AI system linked to diabetes drug de-escalation

An artificial intelligence (AI)-enabled system combining wearable sensors, continuous glucose monitoring, and telecoaching with trained, licensed providers enabled most patients with type 2 diabetes to reach target glycemic goals and reduced glucose-lowering medications in a randomized trial, according to a study published in NEJM Catalyst Innovations in Care Delivery.

The study not only highlighted the potential of AI-guided lifestyle intervention tools to improve future diabetes care, it also helped to dispel certain preconceived notions researchers had about patients' ability to use these AI tools successfully.

“One of the most important things I learned through this study is that health care providers should not let their own personal biases impact what opportunities or treatments they offer patients,” wrote lead study researcher Kevin M. Pantalone, DO, of Cleveland Clinic. “There were many times when I felt a potential subject was not going to be capable of navigating the technology, using the phone application, etc, but time and time again, many of those patients proved me wrong by doing fantastic and achieving astounding results.”

Study Design

Researchers conducted a single-center, unblinded randomized controlled trial at Cleveland Clinic Family Health Center in Twinsburg, Ohio, enrolling 150 adult patients with type 2 diabetes and a body mass index (BMI) of at least 27. Participants were randomly assigned in a 2:1 ratio to the intervention (n = 100) or usual care (n = 50).

Eligible patients were aged 18 to 75 years and met at least one glycemic criterion: hemoglobin A1c (HbA1c) 7.5%–11%; HbA1c 6.5%–7.5% while taking a glucose-lowering medication; or HbA1c below 6.5% while receiving at least one nonmetformin glucose-lowering medication [with or without metformin].

Baseline characteristics included median age 58.5 years, median diabetes duration of nine years, median BMI of 35.1, median HbA1c 7.2%, and the use of two glucose-lowering medications.

The intervention group used the Twin Precision Treatment system (Twin Precision Treatment; Twin Health, Inc., Mountain View, California), which integrates continuous glucose monitoring, an activity sensor, a smart scale, and a blood pressure meter with an AI-driven smartphone application. The system generated personalized recommendations related to diet, physical activity, sleep, and deep breathing, reinforced by telecoaching sessions with trained and licensed providers, nurses, and certified health coaches.

Usual care patients continued diabetes management under their primary care physicians without study-directed influence on management choices.

Primary and Secondary Outcomes

The primary endpoint of the study was the proportion of patients who achieved an HbA1c below 6.5% without glucose-lowering medications other than metformin at 12 months. Overall, 71% of patients assigned to the intervention group reached that hemoglobin A1c target without glucose-lowering medications other than metformin, compared with 2% in the usual care group.

A secondary endpoint requiring sustained target HbA1c for at least 90 days before 12 months was achieved by 53% of intervention patients vs 3% of usual care patients.

Mean HbA1c decreased by 1.3% in the intervention group compared to 0.3% in the usual care. Body weight declined among the two groups by 9% and 5%, respectively.

In a post hoc analysis, reductions in medication use were observed across multiple drug classes in the intervention group over the 12-month period. Use of glucagon-like peptide 1 receptor agonists declined from 41% to 6%, dipeptidyl peptidase 4 inhibitors from 33% to 3%, sodium-dependent glucose cotransporter 2 inhibitors from 27% to 1%, sulfonylureas from 31% to 17%, and insulin from 24% to 13%. Medication use increased or changed little in the usual care group.

Quality-of-life and treatment satisfaction score also improved significantly in the intervention group but not in the usual care group.

Adverse events occurred in 36% of intervention patients and 34% of usual care patients, with three serious events in the intervention group and one in the usual care group.

Subgroup and Post Hoc Findings

Treatment effects were consistent across baseline HbA1c eligibility categories, including patients with HbA1c 7.5%–11%, those with HbA1c 6.5%–7.5% receiving medication, and those with HbA1c below 6.5% taking nonmetformin therapies.

In a post hoc analysis, the median number of glucose-lowering medications in the intervention group decreased from three at baseline to one at 12 months, while the median number remained two in the usual care group.

Limitations and Future Implications

Researchers noted several limitations, including the single-center design, the one-year follow-up period, and limited data on social determinants of health metrics that could also limit the generalizability of the findings. Participants also needed smartphone access and a willingness to use digital tools.

Researchers concluded that the AI-enabled system improved glycemic control and enabled medication de-escalation in patients with type 2 diabetes receiving primary care management.

“The finding that 88% of participants in the intervention group remained engaged with the system intervention for 12 months subjectively affirms favorable usability and adaptation of this technology,” they wrote.

"Leveraging wearable technology and synthesizing that data using AI can make a huge impact on one’s health, particularly through positively influencing patients’ decisions in real-time.  These types of platforms are going to be a very important component of chronic disease management for the foreseeable future," Dr. Pantalone added.

Disclosures: Dr Pantalone reported consulting relationships with Bayer, Boehringer Ingelheim, Corcept Therapeutics, Diasome, Eli Lilly, Merck, Novo Nordisk, and Sanofi; speaker honoraria from AstraZeneca, Corcept Therapeutics, and Novo Nordisk; and institutional research support from Bayer, Novo Nordisk, and Twin Health. Dr. Pantalone also reported a patent application related to hyperglycemia management.

The study was funded by Twin Health. Several co-researchers reported consulting relationships with pharmaceutical companies, research funding to their institutions, or employment and equity holdings with Twin Health.

AACE Endocrine AI is published by Conexiant under a license arrangement with the American Association of Clinical Endocrinology, Inc. (AACE^®). The ideas and opinions expressed in AACE Endocrine AI do not necessarily reflect those of Conexiant or AACE. For more information, see Policies.