Unlocking Equal Ground: AI-Driven DPP Learning from Real-World Trials
In a landmark study, researchers from Johns Hopkins Medicine and the Johns Hopkins Bloomberg School of Public Health report that an AI-powered lifestyle intervention app for prediabetes reduced diabetes risk at rates comparable to traditional, human-led programs. The work, funded by the National Institutes of Health and published in JAMA on Oct. 27, represents a potential turning point in how we scale diabetes prevention.
Prediabetes affects an estimated 97.6 million U.S. adults, a population at elevated risk of progressing to type 2 diabetes within five years. The classic CDC Diabetes Prevention Program (DPP) showed that structured lifestyle changes—diet, weight management, and physical activity—could significantly cut diabetes risk, with adults completing human-led programs 58% less likely to develop the disease. Yet, access barriers such as scheduling and staffing constraints have long limited reach. The new study explores whether an AI-led DPP can fill that gap.
The trial stands out as one of the first phase III randomized controlled trials to compare an AI-powered DPP app directly with a yearlong, group-based human-led program, using the CDC’s risk-reduction benchmarks as the yardstick. This is particularly timely as digital health options multiply, but robust evidence comparing AI-driven options to human coaching has been sparse.
The Study Design: Who Got what, and how we measured success
During the COVID-19 era, researchers enrolled 368 middle-aged adults (median age 58) who were referred to either one of four remote, 12-month human-led DPPs or an AI-driven reinforcement learning app. The app delivered personalized push notifications to guide weight management, physical activity, and nutrition. All participants continued standard medical care through their primary care providers and were screened to exclude other glucose-altering medications or structured programs.
Engagement tracking relied on a wrist activity monitor to measure physical activity for seven consecutive days each month over the 12 months. Importantly, researchers did not actively push participants toward the AI program beyond initial referral; engagement emerged from the program’s design and follow-up cadence.
Key Findings: Similar outcomes, with a boost in initiation and completion for AI
At the 12-month mark, the AI-DPP group and the human-led DPP group met the CDC-defined composite benchmark for diabetes risk reduction at nearly identical rates: 31.7% vs 31.9%, respectively. The benchmarks included at least 5% weight loss, or at least 4% weight loss plus 150 minutes of weekly physical activity, or an absolute A1C reduction of at least 0.2%—reflecting meaningful, clinically relevant improvements in risk profiles.
Beyond comparable outcomes, the AI-DPP demonstrated higher engagement. Initiation rates were 93.4% for the AI group compared with 82.7% for the traditional programs, and completion stood at 63.9% versus 50.3%. This suggests that AI-driven interventions can overcome some barriers that typically hinder entry into, or continuation of, long-term lifestyle programs.
What this means for clinicians and patients
Researchers emphasize that AI-led DPPs can extend reach, particularly for patients facing logistical hurdles such as scheduling, transportation, or time constraints. Fully automated AI systems can operate around the clock, reducing dependence on staff availability while still delivering personalized guidance tailored to individual goals and progress.
“The greatest barrier to DPP completion is initiation,” noted co-first author Benjamin Lalani. “An AI-driven option that patients can access at their convenience may help bridge that gap.” The results suggest primary care providers could consider recommending AI-led DPPs as an alternative or supplement to traditional programs for suitable patients, particularly when human resources are stretched thin.
Looking ahead: Real-world applications and ongoing research
The Johns Hopkins team is exploring how these AI app outcomes translate to broader, underserved populations who may struggle with time or resources. Secondary analyses will examine patient preferences between AI and human modalities, how engagement correlates with outcomes, and the cost implications of AI-led DPPs. While acknowledging the “black-box” concern often cited for AI in medicine, the study demonstrates that AI-driven interventions can offer reliable, personalized support with meaningful clinical benefits.
As healthcare systems seek scalable, effective strategies to curb diabetes incidence, AI-powered DPPs could become a standard option in the toolbox, complementing or even expanding on traditional, human-led programs.
