AI-Driven Lifestyle Interventions for Reducing Diabetes Risk: Systematic Review.

In 30 seconds

This systematic review assessed AI-driven digital lifestyle interventions for people with prediabetes or elevated diabetes risk, including healthy participants in feasibility studies. Eleven studies were included, but only one was randomized. In that trial, an AI-delivered Diabetes Prevention Program was noninferior to human coach-led delivery for the combined 12-month outcome (23/183, 12.6% vs 23/185, 12.4%; risk difference 0.1%, 1-sided 95% CI -5.5%). Initiation and completion were higher with AI, without superior clinical outcomes. Other comparative and uncontrolled evidence did not establish clinical benefit or causal effects.

Key findings

• The randomized trial found noninferiority for the combined 12-month outcome: 23/183, 12.6% vs 23/185, 12.4%; risk difference 0.1%, 1-sided 95% CI -5.5%.
• AI-program initiation was 171/183, 93.4% vs 153/185, 82.7%; completion was 117/183, 63.9% vs 93/185, 50.3%, with no superior clinical outcomes.
• A nonrandomized comparison found no advantage of AI-personalized over static messaging for BMI or physical activity.
• The other studies reported uncontrolled pre-post changes, which do not support causal inference.

Why it matters

The review suggests that a specific AI-delivered prevention program may achieve clinical outcomes comparable to human coaching, while engaging more participants. However, evidence for clinical benefit beyond standard programs remains limited, and findings cannot establish that AI itself caused improvements. Distinguishing engagement outcomes from glycemic and other clinical outcomes is important when evaluating these tools for diabetes prevention.

What to keep in mind

Only one included study was randomized; the remaining evidence was nonrandomized or uncontrolled. Designs and outcome measures were heterogeneous, preventing statistical pooling, and evidence was insufficient to assess differences by population or delivery setting.