The study shows that artificial intelligence can improve intravenous nutrition for premature children

Artificial intelligence has been shown to improve intravenous nutrition for premature children. The study, which he will publish on March 25, is one of the first to show how the AI ​​algorithm can enable doctors to make better clinical decisions for patients with newborns.

The algorithm uses information in electronic preemies medical documentation to predict which nutrients need and in which quantities. The AI ​​tool has been trained in the range of data from almost 80,000 previous prescriptions for intravenous nutrition, which was related to information about how tiny patients were doing.

Scientists say that the use of AI to rewrite IV Nutrition can reduce medical errors, save time and money and facilitate care for preemiejs in low -range conditions. The IV Nutrition, also known as complete parenteral nutrition, is the only way to feed preemomets that are born from their digestive systems, are mature enough to absorb nutrients.

At the moment we come up with a TPN prescription for each child, individually, each day. We do it from scratch and deliver them to them. Complete parenteral nutrition is the largest source of medical error in newborns of intensive care, both in the United States and around the world. “

Dr. Nima Angry, Elder author of studies, Extraordinary professor of anesthesiology, perioperative doctor and pain and pediatrics

Not only is it susceptible to process errors, but it makes it difficult to know if they have obtained the right formula. There is no blood test to measure whether Preemie received, for example, the right number of calories, and unlike full -time children, preemie do not necessarily cry when they are hungry and show satisfaction when they are full.

“Nutrition is one of the areas of intensive neonatal care in which we are the weakest,” said research co -author David Stevenson, MD, neonatologist and professor Harold K. Faber in pediatrics.

“We can’t familiarize what the bearing is doing,” he said.

Slow process

About 10% of children are born prematurely, which means at least three weeks before the deadlines. Infants born over eight weeks earlier are not ready to absorb nutrients by the intestines and require IV feeding. In addition, some of the prioritations experience complications of early delivery and require nutrition IV during intestinal treatment.

Currently, nutrition IV is prescribed every day for these patients individually. Patients need nutritional macronutrients, molecular building building blocks of protein, fat and carbohydrates; microelements such as vitamins, minerals and electrolytes; And drugs such as heparin, which are added to preparation IV to reduce the risk of blood clots. Current prescriptions are based on such factors as the weight of the child, the stage of development and the results of work in the laboratory.

The provision of these prescriptions requires the contribution of six experts cooperating with each other over a multiple process: a neonatologist or pharmacist writes any prescription that is checked by a dietitian for the composition of nutrients and the second pharmacist in terms of safety. The prescription goes to a complex pharmacy, in which it is prepared, and then to the intensive care unit of newborns, where one nurse gives IV and the other nurse twice subordinate, which each patient receives the correct preparation.

“This is a high-risk drug because it is a mix of many different things,” said the co-author of the study Shabnam Gaskari, Pharmd, executive director and Pharmacy Director at Stanford Medicine Children’s Health. “If we produced, ready to use TPN, it would be very beneficial. I think it would be safer for patients.”

Towards standard formulas

Scientists wondered if they could use artificial intelligence to ensure hospitals produced, ready to use the formula of nutrients.

Their AI algorithm has been trained on 10 years of electronic data from medical documentation from tumor intensive care in Lucile Packard Children’s Hospital Stanford, including 79,790 prescriptions for nutrition IV from 5,913 premature patients. The algorithm also had access to information about the medical results of patients, enabling it to find subtle designs that combined the level of nutrients with the health of infants. Although doctors did not always have the right prescription exactly, the amount of data helped overcome this problem, enabling the algorithm to learn in a general way about what works for children in various medical situations.

“It’s AI’s strength: sometimes imperfect data is good enough, if you have a lot,” said Aghaeepour.

After training in a decade of patients, the algorithm grouped similar prescriptions for nutrients to determine how many standard designs would meet the nutritional needs of all patients and what happened to them.

“We wondered: What if we create three standard formulas, 10 or 100?” Aghaeepour said. “It turns out that with 15 separate nutrition patterns IV, what you recommend is quite similar to what doctors, pharmacists and dietitians would do. But then these 15 formulas based on AI can be used to significantly improve speed and safety.”

In addition, scientists have shown that the AI ​​algorithm can use data from patients’ electronic medical documentation to predict which of 15 designs may need every child, and can adapt the recommendations every day when patients have grown and their medical condition. So the algorithm may recommend that a specific child needs formula No. 8 for five days, and then formula No. 3 for a week, and then formula No. 14 for a few days and so on.

To test how this approach was arranged with real prescriptions, the research team created a test for 10 neonatologists: doctors showed clinical information for previous patients, as well as nutrition prescriptions they actually received, and the prescriptions of the algorithm would recommend the algorithm. Doctors were not said which recipe was which; They were asked, which they consider better. Doctors consistently preferred prescriptions generated by AI from real prescriptions.

Scientists also used artificial intelligence to scan electronic medical documentation from previous patients, looking for cases in which the actual recipe of the patient’s nutrition differed from what AI would recommend. From these patients, the risk of mortality, sepsis and intestinal diseases was much higher than in patients whose prescriptions suited what AI would recommend.

The team also approved the AI ​​model using real data from the University of California in San Francisco (including 63 273 nutrition prescriptions from 3,417 patients) and stated that the model did a good job in predicting the needs of nutrients for this population.

Steps to implement

The next step will be a randomized clinical trial in which some patients receive prescriptions for nutrients using a manual method, others receive recipe prescriptions AI, and scientists see how every group is going on.

Assuming that the system is implemented, the team plans to make doctors and pharmacists continue to check AI recommendations and if necessary adapt the prescriptions.

“The AI ​​recommendation is based on all information added to the patient’s electronic medical documentation, so if something is missing in the register, the recommendation will not be accurate,” said Gaskari. “We need a clinician to look at and view it.”

But after approving the medical prescription, one of the 15 standard formulas of nutrients, kept on a hospital shelf, can be immediately handed over to the patient.

Using standard formulas would make intravenous nutrition more accessible and cheaper, because it would no longer require a large engaged team of experts or access to a complex pharmacy. This can bring benefits for hospitals in countries with lower income or in other low resolution conditions.

“This reflects our hope of how AI will strengthen medicine: what he intends to improve and make doctors and make the highest quality care,” said Stevenson. “I hope that it will give our doctors more time to do things that computers cannot do, for example, spending time with children and their families, listening to them and ensuring comfort and assurance.”

Scientists from the University of Southern California Keck School of Medicine and Children’s Hospital of Los Angeles have contributed to research.

These works were supported by the National Institutes of Health (Grant R35GM138353), National Center for Advancing Translational Sciences (Grant Ul1Tr001872), Eunice Fund Kennedy Shriver National Institute of Child Health and Human Development (Grant R42HD115517),), Burrough Wellcome Fund. The Institute through the Stanford translation program, Stanford High Impact Technology Fund and Stanford Biodesign. The project was also supported by the National Center for Advancing Translational Sciences, the National Institutes of Health, through the UCSF Clinical & Translational Science Institute.