American Diabetes Association 78th Scientific Sessions:

Diabetes Care Gets a Boost with Artificial Intelligence Technology

With Lisa Latts, MD, MSPH, MBA

The use of artificial intelligence (AI) and data capture caught the attention of many of the 14,000 attendees at the American Diabetes Association (ADA) 78th Scientific Sessions in Orlando, Florida. Given the desire to achieve precision medicine in diabetes education and treatment, the value of bringing interactivity with diabetes data directly to patients phones may meet a much-needed solution to optimizing meals and physical activity based on individual needs and behaviors.

Sugar.IQ app give real-time insights to insulin use for people with diabetes.

First-Ever Interactive App—Sugar.IQ—is Previewed

"We are excited about bringing artificial intelligence to the diabetes community, which we are demonstrating in a few different ways," said Lisa Latts, MD, MSPH, MBA, deputy chief health officer at IBM Watson Health who discussed the studies being presented during the meeting with EndocrineWeb.

The use of AI employs a computer system that gathers data, understands, reasons, and learns how the information effects outcomes over time. By understands, we mean that the system ingests massive amounts of information—both structured and unstructured data—such as databases that have categorical variables, health records, and the scientific lexicon in research papers all of which is ingested and processed to make connections and then draw conclusions through a process of learning over time to increase in accuracy, she said.

To demonstrate the application of IA technology, the Sugar.IQ diabetes assistant developed by IBM Watson Health in partnership with Medtronic for its investigational use in the Guardian Connect continuous glucose monitor,1 was unveiled at ADA.

This leading-edge interactive app employs cognitive computing and analytic technology developed into a first-ever artificial intelligence-driven algorithm designed to deliver personalized, simplified daily diabetes feedback to individuals.

This diabetes app leverages AI, advanced analytics, and diabetes technology to continually evaluate and update individuals about their glucose levels so they can make adjustments to their food intake, insulin dosages, daily routines, and other factors, said Huzefa Neemuchwala PhD, MBA at Medtronic Diabetes, in Los Angeles, California, presenting an overview of the Sugar.IQ functionality to attendees at an ADA session on Glucose monitoring—Advances, pitfalls, and clinical relevance.

Real-Time Insights Deliver Precision Medicine in Diabetes 

With artificial intelligence technology, Sugar.IQ sends insights to improve diabetes actions.

It can help turn difficult-to-determine patterns that impact glucose levels into meaningful, personalized insights to help people better understand the impact of lifestyle on their diabetes.1

“We have brought AI learning to diabetes care to foster better health outcomes with the introduction of the Sugar.IQ app that employs a food library that contains more than 1 million entries that permit patients to input accurate carbohydrate information into the bolus insulin calculator,” said Dr. Neemuchwala.

“From this volume of data, it becomes much easier for patients to understand the direct impact that meal choices and carb counting have on their insulin requirements. This will be true for patients with type 1 diabetes (T1D) and type 2 diabetes (T2D),” Dr. Latts told EndocrineWeb. 

"For example, I spoke to someone yesterday who has been trialing the Sugar.IQ app. She was trying to understand why her blood sugar rose when she drank coffee on some occasions but not always. The Sugar.-IQ was able to show her that her blood sugars were typically high on weekdays when she was stressed and rushing to get to work, whereas on the weekends when she was relaxed, her blood sugar remained fairly level." Dr. Latts said.

It wasn’t that she was adding different levels or types of sweeteners or cream/milk, as might have been the guess, rather the AI algorithm revealed a behavior that would otherwise have gone unrecognized.

The Sugar.IQ algorithm takes into account repetitive behaviors, user preferences and feedback based on trends to generate personalized, graduated prompts aimed to encourage improvements in care by learning the individual nuances of each user’s behavior and preferences to detect opportunities for improvement and create prompts to guide AI learning to essentially problem-solve for the patient.1

Features and Findings Suggest Benefits to Long-Term Diabetes Care

A key finding was the successful improvements seen in the variability of individual postprandial and post-activity glucose outcomes, suggesting that Sugar.IQ with AI learning may provide an effective, low-cost method for optimizing individual patient management through personalized insights and engagement prompts,according to Dr. Neemuchwala.

Features of the Sugar-IQ app include:2

  • Personalized insights that reveal time and day of behaviors related to glucose levels linked to bolus demands, rapid change in glucose levels such as post-insulin highs and lows, time-based actions, and meal content patterns.
  •  Motivational cues to encourage sustained positive actions.
  • Glucose prompts are created based on patterns and sent to the display panel.
  • Smart food log to make entry of meals and snacks easy and seamless.
  • Stores past glucose metrics to inform future behaviors.

According to patient feedback, while 80-85% of participants liked the prompts, full  acceptance increased over time and exposure to the device feedback, reaching 100% by day 18.1

The researchers indicated that the key finding was less variability in individual postprandial and post-activity glucose outcomes, suggesting that the improvements found following the use of the Sugar.IQ assistant provided an effective method for optimizing individual patient management thought personalized insights and engagement prompts.1

The app currently has the capacity to run on iOS mobile devices; apps for other smartphone platforms are in development. In addition, the Sugar.IQ app also will be integrated into the Guardian Connect CGM system (Medtronic) when it is launched in at the end of the summer 2018.2

First Head-to-Head Cardiovascular Drug Comparison Using AI

Until now, our knowledge of the cardioprotective benefits of certain diabetes medication classes has been limited to placebo-controlled trials, with little data on direct drug comparisons.

“The data we are presenting is the first to provide a head-to-head assessment of the cardioprotective effects of the different classes of diabetes drugs,” said Dr. Latts. Results of an observational cohort study pulling data from the Truven Health Marketscan research databases from 2013 to 2016.3

“We used machine-learning analytics, a type of AI, to compare the risks of cardiovascular events in patients taking sodium-glucose cotransporter-2 inhibitors (SGLT-2s), dipeptidyl peptidase-o4 inhibitors DPP-4s), and glucagon receptor agonists (GLP-1 RAs) against sulfonylureas to look at outcomes for hospitalization for congestive heart failure, stroke, acute myocardial infarction (MI), and a composite of all four indications together,” she said. The relative risks of events in the 12 months prior and following initiation of treatment was compared.3

“What we found was that SGLT-2s and DPP-4s demonstrated a cardioprotective effect, but surprisingly, the DPP-4s showed a slightly greater protective effectiveness  across all four outcome measures, [which differed from placebo-controlled trial findings,] while the SGLT-2s showed a cardioprotective effect for 3 of the four cardiovascular outcomes (but not for stroke),”  Dr. Latts told EndocrineWeb. Propensity score weighting and doubly robust methods including boosting and Poisson regression were used to control for baseline differences.3

"[Findings from ] this study found that machine learning tools may be used to identify potential therapeutic benefits of certain type 2 medication classes by comparing health outcomes reported in patients' electronic health records," said Dr. Latts.

Evaluating Treatment Persistence for T2D Using Artifical Intelligence 

We conducted a retrospective, observational study of 324,136 patients with newly diagnosed type 2 diabetes by examining health claims data from the Truven Health MarketScan Commerical an Medicare Supplemental Databases between 2013-2016,according to Dr. Latts, MD, deputy medical director at IBM Watson Health who co-authored the poster presented at the American Diabetes Association 78th Scientific Sessions in Orlando, Florida. The average age of the patient population was 55 years, and 46% were female, for which all treatment behaviors were followed for 12 months.

Looking at treatment discontinuation among individuals with at least one positive diagnosis of T2D in which a prescription was written for an antidiabetes medication in a real-world setting, 31% of patients stopped their medication within the first three months, and 44% by six months and more than half (58%) discontinued therapy (for at least 45 days) during the study period,4 said Dr. Latts. 

Of the patients who discontinued, 39% had reinitiated treatment at some point during the year with a mean of 107 days without any diabetes-related treatment.

“Since it appears that a majority of patients with type 2 diabetes discontinue their medication during the initial year of therapy, clinicians should initiate interventions aimed at improving treatment persistence to avoid gaps in therapy and reduce the potential for adverse clinical consequences,” said Dr. Latts told EndocrineWeb 

"The artificial intelligence integration will provide real-world context instantaneously, and this promises to improve outcomes for people with diabetes," Dr. Latts said. "Also, clinicians might appreciate that artificial intelligence is a tool that improves with human experiences, augmented by work with experts and clinical data to generate insights. The algorithm the capacity to understand patient data and make it relevant to individual lifestyles."

The authors indicating financial disclosures are Dr. Latts who is employed at IBM Watson and has received funding from Medtronic, Novo Nordisk, and Sanofi. Dr. Neemuchwala and Dr. Kaufman are employed at Medtronic.

Next Summary:
Is Metformin Obsolete? Debating its Future in Managing Type 2 Diabetes
Last updated on

close X