Mathioudakis AI coaching (DPP)

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

Doente / População Intervenção / Exposição Comparação Desfecho

In 368 adults with prediabetes and overweight or obesity at two US sites, referral to a fully automated AI-led Diabetes Prevention Program delivered by app and Bluetooth scale was non-inferior to referral to a remote human coach-led DPP for a 12-month composite of weight loss, physical activity and HbA1c.

N
368
Desenho
Phase 3, pragmatic, parallel-group non-inferiority RCT, 2 sites, USA
Desfecho
Composite at 12 months: HbA1c <6.5% throughout plus ≥5% weight loss, or ≥4% weight loss plus ≥150 min/week physical activity, or HbA1c reduction ≥0.2 percentage points
Relevância
2
ResultadoPrimary composite 58/183 (31.7%) AI-led vs 59/185 (31.9%) human-led (risk difference −0.2%, 1-sided 95% CI lower bound −8.2%; non-inferiority margin −15%). Programme initiation 93.4% vs 82.7%.
Mathioudakis N, et al. JAMA. 2025;334(23):2079-2089. 10.1001/jama.2025.19563
Discussão e crítica

Tests whether an AI can replace the human coach in the best-evidenced diabetes-prevention model — and finds the two referral pathways equivalent, with higher uptake of the automated programme. The non-inferiority margin of −15% is generous, fewer than a third of either arm met the composite, and as a referral-based pragmatic design it measures pathways rather than coaching per se.