JMIR Publications

ABSTRACT

Background:

Prediabetes affects approximately one-third of US adults; however, participation in standard Diabetes Prevention Programs (DPPs) remains low due to barriers of access, scheduling, and program intensity. AI-driven lifestyle interventions have emerged as potentially scalable approaches to expanding diabetes prevention, but their clinical effectiveness relative to standard DPPs is unclear.

Objective:

To evaluate the effectiveness of AI-driven lifestyle interventions on clinical, behavioral, and engagement outcomes, and to assess whether effects differ by AI type, population, or delivery setting.

Methods:

PubMed, Embase, CINAHL, Web of Science, and IEEE Xplore were searched for peer-reviewed studies published from January 2015 through February 2026. Eligible studies included RCTs, quasi-experimental, and observational designs enrolling participants with prediabetes, elevated metabolic risk, or healthy participants in feasibility studies, evaluating a digital lifestyle intervention with an AI-driven adaptive or personalized component. Records were screened using Covidence; study quality was assessed using NHLBI tools.

Results:

Of 4,891 records identified, 11 studies met inclusion criteria: 1 RCT (9%), 7 observational studies (64%), and 3 before-and-after studies (27%), representing 13 to 16,327 participants. Interventions included AI lifestyle coaching systems (64%), predictive AI platforms (27%), and conversational chatbots (9%), predominantly via mobile health applications. The single RCT demonstrated noninferiority of the AI-DPP vs a human coach–led DPP at 12 months (combined ≥4% weight loss and ≥150 min/week physical activity: 12.6% vs 12.4%; RD: 0.1%; 1-sided 95% CI: −5.5%), with higher program initiation (93.4% vs 82.7%) and completion (63.9% vs 50.3%) in the AI arm. Across non-randomized studies, descriptive improvements in weight (1.5%–5.3% reduction), glycemic markers, physical activity, and engagement were reported; however, absent control groups limit causal inference. Effects appeared broadly consistent across populations and settings, with stronger benefits among higher-risk individuals.

Conclusions:

Based on a limited evidence base, AI-driven lifestyle interventions show early promise for diabetes prevention. The single available RCT demonstrated noninferiority to a human coach–led DPP; but replication is needed before definitive conclusions can be drawn. Descriptive evidence suggests improvements in behavioral and engagement outcomes, supporting their potential role as scalable complements rather than replacements for standard DPPs, though causal inference is not possible.