Close
Enter your
text here

Predicting Acute Kidney Injury in Neonates Using Machine Learning Algorithms

Predicting Acute Kidney Injury in Neonates Using Machine Learning Algorithms 480 320 Katie Brind'Amour, PhD, MS, CHES

A new model accurately identifies neonates at risk of acute kidney injury (AKI) — and the factors physicians should address to prevent it — offering the first AKI prediction tool specific to this vulnerable population.

Acute kidney injury (AKI) is common and irreversible, making preventive measures essential to maintaining health. Multiple risk prediction tools exist to predict AKI in pediatric and adult patients admitted to the hospital; by identifying patients at risk of AKI in the coming 2 or 3 days, these tools give clinicians time to intervene and avoid acute damage. But tools for adults and older children with different health needs and short-term admissions don’t translate well to neonates in the neonatal intensive care unit (NICU) for weeks to months. That’s why a team at Nationwide Children’s Hospital set out to create the first neonatal tool for the prediction of AKI.

“Our ultimate goal is to make sure preterm babies will have best outcomes possible,” says Tahagod Mohamed, MD, pediatric nephrologists and director of the Neonatal Nephrology Program at Nationwide Children’s. “These babies are very vulnerable and have a high risk of AKI and chronic kidney disease (CKD) as they age. We want to use any chance to prevent AKI while they’re in the NICU so we can help prevent CKD and hypertension in the future.”

Dr. Mohamed, together with a team of physician-scientists, pharmacists and bioinformaticians at Nationwide Children’s, has made significant progress toward that goal, sharing their initial findings in a poster at the 2024 American Society of Nephrology conference and their most recent data in a submitted manuscript.

The new tool, NEPHRO (NEonatal Protection of Health-Related Outcomes), uses a machine learning algorithm that employs clinical data available in the electronic medical record to put together a risk score for AKI prediction in the coming 48 hours. The risk factors are transparent to users, and the score is continually updated for as long as the patient is admitted.

Dr. Mohamed and the team developed and trained it using a Nationwide Children’s data set of 5,400 NICU patients from 4 years of medical records, then validated it prospectively over the following 2 years with another nearly 3,000 patients. The model performed well in both groups and is currently being built into the hospital electronic record system (EPIC) for further validation.

“There are modifiable factors we can intervene on before kidney injury and potential permanent damage occur,” says Dr. Mohamed, who served as first author on a recent commentary explaining the need for improved population-specific AKI risk prediction in Pediatric Nephrology. “We can optimize blood pressure, address fluid balance and electrolytes, and adjust nephrotoxic medications. It’s a highly individualized approach once we know a patient is at risk.”

With NEPHRO, clinicians will see a risk score and, once the next phase of development is complete, customized best practice alerts that will give them actionable items to evaluate and address to prevent AKI.

Knowing the factors that go into the risk prediction tool and their relevance in the risk score of individual patients is a crucial characteristic of machine learning programs for physicians, increasing the trustworthiness of the model and setting providers at ease about acting on the risk scores.

“NEPHRO has been developed by clinicians and will go through a very rigorous validation stage, which we hope improves its utility and reliability for clinicians,” says Dr. Mohamed. “This algorithm helps us create a therapeutic window before AKI occurs to improve our interventions and diagnostic tools.”

The team’s next step is to test it on an external data set before deploying it in a larger validation study using Dr. Mohamed’s scholarship from PEDSnet, which offers the team access to NICU data from 12 major hospitals. After that comes randomized clinical trials to compare NEPHRO use to standard of care for predicting and avoiding AKI in NICU babies.

Dr. Mohamed believes that by better understanding AKI risk, urine and blood tests can be developed as more reliable biomarkers for neonates moving forward. She also sees the potential to use the team’s collaboration between physician-scientists and informaticians for future machine learning tools.

“As clinicians, we could take forever to look at patient variables, find the best predictors and come up with a risk score,” she says. “But we can harness the power of machine learning to pull from existing data in a highly interpretable and transparent way. Then with a multi-disciplinary team we can evaluate and implement these tools to improve outcomes for patients, and that’s the ultimate goal.”

 

References:

  1. Mohamed T, Asdell N, Ning X, Newland JG, Harer MW, Slagle CL, Starr MC, Spencer JD, Wilson FP, Selewski DT, Slaughter JL. Evidence-based risk stratification for neonatal acute kidney injury: a call to action. Pediatr Nephrol. 2025 Mar 27. Epub ahead of print.
  2. Mohamed T, Slaughter J, Bambach S, Magers J, Rust L, Patel S, Rust S, Spencer JD, Wilson FP. Automated Learning and Early Recognition Technology for Neonatal AKI (ALERT-NAKI): FR-PO708. Journal of the American Society of Nephrology. 2024;35(10S):10.1681/ASN.2024054cj0jp.

 

About the author

Katherine (Katie) Brind’Amour is a freelance medical and health science writer based in Pennsylvania. She has written about nearly every therapeutic area for patients, doctors and the general public. Dr. Brind’Amour specializes in health literacy and patient education. She completed her BS and MS degrees in Biology at Arizona State University and her PhD in Health Services Management and Policy at The Ohio State University. She is a Certified Health Education Specialist and is interested in health promotion via health programs and the communication of medical information.

You might also like