Automated app coaching matched human diabetes prevention
Referring adults with prediabetes to a fully automated, app-based lifestyle program achieved similar 12-month health targets as referral to human coaching.
*Pragmatic noninferiority randomized clinical trial; Level 1b (OCEBM).
Citation
Mathioudakis N, Lalani B, Abusamaan MS, et al; AI-DPP Study Group. An artificial intelligence–powered lifestyle intervention vs human coaching in the Diabetes Prevention Program: a randomized clinical trial. JAMA. 2025;334(23):2079-2089. doi:10.1001/jama.2025.19563
Background
Prediabetes is common, but structured lifestyle programs are underused because of limited capacity and participation barriers. This trial tested whether an automated, app-based program could perform at least as well as standard remote human coaching.
Patients
368 adults (median age 58 years) with prediabetes and overweight or obesity at 2 US sites.
Intervention
Referral to a fully automated, app-based Diabetes Prevention Program with a Bluetooth scale.
Control
Referral to a remote, group-based, human coach–led Diabetes Prevention Program.
Outcome
(Primary) Composite at 12 months: maintained hemoglobin A1c below 6.5% throughout study and met ≥1 of: ≥5% weight loss; or ≥4% weight loss plus ≥150 weekly minutes of physical activity (measured by wearable); or hemoglobin A1c drop ≥0.2%.
Follow-up Period
12 months
Results
Primary analysis was intention-to-treat; participants missing 12-month data were counted as not meeting the outcome.
| Outcome |
Automated app referral |
Human-coach referral |
Difference |
| Primary composite outcome (primary) |
31.7% (58/183) |
31.9% (59/185) |
−0.2%; noninferior (margin −15%) |
| Program initiation |
93.4% |
82.7% |
+10.7%; NNT 10 |
| Program completion |
63.9% |
50.3% |
+13.6%; NNT 8 |
Limitations
Unblinded; used short-term targets rather than diabetes onset. Human coaching was remote (not in-person). Two sites with motivated participants may limit generalizability.
Funding
National Institutes of Health grants; intervention vendors paid; no funder role.
Clinical Application
When access to human coaching is limited, clinicians can refer eligible patients to a fully automated app-based program and expect similar 12-month results.
