Seminars in Fetal & Neonatal Medicine最新文献

Objectives: To explore the applicability of artificial intelligence (AI) in neonatal intensive care units (NICUs), identifying key trends in AI-driven technologies and their roles in the prognosis, classification, monitoring and forecasting of neonatal conditions.

Methods: A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses)-guided systematic review was conducted across MEDLINE, EMBASE, Cochrane, and IEEE Xplore, covering studies published between January 2013 and December 2023. A total of 318 studies were initially retrieved. After removing 61 duplicates and screening 257 articles by eligibility criteria, 64 studies were assessed for full-text eligibility, leading to the final inclusion of 41 studies.

Results: The predominant AI application referred to conditions in the following systems: cardiovascular (n = 9, 21.9 %), neural/brain (n = 8, 19.5 %), respiratory (n = 8, 19.5 %), immune (infections) (n = 6, 14.6 %), gastrointestinal (n = 2, 4.9 %), and microvascular diseases (n = 1, 2.4 %). Additionally, six studies focused on monitoring systems or body positioning (categorized as "Not Disease"), and one study (2.4 %) addressed mortality prediction. Regarding the purposes of AI application, prognosis (n = 23, 56.1 %) was the most common, followed by classification (n = 14, 34.1 %), monitoring (n = 5, 12.2 %), and symptom forecasting (n = 1, 2.4 %). More than 70 % of studies (n = 29, 70.7 %) lacked a validation procedure, highlighting a critical gap in methodological rigor.

Conclusions: Our findings underscore the potential benefits of the use of AI in neonatology, possibly resulting in improved patient outcomes and enhanced operational efficiency. However, data privacy, algorithm interpretability, and ethical considerations must be addressed for responsible AI deployment in neonatal care. We highlight future directions, emphasizing interdisciplinary collaboration, adherence to reporting guidelines, and the need for further research to enhance AI reproducibility and clinical integration in the NICUs. The findings of this study support AI's potential for shaping neonatal health care.

Neonatal sepsis remains a major cause of neonatal deaths globally. Despite advances, accurate and timely diagnosis is hindered by the limited performance of the current clinical approaches, imperfect laboratory biomarkers, and long turnaround time of blood cultures. Artificial intelligence (AI), with its ability to identify patterns and learn continuously (machine learning), seems promising. Basic steps in model development include data filtration, train: test split, feature selection, choosing appropriate algorithms, and evaluating performance using a reference standard. In neonatal sepsis, the role of AI spans from predicting sepsis and related outcomes to formulating an individualized treatment approach for the neonate. Existing models, largely from high-income countries, report encouraging diagnostic accuracy but face methodological limitations, lack external validation, and remain somewhat distant from bedside application. Additional barriers to their generalizability include lack of uniform definition of sepsis, variations in disease and pathogen profiles in different settings (particularly in developing countries), availability of electronic health data, tweaks in feature selection, and ethical and legal challenges. This review synthesizes current evidence, highlights gaps, and outlines priorities for future research. We call for a collaborative effort from AI and neonatal experts to devise robust, context-specific solutions.

Mortality remains a key indicator for the assessment of care quality in medicine. In neonatology, mortality rates are highly variable, both across units and over time. Comparison of crude mortality rates, however, are insufficient for benchmarking, as they fail to account for differences in population case mix and severity of illness. Risk adjustment using artificial intelligence (AI) and machine learning (ML) has emerged as a promising tool to facilitate meaningful comparisons and drive improvement. This review seeks to examine the state of the current literature on the use of AI/ML-based models to predict mortality in the neonatal intensive care unit (NICU). We identified 37 studies describing 242 models. Most studies developed models using single-center data and frequently lacked external validation. Similarly, reporting of performance metrics was heterogenous, limiting evaluation. As a result, further work is necessary before AI/ML-enabled risk adjustment is feasible.

The use of Artificial Intelligence (AI) has the potential to transform healthcare in part by enhancing the accuracy of drug dosing and improving patient safety. However, its use in neonatology and pediatrics has just been started, with limited research exploring its full potential. This scoping review systematically maps the literature on AI applications in pediatric and neonatal pharmacology, analyzing studies published between 2004 and 2024. Searches in databases including MEDLINE, Scopus, and IEEE Xplore identified 412 records, of which 33 met the inclusion criteria. These included neonates (n = 8) and older pediatric patients (n = 25), encompassing 58,864 patients and utilizing various Machine-Learning techniques. The use of AI has demonstrated significant potential for precision dosing, predicting drug efficacy, and decreasing the occurrence of adverse events. Despite these promising findings, however, more rigorous, large-scale studies are essential to validate the results. Future research should prioritize real-world applications and address integration barriers, ensuring safe and effective use of AI in neonatal and pediatric clinical practice.

Background: Non-invasive ventilation (NIV) is a key form of respiratory support in neonatal intensive care units (NICU). Non-invasive ventilation failure, however, can lead to adverse outcomes in preterm infants. This narrative review explores the potential of using artificial intelligence (AI) to improve the prediction of NIV failure, potentially reducing the mortality and morbidity within this population.

Methods: A literature search was conducted using PubMed with terms relating to AI, machine learning, NIV and neonatology. Studies which used AI models to predict NIV failure or the need for intubation, within the neonatal population, were included. Model performance was assessed using area under the receiver operating characteristic curve (AUC).

Results: Six studies, including 3421 infants, were identified. Various AI techniques were used including deep learning models, for example multimodal deep neural networks, as well as simpler machine learning models such as logistic regression and support vector machines. AUC values ranged from 0.78 to 0.93, with most models exhibiting clinically useful performance defined as an AUC >0.8. The modal key predictive factors across the six studies were gestational age, SpO₂ and maximum FiO_2. CONCLUSION: AI- generated models for predicting NIV failure as first intention in the NICU setting show potential. Deep learning models demonstrate particular promise; however, further large multicenter externally validated studies are required to assess generalizability and to aid integration into routine clinical practice. Implementation of AI models to predict NIV failure as first intention and post-extubation could lead to improved clinical decision making and personalized care.

Optimizing neonatal nutrition and diagnosing serious gastrointestinal diseases remains a challenge, as traditional guideline-based approaches often fail to address the individualized needs of preterm and term infants. Advances in artificial intelligence and machine learning provide opportunities for precision diagnostics and therapeutics by incorporating multiomic data and clustering infants based on risk factors and metabolic profiles. For example, machine learning is redefining necrotizing enterocolitis as a spectrum of intestinal injuries rather than a single disease, while digital twin models offer the potential for real-time personalized nutrition optimization. Moreover, integration of advanced gastrointestinal monitoring methods using novel biomarkers and sensor technologies may further enhance early detection and intervention strategies. Altogether, these digital technological advancements may lead to identification of early predictors of nutritional deficiencies and prompt recognition of gastrointestinal pathologies, thereby allowing for proactive interventions and potentially improved outcomes in the neonatal population.