In a groundbreaking study poised to revolutionize the management of prediabetes, researchers from Johns Hopkins Medicine and the Johns Hopkins Bloomberg School of Public Health have demonstrated that an artificial intelligence (AI)-powered lifestyle intervention application can reduce diabetes risk in adults with prediabetes at rates comparable to those achieved by traditional, human-led diabetes prevention programs (DPPs). This research, supported by the National Institutes of Health and published in JAMA on October 27, marks a pivotal advancement in digital health interventions, emphasizing the transformative potential of AI in chronic disease prevention.
Prediabetes, a condition characterized by elevated blood glucose levels below the diagnostic threshold of type 2 diabetes, affects an estimated 97.6 million adults in the United States alone. Without intervention, individuals with prediabetes face a significantly heightened risk of progressing to type 2 diabetes within five years, a trajectory associated with increased morbidity and healthcare expenses. Historically, human-led DPPs have served as the cornerstone for mitigating this risk by facilitating lifestyle modifications in diet and physical activity, demonstrated to reduce progression to diabetes by roughly 58% according to foundational CDC clinical studies. Despite their efficacy, these programs often face logistical and accessibility challenges, limiting widespread participation.
This landmark study sought to interrogate whether AI-driven digital DPPs could surmount these barriers by providing personalized, scalable interventions without compromising clinical effectiveness. Notably, while approximately a hundred CDC-recognized digital DPPs exist, AI-powered models constitute only a small minority, and robust clinical data juxtaposing their efficacy against traditional human-coached programs have been conspicuously absent until now.
In a rigorously designed, phase III randomized controlled trial, 368 middle-aged adults diagnosed with prediabetes and meeting specific overweight or obesity criteria were enrolled during the COVID-19 pandemic. Participants were randomized to receive either one of four remote human-led DPPs or access to an AI-based reinforcement learning algorithm delivered via a mobile application. This AI platform deployed dynamically personalized push notifications tailored to encourage adherence to weight management, physical activity, and nutritional recommendations. The demographic profile of participants reflected diversity, encompassing 61% White, 27% Black, and 6% Hispanic individuals, with a median age of 58 years.
To quantitatively monitor physical activity, all participants wore wrist accelerometers intermittently, providing objective data throughout the year-long intervention. The trial excluded confounding variables such as concurrent enrollments in other structured diabetes programs or the use of medications modifying glucose metabolism or body weight, ensuring the observed effects were attributable solely to the respective interventions. Follow-up assessments were conducted at six and twelve months post-enrollment without enforced engagement strategies to authentically capture naturalistic adherence patterns.
Remarkably, the AI-driven DPP not only equaled the human-led programs in facilitating diabetes risk reduction benchmarks as defined by the CDC—achieving composite endpoints including ≥5% weight loss, or combined ≥4% weight loss with ≥150 minutes per week of physical activity, or a reduction in HbA1c by ≥0.2%—but also exceeded them in participant initiation and completion rates. Specifically, 31.7% of AI-DPP participants met the composite risk reduction endpoint, closely mirroring the 31.9% achievement in the human-led cohort. However, program initiation was significantly higher in the AI group (93.4%) compared to traditional programs (82.7%), and completion rates also favored AI interventions (63.9% versus 50.3%).
These findings posit that AI-driven interventions can effectively address common barriers such as scheduling conflicts and limited program availability, which often hinder engagement in human-coached DPPs. The always-on, fully automated nature of AI programs offers continuous access irrespective of resource limitations like staffing shortages, conferring a scalable solution for broad public health implementation. This study hypothetically establishes a new paradigm whereby AI applications can deliver reliable, personalized health coaching with a consistency previously unattainable in standard clinical settings.
The investigator team, led by Nestoras Mathioudakis, M.D., M.H.S., articulated the novelty of this endeavor, underscoring the paucity of clinical trials directly comparing AI-based, patient-directed interventions against established human-led standards of care. They emphasized that despite concerns regarding the opaqueness often associated with AI (“black-box” phenomena), this study provides empirical evidence affirming that AI methodologies can yield tangible, clinically meaningful outcomes in diabetes prevention.
Looking forward, the research collective aims to extend these findings by exploring real-world application in underserved populations who face disproportionate barriers to diabetes prevention. Concurrent secondary analyses are underway to parse patient preferences relating to modality (AI versus human coaching), to assess the relationship between program engagement and health outcomes, and to elucidate the economic implications of adopting AI-led DPPs at scale.
While the study included collaborations with Sweetch Health, Ltd. and participating DPP providers, the integrity of data analysis and interpretation rested solely with the research team, assuring unbiased results. Notably, Johns Hopkins University and affiliated researchers maintain transparency with conflict-of-interest disclosures, further upholding scientific rigor.
This study epitomizes a pivotal advance at the intersection of artificial intelligence and preventive medicine, offering a scalable, accessible, and efficacious alternative for diabetes risk reduction. Its implications extend beyond diabetes prevention, suggesting broader applications of AI-driven behavioral interventions for chronic disease management in resource-constrained healthcare landscapes.
Subject of Research: AI-powered lifestyle intervention applications for diabetes prevention
Article Title: AI-Powered Lifestyle Intervention App Matches Effectiveness of Traditional Programs in Diabetes Prevention, Johns Hopkins Study Finds
News Publication Date: October 27, 2025
Web References:
- JAMA Article
- CDC Prediabetes Information
- National DPP Overview
References:
10.1001/jama.2025.19563 (Original Study DOI)
Keywords: Artificial intelligence, Diabetes prevention, Prediabetes, Digital health intervention, Lifestyle modification, Chronic disease management
