Researchers want hospitals to use new AI nutrition prescriptions to improve preterm infant health

AI-powered IV nutrition recommendations may improve care for premature infants, save time, and reduce errors, according to a Stanford Medicine study. The research suggests that algorithms can read data from preemies’ electronic medical records to predict what and how much nutrients they would need.

IV nutrition, or total parenteral nutrition, is the only way a premature infant can feed before its digestive system develops to absorb nutrients.

Published in Nature, the study reveals how AI was trained on data from nearly 80,000 previous prescriptions for intravenous nutrition linked to their impacts on patients. The data details the baby’s weight, stage of development, and lab work results.

The researchers say using AI-powered IV nutrition prescriptions can improve care for preemies, especially in low-resource settings. They are also considering using AI to provide hospitals with ready-to-use nutrient formulas.

The next step is to run a randomized clinical trial where some patients receive nutrient prescriptions manually and others receive AI-recommended prescriptions, reveal the authors.

Challenges in formula prescription

The study explains the difficulty of prescribing the correct formula since no blood test measures whether the preterm infant is getting sufficient calories per day. Unlike babies, preemies do not always cry when hungry and show when they are full.

“Nutrition is one of the areas of neonatal intensive care where we are weakest. We can’t approximate what the placenta is doing,” adds study co-author David Stevenson, MD, neonatologist and the Harold K. Faber professor in pediatrics at Stanford Medicine.

The researchers believe their study is among the first to show how AI can help doctors make better clinical decisions for newborns.

“Right now, we come up with a total parenteral nutrition (TPN) prescription for each baby, individually, every day. We make it from scratch and provide it to them,” says senior study author Nima Aghaeepour, PhD, associate professor of anesthesiology, perioperative and pain medicine, and pediatrics at Stanford University.

The researchers believe their study is among the first to show how AI can help doctors make better clinical decisions for newborns.“TPN is the single largest source of medical error in neonatal intensive care units, both in the US and globally.”

Long prescription chain

The researchers reveal that 10% of babies are born prematurely and require IV nutrition due to gastrointestinal complications and help gut healing.

The patients receiving prescribed IV nutrition daily require macronutrients to help boost protein, fat, and carbohydrates. They also need micronutrients such as vitamins, minerals, and electrolytes, along with electrolytes and medications, to reduce the risk of blood clots.

The researchers explain that six experts need to work over hours to make the prescriptions. Each prescription is written by a pharmacist or neonatologist and reviewed by a second pharmacist for safety and a dietitian for nutritional composition. 

Once formulated, the prescription is sent to a compounding pharmacy and then to the neonatal intensive care unit once prepared. A nurse administers the IV, and a second nurse checks that each patient has received the right preparation.

“It’s a high-risk drug because it is a mixture of many different things. If we had manufactured ready-to-use TPNs, that would be very beneficial. I think it would be safer for patients,” says study co-author Shabnam Gaskari, PharmD, executive director and chief pharmacy officer at Stanford Medicine.

Improving prescription accuracy 

To improve the prescription process, the researchers trained their AI algorithm on 10 years of medical record data of 5,913 preemies from the neonatal intensive care unit at Lucile Packard Children’s Hospital Stanford.

The researchers reveal that 10% of babies are born prematurely and require IV nutrition due to gastrointestinal complications and help gut healing.The AI tracked patterns tied to prescriptions and outcomes and found that doctors did not always get the prescription exactly right, but the volume of data helped overcome the problem.

“This is a strength of AI — sometimes imperfect data is good enough as long as you have a lot of it,” says Aghaeepour. “We wondered, ‘what if we make three standard formulas, or 10, or 100?’”

“It turns out that with 15 distinct formulas for IV nutrition, what you are recommending is pretty similar to what the physicians, pharmacists, and dietitians would have done anyway. But then these 15 AI-based formulas can be used to significantly improve speed and safety.”

The researchers demonstrated how the AI could predict which of the 15 formulas each baby might require and adjust the recommendations daily as preterm infant growth changes medical conditions. 

They tested 10 doctors prescribing IV nutrition. Researchers gave them clinical information for past patients, IV nutrition prescriptions they had actually received, and the prescriptions the algorithm would recommend. Not knowing which prescription was which, the doctors consistently preferred the AI-generated prescriptions.

Moreover, when looking at patients who received prescriptions different from what the AI would have recommended, researchers found that the risk of mortality, sepsis, and bowel disease was significantly higher.

Researchers trained their AI algorithm on 10 years of medical record data of 5,913 preemies from the neonatal intensive care unit at Lucile Packard Children’s Hospital Stanford.The AI model was also validated using data from the University of California, San Francisco, US, (including 63,273 nutrition prescriptions from 3,417 patients) and found the model was good at predicting the nutrient needs of the population.

Using AI in hospitals 

The researchers want to run a randomized clinical trial next to see if the AI prescriptions really improve preterm health.

If the system is implemented, they want doctors and pharmacists to continue checking AI recommendations and change prescriptions if needed. 

“The AI recommendation is based on whatever information has been added to a patient’s electronic medical record, so if something is missing from the record, the recommendation won’t be accurate,” says Gaskari. “We need a clinician to look at it and review it.”

She adds that if the prescription has medical approval, one of the 15 standard nutrient formulas can be given to patients immediately. 

The researchers also believe standard formulas would make IV nutrition more accessible and less expensive, as they would not require a large expert team and compounding pharmacy to be involved.

“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,” Stevenson says. “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.”