Joseph R. Hageman, MD, Lolita Alkureishi, MD, Walid Hussain, MD, Mitchell Goldstein, MD, MBA, CML

Artificial intelligence (AI) and machine learning (ML) models are being utilized in several clinical issues in neonatology, including necrotizing enterocolitis, retinopathy of prematurity, and now acute kidney injury (AKI) (1-5). 

AI is a “branch of computer science that develops systems capable of human-like intellectual processes to solve problems. ML, a subset of AI, uses large data sets and some human input (adding or changing parameters) to teach itself patterns and to make predictions (6)”. 

Neonatal acute kidney injury (AKI) is a common clinical problem in neonates, especially in both premature infants and term infants who have undergone cardiac or abdominal surgery (2-4), and it holds great promise for the use of AI and ML clinically. The most recent gold standard definition of neonatal AKI is the neonatal-modified Kidney Disease: Improving Global Outcomes (KDIGO) definition, which is summarized in the table from a review by Coleman et al. (3) as well as by Starr and colleagues and involves AKI staging (0,1,2,3) based on serum criteria and hourly urine output (2,4). 

It is also vital to understand embryology and that nephrogenesis begins at five weeks gestation and continues until 34 to 36 weeks gestation (2). The nephron number is highly variable, particularly in premature infants (2). Renal blood flow and perfusion pressure also increase over the first few weeks of post-natal life, as does the proportion of cardiac output to the kidney (2). Several factors affect renal blood flow and perfusion pressure, including hypotension, hemorrhage, hypoxic-ischemic encephalopathy, nephrotoxic medications, sepsis, and congenital heart disease, for example (2). 

There have been a couple of projects using AI and ML models to predict neonatal AKI, as well as to reduce the incidence of AKI in neonates by making clinicians more aware of these factors that contribute to the development of neonatal AKI: “STARZ risk stratification AI model which was incorporated in the electronic medical record (EMR) and showed a predictive ability of AKI within seven days of NICU admission of the area under the curve (AUC) of 0.93 and AUC of 0.96 in the validation and derivation cohorts, respectively (3)”. Also, “in the neonatal population, using the “Baby NINJA” model showed a decrease in nephrotoxic medication exposure by 42%, a decrease in the rate of AKI by 78%, and decreased number of days with AKI by 68% (3)”. By harnessing the power of information already readily available in the EMR, AI and ML models hold great promise in flagging these at-risk infants, thus helping to stave off AKI injury before it begins. 

Of note, many studies have reported the superiority of using biomarkers to predict AKI in pediatric patients and neonates as well. Future directions include the application of AI along with biomarkers (neutrophil gelatinase-associated lipcalin (NGAL), for example, in a Labelbox configuration to create a more robust and accurate model for predicting and detecting pediatric/neonatal AKI (3,4). AI and ML models may be able to take these biomarkers into account, in addition to other clinical predictors, in order to identify at-risk infants correctly. 

Finally, methylxanthines, theophylline, and caffeine have demonstrated reno-protective effects by inhibiting adenosine-induced renal vasoconstriction, thereby preventing neonatal AKI (2). The thought is that if the AI or ML model predicts the possibility of the development of neonatal AKI in a patient, an EMR notification would appear, suggesting that possibility with a suggestion of using caffeine therapy to prevent the development of AKI. The same notification could be used for avoiding nephrotoxic medications (2,3). 

There is a wealth of information in the EMR that can be used for the benefit of our patients; however, doing so can be cumbersome and inefficient. With the entry of AI and ML into the healthcare setting, the EMR can be leveraged in a much more simplified and accessible way such that it can be used to help guide real-time, informed, evidence-based medical decisions (7). The intersection of the EMR platform with AI makes for an inspiring time for health care, and we look forward to seeing how these translate into efficient and intuitive management workflows for clinicians in AKI and beyond. 

References: 

1. McElroy SJ and Lueschow SR> State of the art on machine learning and artificial learning in the study of neonatal necrotizing enterocolitis . Frontiers in Pediatrics 2023. Doi: 10.3389/fled.2023.1182597. 
2. Starr MC, Charlton JR, Gulliet R et al. Advances in neonatal kidney injury. Pediatrics 2021; 148(5);32021051220. 
3. Raina R, Nada A, Aly I et al. Ariticial intelligence in early detection of pediatric/neonatal acute kidney injury: current status and future directions. Pediatr Nephrol. 2023; DOI: 10.1007/s00467-023-06191-7. 
4. Coleman C, Tambay Perez A, Selewski DT, Steflik HJ. Neonatal Acute Kidney Injury. Front Pediatr. 2022 April 7;10:842544. doi: 10.3389/fped.2022.842544. PMID: 35463895; PMCID: PMC9021424. 
5. Jeong H, Kamaleswaran R. Pivotal challenges in artificial intelligence and machine leaning applications for neonatal care.Semin Fetal Neonatal Med. 2022;Oct,27(5):101393. Epub.2022 October 13. Doi: 10.1016/jsiny.2022.101393. Epub 2022 October 13.
6. Shah S, Slaney E, VerHage E et al. Application of artificial intelligence in the early detection of retinopathy of prematurity: Review of the literature. Neonatology 2023; 120(5):558-565. https://doi.org/10.1159/000531441 
7. Guide to unlocking value in EHR & EMR using. October 4, 2023. AI https://www.hippocraticpost.com/innovation/guide-to-unlocking-value-in-ehr-emr-using-ai/ Accessed January 29, 2024. 

Disclosures: The authors have no disclosures