AI-Powered Diabetes Precision Health: From Data to Action

Abstract

Diabetes has become a major public health challenge due to its high prevalence and chronic nature, with many individuals managing the condition for decades. The vast heterogeneity in diabetes necessitates personalized approaches to its prevention, diagnosis, treatment, and prognosis. Recently, the National Institute of Diabetes and Digestive and Kidney Diseases at the National Institutes of Health convened experts from the fields of diabetes and AI to identify and discuss existing gaps, as well as potentially transformative opportunities and actionable items enabled by recent advancements in AI. One prominent theme that has emerged from this discussion was the considerable potential of AI in Diabetes Precision Health, a field that warrants greater attention. The purpose of this article is to describe the opportunities and challenges identified during the workshop and outline potential strategies recommended by workshop attendees to advance this promising field.

Introduction

Chronic conditions such as diabetes and obesity are affecting people in the United States and globally at unprecedented rates. Without sustained and effective lifestyle changes, about 70% of people with prediabetes eventually develop diabetes. The transition typically occurs within 5 to 10 years, but it may be delayed or prevented. The extensive heterogeneity of this disease necessitates personalized prevention and treatment approaches. Recognizing the potential of AI, the U.S. National Institute of Diabetes and Digestive and Kidney Diseases at the National Institutes of Health (NIH) recently organized a workshop to address this.

Workshop attendees generally agreed that the transformative potential of AI is particularly significant in precision phenotyping; overcoming barriers in delivery and implementation (e.g., access to information, quality care, and language barriers); broadening unbiased data collection and representation of all populations; and learning the joint representation of multiple data modalities and thus obtaining a holistic picture of complex diseases.

A second critical issue lies in the gaps and limitations inherent in the data. Have we included enough modalities in phenotyping the broad disease spectrums to understand the underlying pathophysiology? Do we have sufficient data in any modality to capture the progression and heterogeneity of diabetes at both individual and population levels? Are the data AI-ready? And how to balance data sharing with data protection? One significant gap is data “missingness,” which manifests not only in data sparsity for certain data types (such as electronic health records) but also in biased representations and a lack of critical data types and modalities, such as longitudinal and lifestyle data.

Finally, the rapid rise in diabetes and obesity rates over recent decades, especially among younger populations, combined with the chronic, slow-progressing nature of diabetes and increasing cost of disease management, highlights the urgent need for more effective strategies in early detection, prevention, and long-term disease management.

Diabetes Precision Health

Precision medicine tailors medical care to individual variability by considering genetics, environment, and lifestyle factors. In contrast, precision health is an emerging field that places greater emphasis on disease prevention and early detection.¹ It is a broader concept that encompasses precision medicine while also incorporating approaches outside of traditional health care and medical settings. By monitoring an individual’s unique risk factors over time, precision health develops proactive and personalized solutions aimed at improving overall wellness, as well as addressing specific health issues.^2,3 Diabetes Precision Health (DPH) (Fig. 1) may still be a new term, though the concept of applying precision health strategies to diabetes care and management has been appreciated for some years.^4,5

Figure 1.

AI-Powered Diabetes Precision Health.

DPH and Diabetes Precision Medicine share some common objectives. However, there are differences in scope, primary goals, the role of AI, and several other features, as summarized in Table 1. DPH focuses more on early detection and prevention through proactive engagement of all individuals and the health care sector.

Table 1.

Comparison between Diabetes Precision Health and Precision Medicine.^*

Aspect of Comparison	Diabetes Precision Health	Diabetes Precision Medicine
Scope	Emphasizes prevention and early detection of disease, and promotion of overall wellness.	Focuses on diagnosis and treatment.
Target population	Both prediabetic and diabetic individuals focusing on risk reduction and sustainable health habits.	Mostly people with diagnosed diabetes who need personalized medication and management.
Primary goal	Prevention and holistic management of diabetes by optimizing lifestyle interventions that promote overall health.	Optimizing medications and therapies for diabetes based on molecular and clinical data of individuals.
Main approaches	Nonpharmacologic prevention and interventions, glucose prediction models, and personalized lifestyle coaching. Both individual and population-level risk factors are considered.	Pharmacological interventions and precision insulin therapy by matching the right drug or insulin regimen to the right patient.
Key technologies	Wearables (CGM, fitness trackers), digital twins for behavior modeling, AI-driven meal and exercise planning, and stress tracking.	Omics-based analysis (genomics, proteomics, metabolomics), AI-driven drug discovery, digital twins for therapy testing.
Key data sources	Glucose monitoring data; behavioral patterns; stress markers; physical activity, sleep, exercise, social, environmental, and genetic data.	Genetics, biomarkers, pharmacogenomics, multiomics, clinical history.
The role of AI	AI provides behavioral insights, personalized coaching, and real-time health feedback.	AI helps with drug response prediction, precision insulin dosing, and metabolic profiling.

^*Both aim to improve diabetes care through personalized strategies, but they focus on different aspects of disease management.

CGM denotes continuous glucose monitoring.

OPPORTUNITIES FOR AI IN DPH

Workshop attendees discussed several opportunities for AI to support a more proactive and individualized approach to diabetes prevention and management.

Conventional health care systems primarily interact with patients after diabetes onset, capturing only a limited snapshot of the disease’s natural progression. For example, emerging evidence suggests that sporadic hyperglycemia may occur decades before a diagnosis of diabetes — even before prediabetes. By involving individuals in continuous and longitudinal health monitoring and real-world data collection, a proactive approach might gather critical early-stage information and detect subtle health changes. It might also provide valuable insights into lifestyle factors and environmental influences that contribute to the onset or the prevention of chronic diseases such as diabetes.

Wearable technologies, already widely used by the public, have the potential to revolutionize physiological measurements, extending far beyond the static data points traditionally captured in clinical and research settings. These continuous, real-time data streams — such as minute-by-minute heart rate, physical activity, and interstitial glucose levels — are already being leveraged to estimate sleep quality, assess stress levels, and predict hypoglycemia. Although inherently noisy, these real-world data sources may provide insights into biological and pathophysiological processes that were previously unobservable.

Finally, AI is poised to play a uniquely powerful role in DPH. Many cutting-edge diabetes technologies, including continuous glucose monitors, insulin pumps, smart insulin pens, and hybrid closed-loop systems, are already powered by U.S. Food and Drug Administration–approved AI algorithms. In fact, AI-driven advancements in recent years have significantly improved glycemic control, reduced complications, and enhanced overall diabetes management and quality of life. Looking ahead, AI has the potential to extract meaningful insights from complex real-world data types related to human physiology and overall wellness, potentially moving diabetes management from a reactive model to a proactive, highly individualized approach that could revolutionize care and improve long-term outcomes.

These recommendations echo the general community interest in precision health, as reflected in several recent perspective articles in NEJM AI.^6–9

CHALLENGES

As an emerging field, precision health faces numerous challenges. In the realm of AI, challenges persist in integrating diverse data sources and in ensuring AI models meet the high standards required for health care applications. Some examples of technical issues include data leakage, where information from outside the training dataset slips into the training process, and overfitting, where the model memorizes the training data instead of learning general patterns. With the growing prevalence and power of AI, data protection is also an increasing concern. Robust privacy safeguards must be implemented to protect patient information and ensure ethical data collection and AI usage. Finally, given the unique scientific questions and data complexities in DPH, new innovative AI research and development are needed to address DPH-specific needs in causal relationship inference, predictive modeling, and personalized decision-making.

The implementation of AI in personal health monitoring also raises significant ethical and policy concerns. AI tools must be designed to ensure fair access for all individuals, regardless of socioeconomic status. Concerns have been raised that AI-generated health risk assessments could be misused by insurance companies to adjust premiums, potentially penalizing individuals based on predictive health insights. To prevent misuse, policies must be established to safeguard individuals’ rights, ensuring AI benefits enhance health outcomes without compromising privacy or increasing financial burdens.

Although developing effective strategies to overcome these challenges will require broad collaboration across the entire biomedical and AI community, workshop attendees recommended several key starting points. These include strengthening partnerships across academia, government, and industry to enhance data sharing and interoperability, particularly through wearable technologies, while safeguarding privacy; fostering cross-disciplinary initiatives, such as citizen science, to collect real-world evidence and promote community engagement; leveraging large biobank programs, such as All of Us; and encouraging federal agencies to take a more proactive role in supporting infrastructure development, education, and coordination of community efforts. Attendees also underscored that AI itself should be at the core of all strategic efforts.

Conclusion

The prevalence of diabetes and obesity has risen rapidly over the past several decades, becoming a significant public health challenge. These conditions are driven by complex and heterogeneous risk factors and disease subtypes, making effective prevention and management difficult. Reversing this trend will require engaging and empowering individuals to better understand and manage their personal risk. Achieving this goal will depend on individualized preventive care, and personalized decision support for health information, lifestyle choices, risk monitoring, and chronic condition management — ultimately reducing reliance on pharmaceutical interventions. This vision aligns with the new U.S. Department of Health and Human Services Secretary’s goal of “unleashing the potential in every one of us to make good personal choices,” as stated in his welcome remarks to staff (https://www.youtube.com/@HHS). Several recent efforts at the NIH reflect movement in this direction.^10,11 Combining AI with advanced technologies will be uniquely enabling and transformative for DPH, making personalized prevention the norm and enhancing diabetes management. This shift from reactive to proactive care will improve health outcomes and ease the burden on health care systems, leading to healthier populations and more sustainable health care.