Machine Learning Powers CDS Tool for Diabetes Management

17Jun - by aiuniverse - 0 - In Machine Learning


June 16, 2020 – A clinical decision support system that leverages machine learning techniques could help patients control their glucose levels and enhance type 1 diabetes management, according to a study published in Nature Metabolism.

People with type 1 diabetes do not produce their own insulin, so they have to take it continuously throughout the day using an insulin pump or with multiple daily injections. Dosing errors can result in life-threatening hypoglycemia events and hyperglycemia, which increases an individual’s risk of neuropathy, retinopathy, and nephropathy.

Additionally, because patients with type 1 diabetes can typically go three to six months between appointments with their endocrinologist, they can be at high risk of these dangerous complications if their glucose levels rise too high or fall too low.

To improve diabetes management, researchers from Oregon Health & Science University (OHSU) leveraged a machine learning algorithm that could generate insulin injection recommendations.

The team trained the algorithm using over 50,000 glucose observations. The algorithm was trained to identify causes of hypoglycemia or hyperglycemia and determine necessary insulin adjustments from a set of 12 potential recommendations. When paired with a smartphone app called DailyDose, the recommendations from the algorithm were shown to be in agreement with physicians 67.9 percent of the time.

Researchers then validated the system by monitoring 16 people with type 1 diabetes over the course of four weeks, showing that the model can help reduce hypoglycemia.

“Our system design is unique,” said lead author Nichole Tyler, an MD-PhD student in the OHSU School of Medicine. “We designed the AI algorithm entirely using a mathematical simulator, and yet when the algorithm was validated on real-world data from people with type 1 diabetes at OHSU, it generated recommendations that were highly similar to recommendations from endocrinologists.”

The researchers noted that their study advances previous findings on using machine learning tools to help patients manage glucose levels.

“There are other published algorithms on this, but not a lot of clinical studies,” said Peter Jacobs, PhD, associate professor of biomedical engineering in the OHSU School of Medicine and senior author on the study.

“Very few have shown a statistically relevant outcome – and most do not compare algorithm recommendations with those of a physician. In addition to showing improvement in glucose control, our algorithm-generated recommendations that had very high correlation with physician recommendations with over 99 percent of the algorithm’s recommendations delivered across 100 weeks of patient testing considered safe by physicians.” 

Investigators have recognized the potential for artificial intelligence to improve diabetes management. A 2019 study from Rensselaer Polytechnic Institute leveraged AI and big data analytics to evaluate information from thousands of glucose monitors and insulin pumps. The team will use the data to enhance the algorithms that control these devices, resulting in better quality of life for people with type 1 diabetes.

“If we look at hundreds of people we can say, ‘Oh, certain problems occur more often in this age group, this type of population, or with this particular type of sensor,’” said Wayne Bequette, professor of chemical and biological engineering at Rensselaer Polytechnic Institute.

“If, for example, you find that it’s more likely that people 8 to 12 years old have these types of irregularities, then you can account for that in your algorithm, and provide more personalized control while reducing burden.”

Going forward, the OHSU team will continue to refine and develop the clinical decision support tool to further improve patients’ management of type 1 diabetes.

“We have plans over the next several years to run several larger trials over eight and then 12 weeks and to compare DailyDose with other insulin treatment strategies, including automated insulin delivery,” said co-author Jessica Castle, MD, associate professor of medicine (endocrinology, diabetes and clinical nutrition) in the OHSU School of Medicine.

Facebook Comments