Summary: In a phase 3 pragmatic noninferiority trial, referral to a fully automated, AI-led Diabetes Prevention Program was noninferior to referral to a human coach-led program for a 12-month composite of weight loss, physical activity and HbA1c outcomes in adults with prediabetes and overweight or obesity.
PICO Summary
| Element | Detail |
|---|---|
| Population | 368 adults with prediabetes and overweight or obesity at two US clinical sites. |
| Intervention | Referral to a fully automated AI-led Diabetes Prevention Program delivered through a mobile app and Bluetooth-enabled scale for 12 months. |
| Comparator | Referral to a remotely delivered human coach-led Diabetes Prevention Program for 12 months. |
| Outcomes | The composite outcome was achieved by 31.7% with AI versus 31.9% with human coaching; risk difference -0.2 percentage points (one-sided 95% CI -8.2%), meeting the prespecified noninferiority criterion. Program initiation was 93.4% versus 82.7%. |
| Study design | Phase 3, parallel-group, pragmatic, randomized noninferiority clinical trial. |
AI vs Human Diabetes Prevention
Phase 3 noninferiority RCT - 12 months
A fully automated AI-led Diabetes Prevention Program was noninferior to human coaching for a 12-month composite of weight, activity and HbA1c outcomes in adults with prediabetes and overweight or obesity.
Expert Commentary
This trial tests a service-delivery question rather than a new drug: can an automated digital Diabetes Prevention Program perform comparably to human coaching? Among adults with prediabetes and overweight or obesity, the 12-month composite outcome was achieved by 31.7% in the AI group and 31.9% in the human-coaching group. The one-sided confidence interval met the prespecified noninferiority margin. More participants initiated the AI referral, although initiation is not the same as sustained engagement or clinical benefit.
Can I use this with my patients? An automated program can be a reasonable access option for patients who prefer digital support, have compatible devices and can engage without frequent human contact. It should expand choice rather than displace coaching for people who need accountability, language support, behavioral health care or help overcoming social barriers. The study was conducted at two US sites, used referral rather than intensive clinician management, and was not designed to compare long-term diabetes incidence, cost or equity across health systems. Clinicians should confirm that the platform is evidence-based, protects data, supports escalation and provides a route to human care when needed. The result supports noninferiority for this composite trial outcome, not equivalence for every patient or every prevention service.
References
Mathioudakis N, Lalani B, Abusamaan MS, Alderfer M, Alver D, Dobs A, et al. An AI-Powered Lifestyle Intervention vs Human Coaching in the Diabetes Prevention Program: A Randomized Clinical Trial. JAMA. 2025. doi:10.1001/jama.2025.19563. PMID: 41144242. ClinicalTrials.gov: NCT05056376.