March 12, 2026
Melissa Weber
A recently published scoping review study started with a seemingly simple question: If using AI in hospital settings to continuously monitor blood glucose levels offers better outcomes for diabetes patients, why aren’t more hospitals using AI?
In order to answer that question, study authors needed to answer another important question first: how do you define the use of AI in clinical practice to find the studies that offer best-practice solutions for patients? Scoping reviews require rigorous criteria to assure best practices are followed, clearly stating what terminology is used and how terms are measured.
Study co-author and HRS professor Emily Patterson, PhD enlisted the input of an engineering statistician co-investigator to assure the researchers were looking for the right words.
“We think we are one of the first groups to try to define AI for the purpose of reviewing medical studies,” Patterson said. While some users think of AI as large language models (LLMs), Patterson noted that such a limited perspective overlooks algorithms and neural nets used for images, for example.
“How do you conduct a literature review when people can't decide the definition of the words you’re reviewing?” she explained.
Their study, Artificial Intelligence Algorithms for Insulin Management and Hypoglycemia Prevention in Hospitalized Patients—A Scoping Review, benefitted from input by their interdisciplinary team. The librarian who searched medical study databases for which studies to include in their scoping review initially identified 13,768 citations. The team narrowed the scope to just 26 studies that fit their criteria. The result? They noted that machine learning models resulted in “strong predictive accuracy” for hypoglycemia risk, which is heightened for diabetes patients in hospitals.
“I was convinced it’s safer to have [AI-assisted monitoring] than not to have it,” Patterson said. “There are very few clinical indicators for hypoglycemia in some patients.”
Being able to identify patients who are at risk for a hypoglycemic reaction could prevent a host of complications. Hypoglycemia can cause severe cognitive dysfunction, falls, and cardiovascular stress, including arrhythmias. Without treatment, prolonged low blood sugar can lead to permanent brain damage and even death.
Patterson noted several possible reasons why clinicians are not more quickly adopting AI-managed continuous glucose monitors. While cost is one possibility, another might be that clinicians don’t believe the data from the trials and want to see first-hand how their patients might be helped. One way to improve adoption might be to implement monitor alarms. If nurses receive an alert from the AI monitor that indicates, “this patient is a risk,” they might be more open to adopting this new method.