Source/Disclosures
Disclosures: Weinzimer reports he has received grants from the Helmsley Charitable Trust, grants and personal fees from Medtronic and personal fees from Eli Lilly, Insulet, Sanofi, Tandem and Zealand. Please see the study for all other authors’ relevant financial disclosures.
September 21, 2020
3 min read
Save

AI-based decision support tool optimizes insulin use for youths with type 1 diabetes

Source/Disclosures
Disclosures: Weinzimer reports he has received grants from the Helmsley Charitable Trust, grants and personal fees from Medtronic and personal fees from Eli Lilly, Insulet, Sanofi, Tandem and Zealand. Please see the study for all other authors’ relevant financial disclosures.
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

An automated decision support tool for optimizing insulin pump settings was noninferior to intensive insulin titration provided by physicians from specialized academic diabetes centers, according to data from a randomized controlled trial.

“Despite the use of advanced pumps, which deliver insulin doses precisely and reliably, and continuous glucose monitoring, which reports glucose levels every 5 minutes, the majority of people with type 1 diabetes don’t reach recommended glucose targets, for many reasons,” Stuart A. Weinzimer, MD, professor of pediatrics at Yale University School of Medicine and attending endocrinologist at Yale New Haven Children’s Hospital, told Healio. “The most common reason that diabetes control remains suboptimal is insulin needs can vary widely depending on the meal, activity level and a host of other circumstances. Automated closed-loop systems improve time in desired glucose ranges and overall control, but even they are not perfect. Furthermore, access to these systems at the present time is limited, either due to financial constraints, insurance constraints or reluctance by the practitioner or user to adopt these systems.”

Stuart A. Weinzimer, MD, professor of pediatrics at Yale University School of Medicine and attending endocrinologist at Yale New Haven Children’s Hospital.

AI-supported insulin doses

Weinzimer and colleagues assessed whether frequent insulin dose adjustments guided by an automated artificial intelligence-based decision support system — the DreaMed Advisor Pro — is as effective and safe as those guided by physicians in controlling glucose levels.

The decision support tool consists of two parts: a communications system that pulls in all the data from the insulin pump, meal size and timing, and CGM data uploaded by the user from Glooko; and an algorithm, which is responsible for analyzing all of the data and making the specific recommendations to improve diabetes control.

“The algorithm uses artificial intelligence that detects and analyzes glucose patterns and insulin dosing events in a similar approach to that used by a health care provider based on expert knowledge, recommendations and data acquired from various clinical studies,” Weinzimer said. “The recommendations themselves may involve increasing or decreasing meal insulin doses, correction doses or basal doses, or may focus instead on other behavioral aspects of diabetes management, such as regularity of activities or timing of dosing.”

For the study, researchers analyzed data from 108 children and young adults (aged 10 to 21 years) with type 1 diabetes using insulin pump therapy. Researchers randomly assigned participants 1:1 to remote insulin dose adjustment every 3 weeks guided by either the decision support system (n = 54) or by physicians (n = 54) for 6 months. The findings were published in a letter in Nature Medicine.

Researchers found that the percentage of time spent within the target glucose range of 70 mg/dL to 180 mg/dL for participants in the decision support tool arm was statistically noninferior compared with participants in the physician arm (mean, 50.2% vs. 51.6, respectively; P < 1 × 107). Similarly, the percentage of readings below 54 mg/dL among participants in the decision support tool arm was statistically noninferior compared with participants in the physician arm (mean, 1.3% vs. 1%, respectively; P < .0001).

Three severe adverse events related to diabetes (two severe hypoglycemia events; one diabetic ketoacidosis event) were reported in the physician arm and none in the decision support tool.

There were no between-arm differences in the percentage of time above and below glucose target ranges. Mean total daily insulin and daily basal insulin doses were not statistically different between groups. Daily bolus insulin doses among participants in the decision support arm were higher vs. the physician arm (mean, 29.6 vs. 26.6; P = .03).

“I was not terribly surprised by the findings so much as I was a little humbled, rather like a chess master who can’t defeat the computer,” Weinzimer said. “If anything was surprising, it was that the engineers who designed the DreaMed Advisor Pro managed to create an algorithm that indeed ‘thinks’ like a diabetes expert, considering multiple variables and unknowns and making sound decisions.”

Reducing disease burden

Weinzimer said data from the study show that the support tool is as safe and effective as world-renowned diabetes specialists in managing type 1 diabetes.

“This by no means replaces the provider, but instead frees the provider to spend valuable time on education, counseling and other important facets of care, instead of the mundanities of dose adjustment,” Weinzimer said. “It will also empower those providers who may have not previously had the experience or comfort level with continuous glucose sensors and insulin pumps, to offer these technologies to their patients and use DreaMed Advisor as a tool to optimize their care.”

The system could also be used in tandem with telemedicine approaches to deliver expert knowledge and reduce disease burden for patients and to alleviate burdens on caregivers, he said.

“They can also lower the costs associated with clinical visits and overcome missed visits, particularly among people who live in rural areas, or during other circumstances, like COVID-19, under which access to face-to-face visits with physicians is limited,” Weinzimer said.

For more information:

Stuart A. Weinzimer, MD, can be reached at email: stuart.weinzimer@yale.edu.