AI Algorithm Improves Intravenous Nutrition for Premature Babies
Posted on 27 Mar 2025
Each year, around 10% of babies are born prematurely, which means they arrive at least three weeks before their due date. Babies born more than eight weeks early are often unable to absorb nutrients through their intestines and require intravenous (IV) feeding. Some premature babies also face gastrointestinal complications, necessitating IV nutrition, also known as total parenteral nutrition (TPN), while their digestive systems heal. Currently, IV nutrition is prescribed daily based on factors such as the baby's weight, developmental stage, and lab results. This process requires input from six experts, including a neonatologist or pharmacist, a dietitian, and two pharmacists for verification. The prescription is then sent to a compounding pharmacy to be prepared before it reaches the neonatal intensive care unit (NICU), where a nurse administers the IV while another nurse checks the accuracy of the prescription. This multi-step procedure is prone to errors and makes it difficult for doctors to confirm whether a premature baby has received the correct number of calories, since no blood test can measure this, and unlike full-term infants, premature babies do not typically cry when hungry or show contentment when full. As a result, total parenteral nutrition is a leading cause of medical errors in NICUs worldwide.
A new study at Stanford Medicine (Stanford, CA, USA) suggests that artificial intelligence (AI) could improve IV nutrition for premature infants. The study, published in Nature Medicine, is among the first to demonstrate how AI can aid clinicians in making better decisions for newborns. The AI algorithm uses data from preemies' electronic medical records (EMRs) to predict the necessary nutrients and their amounts. The algorithm was trained on ten years of EMR data from Lucile Packard Children’s Hospital Stanford, encompassing 79,790 IV nutrition prescriptions for 5,913 premature infants. The system also had access to patient outcomes, which enabled it to uncover patterns that linked nutrient levels to health outcomes. Although the doctors had not always been able to get each prescription exactly right, the AI was able to analyze a large volume of data and generalize its findings to improve nutrient recommendations for babies in different medical situations.
After processing this data, the AI grouped similar nutrient prescriptions to identify how many standard formulas could meet all patients' nutritional needs and what should be included in each formula. The researchers demonstrated that the AI could use patient data to recommend which of the 15 formulas a baby might need on any given day, adjusting recommendations as the baby’s condition and needs evolved. For instance, the algorithm might recommend formula No. 8 for five days, then switch to formula No. 3 for a week, and so on. To test the effectiveness of this approach, the research team asked 10 neonatologists to evaluate clinical information from past patients, along with the actual prescriptions the babies had received and the AI-generated recommendations. The doctors did not know which prescription was which and were asked to choose the better option. The results showed that doctors consistently preferred the AI-generated prescriptions over the real ones.
The researchers further validated their findings by comparing the AI-generated prescriptions with actual prescriptions using data from the University of California, San Francisco, which included 63,273 nutrition prescriptions for 3,417 patients. The AI model successfully predicted the nutrient needs of this new group of patients as well. The next step for the team is to conduct a randomized clinical trial, where some patients will receive prescriptions using the traditional manual method, while others will follow the AI recommendations, to assess the impact of each method on patient outcomes.
Once implemented, the AI model would allow doctors and pharmacists to continue reviewing the recommendations and adjusting them as necessary. When a prescription receives medical approval, the baby can immediately receive one of the 15 standard nutrient formulas, which would be stored on a hospital shelf, eliminating the need for a compounding pharmacy. Using standard formulas could significantly reduce costs and improve accessibility, especially in low-resource settings or hospitals in lower-income countries. The introduction of AI-driven IV nutrition prescriptions could reduce medical errors, save time and money, and streamline care for premature babies in settings with fewer resources.
“This reflects our hope for how AI will enhance medicine: What it’s going to do is make doctors better and make top-notch care more accessible,” said study coauthor David Stevenson, MD, a neonatologist and the Harold K. Faber Professor in Pediatrics. “Hopefully, it will also give our physicians more time to do the things computers can’t do, such as spending time with babies and their families, listening to them, and providing comfort and reassurance.”
Related Links:
Stanford Medicine
