An AI-Powered Lifestyle Intervention vs Human Coaching in the Diabetes Prevention Program: A Randomized Clinical Trial.

Summary

In a 12-month pragmatic noninferiority RCT (n=368), referral to a fully automated AI-led DPP achieved the primary composite outcome in 31.7% vs 31.9% with human-led DPP, meeting the prespecified noninferiority margin. Program initiation was higher in the AI group (93.4% vs 82.7%), and results were consistent across composite components and sensitivity analyses.

Key Findings

• Primary composite outcome achieved in 31.7% (AI-led) vs 31.9% (human-led); noninferiority met with 1-sided 95% CI lower boundary of the risk difference at -8.2% (margin -15%).
• Program initiation after referral was higher with AI-led DPP (93.4%) than with human-led DPP (82.7%).
• Findings were consistent across composite components (weight loss, HbA1c reduction, physical activity) and in sensitivity analyses.

Clinical Implications

Health systems can consider referring eligible adults with prediabetes to AI-led DPPs to expand access without compromising effectiveness, potentially reducing costs and workforce burden while maintaining weight, HbA1c, and activity outcomes.

Limitations

• Conducted at two US sites, which may limit generalizability
• Open-label referral comparison with 12-month follow-up; intervention delivery was external to the study team

Future Directions

Evaluate long-term diabetes incidence, cost-effectiveness, and equity impacts of AI-led DPP at scale across diverse health systems and populations.