Supported by the National Institutes of Health, this study is in its inaugural phase III randomized controlled clinical trial that shows an AI-driven diabetes prevention program (DPP) app assists patients in achieving diabetes risk-reduction targets set by the Centers for Disease Control and Prevention (CDC) at rates similar to those found in programs led by humans.
It's estimated that approximately 97.6 million adults in the United States have prediabetes, a condition characterized by blood sugar levels that are elevated but not high enough to be classified as type 2 diabetes. This diagnosis means these patients are at increased risk of developing type 2 diabetes within the next five years.
Prior studies have indicated that adults with prediabetes who participate in a human-led DPP, which assists individuals in making dietary and exercise modifications, are 58 % less likely to progress to type 2 diabetes, as evidenced by the original clinical study conducted by the CDC on the Diabetes Prevention Program (DPP).
Nonetheless, barriers to access, including scheduling issues and availability, have restricted the outreach of these programs. Among the nearly 100 digital DPPs acknowledged by the CDC, AI-driven DPPs constitute merely a small fraction, and there is a deficiency of data illustrating their effectiveness in comparison to programs led by humans.
In this research, the investigators examined whether a completely AI-operated program could offer adults with prediabetes health advantages comparable to those provided by yearlong, group-oriented programs facilitated by human coaches.
Even beyond diabetes prevention research, there have been very few randomized controlled trials that directly compare AI-based, patient-directed interventions to traditional human standards of care.
Nestoras Mathioudakis, M.D., M.H.S., Co-Medical Director, Diabetes Prevention & Education Program, Johns Hopkins Medicine
During the COVID-19 pandemic, 368 middle-aged participants (with a median age of 58 years) volunteered to be assigned to one of four remote, 12-month, human-led programs or to a reinforcement learning algorithm application that provided personalized push notifications to assist with weight management behaviors, physical activity, and nutrition.
All participants in the study met the race-specific body mass index cutoffs for being overweight or obese and had received a diagnosis of prediabetes before its commencement.
In both groups, a wrist activity monitor was used to monitor the physical activity of participants for seven consecutive days each month throughout the 12-month study.
During their participation, study volunteers maintained their medical care with their primary care providers. However, they were prohibited from engaging in other organized diabetes programs or using medications that could influence glucose levels or body weight, including metformin or GLP-1 agonists.
After the referral, the researchers did not encourage participation in the program and only conducted follow-ups with both groups at six and 12-month intervals.
The greatest barrier to DPP completion is often initiation, hindered by logistical challenges like scheduling. So, in addition to clinical outcomes, we were interested in learning whether participants were more likely to start the asynchronous digital program after referral.
Benjamin Lalani, Study Co-First Author and Medical Student, Harvard Medical School
After 12 months, the research team discovered that 31.7 % of participants in the AI-DPP and 31.9 % of those in the human-led DPP achieved the CDC-defined composite benchmark for diabetes risk reduction.
This benchmark includes at least a 5 % weight loss, a minimum of 4 % weight loss combined with 150 minutes of physical activity weekly, or a reduction in absolute A1C of at least 0.2 %.
The findings indicated that comparable results can be attained through both a human coach-based program and an AI-DPP. Additionally, the AI-DPP cohort exhibited higher rates of program initiation (93.4 % compared to 82.7 %) and completion (63.9 % versus 50.3 %) when contrasted with traditional programs.
The researchers believe improved accessibility has increased participant involvement in the AI group, indicating that AI interventions might serve as a viable substitute for traditional human-coached programs. Consequently, primary care providers might contemplate implementing AI-led Diabetes Prevention Programs (DPPs) for patients requiring a lifestyle modification program, particularly those facing significant logistical challenges.
Unlike human-coached programs, AI-DPPs can be fully automated and always available, extending their reach and making them resistant to factors that may limit access to human DPPs, like staffing shortages. So, while the black-box nature of AI is a commonly cited barrier to clinical adoption, our study shows that the AI-DPP can provide reliable personalized interventions.
Benjamin Lalani, Study Co-First Author and Medical Student, Harvard Medical School
Looking forward, the team aims to investigate how the outcomes of the AI app they have observed can be applied to larger, underserved real-world patient populations who may lack the time or resources to participate in conventional lifestyle intervention programs.
Several secondary analyses are currently in progress, which aim to examine patient preferences between AI and human modalities, the effects of engagement on outcomes in each intervention, and the costs related to AI-led DPPs.
Journal Reference:
Mathioudakis, N., et al. (2025) An AI-Powered Lifestyle Intervention vs Human Coaching in the Diabetes Prevention Program. JAMA. doi.org//10.1001/jama.2025.19563