Type 2 Diabetes Pharmacotherapy De-Escalation Through AI-Enabled Lifestyle Modifications: A Randomized Clinical Trial

Optimal glycemic control in type 2 diabetes (T2D) remains an elusive goal despite an expanding range of effective treatments and technologies. Sustained, disciplined adherence to lifestyle modifications leads to positive outcomes in T2D, although converting medical recommendations into effective patient action is challenging in real-world practice. The emergence of AI and machine learning (ML) presents innovative opportunities to prompt effective lifestyle modifications and diabetes goal achievements in highly user-specific and user-friendly ways. The Twin Precision Treatment system is an AI-enabled bundled system of sensors and coaching that generates personalized recommendations through wearable Bluetooth-enabled technologies (including continuous glucose monitors), targeted laboratory data, Internet of Things, AI-ML algorithms, and human input. The authors designed a study to explore whether or not this bundled system intervention (INT) could help individuals — with T2D and managed in a primary care setting — achieve glycemic targets while concurrently de-escalating glucose-lowering medications. This single-center trial randomly assigned 150 adults with T2D and a body mass index (the weight in kilograms divided by the square of the height in meters) ≥27 to the INT (N=100) or a usual care (UC) control group (N=50). The primary end point was a hemoglobin A1c level (HbA1c) <6.5% (<48 mmol/mol; “target”) without glucose-lowering medications, except metformin, at 12 months; a secondary end point was the same outcome sustained for ≥90 days prior to 12 months. Other secondary end points included target attainment without any glucose-lowering medications at 12 months and ≥90 days prior to 12 months, as well as HbA1c and weight changes at 12 months. The primary end point was achieved by 71.0% (95% confidence interval [CI], 60.1 to 80.0) of INT participants versus 2.4% (95% CI, 0.5 to 11.6) of UC participants (P<0.001). Significantly more INT participants than UC participants achieved or sustained target ≥90 days prior to 12 months without glucose-lowering medications, except metformin (52.5% vs. 2.8%; P<0.001). The mean changes in HbA1c levels (−1.3% vs. −0.3%; P<0.001) and body weight (−8.6% vs. −4.6%; P<0.001) were significantly greater in the INT group than in the UC group. Other secondary end points were not significantly different. In post hoc analysis, overall use of glucose-lowering pharmacotherapy decreased markedly in the INT group but not the UC group. Quality-of-life and treatment satisfaction scores improved significantly from baseline in the INT group but not in the UC group (exploratory analyses). Overall, the AI-enabled bundled system of sensors and coaching facilitated significant improvements in glycemic control, weight loss, and quality of life versus UC, while allowing marked de-escalation of glucose-lowering pharmacotherapy.