practice, detailed oxygenation data are hardly used and crucial patterns may be missed due to the snapshot presentations and subjective observations. This study aimed to develop a web-based dashboard with both detailed and summarized oxygenation data in real-time and to test its feasibility to support clinical decision making. Methods: Data from pulse oximeters and ventilators were synchronized and stored to enable real-time and retrospective trend visualizations in a web-based viewer. The dashboard was designed based on interviews with clinicians. A preliminary version was evaluated during daily clinical rounds. The routine evaluation of the respiratory condition of neonates (gestational age < 32 weeks) with respiratory support at the NICU was compared to an assessment with the assistance of the dashboard. Results: The web-based dashboard included data on the oxygen saturation (SpO₂), fraction of inspired oxygen (FiO₂), SpO₂/FiO₂ ratio, and area < 80% and > 95% SpO₂ curve during time intervals that could be varied. The distribution of SpO₂ values was visualized as histograms. In 65% of the patient evaluations (n = 86) the level of hypoxia was assessed differently with the use of the dashboard. In 75% of the patients the dashboard was judged to provide added value for the clinicians in supporting clinical decisions. Conclusions: A web-based customized oxygenation dashboard for preterm neonates at the NICU was developed and found feasible during evaluation. More clear and objective information was found supportive for clinicians during the daily rounds in tailoring treatment strategies.

The aim of this study was to investigate the association between the implementation of a local heart rate variability (HRV) monitoring guideline combined with determination of inflammatory biomarkers and mortality, measures of sepsis severity, frequency of sepsis testing, and antibiotic usage, among very preterm neonates. In January 2018, a guideline was implemented for early detection of late-onset neonatal sepsis using HRV monitoring combined with determination of inflammatory biomarkers. Data on all patients admitted with a gestational age at birth of < 32 weeks were reviewed in the period January 2016–June 2020 (n = 1,135; n = 515 pre-implementation, n = 620 post-implementation). Outcomes of interest were (sepsis-related) mortality, sepsis severity (neonatal sequential organ failure assessment (nSOFA)), sepsis testing, and antibiotic usage. Differences before and after implementation of the guideline were assessed using logistic and linear regression analysis for binary and continuous outcomes respectively. All analyses were adjusted for gestational age and sex. Mortality within 10 days of a sepsis episode occurred in 39 (10.3%) and 34 (7.6%) episodes in the pre- and post-implementation period respectively (P = 0.13). The nSOFA course during a sepsis episode was significantly lower in the post-implementation group (P = 0.01). We observed significantly more blood tests for determination of inflammatory biomarkers, but no statistically significant difference in number of blood cultures drawn and in antibiotic usage between the two periods.

Conclusion: Implementing HRV monitoring with determination of inflammatory biomarkers might help identify patients with sepsis sooner, resulting in reduced sepsis severity, without an increased use of antibiotics or number of blood cultures.