AI-Powered Diabetes Prevention Program Intervention Matches Human Coaching in Landmark Trial

An artificial intelligence (AI)-driven intervention based on the Diabetes Prevention Program (DPP) achieved outcomes comparable to traditional human-led programs in helping adults with prediabetes reduce their risk of progressing to diabetes, according to results from the first phase 3 randomized controlled trial to directly compare the 2 approaches.¹

The study, published October 27 in JAMA, found that 31.7% of participants referred to the AI-powered program and 31.9% referred to human coaching met CDC benchmarks for diabetes risk reduction at 12 months, a difference that met the prespecified criterion for noninferiority. Notably, the AI program demonstrated higher rates of both initiation (93.4% vs 82.7%) and completion (63.9% vs 50.3%) compared with traditional programs, according to the results.¹

"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, MD, MHS, principal investigator and comedical director of the Johns Hopkins Medicine Diabetes Prevention & Education Program, said in a Johns Hopkins statement.²

Study Design and Population

For the parallel-group, pragmatic noninferiority trial, conducted from October 2021 to December 2024, Mathioudakis and colleagues enrolled 368 middle-aged adults (median age 58 years) with prediabetes and overweight or obesity from 2 US clinical sites in Maryland and Pennsylvania. The cohort was 71% women; 61% self-identified as White, 27% as Black, and 6% as Hispanic.¹

Participants were randomly assigned to receive referral to either a fully automated AI-powered DPP mobile app powered by reinforcement learning algorithms or 1 of 4 CDC-recognized human coach-led programs. A critical component of the study, according to the authors, the research team did not promote engagement after referral, following up only at 6 and 12 months to assess real-world effectiveness.¹

Primary and Secondary Outcomes

The primary composite outcome required maintaining HbA1c below 6.5% throughout the study plus achieving at least one of the following:

• 5% weight loss
• 4% weight loss combined with 150 minutes of weekly moderate-to-vigorous physical activity
• Absolute HbA1c reduction of 0.2 percentage points

Clear Noninferiority for the AI-Based Program

The risk difference of −0.2% (one-sided 95% CI, −8.2%) fell well within the noninferiority margin of −15%, demonstrating the AI program was not worse than human coaching. Individual components of the composite endpoint showed similar directional consistency: 16.9% of AI participants versus 20.0% of human-coached participants achieved 5% weight loss, while 12.6% versus 12.4% achieved the combined weight loss and physical activity target.¹

"The greatest barrier to DPP completion is often initiation, hindered by logistical challenges like scheduling," study cofirst author Benjamin Lalanim now a medical student at Harvard Medical School, said in the statement.² "So, in addition to clinical outcomes, we were interested in learning whether participants were more likely to start the asynchronous digital program after referral."²

Access and Engagement Patterns

When Mathioudakis, Lalanim, et al reviewed engagement data, they observed distinct patterns between groups. Among AI participants who achieved the primary outcome, 74% were program completers compared with 56% in the human-led group.¹ “This suggests intervention exposure may play a more central role in the AI-led DPP, which depends on continuous self-directed engagement,” the team wrote. The structured coach-led formats, on the other hand, may allow participants to benefit with less concentrated time investment.¹

"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," Lalani said. "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."²

The study's 85.1% retention rate and alignment with expected community-based DPP outcomes (20% achieving 5% weight loss, within the typical 13.5%-35.5% range) support the validity of the findings and the adequacy of program delivery in both arms.¹

Implications for Clinical Practice

The findings carry particular significance given that only 3% of US adults with prediabetes currently participate in DPPs, despite an estimated 97.6 million Americans having the condition.³ With approximately 1,549 CDC-recognized DPP programs nationwide, which translates to 1 program per 63,000 adults with prediabetes, scalability remains a critical barrier. Further, the authors cite research showing that only 35% of those referred to a DPP ultimately participate. ⁴

Limitations and Future Directions

Among the study’s limitations, Mathioudakis and fellow investigators acknowledge the unmasked design, use of surrogate rather than diabetes incidence outcomes, and recruitment of motivated volunteers with high baseline physical activity levels and educational attainment from only two sites, limiting generalizability.¹

As AI continues to evolve, these findings suggest that fully automated behavioral interventions may offer a viable complement, though not necessarily a replacement, for traditional human-coached diabetes prevention programs, the authors said.¹

Looking ahead, the study team is interested in exploring how the AI app outcomes they observed translate to broader, underserved, real-world patient populations who may not have the time or resources to engage in traditional lifestyle intervention programs. They emphasize that primary care providers, in particular, may consider AI-led DPPs for patients in need of a lifestyle change program, especially those with considerable logistical constraints.

References

1. Mathioudakis N, Lalani, B, Abusamaan MS, et al. An AI-powered lifestyle intervention vs human coaching in the Diabetes Prevention Program: a randomized clinical trial. JAMA. Published online October 27, 2025. doi:10.1001/jama.2025.19563

2. AI-powered Diabetes Prevention Program shows similar benefits to those led by people. News release. Johns Hopkins Medicine. October 27, 2025. Accessed October 27, 2025. https://www.hopkinsmedicine.org/news/newsroom/news-releases/2025/10/ai-powered-diabetes-prevention-program-shows-similar-benefits-to-those-led-by-people

3. About prediabetes and type 2 diabetes. CDC National Diabetes Prevention Program. Updated May 15, 2024. Accessed October 27, 2025. https://www.cdc.gov/diabetes-prevention/about-prediabetes-type-2/index.html

4. Ackermann RT, O’Brien MJ. Evidence and challenges for translation and population impact of the Diabetes Prevention Program. Curr Diab Rep.
   2020;20(3):9. doi:10.1007/s11892-020-1293-4