Opportunities and Challenges of Using Artificial Intelligence in Predicting Clinical Outcomes and Length of Stay in Neonatal Intensive Care Units: Systematic Review.

Background: The use of artificial intelligence (AI) in health care has been steadily increasing for over 2 decades. Integrating AI into neonatal intensive care units (NICUs) has promise as it has the potential to reshape neonatal care and improve outcomes. However, challenges such as data quality, clinical interpretation, and ethical considerations may hinder AI's practical implementation in NICUs.

Objective: This study aims (1) to analyze the current AI research landscape for predicting clinical outcomes and length of stay in the NICU and (2) to explore the benefits and challenges of using AI in the NICU for these predictions.

Methods: A systematic review was conducted across 6 databases-PubMed, Embase, CINAHL, Cochrane Library, Informit, and La Trobe Library-to identify English-language peer-reviewed articles published between January 2017 and March 2023 that focused on the use of AI for predicting length of stay and clinical outcomes for NICU patients. Eligibility criteria excluded studies outside the NICU context or lacking predictive focus. Both prospective and retrospective designs were included. A thematic analysis of AI applications in NICUs from the articles identified was conducted.

Results: A total of 24 studies were included in the review, comprising 15 retrospective and 9 prospective designs. These studies primarily originated from the United States (13 studies), with others from Austria, Taiwan, and other countries. The studies evaluated AI applications in NICU settings to predict comorbidities (18/24), mortality (4/24), and length of stay (2/24). Sixteen studies were in the exploration stage, lacking cohesive AI strategies, while 8 demonstrated systematic exploration but no fully integrated solutions. The synthesis of results identified key applications of AI in NICU care, including data-driven insights and predictive models, advancements in medical imaging, improved risk stratification, and personalized neonatal care. AI showed promise in enhancing diagnostic accuracy and care planning, but significant challenges persist, such as data quality, model generalization, and ethical concerns. No studies reported a fully integrated AI ecosystem, highlighting the need for further research to bridge gaps and realize AI's transformative potential in neonatal care.

Conclusions: This review highlights the potential of AI in improving NICU care, particularly through predictive models, medical imaging, and personalized interventions. However, the evidence is limited by significant methodological variability, small sample sizes, risk of bias, and a lack of external validation in included studies. Many studies remain in exploratory phases without cohesive AI strategies or integration into clinical practice, limiting the practical applicability of findings. These results underscore the importance of addressing challenges such as data quality, model generalization, and ethical considerations to fully realize AI's potential in neonatal care. Future research should focus on robust validation, comprehensive implementation strategies, and ethical frameworks to ensure AI's effective and responsible integration into NICU settings.