Enterprise AI Analysis

Introduction

Chronic diseases lead to significant morbidity and mortality, and account for the highest amounts of healthcare spending worldwide. Artificial Intelligence (AI) is currently positioned to transform clinical medicine, where AI tools derived for working with Machine Learning (ML), Deep Learning (DL), and more recently Large Language Models (LLMs) algorithms, are being used to analyze complex datasets to predict individual health risks accurately, provide early diagnosis, and deliver personalized interventions. Although significant developments have been made to promote transparency and completeness in the reporting of clinical trials involving AI interventions, numerous challenges still hinder the adoption of AI tools to enhance chronic disease management. Amid growing recognition of the capabilities of generative LLM-AI tools, such as the Generative Pre-Trained Transformer (GPT), in predicting, diagnosing, and managing chronic diseases by the general public, it becomes increasingly essential to review the current evidence, challenges, and future directions of AI in the management of chronic diseases.

This scoping review attempts to summarize the current evidence on the use of AI in the early prediction, detection, and management of chronic diseases. This review followed the Standards for Reporting Narrative Reviews (SANRA) to ensure transparency and methodological rigor. The literature search covered databases such as PubMed, Scopus, and Google Scholar to access relevant articles on the topic published from January 2015 to September 2025. Sources were thematically synthesized rather than quantitatively pooled. The SANRA checklist was used to assess quality and comprehensiveness across six criteria. We describe how AI tools are being utilized for health risk prediction, screening, and, in some cases, even for the reversal or management of chronic diseases. It then reviews the quality control issues governing randomized controlled trials (RCTs) involving AI interventions, followed by the strengths and limitations of existing AI approaches, the challenges that hinder the successful adoption of AI in the management of chronic diseases, its integration into clinical workflows, and finally, the economics and implementation (or lack thereof) of AI in chronic disease prevention, detection, and management.

Methodology

This review was conducted as a scoping review to map and synthesize existing evidence on how artificial intelligence (AI), including machine learning (ML), deep learning (DL), and large language models (LLMs), is being used to enable or amplify behavior change within lifestyle and integrative medicine frameworks. The approach was informed by guidance for scoping reviews (Arksey & O'Malley; PRISMA-ScR) and adhered to the Standards for Reporting Narrative Reviews (SANRA) to ensure transparency and reproducibility.

To identify relevant literature, we searched PubMed, Scopus, Google Scholar, and PsycINFO for studies published between January 2015 and September 2025. Search terms included combinations of “artificial intelligence,” “machine learning,” “behavior change,” “lifestyle medicine,” “integrative medicine,” “digital health,” “LLMs,” “coaching,” “nudging,” and “personalization.” Reference lists of included studies and recent reviews were screened to identify additional sources. Eligible studies described (1) AI-enabled interventions targeting at least one pillar of lifestyle or integrative medicine (nutrition, physical activity, sleep, stress management, social connection, or purpose) or (2) AI systems influencing behavioral mechanisms such as motivation, adherence, or self-efficacy. Editorials, non-English publications, and conference abstracts without data were excluded.

Data were charted and synthesized qualitatively to characterize themes, conceptual models, and implementation considerations; no formal effect-size pooling or meta-analysis was performed. Consistent with scoping-review methodology, the objective was to map the breadth of evidence and identify gaps, rather than evaluate intervention effectiveness. SANRA criteria were applied to support structured reporting of aims, search methods, data handling, and synthesis.

Burden of Chronic Diseases Globally

For several decades, chronic diseases such as cancer, diabetes, heart disease, and stroke have been major causes of morbidity and mortality worldwide. A report from the World Health Organization revealed that chronic diseases are the leading cause of death globally. These diseases not only lead to death but also account for nearly half of the global burden of disability and ill health, representing about 40–45% of all disability-adjusted life years (DALYs). According to the WHO, chronic diseases caused 43 million deaths in 2021, accounting for 75% of all deaths globally. Approximately 33% of adults globally live with either one or more than one chronic disease or multiple chronic conditions (MCC). In developed nations, the situation is even worse, as every three in four adults have either one chronic disease or MCC.” This situation is only expected to worsen, leading to a disproportionate health and economic burden in the years to come. The estimated cost of chronic disease is estimated to reach $47 trillion globally by 2030.

The burden of chronic diseases not just affects individuals, but also families and communities as they overwhelm health systems and increase financial burdens. A staggering 75% of deaths occurring from chronic diseases occur due to modifiable risk factors such as tobacco use, lack of physical activity, unhealthy diets, and excessive alcohol. Many of the chronic conditions, such as diabetes, heart disease, and stroke, are preventable. However, investments in prevention remain vastly underfunded compared to those in treatment. This is true both from a perspective of social determinants of health and lifestyle. With this in mind, investing in AI interventions for early detection, screening, and personalized interventions could be an effective way to reduce the burden of chronic diseases in the years to come.

Why is AI Promising in Prevention and Management?

The use of AI in the field of medicine was first conceived in the 1970s, nearly one and a half decades after the term Artificial Intelligence (AI) was coined. Artificial intelligence (AI) is technology that enables computers and machines to simulate human learning, comprehension, problem-solving, decision-making, creativity, and autonomy. Since then, AI has and continues to transform the field of medicine. It has transformed health promotion by improving early detection and screening of chronic diseases while also encouraging healthy lifestyle modifications, providing personalized health interventions, and ultimately reducing the financial or economic burden on individuals as well as health systems. This is evidenced by the fact that today, AI (encompassing ML, DL, and LLMs) is being used to predict individual health risks, detect diseases early, and deliver responsive, personalized care interventions for many chronic diseases, such as diabetes, cardiovascular diseases, hypertension, and others.

To clarify how different AI systems contribute to health behavior interventions, Table 2 compares major AI modalities by their data inputs, clinical use cases, strengths, and current limitations.

Comparative Overview of AI Modalities

Modality	Typical Data Inputs	Clinical Use Case	Strengths	Limitations
Machine Learning (ML)	Structured EHR, labs	Risk prediction	Fast, interpretable	Limited by feature design
Deep Learning (DL)	Imaging, ECGs	Diagnostics	High accuracy	Low transparency
Large Language Models (LLMs)	Text, conversations	Coaching, education	Accessible, empathetic	Hallucinations, bias
Multimodal AI	Combined (EHR + imaging + text)	Prognostics, precision care	Holistic view	Complex validation

Diabetes: Risk Prediction, Digital Twins, Glucose Monitoring Support

Cardiovascular Disease: Early Detection, Personalized Risk Calculators

In addition to diabetes, AI techniques have demonstrated tremendous potential in detecting high-risk cardiovascular disease at an early stage. This is evident from a recent study, which applied AI models trained on EHRs and developed a transformer-based model (TRisk) to predict the 10-year risk of cardiovascular disease in heterogeneous primary care settings. The study demonstrated that the novel TRisk model significantly outperformed the widely recommended benchmark models used in identifying individuals at risk of cardiovascular disease. Additionally, AI-enabled electrocardiography has been shown to screen for left ventricular systolic dysfunction, outperforming existing screening methods that require prolonged monitoring. This highlights the potential of AI in enabling discussions on developing a rapid and inexpensive method to identify atrial fibrillation.

Hypertension, Obesity, Metabolic Syndrome: AI-Driven Monitoring & Feedback

AI has also been found to help manage hypertension, obesity, and metabolic syndrome. For example, metabolic syndrome (MetS) – a cluster of interrelated risk factors such as hypertension, insulin resistance, and central obesity increases the likelihood of cardiovascular diseases and type 2 diabetes. As a result, early identification of hypertension is vital for offering timely intervention and disease management. A study by Asoka et al. (2025) found that a hybrid ML model integrating XGBoost classification with K-means clustering strengthens hypertension prediction and identifies patients based on metabolic risk factors. In another study by Samal et al. (2024), the management of hypertension was found to be improved through the use of a computerized clinical decision support (CDS) intervention, leading to better clinical outcomes for patients with chronic kidney disease. AI and ML have emerged as powerful tools for predicting and managing obesity risk, benefiting both patients and healthcare providers. For example, from the healthcare provider's perspective, AI/ML-driven interventions help gain real-time data from EMRs, wearables, and health apps and provide customized treatment plans. On the other hand, from the patient's perspective, AI/ML-driven interventions offering personalized coaching are proving effective in obesity management. The findings of these studies highlight the potential of AI/ML-driven decision support systems in detecting hypertension, managing MetS, and treating patients with obesity.

Table 1 shows AI models across chronic diseases use data from EHRs, wearables, and glucose monitors to predict risk, personalize care, and improve outcomes. These approaches demonstrate measurable benefits in diabetes, cardiovascular disease, hypertension, obesity, and metabolic syndrome.

Summary of AI Approaches and Evidence Across Major Chronic Disease Conditions

Condition	AI Approach	Data Source	Key Function	Main Findings	References
Diabetes	Digital twin modeling, ML	CGM, EHR, lifestyle data	Predicts glycemic patterns; supports personalized nutrition	Improved HbA1c and reduced medication use	Joshi 2023; Shamanna 2020
Cardiovascular disease	Transformer model (TRisk)	EHRs	10-year risk prediction	Outperformed pooled cohort equations	Rao 2025
Hypertension	Clinical decision support	EHR + BP monitoring	Medication adjustment, patient alerts	Improved BP control in CKD patients	Samal 2024
Obesity	AI-coaching and ML prediction	Wearables, apps	Personalized activity and diet feedback	Higher adherence and weight reduction	Huang 2025
Metabolic syndrome	Hybrid XGBoost + K-means	Lab + anthropometric data	Early risk stratification	Identified clusters of high-risk patients	Asoka 2025

What RCTs, Meta-Analyses, and Real-World Studies Show So Far

The number of randomized controlled trials (RCTs) on AI in clinical practice has grown over the years. Narrative and systematic reviews examining the design, reporting standards, risk of bias, and inclusivity report positive primary endpoints, particularly in relation to diagnostic yield or performance. However, the dominance of single-centre trials, the scarcity of demographic reports, and varying reports of operational efficiency have limited the generalizability and practicality of the results of these RCTs. A statistically significant result does not inherently equate to clinical importance. Many interventions show p-values below conventional thresholds, yet the observed differences are too small to translate into tangible improvements in patient health or everyday functioning.

Additionally, many of the published RCTs have been found to show high variability in adherence to reporting standards, risk of bias, and a lack of representation of minority groups. It is mainly due to these reasons that there has been a growing recognition that all AI-related interventions should undergo rigorous, prospective evaluation if they are to demonstrate an impact on health outcomes. Recent developments in guidelines, such as the Consolidated Standards of Reporting Trials-Artificial Intelligence (CONSORT-AI) and the Standard Protocol Items: Recommendations for Interventional Trials-Artificial Intelligence (SPIRIT-AI), extend the minimum guidelines for reporting randomized trials, and provide clear instructions for use, improve transparency, and ensure completeness in reporting clinical trials for AI interventions.

Strengths & Limitations of Existing Approaches

AI is reshaping clinical medicine by presenting novel solutions to the diagnosis, treatment, and management of patients with chronic diseases. Existing approaches of AI systems, such as ML, Deep Learning (DL), and Natural Language Processing (NLP), enable AI systems to process vast amounts of data more efficiently than human beings. As such, a significant strength of existing approaches in AI is that they allow healthcare providers to make efficient decisions based on data. They are now able to conduct more accurate diagnoses, offer precision in treatment, and improve overall patient care. For example, a significant breakthrough in AI-driven medicine is the application of Convolutional Neural Networks (CNNs) in medical imaging, which provides accurate diagnostic functionality. At the same time, NLP algorithms have been found to provide valuable insights from unstructured data, including EHRs, medical literature, and even physician notes.

One of the more significant developments in recent years is the emergence of Large Language Model (LLM) tools, such as ChatGPT, Qwen, Llama, Gemini, and Claude. These developments are significant because the general public is widely using these tools for self-care. A scoping review synthesizing the potential uses of LLMs in patient education and engagement revealed that these LLM tools were found to provide accurate responses to patients' questions and even translate medical information into patient-friendly language, as well as enhance existing educational materials. More specifically, in chronic disease management, these tools were effective in providing personalized guidance and lifestyle coaching, thereby reducing the gap between clinical visits and continuous care.

The Potential of LLMs in Transforming Chronic Disease Management

LLM tools demonstrate promising potential for enhancing chronic disease management across patient-centered tasks. For example, a scoping review investigating the applications of LLMs, such as ChatGPT and Llama, in chronic disease management revealed that patients used these tools to obtain personalized health advice and support symptom monitoring in diabetes management, as well as to access mental health support and for treatment coordination, patient education, and assess individual risk profiles for different chronic conditions such as blood pressure, sickle cell anaemia, and even bowel disease management. In mental health support, these tools were used for emotional support, facilitating therapeutic conversations, and providing personalized recommendations. The emergence of sophisticated models, such as ChatGPT, Gemini, Llama, and Qwen, has fueled research examining the transformative impact these LLMs have on various aspects of healthcare, ranging from enhancing patient care to providing personalized mental health support and assisting in the management of chronic conditions.

The use of ChatGPT as a Virtual Health Coach

Known for its advanced NLP capabilities, numerous studies have examined the role of ChatGPT in healthcare, especially its role in serving as a virtual health coach in managing chronic diseases. For example, quasi-experimental studies investigating the use of ChatGPT as a virtual health coach, conducted by Alanezi (2024) and AI-Anezi (2024), revealed that patients were using ChatGPT for continuous learning and education on health topics. The studies also showed that ChatGPT assisted patients in managing chronic diseases by helping them modify their behavior (e.g., adopting healthy diets and engaging in physical activity), and provided emotional support by offering empathetic responses and a non-judgmental space for patients with mental health conditions. The potential of ChatGPT in encouraging patient engagement and adherence to chronic disease management has also led to its integration in real-world EHRs to support healthcare professionals in diagnosing diseases as well as crafting personalized treatment regimes. This integration is effective in addressing primary health concerns and serves as a vital resource for patients, particularly in low-resource settings and remote areas. That said, leaders at OpenAI (makers of ChatGPT) have warned against using the tool for health and therapy purposes, and it has not been trained specifically for these use cases.

The Use of Gemini Advanced in Early Disease Detection and Medical Decision-Making

Although ChatGPT has demonstrated promising potential in providing mental health support and enhancing chronic disease management, Google's Gemini Advanced has also shown promise in facilitating early disease detection and supporting medical decision-making through data analytics. A comprehensive review of literature examining the impact of Gemini as a virtual doctor in medical services by Irshad and colleagues revealed that Gemini holds may attempt to act as a “virtual doctor” by offering instant access to patient information, individualized learning experiences, personalized patient interaction, and providing up-to-date information to support clinical decision-making. In cardiovascular health, Gemini has also been found to enhance patient education by providing personalized, round-the-clock access to information, thereby improving health literacy. For example, Gemini may be useful for creating patient education guides on cardiac health, sleep, and dietary habits, reshaping how information is delivered and managed, and is increasingly used to enhance patient education. However, incident reports for Google's earlier healthcare model warrant concern regarding faulty output. As well recent attempts, namely the Verily ME app (by Google's parent company, Alphabet) have come under fire by scholars and experts in the field, for illustrating the dangers of releasing a consumer product too early, without the necessary guardrails for privacy and without demonstrated efficacy or effectiveness of the model itself or the user experience.

Given the increasing use of LLMs by the general public, early studies suggest that the potential benefits these tools offer include increased accessibility and efficiency. However, these tools also have their own limitations. For example, the application of LLMs in medical chatbots raises questions about safety, reliability, and ethical implications. Additionally, significant challenges persist regarding the use of LLMs in medicine. These include concerns related to algorithmic bias and broader issues such as regulatory uncertainty, data privacy, and a lack of ethical considerations.

Integration Into Care Workflows

AI is also becoming an integral part of the healthcare workflow. A review of RCTs by Han and colleagues revealed that AI systems, when integrated with healthcare workflows, have the potential to either streamline or complicate workflows. As a result, the successful integration of AI tools into workflows largely depends on operational efficiency, cost-effectiveness, and the training of the workforce. Today, AI is being integrated into healthcare workflows in various tangible forms, such as medical devices, and has been found to improve diagnostic precision, treatment personalization, and patient care outcomes by analyzing large, complex medical datasets, including EHRs. However, the use of EHRs has raised concerns about data privacy, algorithmic bias, and other issues, thereby increasing barriers to the widespread integration of AI into healthcare workflows. The integration of AI into healthcare workflows is still in its early stages. Still, it has been found to have the potential to transform the work of healthcare personnel, particularly in reducing the burden of clinical documentation in EHRs. One critical development in this regard is the development of ambient AI scribes, which use ML to facilitate scribe-like capabilities in real-time. Ambient AI scribes may reduce the documentation burden, thereby enhancing the physician-patient relationship and indirectly fostering clinicians' preventive care capabilities. Reported incidents have indicated that AI scribes are still subject to error.

The Role of Multi-Modal AI and Patient-Centered Integration

Due to the advantages that AI may offer in preventing, diagnosing, and managing diseases, the research on the trajectory of AI application in medicine has shifted from singular models to examining the role of using multi-modal AI. Multi-modal AI refers to models that integrate heterogeneous or multiple data sources, such as images, texts, EHRs, and other data types, and present a multidimensional perspective on patient health to enhance diagnostic accuracy, patient treatment, and management of various medical conditions. A review by Jandoubi and Akhloufi examining the performance of comprehensive and up-to-date ML and DL-based multi-modal AI systems in various diagnostic tasks revealed that multi-modal AI approaches outperformed unimodal systems that rely solely on radiological images, clinical notes, or physiological signals in terms of diagnostic performance. Similarly, a review conducted by Schouten and colleagues examining multi-modal AI approaches across various medical disciplines found that these approaches outperformed their unimodal counterparts in providing accurate diagnoses. In addition to offering superior diagnostic accuracy and clinical decision-making capabilities over unimodal AI approaches, multi-modal AI approaches have also been found to help clinicians focus more on patient care as the AI approaches automate routine tasks. These approaches have shown potential to improve diagnosis and provide personalized treatment for patients. However, despite the exponential growth in interest in multi-modal AI approaches in recent years, these approaches are currently largely limited to a research context.

Ethics, Privacy, and Risk Considerations

In addition to examining the advancements of AI, research investigating the use of AI in chronic disease management raises various ethical and privacy concerns. These encompass data privacy concerns, the issue of confidentiality, and a lack of understanding regarding data protection guidelines. One primary ethical concern highlighted by this study revolves around the privacy and security of patient data used to train some of the AI systems. Regarding the use of LLMs in chronic disease management, research has also highlighted various critical concerns, including privacy issues, compliance with healthcare laws across different jurisdictions, and the lack of regulatory approval and standardized guidelines. This research also highlighted ethical challenges related to accountability for errors and limitations in addressing ethical dilemmas that may arise from the use of LLM tools in clinical care. As such, to ensure the safety and effective use of AI in chronic disease management, it is essential to consider these ethical and privacy concerns and develop relevant strategies that mitigate these concerns. For example, researchers are exploring multifaceted solutions that incorporate robust data security, multimodal data integration, and advanced domain-specific fine-tuning techniques for LLM models to ensure adherence to evidence-based practice.

Finally, stitched, multi-component AI (direct to consumer companies that ‘stitch together' models from various companies – DeepSeek, Gemini, OpenAI, Llama, while calling it 'proprietary' without provenance or a Software Bill of Materials/SBOM) creates unknown security/licensing/quality risks. Routing PHI through external or mixed models in such a patchwork increases privacy, repurposing, and breach exposure. In healthcare, labeling/transparency are expected; obscuring components undermines FDA GMLP principles and liability clarity.

Data Quality, Equity, Bias

Notwithstanding the potential of AI in healthcare, the risk of AI bias has also garnered immense attention. Several studies have mainly described the risk of AI bias in healthcare and its associated sources. For example, a significant source of AI bias stems from the data generated by digital health technologies. This data, which is often embedded with biases in healthcare, is used to train AI algorithms. The training of AI algorithms based on data from these repositories, which usually reflects biased clinical decision-making stemming from long-standing inequities in risk identification, diagnosis, and treatment of patients, has been found to contribute to AI bias. As a result, many AI models or systems have been found to increase the risk of misrepresentation and potentially exacerbate health problems in historically underserved patients or marginalized groups.

In fact, a study conducted by Obermeyer and colleagues found evidence of racial bias in one widely used algorithm. According to the findings of this study, the widely used algorithm assigned the same level of risk to Black patients as to White patients. This racial bias occurred because the algorithm used health costs as a proxy for health needs without considering the reality that Black patients, on average, spend less on the same level of health needs as compared to White patients. Biases in AI are not limited to algorithm development, but are also evident during implementation and post-implementation practices that inform AI learning. Given the high use of LLMs such as ChatGPT and Gemini, despite concerns about accuracy and bias, the negative impact of AI bias should not be ignored, as output can result in inaccurate risk prediction and disease classification, but also in biased decision-making. It can also lead to inaccurate diagnosis and treatment recommendations, especially among historically underserved patients such as female patients, Black patients, or patients from low socioeconomic status and people in low- and middle-income countries where gaps in access to even primary healthcare are much more profound.

Finally, a critical yet underrecognized determinant of AI performance in health is the quality and granularity of the input data. To realize the full promise of AI-enabled prevention and personalized care, core lifestyle measures—including diet quality, sleep patterns, and physical activity—must be captured with the same rigor of clinical vital signs, ensuring that intelligence systems are grounded in the behaviors most responsible for chronic disease risk. The concept of holistic onboarding, proposes that authentic integrative care requires equally comprehensive data. Effective holistic practice begins with capturing the full spectrum of lifestyle information, then deconstructing that whole into actionable components applied in an optimal sequence. At present, sequencing decisions—such as determining which domain to address first to enable subsequent gains—rely primarily on clinician judgment, yet they could be refined through machine learning to generate individualized, adaptive care pathways. This framework reinforces the broader imperative to enhance the quality of population-level inputs—diet, activity, sleep, stress—so that AI systems can deliver truly holistic, efficient, and personalized guidance for health and vitality.

Regulatory and Reimbursement Landscape

In addition to addressing issues related to data quality, data privacy, and algorithm bias, another essential factor for the successful adoption of AI is the need for effective regulation and reimbursement strategies. The use of AI in medical practice also raises potential ethical concerns and regulatory ambiguities that may hinder the widespread adoption of AI. Even though regulations such as the General Data Protection Regulation (GDPR), the Health Insurance Portability and Accountability Act (HIPAA), and the U.S. Food and Drug Administration (FDA) AI/ML guidance have addressed issues of fairness and provided limited pathways for legal compliance in areas at the intersection of health and AI, there is a lack of global consensus on how the clinical risks should be assessed and an absence of established regulatory frameworks geared to risks and benefits of AI models such as LLMs that continue to evolve through online learning. This highlights the need for the development of robust regulatory frameworks that are developed with the coordination and combination of external expert inputs from all stakeholders, such as technologists, clinicians, and policymakers.

In addition to having a robust regulatory framework in place, the widespread adoption of AI in medicine also depends on the existence of reimbursement models that recognize the scalability and automation capabilities of AI in chronic disease prevention, diagnosis, and management. Over the years, healthcare systems, especially in the United States, have adopted a per-use payment for AI. However, experience with traditional medical devices suggests that the use of per-use payment may lead to the overuse of AI. As a result, alternative and complementary reimbursement approaches are needed, for instance those that provide payment after the AI health systems generate additional revenue or significant savings through reduced preventable utilization for offering earlier diagnosis of diseases, incentivizing outcomes instead of reimbursing based on volume of use, and rewarding for bias mitigation. As AI continues to be rapidly integrated into healthcare, careful design of reimbursement is crucial, not just to maximize cost-effectiveness and equity, but also to improve patient care outcomes.

Key Implementation Challenges and System-Level Barriers to AI Integration in Healthcare

Domain	Challenge	Implication	Example Reference
Data Quality	Fragmented or low-fidelity lifestyle data	Limits accuracy of risk prediction	Katz (2013)
Bias & Equity	Underrepresentation of marginalized groups	Exacerbates disparities	Obermeyer 2019; Ganapathi 2022
Workflow Integration	Alert fatigue, unclear roles	Low clinician adoption	Han 2024
Privacy & Regulation	Ambiguous oversight of LLMs	Hinders deployment	Chow & Li 2025
Reimbursement	Fee-per-use misaligned with outcomes	Encourages overuse	Parikh & Helmchen 2022

Social Determinants of Health

Addressing behavior change and AI-enabled prevention also requires attention to the social determinants of health, which shape individuals' lived environments, resources, and capacity to make and sustain healthy choices. Structural factors such as food access, socioeconomic conditions, neighborhood safety, and time constraints often limit the feasibility of recommended behaviors, underscoring that personalization must extend beyond physiology to context. Thus, AI-driven approaches must account for both biological and social inputs to ensure equity and avoid reinforcing structural disadvantage.

Emerging efforts to address this social context include predictive risk-scoring tools embedded in EMRs that integrate with community referral platforms such as Unite Us (which acquired NowPow). These integrations allow EMR-based screening to identify at-risk individuals and automatically notify clinical teams for streamlined referrals to community-based organizations (CBOs). However, evidence on their real-world effectiveness remains limited. Early evaluations highlight issues with SDOH data quality and completeness, uneven uptake among community partners, and inconsistent access to such technologies across health systems. Broader structural and economic constraints continue to limit widespread implementation and sustainability of these tools.

Despite the availability of technological advances, fundamental inequities persist for many vulnerable populations. For individuals with housing insecurity, unreliable internet access, or consistent primary care, AI-driven interventions remain largely out of reach. Until these foundational barriers are addressed, the promise of AI-enabled precision prevention will be confined to those already well-connected to the healthcare system, leaving behind those whose health might benefit most from these innovations.

Translational pathway for AI in chronic disease care

Future Directions

The application of AI continues to increase rapidly, with use cases demonstrating its ability to enable early diagnosis, offer personalized treatment, and improve patient care outcomes. Hence, the future of AI in prevention, diagnosis, and management remains promising. One of these promising trends involves moving towards offering more personalized medicine, which will harness the power of predictive analytics of AI, entail the development of AI-driven tools for diagnosis, monitoring, and the provision of more precise treatments tailored to individual patient profiles. Another future direction for integrating AI in the prevention, diagnosis, and management of chronic diseases involves training and deploying AI models that accurately and explicitly reflect patient values and goals. This is because human values enter at every stage of training and deploying AI models. As a result, the future of AI in the prevention, diagnosis, and management of chronic diseases lies in embedding the “human values" that reflect human goals as well as guide human behavior when creating and using AI models.

Further, emerging frameworks in Human-Centered Artificial Intelligence emphasize that AI systems must be designed not simply to automate clinical tasks, but to augment human judgment, respect patient autonomy, and enhance the therapeutic relationship. This paradigm, advanced by institutions such as Stanford's Human-Centered AI initiative, prioritizes transparency, trust, context awareness, and equitable benefit distribution in the deployment of intelligent health technologies. In the context of chronic disease prevention and management, a human-centered approach ensures that AI tools support patient engagement, shared decision-making, and sustained behavior change rather than replacing clinician or coach insight. As AI becomes more deeply integrated into healthcare, adherence to human-centered principles will be essential to achieving outcomes that are not only effective and scalable, but also ethical, empathetic, and patient-aligned.

Finally, AI models enriched with social and environmental data can help determine when individual-level coaching is likely to drive meaningful behavior change—and when it will not. As population-level evidence accumulates, these systems can guide coaching and skill-building efforts toward contexts where they are most effective, while signaling when broader actions (e.g., public policy, environmental change, or community supports) are more likely to influence behavior and improve health. Today, coaching is often deployed under the assumption that it can help everyone, everywhere. Developing systematic insight into when agency rests with the choices people make versus the choices available to them would represent a major advance in behavioral health promotion.

Conclusion

This scoping review examined the role of AI in the prevention, early diagnosis, and management of chronic diseases, detailing the application of various AI models in predicting individual risks, detecting diseases early, and delivering personalized interventions for chronic diseases such as cancer, diabetes, and cardiovascular diseases. AI has introduced new capabilities for improving detection and management in specific preventive tasks and holds promise for bringing about the lifestyle and behavioral changes required for the prevention and management of chronic diseases. However, reaping the benefits that AI offers at the population level requires thoughtful attention to integrating human values, conducting rigorous trials, ensuring the careful integration of AI into workflows, and utilizing equitable datasets. By having these safeguards in place, we can conclude that prudent use of AI may strengthen the prevention of chronic diseases in targeted settings, support disease reversal in some cases, help close the widening disparities, and reduce healthcare cost burden by reducing the gap between clinical visits and continuous care.

Limitations (of the review itself)

Since this is not a systematic review, potential selection bias and incomplete retrieval of studies cannot be excluded. The heterogeneous nature of AI research and inconsistent reporting across behavioral domains limited cross-study comparability. Because many interventions are proprietary or unpublished, evidence may under-represent negative or null findings. However, the use of SANRA standards, explicit inclusion criteria, and transparent synthesis methods strengthens reliability. Future systematic or scoping reviews could further quantify effect sizes and explore longitudinal outcomes once a more standardized evidence base emerges.