Search

Artificial intelligence (AI) is a rapidly advancing area with growing clinical applications in healthcare. The neonatal intensive care unit (NICU) produces large amounts of multidimensional data allowing AI and machine learning (ML) new avenues to improve early diagnosis, enhance monitoring, and provide highly-targeted treatment approaches. In this article, we review recent clinical applications of AI to important neonatal problems, including sepsis, retinopathy of prematurity, bronchopulmonary dysplasia, and others. For each clinical area, we highlight a variety of ML models published in the literature and examine the future role they may play at the bedside. While the development of these models is rapidly expanding, a fundamental understanding of model selection, development, and performance evaluation is crucial for researchers and healthcare providers alike. As AI plays an increasing role in daily practice, understanding the implications of AI design and performance will enable more effective implementation. We provide a comprehensive explanation of the AI development process and recommendations for a standardized performance metric framework. Additionally, we address critical challenges, including model generalizability, ethical considerations, and the need for rigorous performance monitoring to avoid model drift. Finally, we outline future directions, emphasizing the importance of collaborative efforts and equitable access to AI innovations.

Background: Predicting mortality risk in neonatal intensive care units (NICUs) is challenging due to complex, variable clinical and physiological data. Machine learning (ML) offers potential for more accurate risk stratification. Objective: To compare the performance of various ML models in predicting NICU mortality using a team-based modeling competition. Methods: We conducted a modeling competition with five neonatologist-led teams applying ML techniques—logistic regression, CatBoost, neural networks, random forest, and XGBoost—to a shared dataset from over 6,000 NICU admissions. The dataset included static demographic and clinical variables, alongside daily samples of heart rate and oxygen saturation. Each team developed models to predict mortality risk at baseline and within 7 days. Models were evaluated using the area under the receiver operator characteristic curve (AUC). Results were presented at a national meeting, where an audience poll ranked models before AUC results were revealed. Results: The audience favored the most complex model (CNN) for real-world application, though logistic regression achieved the highest AUC on test data. Teams employed varied feature selection, tuning, and evaluation strategies. Conclusions: Logistic regression outperformed more complex models, highlighting the importance of selecting modeling methods based on data characteristics, interpretability, and expertise rather than model complexity alone. Impact:

By demonstrating that model complexity does not necessarily equate to better predictive performance, this research encourages the careful selection of modeling approaches.

AbstractObjective:Antibiotic exposure increases the risk of morbidity and mortality in premature infants. Many centers use at least 48 hours of antibiotics in the evaluation of early-onset sepsis (EOS, <72 hours after birth), yet most important pathogens grow within 24 hours. We investigated the safety and efficacy of reducing empiric antibiotic duration to 24 hours.Design:Quality improvement study.Setting:A tertiary-care neonatal intensive care unit.Patients:Inborn infants <35 weeks gestational age at birth (ie, preterm) admitted January 2019 through December 2020.Intervention:In December 2019, we changed the recommended duration of empiric antibiotics for negative EOS evaluations from 48 hours to 24 hours.Results:Patient characteristics before and after the intervention were similar. After the intervention, 71 preterm infants (57%) with negative EOS evaluations received ≤24 hours of antibiotics, an increase from 15 (10%) before the intervention. These 71 infants comprised 77% of infants with negative EOS blood cultures after excluding those treated as clinical sepsis (≥5 days of antibiotics). For all negative EOS blood cultures, the mean treatment duration decreased by 0.5 days from 3.9 days to 3.4 days. This finding equated to 2.4 fewer antibiotic days per 100 patient days for negative EOS blood cultures but similar antibiotic days per 30 patient days (7.2 days vs 7.5 days). This measure did not change over time. Subsequent sepsis evaluations <7 days after a negative EOS blood culture did not increase. Excluding contaminants, the median time to positivity was 13.2 hours (range, 8–23) in 8 positive blood cultures.Conclusion:Implementation of a 24-hour antibiotic course for negative EOS evaluations safely reduced antibiotic exposure in 77% of infants <35 weeks gestational age at birth in whom EOS was ruled out. All clinically significant pathogens grew within 24 hours.

AbstractBackground:Antibiotics are widely used in very low-birth-weight infants (VLBW, <1500 g), and excess exposure, particularly to broad-spectrum antibiotics, is associated with significant morbidity. An antibiotic spectrum index (ASI) quantifies antibiotic exposure by relative antimicrobial activity, adding information to exposure measured by days of therapy (DOT). We compared ASI and DOT across multiple centers to evaluate differences in antibiotic exposures.Methods:We extracted data from patients admitted to 3 level-4 NICUs for 2 years at 2 sites and for 1 year at a third site. We calculated the ASI per antibiotic days and DOT per patient days for all admitted VLBW infants <32 weeks gestational age. Clinical variables were compared as percentages or as days per 1,000 patient days. We used Kruskal-Wallis tests to compare continuous variables across the 3 sites.Results:Demographics were similar for the 734 VLBW infants included. The site with the highest DOT per patient days had the lowest ASI per antibiotic days and the site with the highest mortality and infection rates had the highest ASI per antibiotic days. Antibiotic utilization varied by center, particularly for choice of broad-spectrum coverage, although the organisms causing infection were similar.Conclusion:An antibiotic spectrum index identified differences in prescribing practice patterns among 3 NICUs unique from those identified by standard antibiotic use metrics. Site differences in infection rates and unit practices or guidelines for prescribing antibiotics were reflected in the ASI. This comparison uncovered opportunities to improve antibiotic stewardship and demonstrates the utility of this metric for comparing antibiotic exposures among NICU populations.

Neonatal infections pose a significant threat to the health of newborns. Associated morbidity and mortality risks underscore the urgency of prompt diagnosis and treatment with appropriate empiric antibiotics. Delay in treatment can be fatal; thus, early detection improves outcomes. However, diagnosing early is a challenge as signs and symptoms of neonatal infection are non-specific and overlap with non-infectious conditions. Machine learning (ML) offers promise in early detection, utilizing various data sources and methodologies. However, ML models require rigorous validation and consideration of various challenges, including false alarms and user acceptance requiring careful integration and ongoing evaluation for successful implementation.