Researchers from Johns Hopkins Medicine and the Johns Hopkins Bloomberg School of Public Health have found that an AI-based lifestyle intervention app designed for diabetes prevention shows comparable benefits to traditional, human-led programs. This significant study, funded by the National Institutes of Health and published in JAMA on October 27, marks what is believed to be the first phase III randomized controlled clinical trial evaluating an AI-driven diabetes prevention program (DPP).
Approximately 97.6 million adults in the United States are affected by prediabetes, a condition characterized by elevated blood sugar levels that are not yet high enough to warrant a diabetes diagnosis. Those with prediabetes face an increased risk of developing type 2 diabetes within a five-year period. Earlier research indicates that participants in traditional human-led DPPs can reduce their risk of type 2 diabetes by 58% through lifestyle modifications in diet and exercise, as highlighted in the original DPP clinical study conducted by the Centers for Disease Control and Prevention (CDC).
Despite the proven efficacy of human-led programs, barriers such as scheduling conflicts and limited availability often hinder access. Out of nearly 100 CDC-recognized digital DPPs, AI-driven options represent only a small fraction, and evidence of their effectiveness compared to human-led programs has been scarce.
The study focused on whether a fully automated, AI-driven intervention could provide health benefits equivalent to those offered by year-long, group-based programs facilitated by human coaches. “There have been very few randomized controlled trials that directly compare AI-based, patient-directed interventions to traditional human standards of care,” stated Dr. Nestoras Mathioudakis, co-medical director of the Johns Hopkins Medicine Diabetes Prevention & Education Program and principal investigator of the study.
During the COVID-19 pandemic, 368 middle-aged participants were recruited to either join one of four remote, 12-month human-led programs or utilize an app that employed a reinforcement learning algorithm to deliver personalized notifications aimed at managing weight, physical activity, and nutrition. The participants were predominantly female (71%), with a racial composition of 61% white, 27% Black, and 6% Hispanic. All participants had a prediabetes diagnosis and met specific body mass index criteria.
The study utilized wrist activity monitors to track physical activity over seven consecutive days each month throughout the year-long study. Participants were allowed to continue receiving care from their primary healthcare providers but were prohibited from engaging in other structured diabetes programs or using medications that could influence blood glucose levels or body weight.
After 12 months, the results indicated that 31.7% of participants in the AI-driven DPP and 31.9% of those in the human-led DPP achieved the CDC-defined benchmark for diabetes risk reduction, which includes criteria such as a minimum of 5% weight loss or 150 minutes of physical activity per week. This suggests that both the AI and human-led programs yield similar outcomes in diabetes prevention.
Furthermore, the AI-DPP group demonstrated higher rates of program initiation at 93.4% compared to 82.7% in the traditional programs, and completion rates were also better at 63.9% versus 50.3%. These findings imply that the accessibility of AI interventions may enhance participant engagement, presenting AI-driven DPPs as a viable alternative for individuals requiring lifestyle change programs, particularly those facing logistical challenges.
“AI-DPPs can be fully automated and available at all times, making them resilient to access barriers that often affect human-led programs, such as staffing shortages,” remarked Benjamin Lalani, co-first author and medical student at Harvard Medical School. The research team aims to further investigate how the outcomes observed with the AI app can be applied to broader, underserved populations who may lack the time or resources for conventional lifestyle interventions.
Additionally, several secondary analyses are planned to examine patient preferences between AI and human-led modalities, the impact of engagement on outcomes, and the costs associated with AI-driven DPPs.
As part of the study, Sweetch Health, Ltd. and the participating DPPs received compensation for their services to participants. Importantly, the DPPs did not have access to the overall results of the cohort and did not analyze the data.
The study received funding from the National Institute of Diabetes and Digestive and Kidney Diseases and the National Institute on Aging, with additional support from the Johns Hopkins Institute for Clinical and Translational Research.
Other researchers involved in the study include Mohammed S. Abusamaan, Defne Alver, Adrian Dobs, John McGready, Kristin Riekert, Benjamin Ringham, Aliyah Shehadeh, Fatmata Vandi, Amal A. Wanigatunga, Daniel Zade, and Nisa M. Maruthur from Johns Hopkins, along with Brian Kane from Tower Health Medical Group Family Medicine and Mary Alderfer from Reading Hospital Tower Health.
The study”s DOI is 10.1001/jama.2025.19563.
