Exploring the Role of AI in Predicting Chronic Disease Progression: Diabetes and Cardiovascular Diseases

Azza Moustafa Fahmy
Theodor Bilharz Research Institute, Giza, Egypt
Correspondence to: Azza Moustafa Fahmy, azzafhmy@gmail.com

DOI: https://doi.org/10.70389/PJPH.100021

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Azza Moustafa Fahmy – Conceptualization, Writing – original draft, review and editing
• Guarantor: Azza Moustafa Fahmy
• Provenance and peer-review:
  Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: ai in healthcare, Diabetes prediction, Cardiovascular disease prediction, Algorithmic bias, Multimodal data integration.

Peer Review
Received: 13 April 2025
Revised: 12 July 2025
Accepted: 12 July 2025
Published: 25 July 2025

Plain Language Summary Infographic

Introduction

Many aspects of medical practice have been altered by the rapid integration of artificial intelligence (AI) technology into health care, significantly augmenting diagnosis, treatment design, and operational effectiveness.^1,2 The implementation of AI-powered models has permitted earlier detection of chronic diseases, more precise risk stratification, and the potential for personalized treatment approaches.³ These comprise preventive complications reduction, tailored risk estimation, and early disease identification,⁴ consequently expanding patient outcomes as well as the efficiency of medical resources. However, the integration of AI into clinical practice is accompanied by significant challenges, particularly regarding the fairness and equity of these technologies.⁵

Algorithmic bias arises when AI models are trained on nonrepresentative or skewed data, resulting in systematically unfair or discriminatory outcomes, often leading to unequal care across demographic groups such as ethnicity, gender, or socioeconomic status.^2,3 Such prejudice makes it much harder to use AI fairly in health care because it could make health inequities worse.⁵ When AI systems are built or trained with data that does not include sufficient numbers of individuals from certain groups, their predictions for those groups may not be as accurate, which can lead to misdiagnosis, undertreatment, or overtreatment.^6,7 For instance, an AI model that was mainly trained on data from White patients could not take into account the higher risk of heart disease in Black patients. This could mean fewer follow-up scans or perhaps more cases that remain undiagnosed.^2,6 Similarly, algorithms for skin cancer detection trained primarily on images of lighter skin tones, when applied to patients with darker skin tones, may not perform well.²

A widely used cardiovascular risk scoring algorithm was found to be significantly less accurate for African American patients because 80% of its training data consisted of Caucasian individuals.² Another study showed that commercial risk prediction tools for high-risk care management in the United States systematically underestimated the health needs of Black patients, exacerbating disparities in access to care.^3,7 Given these risks, addressing algorithmic bias is essential to building AI systems that are generalizable, trustworthy, and equitable.^2,3,5 Strategies to detect and mitigate bias—such as ensuring diverse, representative training data and conducting fairness audits—are pivotal to improving health care outcomes for all populations.^7,8

Beyond addressing bias, a critical frontier in AI-driven health care is the integration of multimodal data sources, including electronic health records (EHRs), wearable sensor data, and genetic profiles.⁸ Combining these diverse data streams has the potential to enhance predictive accuracy, enable earlier intervention, and support more personalized care for chronic diseases such as diabetes and cardiovascular disease (CVD). However, challenges remain in harmonizing these data types, ensuring data quality, and validating AI models across varied clinical environments.^6,8 Despite rapid technical advances, many existing reviews have focused primarily on isolated aspects of AI—such as algorithmic performance or single-disease applications—without comprehensively addressing the intersection of fairness, data integration, and real-world validation.^2,6,8 As a result, there is a critical need for a synthesis that evaluates not only technical innovations but also examines operational, ethical, and translational barriers to equitable AI adoption in chronic disease management.^3,6,8 This review is the first to look at how federated learning frameworks (e.g., FedIDA) and fairness audits can jointly address data privacy and equity in real-world chronic disease prediction.

This review addresses these gaps by integrating recent research on multimodal AI cooperation, demonstrating structured frameworks for fairness evaluations, and analyzing the real-world impact of AI-driven solutions across different clinical environments. By highlighting both the improvements and the current barriers, we aim to provide practical suggestions for researchers, doctors, and policymakers trying to foster equitable, beneficial, reliable, and trustworthy AI in health care.

Methodology

This work builds on current literature on the role of AI in predicting chronic diseases such as diabetes and CVD using a narrative review approach. The author conducted an extensive search up to April 2025 to identify relevant literature. The aim is to provide a thorough overview of current developments, obstacles, and future directions in AI-driven health care applications.

Literature Search Approach

• Searches in databases like PubMed, IEEE Xplore, Scopus, and Web of Science (2021–2025) directed attention to related studies. Studies were selected based on several considerations. First, such databases are favored by the academic community. Second, students and researchers enjoy full access to Scopus and Web of Science under institutional agreements. Third, both databases are known for producing comprehensive and reliable search results that support reproducibility across multiple studies.
• “Type 2 diabetes,” “cardiovascular disease,” “machine learning,” “predictive analytics,” and “chronic disease management” were among the keywords.
• The search was driven by topic relevance rather than applying rigorous inclusion/exclusion standards.

Study Selection

• With preference given to those offering experiential perspectives or real-world implementation insights, studies were considered if they emphasized their relevance to AI applications in chronic disease predictions and management.
• Methodological Guidelines: With the exception of non-English articles and nonempirical commentaries, inclusion criteria provided priority to peer-reviewed studies on AI applications in early illness diagnosis, personalized treatment optimization, or operational efficiency enhancement.
• The Ethical Issues: The review focused on studies addressing techniques for verifying algorithmic bias, privacy-preserving approaches (e.g., federated learning), and transparency tools like explainable AI (XAI).
• Clinical Utility: Clinical utility was assessed based on prediction accuracy (e.g., AUROC scores), hospitalization decline, and medication adherence enhancements. Studies including many data sources—EHRs, wearable devices, genetic profiles, and environmental factors—were evaluated for their impact on model generalizations and precision.
• Multimodal Data Incorporation: Studies that involve multiple data sources—EHRs, wearable devices, genetic profiles, and environmental elements—were assessed for their influence on model generalizability and precision.

Guidelines for Reporting

Although representing a narrative study rather than a systematic assessment, we adopted clear standards of reporting where required. The PRISMA 2020 guidelines⁹ were implemented for maintaining the structure and extensiveness of the procedure for searching, inclusion criteria, and synthesis of the findings. A formalized risk-of-bias assessment (e.g., RoB 2) was not conducted due to the diverse features of the reviewed research.

Data Synthesis

A qualitative evaluation among the selected studies helped to highlight key themes, trends, and discrepancies in the existing research. Results were structured around three fundamental goals:

• Mechanistic Insights: Investigating how AI combines multimodal data to outperform traditional risk scores.
• Clinical Impact: Evaluating the efficacy of AI in lowering complications through early intervention.
• Translational Barriers: Addressing challenges such as algorithmic bias and lack of external validation to recover equity and regulatory compliance.

Unlike prior reviews, we focused our search on studies addressing underrepresented themes such as federated learning for privacy-preserving collaboration, fairness audits in model development, and the operational impact of AI in both diabetes and CVD management. This approach enabled us to identify cross-cutting innovations and practical barriers to real-world implementation. By synthesizing studies across diabetes, CVD, and comorbidities, we reveal how federated learning and fairness audits can be generalized beyond single-disease contexts—a critical advance for scalable AI deployment.

Software and Implementation Tools Utilized in Analyzed Studies

This paper reviews studies that utilized several programming environments and machine learning (ML) libraries for model development and evaluation. Most solutions utilized Python (versions 3.9–3.11) and employed popular libraries such as Scikit-learn, TensorFlow, PyTorch, and XGBoost. Optimization methods, such as particle swarm optimization (PSO), were frequently executed utilizing the PySwarms package. Model validation across studies commonly utilized 10-fold cross-validation to evaluate performance indicators, including accuracy and AUROC. Matplotlib and Seaborn were consistently employed for visualization and result presentation. It is essential to acknowledge that these techniques were documented in the original research and are presented here for contextual clarity; no independent model implementation was performed in this narrative review.

Results

Multimodal Data Integration: Comparative Performance and Impact

Recent advances demonstrate that integrating EHRs, wearable sensor data, genetic profiles, and imaging consistently enhances predictive accuracy and clinical utility in diabetes, CVD, and comorbidities.10,11 Models incorporating multiple data types scored AUROCs between 0.88 and 0.94 for diabetes and 0.85–0.91 for CVD, outperforming unimodal approaches by 12–18%. Multimodal integration also yielded earlier risk detection, improved operational efficiency (e.g., clinician time savings of 4.7 hours/week), and reduced demographic performance disparities by up to 30%.12,13 The comparative predictive performance and key limitations of unimodal and multimodal data approaches are summarized in Table 1.

Table 1: Predictive performance and limitations of data modalities in chronic disease management.
Data Modality	Diabetes (AUROC)	CVD (AUROC)	Comorbidities (Accuracy) (%)	Key Limitations
EHRs (Unimodal)	0.72–0.78	0.70–0.75	85	Misses trends, limited context10
Wearables (Unimodal)	0.65–0.70	0.68–0.72	78	Lacks lab/imaging correlation12
Multimodal Integration	0.88–0.94	0.85–0.91	99.67	Requires data harmonization13

Key Cross-Cutting Outcomes

• Improved Accuracy: XGBoost, artificial neural network (ANN), and hybrid ML models using multimodal data consistently outperformed traditional predictive methods.
• Early Detection: Multimodal models identified high-risk patients 12–18 months earlier than unimodal models.
• Operational Efficiency: Integration of wearables and EHRs reduced manual data review and hospitalizations and enabled tailored interventions.
• Equity: Federated learning and representative datasets enhanced generalizability and reduced demographic bias.

AI in Diabetes Progression Prediction

AI models have shown significant advances in predicting both the onset and progression of diabetes as well as in complication management by leveraging multimodal data. These models provide practical suggestions for individual therapy and earlier interventions through the application of deep learning (DL) and ML.^14,15

Risk Stratification and Early Detection

Using ML models applied to standardize common data models (CDMs), diabetes prediction has excellent performance. In 2022, research utilizing CDM-structured data from the National Health Insurance Corporation found that predicting 10-year diabetes growth using XGBoost and random forests outperformed logistic regression by 12–18% achieving AUROCs of 0.84–0.93.13 These models incorporated variables such as BMI, HbA1c, and genetic predispositions, enabling preemptive lifestyle interventions. Feature selection techniques like PSO further enhance accuracy, with ANNs attaining 99.76% accuracy in multidisease prediction by prioritizing biomarkers, including fasting glucose and insulin resistance.¹⁵ Multimodal integration (EHRs, genetics, wearables) enabled preemptive lifestyle interventions and earlier identification of high-risk individuals.¹⁶

Complication Prevention and Personalized Care

Through proficient data analysis, AI systems show a strong capacity to anticipate and reduce acute and chronic diabetic problems. Real-time ML models analyze blood glucose patterns, medication adherence, and physical activity data to trigger automated alerts, which help reduce sudden health issues by 40%.¹⁷ Integrated systems improve these predictions by incorporating meal logs and continuous glucose monitoring (CGM), enabling dynamic insulin dosage adjustments tailored to demographic factors like age and ethnicity.¹⁷ Ensemble models achieved AUROCs exceeding 0.8 for diabetic kidney disease prediction by recognizing biomarkers such as albuminuria and eGFR decline, with XGBoost outperforming conventional logistic regression.^18,19 Building on the advances in diabetes prediction, AI systems are progressively applied to manage comorbidities, including diabetes-hypertension and diabetic cardiomyopathy (DbCM), leveraging multimodal data integration for enhanced diagnostic accuracy.^12,18 Performance metrics and clinical outcomes of these AI-driven predictive analytics for comorbidity management are summarized in Table 2.

Table 2: AI-driven predictive analytics for comorbidity and complication management.
Application/Complication	AI Model	Performance Metric	Clinical Impact
Diabetes-Hypertension	PSO-ANN (multimodal)	99.67% diagnostic accuracy	Early, presymptomatic diagnosis enabling lifestyle interventions and monitoring
DbCM	deep neural network (DeepNN) classifier	AUC: 0.88	Targeted therapy (e.g., SGLT2 inhibitors), 12.1% 5-year heart failure (HF) incidence reduction
Diabetic Retinopathy (DR)	DeepDR Plus	C-index: 0.823	Cost reductions, tailored screening times
Diabetic Foot Ulcers (DFUs)	Convolutional neural networks (CNN) + thermal imaging	AUC: 0.89	Lower amputation rates, early identification

Management of Diabetes Complications

DR Progression Prediction: By means of early interventions to stop vision loss, ML algorithms have shown tremendous promise in forecasting DR development. Predicting 10-year DR, achieved AUROCs ranging from 0.84 to 0.93. ML models use standardized CDM datasets. The training process entailed the application of these algorithms to large-scale, structured health data from sources such as the National Health Insurance Corporation.^14,18,20

DFU Severity Classification: CNNs coupled with thermal imaging have advanced the categorization of DFU severity, which is vital for avoiding amputations. Using thermal imaging data, research by Tao et al.¹⁸ found AUCs of 0.89 for diagnosing DFU severity. The performance indicators and clinical advantages of these AI-driven innovations in diabetes complications management are summarized in Table 2.

Comorbidity and Cross-Disease Management

 Integrated Predictive Frameworks for Comorbid Diabetes and Hypertension

To control comorbidities such as concurrent diabetes and hypertension, AI systems are using multimodal data integration. To reach a diagnosis accuracy of 99.67%, PSO-enhanced neural networks combined genetic data, wearable-derived biometrics (such as blood pressure, glucose levels), and EHRs.²¹ Early diagnosis of metabolic dysfunction was given priority by biomarkers, including renal function indicators and lipid-protein ratios (e.g., LDL/HDL balance).²² The reviewed studies revealed that PSO methods for feature selection and weight tuning are usually utilized to optimize PSO-enhanced ANNs. Learning rates (0.001–0.01), two to four hidden layers, and activation features like ReLU and Sigmoid were among the most recorded hyperparameters. 10-fold cross-validation was a standard part of confirmation techniques.

The above settings allowed the diagnostic performance to be highly effective, with some studies stating rates of precision as high as 99.67% when predicting comorbidity. In a comparable way, research studies involving XGBoost usually use grid search or random searches to fine-tune parameters, which include the learning rate, maximum tree depth, and subsample ratios. CNNs were commonly used in image-based diagnosis tasks (e.g., DR, foot ulcers), using designs such as VGG¹⁶ or ResNet adapted for medical image categorization. AUROC, sensitivity, and specificity always served as evaluation metrics for these models.^15,20,21 Based on the research we browsed, some AI models have demonstrated potential for anticipating chronic diseases. XGBoost, CNN, and PSO-ANN are some of these models. Figure 1 shows their AUROC scores for predicting diabetes and CVD. PSO-ANN models show notably higher performance in diabetes prediction, while CNN and XGBoost offer robust results across both domains.

DbCMy Risk Prediction

A major comorbidity in diabetes patients, DbCM is defined by structural and functional changes of the heart that lead people to HF. Recent advances in AI have allowed for the creation of a DeepNN classifier that identifies high-risk DbCM phenotypes using echocardiographic parameters and cardiac biomarkers, enabling focused treatments like SGLT2 inhibitors and enhancing patient outcomes by early intervention.²³

AI in CVD Prediction

AI has emerged as a powerful tool for CVD risk assessment, diagnosis, and personalized therapy, particularly as conventional risk scores (e.g., Framingham) often fail to capture complex interactions between genetic, environmental, and clinical variables.²⁴ By integrating multimodal data—including genomics, wearable biometrics, and advanced imaging—AI-driven models enable earlier detection, precise risk stratification, and tailored interventions, addressing critical gaps in traditional methodologies.²⁵

Innovations in CVD Risk Prediction Diagnostics

• AI-enhanced cardiac imaging identified structural abnormalities in echocardiograms and MRIs with greater than 95% accuracy; CNNs reached 99.5% accuracy in congenital heart disease defect detection.²⁶
• Wearables with edge AI monitored blood pressure abnormalities and heart rate variability, leading to a 22% reduction in hospitalizations via preventive actions.²⁷
• VGG16-RF hybrid models achieved >92% accuracy for CVD prognosis.²⁸ These improvements in diagnostic innovations for CVD risk assessment are summarized in Table 3.

Table 3: Improvements in diagnostic innovations for cvd risk assessment.
Diagnostic Innovation	AI Model/Approach	Performance Metric	Clinical Advantages
AI-Based Risk Prediction Tools	ML Models	Sensitivity: 85%, Specificity: 80%	Improved early identification of high-risk patients compared to traditional methods (e.g., Framingham Risk Score)
Cardiac Imaging Analysis	CNNs for Echocardiograms and MRIs	>95% precision	Enhanced identification of congenital cardiac anomalies
Wearable Technology Integration	Edge AI in Wearables	22% reduction in hospitalizations	Proactive arrhythmia management and reduced admissions
Noninvasive Biomarker Detection	AI-Driven Troponin Analysis	Improved 10-year risk prediction accuracy	Better forecasting of myocardial infarction and stroke
Hybrid ML Models	VGG16-Random Forest Hybrid	92% accuracy, 91.3% precision, 92.2% recall	Precise prognosis of CVD

Precision Medicine

• Tools such as the GENinCode’s CARDIO inCode-Score combine polygenic risk with clinical variables to assess the risk of coronary heart disease. It moves 14% of patients from the intermediate risk category to the higher risk category, opening the door to individualized preventive measures, including lifestyle changes and targeted treatments.²⁹
• Comorbidity Management: Unified AI systems can control diabetes, chronic kidney disease (CKD), and CVD comorbidities by including environmental variables, wearable data, and EHRs. With a diagnosis accuracy of 99.67% for concurrent diseases, these systems enable bridging health care gaps across various groups.^26,27

Key Innovations and Operational/Ethical Impact

Multimodal integration—combining EHRs, wearables, genetic information, and imaging—consistently improved predictive accuracy (AUROCs 0.88–0.94) and enabled earlier detection of chronic diseases by 12–18 months compared to unimodal approaches, as demonstrated across diabetes, CVD, and comorbidities.²⁵ Operational benefits included reduced clinician workload (4.7 hours/week saved through wearable-EHR integration), 22–25% lower hospitalization rates in diabetes/CVD cohorts, and more efficient resource allocation via AI-optimized workflows.²⁷ Equity and generalizability were enhanced through federated learning frameworks and diverse datasets, which reduced demographic performance gaps by up to 30% in underrepresented populations.³⁰ Finally, personalized care advanced through the integration of genomics (e.g., polygenic risk scores), real-time biometrics, and imaging, supporting precision medicine initiatives such as tailored SGLT2 inhibitor therapies for high-risk DbCM patients.^23,31,32 These innovations collectively address longstanding challenges in chronic disease management, balancing technical performance with ethical responsibility and practical scalability.²⁵

Clinical Applications and Real-World Impact of AI in Chronic Disease Management

AI incorporated into the management of chronic illnesses has shown notable clinical uses and real-world influence in several areas, including diabetes, CVD, and other chronic ailments. This part investigates the real advantages and results of AI use in medical circumstances.

Glycemic Control and Personalized Interventions

AI-driven platforms that integrate wearable devices, CGM, and EHRs have shown notable improvements in glycemic management. A 24-week randomized study of an AI-powered nutrition management system, for example, found that persons with type 2 diabetes (T2D) lost 1.5–2.3 kg and HbA1c dropped 0.32–0.49% when compared to usual treatment. By means of real-time feedback, these systems allow dynamic changes to meal planning and insulin dosage by analyzing patterns in dietary intake, physical activity, and glucose dosing.³³

Key Innovations

• Automated Insulin Delivery: Hybrid closed-loop systems incorporating AI algorithms (e.g., Medtronic’s MiniMed 780G) alter basal insulin rates in real time, lowering hypoglycemia by 40% and time-in-range improvements by 2.1 hours/day.³⁴
• Noninvasive Monitoring: Devices such as the Eversense CGM system employ AI to examine interstitial fluid via subdermal sensors, providing 90-day glucose trends with 92% accuracy and removing fingerstick calibrations.³⁴

Comorbidity Prevention and Risk Stratification

Models of AI are particularly effective in anticipating and managing diabetes-related comorbidities, including CKD, HF, and DR. AI facilitates early diagnosis as well as personalized therapies for high-risk patients through the integration of multimodal data sources, including EHRs, wearable devices, and genetic profiles. By giving preventive approaches priority over reactive therapies, these developments not only enhance patient outcomes but also save health care expenditures.

• HF and CKD: Gradient boosting models trained on Danish health registries forecasted 5-year CKD risk with AUC 0.85 and prioritized early SGLT2 inhibitor medication for high-risk patients (observed incidence ratios >4.71).³⁵
• DR: Another creative idea is the creation of the Retinopathy Severity Grading Network (RSG-Net), which accurately categorizes DR into four phases using a multiclass classification system. RSG-Net beat other state-of-the-art models while maintaining computational efficiency by addressing class imbalances in datasets and using strong preprocessing methods.³⁶

Integrated Care Platforms

By involving various data sources such as EHRs, wearable devices, and telehealth platforms, AI-enabled digital health ecosystems integrate diverse data sources. Particularly in resource-limited environments, these technologies simplify chronic illness management by means of real-time monitoring, personalized interventions, and enhanced care coordination:

• China’s AI-Driven Management: Through automated alarms for glucose anomalies and medication nonadherence, platforms integrating CGM, smart insulin pens, and telemedicine, reduce hospitalizations by 25%.³⁴
• Resource-Limited Settings: By integrating EHRs, wearable data, and socioeconomic variables, frameworks such as the Comorbidity Ontological Framework for Intelligent Prediction (COFIP) in sub-Saharan Africa triage high-risk T2D patients, hence enhancing access to ACE medications and statins.³⁷

Economic and Operational Efficiency

Cost Savings Impact of AI in Diabetes and CVDs Management: By means of proactive treatments, real-time monitoring, and simplified processes, AI is changing cost control in diabetes and CVD management. These developments enhance long-term results across both diseases, maximize resource usage, and lower hospitalizations.

Managing Diabetes: Through reduced emergency department visits, improved glycemic management, and elimination of unnecessary laboratory testing, AI-driven digital health technologies have reduced yearly health care costs.¹⁶ By means of constant glucose monitoring and automatic notifications for glycemic abnormalities, telemedicine systems driven by AI decrease hospitalizations by 25%, enabling doctors to modify treatment strategies remotely and avoid acute problems.³⁴ Closed-loop insulin devices reduce hypoglycemia incidents by 40% and increased time-in-range by 2.1 hours/day, hence lowering emergency care expenses.³⁸ AI-enhanced preventive screening—such as DR programs in low- and middle-income countries (LMICs)—also offers scalable solutions for early diagnosis and intervention.³⁹

Managing CVD: By identifying early warning signals such as abnormal heart rhythms or pulmonary congestion, AI-enabled cardiac monitoring systems avoided 200 yearly readmissions, hence saving $5 million per health care network.⁴⁰ Data-driven discharge choices reduce hospital stays, which saved millions of USD yearly.⁴¹ Edge AI technologies decentralize the processing of wearable data in resource-limited environments, hence lowering infrastructure expenses while preserving diagnostic accuracy. For instance, AI-enhanced ECG scans in outpatient clinics raised early HF identification at a reasonable cost.⁴²

Common Operational Efficiencies: AI-optimized EHRs saved doctors 4.7 hours a week in data review, giving priority to high-risk situations.⁴³ Automated systems raised information retrieval efficiency by 18%, hence releasing time for direct patient care.⁴⁴ Proactive risk stratification tackled medical and social determinants of health—e.g., medication availability, housing instability—thereby lowering unnecessary readmissions in both diabetic and CVD populations.^45,46 AI-driven telemedicine and remote monitoring in LMICs reduce unexpected hospitalizations by 22%, hence improving treatment availability without affecting results.¹⁶ Real-time monitoring systems provided actionable insights into glucose trends and vital signs like heart rate and blood pressure, minimizing the need for frequent in-person consultations while improving patient outcomes.⁴⁷ The clinical applications and real-world impact of AI in chronic disease management—including diabetes and CVDs—are summarized in Table 4.

Table 4: Real-world impact of ai in chronic disease management.
Category	Key Insights	Source
Diabetes Management	In US cohorts, AI-based digital health solutions reduce yearly health care expenses by $1200 per patient.	16
Diabetes Management	Real-time blood glucose monitoring and automatic alerts helped AI-driven telemedicine systems save hospitalizations by 25%.	34
Diabetes Management	Improving supply use by reducing superfluous diagnostic tests and simplifying patient care routes.	34
CVD Management	AI-powered remote cardiac monitoring devices saved $5 million yearly by preventing 200 readmissions.	40
CVD Management	By reducing operating expenses by 25% through fewer admissions and shorter hospital stays, AI-based solutions eliminated needless diagnostic tests.	48
Clinician Workload	By means of data review, AI-optimized EHRs let doctors save an average of 4.7 hours each week, hence allowing attention on high-risk situations.	43
Clinician Workload	By 18%, AI-driven EHR systems increased information retrieval efficiency during patient record checks, hence enabling more direct patient treatment.	44
Broader Economic Impact	AI-led DR screening in Trinidad and Tobago might save up to $60 million yearly because of labor cost savings and enhanced screening efficiency.	44
Broader Economic Impact	In China, AI-driven telemedicine platforms significantly lowered health care costs by reducing in-person visits while maintaining glycemic control.	34
Patient Outcomes	AI-powered remote monitoring improves HbA1c, blood pressure, medication adherence, and quality of life.	1,3,4

Challenges and Limitations of AI in Chronic Disease Prediction

The extensive use of AI in chronic illness prediction faces several obstacles, including data quality, model design, ethical issues, and practical application. Overcoming these obstacles is essential to seeing AI’s revolutionary power in medicine.

Data-Related Challenges

While health care data is scattered among institutions, especially in LMICs, AI models rely on high-quality, standardized information. Limited access to standardized EHRs and diversity in data collecting techniques, population demographics, and outcome definitions create bias, hence lowering model generalizability.^49,50 Privacy issues further complicate data use, as laws like HIPAA and GDPR require safe storage and anonymization of sensitive patient data. Data breaches undermine patient confidence and may incur substantial financial penalties, hence highlighting the importance of strong cybersecurity systems.⁵¹ These key data challenges in AI for chronic disease prediction are illustrated in Figure 2.

Model-Related Constraints

Most AI models fo chronic disease prediction lack external validation across diverse clinical settings.⁵² For instance, in minority populations, AUROC declines varied from 0.10 to 0.15, and 81% of AI models for CVD prediction exhibited decreased accuracy in external datasets, according to a systematic review. This renders them less valuable in the real world and drives up queries about their generalizability.⁵³

Social and Ethical Issues

With models trained on nondiverse datasets showing worse performance in underrepresented groups, the issue of algorithmic bias is still a major concern. For instance, CVD risk prediction models revealed AUC declines of 0.08–0.15 in elderly people (>81 years) and Black patients relative to younger, White groups.^49,54 Geographic and socioeconomic prejudices aggravate inequalities as models frequently exclude resource-limited areas.⁵⁵ Moreover, patients often lack a clear understanding of how AI affects their treatment, which raises ethical concerns related to informed consent and autonomy.⁵⁶ The primary ethical and social concerns in AI health care applications and strategies for addressing them are illustrated in Figure 3.

Practical Implementation Challenges

Integrating AI into clinical processes faces problems like EHR systems’ interoperability and workflow disturbances caused by system redesign. For example, EPIC Systems’ sepsis model failed to identify 67% of actual sepsis cases in real-world deployments, hence underlining performance and implementation issues.⁵⁷ Particularly in LMICs, where fragmented health care infrastructures and high prices of modern technologies like genomic integration limit scalability and adoption, financial and infrastructural obstacles restrict scaling even more.^49,50

Future Directions for AI in Chronic Disease Management

Future research should focus on:

• Improving prediction Capabilities: Multimodal data integration will increase prediction accuracy by including EHRs, wearable measures, genetic information, and environmental variables.^11,58 By combining genetic and lifestyle data to find complicated relationships, generative AI models will improve risk stratification.⁵⁹
• Ethical Considerations: Building on recent international guidance, future research should prioritize fairness audits throughout model development to address algorithmic biases affecting marginalized populations^25,60 and ensure compliance with emerging regulatory frameworks.^61,62
• XAI techniques (e.g., SHAP, LIME) should be adopted to enhance transparency for both clinicians and patients. Large-scale, multicenter clinical trials and longitudinal studies are essential to validate AI models across diverse populations, ensuring generalizability and sustained impact.^63,64

Improved Predictive Capacity

To increase predictive accuracy for chronic diseases, including diabetes and CVD, future AI systems will include several data sources, including environmental variables, genetic data, wearable measurements, and EHRs. Advanced ML models analyzing longitudinal data will facilitate earlier prediction of disease onset and complications, thereby enabling timely clinical interventions.^11,58 By combining genetic and lifestyle data to find complicated relationships, generative AI models will provide even more risk stratification. For example, neural networks that evaluated risk scores that were polygenic performed superior to traditional models by including genetic markers like APOE-ε-4 first, leading to AUROCs of 0.80–0.84 for predicting Alzheimer’s disease.⁶⁵ These techniques can help doctors find high-risk patients more accurately, specifically if they have other medical issues.^29,59

Customizing Treatment Strategies

By customizing treatment regimens depending on personal genetic profiles, lifestyle choices, and real-time monitoring data, AI-powered systems are pushing precision medicine. Tools like GENinCode’s CARDIO inCode-Score are expanding their clinical applications for individualized cardiovascular treatment.²⁹ AI algorithms also help to balance efficacy and toxicity for chronic diseases like diabetes and cancer, thereby optimizing drug doses.^66,67 Real-time feedback on medication adherence, dietary changes, and exercise routines helps patients with chronic diseases manage their conditions by means of behavioral interventions driven by conversational AI applications. Research indicates that by over 30%, these tools help people follow their medications, substantially enhancing health outcomes.^63,68

Incorporation into Health Care Systems

By using open APIs and uniform frameworks that enable smooth interaction with current EHR systems, future AI systems will solve interoperability issues.^69,70 AI-driven automated workflows streamline administrative processes such as patient intake, scheduling, and invoicing, hence enabling doctors to have more time for direct patient treatment. Scalable solutions designed for LMICs use cloud-based platforms and mobile health technology to enhance access to care. Frameworks such as COFIP extend their utility to underserved regions to rectify inequalities in health care provision.^37,71

Digital Health Innovations and Remote Monitoring

Providing real-time alerts for early intervention, wearable devices with edge computing provide continuous monitoring of vital indicators like heart rate variability and blood glucose levels.^72,73 AI-integrated diagnostic tools in telehealth kiosks help telemedicine expand by providing instant health advice in rural areas and workplace settings, hence lowering emergency visits and guaranteeing fair access to health care.4^7,74

Ethical and Regulatory Issues

Future initiatives will prioritize transparency through the use of XAI technologies such as SHAP values or LIME frameworks, therefore guaranteeing ethical deployment of AI systems by enhancing model interpretability for patients and doctors alike.^61,62 By means of secure exchange of sensitive patient data between institutions without sacrificing confidentiality, federated learning methods enhance data privacy protections.⁷⁴ Fairness audits during model building guarantee fair health care delivery across various demographics by addressing algorithmic biases that disproportionately harm disadvantaged populations.^25,60 Addressing ethical issues and scalability issues, AI in chronic illness management will be able to combine sophisticated prediction skills with individualized care delivery. AI has the potential to change world health care systems and improve outcomes for patients with chronic diseases by focusing on transparency, equity-focused development processes, interoperability criteria, and thorough clinical validation initiatives. The principal areas of concentration for these AI-driven advances are summarized in Figure 4.

Discussion

Main Findings and Comparison with Prior Studies

This review demonstrates that integrating multimodal data sources-EHRs, wearable, and genomic data-enables earlier and more precise risk stratification for comorbid diabetes and CVD. Our approach is different from previous evaluations that mostly looked at technical metrics (such as AUROC and sensitivity).³ It looks at both the operational and ethical effects of AI advancements. For example, we suggest a formal approach for carrying out fairness audits,²⁵ which directly addresses algorithmic bias—a major gap in predicting chronic diseases—and encourages fair care for people from different backgrounds.

Federated learning frameworks not only protect patient privacy but also reduce performance gaps by 30% in groups that are not well represented, making it possible to use AI on a large scale and in any institution.³⁰ These improvements deal with systemic problems, including Black patients not getting enough diagnoses of their cardiovascular risk, and they also give people the means to put them into action in the actual world.^2,3 Previous studies, like Lee et al. and DeGroat et al.,^11,14 showed that multimodal AI improved chronic disease prediction in single-center or limited-cohort settings. Our synthesis shows that these benefits are strong across many disease domains and real-world populations, with AUROCs of 0.88–0.94 for diabetes and 0.85–0.91 for CVD. For instance, DeGroat et al.¹¹ said that using multiomics data in a single cohort could predict CVD with 100% accuracy. Our results show that similar predictive performance can be achieved on a larger scale by using a wider range of data types and federated frameworks, which can reduce demographic differences by up to 30%.³⁰

Our study also builds on the work of Hasanzadeh et al.,² which looked at federated learning to reduce bias in one institution. We show that cross-institutional federated frameworks, like WFDS, can increase fairness by as much as 30% while still keeping privacy and scalability in multicenter networks.³⁰ This is especially important because models trained on homogenous datasets like the UK Biobank have shown persistent algorithmic bias, with AUC drops of 0.10–0.15 in non-White groups. This is in line with Hasanzadeh et al. and Mihan et al.^2,54 We have tested XAI methods like SHAP and LIME on single-disease models (for example, DR by Panda and Mahanta, 2023).⁶¹ Our synthesis shows that they can also be used to predict cross-disease comorbidity, which is important for clinical use where understanding is necessary. Only 10% of neuroradiology research used XAI techniques, which shows a big gap in how they are used. Our review fills this gap by suggesting that explainability frameworks be used more widely.

Our findings support those of Guan et al.¹⁶ and He and Li³⁴ in terms of operational and economic impact. They show that AI-driven interventions can reduce hospitalizations by up to 25% in diabetes management and save clinicians about 4.7 hours a week on data review tasks. Importantly, our review adds to the evidence that these benefits are not just for places with a lot of resources. Ciecierski-Holmes et al.⁷¹ show that AI-enabled triage and remote monitoring can help people get better care in places with fewer resources. From an ethical and legal point of view, our synthesis draws on the World Health Organization’s (WHO) foundational principles for AI in health care and new guidelines for fairness audits.²⁵ To encourage openness, responsibility, and fair distribution of rewards, we suggest using fairness audits and XAI tools throughout the AI lifecycle. This is something that previous assessments did not systematically cover.

In spite of these developments, there are still numerous significant obstacles. Most studies still do not have long-term, multicenter validation, and many models are based on data from high-income nations, which makes them less useful for communities that are underrepresented or lack resources. Like You et al.,⁶⁴ our review recommends large-scale, forward-looking clinical trials and investigations in the real world to find out how long AI-driven interventions last and how fair they are. In summary, by directly comparing our findings to recent studies, we demonstrate that the integration of federated learning, fairness audits, and multimodal data not only advances predictive accuracy but also addresses systemic inequities and operational barriers in chronic disease management. These innovations, if implemented with robust validation and ethical oversight, have the potential to transform global health care delivery.

Interpretation of Findings

The examined research shows the great prediction accuracy of AI-driven models for chronic illnesses like diabetes and CVDs. By combining EHRs, wearable sensor data, and genetic data, XGBoost models, for example, attained AUROCs of 0.84–0.93 in diabetes forecasting.¹⁴ Likewise, by combining calcium scoring and cardiac architecture, AI-enabled CT scans forecasted HF risk with over 90% accuracy.³ These results fit with more general patterns in health care AI, where advanced analytics outperform traditional risk scoring systems like Framingham by identifying nonlinear interactions between genetic, environmental, and lifestyle elements.^4,11 Biases in training datasets restrict the generalizability of these models, however. Models trained on homogenous datasets like the UK Biobank (80% White participants), for instance, show worse accuracy when applied to underrepresented populations, including Black or Hispanic people.⁴⁹ These differences draw attention to the need to tackle algorithmic prejudice to ensure equitable health care delivery across all demographic categories.

Investigation of Other Hypotheses

Many AI experiments depend on tiny or nondiverse datasets, which increases the risk of overfitting. For instance, a DR model performed well in controlled studies with a C-index of 0.754–0.846 but could not translate to rural areas with restricted access to health care. Class imbalance, in which “healthy” instances predominate in training data and cause misclassification of minority cases like DFUs, aggravates overfitting even further.²⁰ Compared to younger, White populations, AI models for CVD prediction revealed AUC declines of 0.10–0.15 in elderly individuals (>81 years) and Black patients.^49,54 Moreover, just 10% of research is done outside validation across different geographic or demographic populations, so severely limiting the generalizability and practical relevance.¹¹

Consequences for Health Care Systems

Our findings underscore that the real-world impact of AI in chronic disease management extends beyond predictive accuracy to include substantial clinical and operational benefits. For instance, AI-driven interventions have saved up to $1200 per patient per year, while digital health solutions have reduced hospital stays by up to 25% for people with diabetes.^16,34 AI-optimized EHR systems also make workflows more efficient by saving doctors an average of 4.7 hours/week on data review, which lets them spend more time caring for patients.⁴³ These operational improvements are especially important in places where resources are scarce, where AI-enabled triage and remote monitoring can assist in improving access to and quality of treatment.⁴⁷ From an ethical point of view, our review shows how important it is to build justice, openness, and responsibility into AI systems. Based on the WHO’s foundational principles for ethical AI, which include protecting autonomy, being open, and promoting justice, our synthesis suggests that all parties involved in AI (developers, clinicians, and policymakers) should use fairness audits and XAI tools throughout the AI lifecycle.^2,61,62 This is important for developing trust, protecting patients’ rights, and making sure that AI-driven health care benefits are shared equitably.

Technical Suggestions

Advanced methods like NLPNN algorithms can increase feature spaces and handle class imbalance via attention processes, thereby reducing overfitting and improving generalizability.⁴ Federated learning systems enable decentralized training across several datasets while maintaining privacy.²⁹ Fairness checks during model creation can help to guarantee fair representation across demographic groups.^25,62 Regulatory systems should require demographic reporting in AI projects to guarantee fair representation. A study by Hasanzadeh et al.² underscores the importance of addressing prejudice throughout the AI model’s lifetime, from conception to postdeployment monitoring. The paper describes methodical strategies, including fairness audits, federated learning, and XAI tools to guarantee fair results across many different patient groups.2^5,61,62 Techniques of XAI, like SHAP values or LIME frameworks, improve transparency by allowing clinicians to interpret model predictions and handle biases properly.⁶¹

Limitations and Future Directions

This review highlights several key limitations in the investigation of AI’s influence on chronic illness management. Particularly in forecasting illness development and consequences, the absence of longitudinal data restricts understanding of the long-term effectiveness and sustainability of AI therapies. Most research was also done in high-income nations, which limits knowledge of AI scalability in resource-limited environments where infrastructural deficits and labor shortages continue to be major concerns. The generalizability of results to different or underprivileged communities is further compromised by the dependence on uniform datasets, mostly reflecting metropolitan or wealthy people. Ethical challenges, such as algorithmic bias, privacy concerns, and the opacity of AI systems, continue to be under-discussed, raising concerns around public trust and regulatory compliance. Future research must address these limitations to guarantee fair and efficient deployment of AI technology all over the world.

Main Limitations

Limited Data Diversity and Generalizability: Most AI models are developed and validated using datasets from high-income countries or urban populations, limiting their generalizability to underrepresented groups and resource-limited settings.^6,13 For example, models trained on the UK Biobank have demonstrated AUC drops in non-White groups.⁵⁴ Additionally, CVD models show AUC declines of 0.10–0.15 in elderly and Black patients.^63,65

Inadequate External and Longitudinal Validation: A lot of research in this subject relies on retrospective or cross-sectional data, and there are not many large-scale, prospective designs or external validation cohorts. A systematic analysis demonstrated that 81% of AI models for radiologic diagnosis did worse when tested on data from outside sources, and 24% had big drops in accuracy.^3,11,54

Data Quality and Standardization Issues: Variability in EHR formats, wearable device outputs, and genetic data poses a major barrier to seamless data integration. Harmonizing multimodal inputs (EHRs, wearables, genomics) remains challenging due to inconsistent formats and missing data.^6,13

Concerns About Fairness and Bias in Algorithms: Nonrepresentative training data that is biased over time makes predictions less accurate for older people, racial and ethnic minorities, and people with few resources. AI models trained on datasets that are all the same (like the UK Biobank) have shown AUROC drops of 0.10–0.15 in groups that are not White.^13,54

Privacy, Security, and Regulatory Barriers: Data privacy regulations (e.g., HIPAA, GDPR) and cybersecurity risks complicate cross-institutional collaboration.⁵ Synthetic data generation⁵⁹ and federated learning30 are promising but require robust encryption and alignment with evolving legal and ethical frameworks.⁷⁵

Clinical Workflow Integration Challenges: AI tools can disrupt established clinical workflows, lack interoperability with EHRs, or require additional training. For example, EPIC Systems’ sepsis prediction model failed in 67% of real-world deployments due to misalignment with clinical workflow and limited clinician engagement.⁵⁷

Explainability and Trust: The “black-box” nature of many AI models impedes clinician trust. Only 10% of studies in neuroradiology used XAI tools like SHAP or LIME to enhance interpretability and clinical transparency.^61,62

Future Research Directions

Expand Data Diversity and Global Collaboration: Future research should prioritize the use of datasets that reflect racial, ethnic, and socioeconomic diversity. Federated learning frameworks (e.g., FedIDA) can aggregate data from underrepresented regions while maintaining patient privacy and data sovereignty.⁵

Longitudinal, Prospective, and Multicenter Trials: Conduct real-world studies with multiyear follow-ups. The FUTURE-AI guidelines recommend external validation across ≥3 clinical sites to ensure generalizability.^30,64

Advanced Multimodal and Interoperable Data Integration: Develop standardized frameworks (e.g., FHIR APIs) to harmonize EHRs, wearables, and genomics. Hybrid models like VGG16-RF improved diabetes prediction accuracy by 12–15% through multimodal fusion.^28,49

Implement Fairness Audits and Federated Learning: Integrate fairness audits (e.g., IBM’s AI Fairness 360)²⁵ and federated learning (e.g., WFDS framework)⁷⁵ to reduce demographic bias by 30% in underrepresented groups.^3,12,55

Enhance Explainability and Transparency: Adopt XAI methods (e.g., SHAP, LIME) to visualize model decisions. For example, post hoc explanations of lab test data improved clinician trust in diabetes prediction models.⁶¹

Strengthen Regulatory and Ethical Frameworks: Collaborate with policymakers to mandate bias reporting and transparency. The WHO’s ethical AI principles emphasize accountability and equitable benefit distribution.²⁴

Focus on User-Centered and Scalable Solutions: Design affordable, adaptable tools for low-resource settings. AI-optimized telehealth kiosks reduced emergency visits by 22% in rural areas through real-time monitoring.^47,71 Cloud computing platforms need to be utilized as well to render it accessible to institutions to use AI in a scalable and real-time manner, particularly in rural or low-resource locations, since there may not be sufficient computing resources nearby.⁷⁶

Promote Interdisciplinary Collaboration: Foster partnerships among clinicians, engineers, and ethicists. Nurses’ involvement in AI development improved usability and workflow integration in chronic disease management.^57,71 The main gaps in validation and generalizability, along with future recommendations, are summarized in Table 5.

Table 5: Validation and generalizability of ai models in chronic disease prediction: current gaps and future recommendations.
Aspect	Current Gap	Future Recommendation
External Validation	81% of models lack external validation	Mandate validation across ≥3 diverse clinical sites (FUTURE-AI) and populations
Model Explainability	Restricted validation of AI model against established medical knowledge decreases confidence	To acquire trust, use approaches that make things clear and validate explanations against medical literature
Dataset Diversity	A lack of lifestyle, socioeconomic, and experiential data makes personalization harder	Add more than just clinical information about patients to improve the model’s relevance and fairness
Data Quality and Bias	Data that is broken up, prejudiced, or missing can make things less accurate and keep unfairness going	Be sure that datasets are of high standard, have an extensive variety of demographics, and are in identified formats (like FHIR)
Multicenter Trials	Single-center studies dominate	Collaborate with LMIC institutions to test scalability and equity
Longitudinal Data	Short-term outcomes limit durability assessment	Conduct 5-year follow-ups to evaluate real-world impact on complications and hospitalizations
Clinical Integration	Adoption is limited given that it affects workflows, clinicians are not trained, and individuals do not want to change	With proper training and guidance from humans, it can easily add AI to medical procedures
Regulatory and Ethical	Unclear accountability and gaps in regulation make it hard to deploy safely	Be sure that all AI in health care aligns with rules and regulations and with ethical standards

Plans for LMICs’ Policies: To make sure that AI is used fairly in managing chronic diseases in LMICs, policymakers need to initiate specific action to solve challenges with infrastructure, data availability, and institutional unfairness.

Funding in Data Infrastructure: The government and health systems in LMICs must categorize digital infrastructure development as a top priority. That includes cloud platforms, mobile broadband connectivity, and EHR systems. To enable and facilitate sustainable and scalable AI deployments, these fundamental investments are mandatory.

Federated Learning for Privacy-preserving Collaboration: Federated learning frameworks offer an ethical manner for sharing data as they allow you to train models locally without delivering private patient information. This preserves the sovereignty of data and promotes involvement from areas that are not well represented, without violating privacy rules.⁷⁶

Require Fairness Audits and Validation that Includes Everyone: Before AI tools are used, national health authorities should undertake fairness audits. IBM’s AI Fairness 360 and other tools can help find demographic biases. Also, all models should be tested on records that include local people to make sure that individuals with vulnerabilities ae not excluded or misclassified.

Encouraging the Use of Open-Source and Affordable AI Tools: In order to be sure that systems with limited resources may use AI despite having to pay a lot for permits or infrastructure, open-source platforms (like TensorFlow and Scikit-learn) and low-compute models should be prioritized.

Give Local Health Workers the Training and Tools They Need to Do Their Jobs: AI literacy initiatives created for doctors, nurses, and other health workers are significant to ensure that AI is used wisely and ethically. Building capacity should focus on knowledge of AI outputs, discovering the limits of models, and taking part in co-design processes.

Strengthen Regulatory and Ethical Oversight: LMICs need to establish national AI governance frameworks that adhere to global standards. The WHO 2021 guidelines on AI ethics for health involve principles such as openness, responsibility, inclusion, and fair sharing of benefits.76 In brief, policies that place people first and take into account the local context should lead AI in LMICs. These policies should not just focus on efficient technology, but also on terminating digital gaps along with encouraging health equity.

Conclusion

By combining multimodal data sources to improve diagnosis accuracy, early intervention, and individualized treatment, AI holds transformative potential in chronic illness prediction and management. Across diabetes and CVDs, AI-driven technologies have demonstrated notable progress in lowering complications and enhancing operational efficiency. However, widespread implementation remains limited by critical issues, including algorithmic bias, overfitting caused by homogenous datasets, lack of external validation, and ethical questions about openness and privacy. Ensuring fair implementation across different populations depends on addressing these constraints by means of technical advances such as federated learning systems, fairness audits, and XAI methods. Future research should prioritize multicenter trials and longitudinal research to validate AI models in real-world health care settings while bridging the gap between clinical reality and algorithmic promise. Addressing these challenges is needed to entirely unlock the transformative power of AI and bring equitable, effective solutions for controlling chronic disease on a worldwide scale.

References

1. Fahmy A. Opportunities and challenges of applying artificial intelligence in healthcare. Prem J Artif Intell. 2024;1:100003. https://doi.org/10.70389/PJAI.100003
2. Hasanzadeh F, Josephson CB, Waters G, Adedinsewo D, Azizi Z, White JA. Bias recognition and mitigation strategies in artificial intelligence healthcare applications. NPJ Digit Med. 2025;8(1):154. https://doi.org/10.1038/s41746-025-01503-7
3. Nazer LH, Zatarah R, Waldrip S, Ke JXC, Moukheiber M, Khanna AK, et al. Bias in artificial intelligence algorithms and recommendations for mitigation. PLoS Digit Heal. 2023;2(6):e0000278. https://doi.org/10.1371/journal.pdig.0000278
4. Arefin S. Chronic disease management through an AI-powered application. J Serv Sci Manag. 2024;17(04):305–20. https://doi.org/10.4236/jssm.2024.174015
5. Gurevich E, El Hassan B, El Morr C. Equity within AI systems: what can health leaders expect? Healthc Manag Forum. 2023;36(2):119–24. https://doi.org/10.1177/08404704221125368
6. Kumar D, Malin BA, Vishwanatha JK, Wu L, Hedges JR. AI in diomedicine—a forward-looking perspective on health equity. Int J Environ Res Public Health. 2024;21(12):1642. https://doi.org/10.3390/ijerph21121642
7. Huang Y, Guo J, Chen WH, Lin HY, Tang H, Wang F, et al. A scoping review of fair machine learning techniques when using real-world data. J Biomed Inform. 2024;151:104622. https://doi.org/10.1016/j.jbi.2024.104622
8. Simon BD, Ozyoruk KB, Gelikman DG, Harmon SA, Türkbey B. The future of multimodal artificial intelligence models for integrating imaging and clinical metadata: a narrative review. Diagn Interv Radiol. 2025;31(4):303–12. https://doi.org/10.4274/dir.2024.242631
9. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. Int J Surg. 2021;88:105906. https://doi.org/10.1016/j.ijsu.2021.105906
10. Zhou T, Wang Y, Xu Y, Xu L, Tang L, Yang Y, et al. Multimodal data integration for enhanced longitudinal prediction for cardiac and cerebrovascular events following initial diagnosis of obstructive sleep apnea syndrome. J Glob Health. 2024;14:04103. https://doi.org/10.7189/jgh.14.04103
11. DeGroat W, Abdelhalim H, Peker E, Sheth N, Narayanan R, Zeeshan S, et al. Multimodal AI/ML for discovering novel biomarkers and predicting disease using multi-omics profiles of patients with cardiovascular diseases. Sci Rep. 2024;14(1):26503. https://doi.org/10.1038/s41598-024-78553-6
12. Amal S, Safarnejad L, Omiye JA, Ghanzouri I, Cabot JH, Ross EG. Use of multi-modal data and machine learning to improve cardiovascular disease care. Front Cardiovasc Med. 2022;9:840262. https://doi.org/10.3389/fcvm.2022.840262
13. Sufian MA, Alsadder L, Hamzi W, Zaman S, Sagar ASMS, Hamzi B. Mitigating algorithmic bias in ai-driven cardiovascular imaging for fairer diagnostics. Diagnostics. 2024;14(23):2675. https://doi.org/10.3390/diagnostics14232675
14. Lee C, Jo B, Woo H, Im Y, Park RW, Park C. Chronic disease prediction using the common data model: development study. JMIR AI. 2022;1(1):e41030. https://doi.org/10.2196/41030
15. Rashid J, Batool S, Kim J, Wasif Nisar M, Hussain A, Juneja S, et al. An augmented artificial intelligence approach for chronic diseases prediction. Front Public Heal. 2022;10:860396. https://doi.org/10.3389/fpubh.2022.860396
16. Guan Z, Li H, Liu R, Cai C, Liu Y, Li J, et al. Artificial intelligence in diabetes management: advancements, opportunities, and challenges. Cell Reports Med. 2023;4(10):101213. https://doi.org/10.1016/j.xcrm.2023.101213
17. Pan M, Li R, Wei J, Peng H, Hu Z, Xiong Y, et al. Application of artificial intelligence in the health management of chronic disease: bibliometric analysis. Front Med. 2025;11:1506641. https://doi.org/10.3389/fmed.2024.1506641
18. Tao Y, Hou J, Zhou G, Zhang D. Artificial intelligence applied to diabetes complications: a bibliometric analysis. Front Artif Intell. 2025;8:1455341. https://doi.org/10.3389/frai.2025.1455341
19. Dholariya S, Dutta S, Sonagra A, Kaliya M, Singh R, Parchwani D, et al. Unveiling the utility of artificial intelligence for prediction, diagnosis, and progression of diabetic kidney disease: an evidence-based systematic review and meta-analysis. Curr Med Res Opin. 2024;40(12):2025–55. https://doi.org/10.1080/03007995.2024.2423737
20. Dai L, Sheng B, Chen T, Wu Q, Liu R, Cai C, et al. A deep learning system for predicting time to progression of diabetic retinopathy. Nat Med. 2024;30(2):584–94. https://doi.org/10.1038/s41591-023-02702-z
21. Jain S, Sharma SP. Diagnosis of diabetes using modified PSO and FFNN technique their application to WSD. Turk Online J Qual Inq. 2021;12(3):2671–7.
22. Bellamy RKE, Dey K, Hind M, Hoffman SC, Houde S, Kannan K, et al. AI Fairness 360: an extensible toolkit for detecting and mitigating algorithmic bias. IBM J Res Dev. 2019;63(4/5):4:1–4:15. https://doi.org/10.1147/JRD.2019.2942287
23. Segar MW, Usman MS, Patel K V., Khan MS, Butler J, Manjunath L, et al. Development and validation of a machine learning-based approach to identify high-risk diabetic cardiomyopathy phenotype. Eur J Heart Fail. 2024;26(10):2183–92. https://doi.org/10.1002/ejhf.3443
24. Hasan R, Dattana V, Mahmood S, Hussain S. Towards transparent diabetes prediction: combining automl and explainable AI for improved clinical insights. Information. 2024;16(1):7. https://doi.org/10.3390/info16010007
25. Lu J, Sattler A, Wang S, Khaki AR, Callahan A, Fleming S, et al. Considerations in the reliability and fairness audits of predictive models for advance care planning. Front Digit Heal. 2022;4:943768. https://doi.org/10.3389/fdgth.2022.943768
26. Ferdowsi M, Goh CH, Liu H, Tse G, Ho Hui JM, Wang X. Clinical application of artificial intelligence in the diagnosis, prediction, and classification of coronary heart disease. Cardiovasc Innov Appl. 2025;10(1):3–12. https://doi.org/10.15212/CVIA.2025.0009
27. Elvas LB, Almeida A, Ferreira JC. The role of AI in cardiovascular event monitoring and early detection: scoping literature review. JMIR Med Inf. 2025;13:e64349. https://doi.org/10.2196/64349
28. Addisu EG, Yirga TG, Yirga HG, Yehuala AD. Transfer learning-based hybrid VGG16-machine learning approach for heart disease detection with explainable artificial intelligence. Front Artif Intell. 2025;8:1504281. https://doi.org/10.3389/frai.2025.1504281
29. GENinCode. CARDIO inCode-Score UK: genetic testing for personalised cardiovascular risk assessment. CARDIO in Code-Score UK | Genetic Testing For Personalised Health Care and Medicine; 2023.
30. Zhang F, Shuai Z, Kuang K, Wu F, Zhuang Y, Xiao J. Unified fair federated learning for digital healthcare. Patterns. 2024;5(1):100907. https://doi.org/10.1016/j.patter.2023.100907
31. Karnain Wadoo O, Ahmad I, Sayeed SI. Reduced lung function and progression to prediabetes: a prospective study. Iran J Diabetes Obes. 2021;13(4):178–83. https://doi.org/10.18502/ijdo.v13i4.7994
32. Huang K, Luo X, Liao B, Li G, Feng J. Insights into SGLT2 inhibitor treatment of diabetic cardiomyopathy: focus on the mechanisms. Cardiovasc Diabetol. 2023;22(1):86. https://doi.org/10.1186/s12933-023-01816-5
33. Lee YB, Kim G, Jun JE, Park H, Lee WJ, Hwang YC, et al. An integrated digital health care platform for diabetes management with AI-based dietary management: 48-week results from a randomized controlled trial. Diabetes Care. 2023;46(5):959–66. https://doi.org/10.2337/dc22-1929
34. He Z, Li W. AI-driven management of type 2 diabetes in China: opportunities and challenges. Diabetes Metab Syndr Obes. 2025;18:85–92. https://doi.org/10.2147/DMSO.S495364
35. Dworzynski P, Aasbrenn M, Rostgaard K, Melbye M, Gerds TA, Hjalgrim H, et al. Nationwide prediction of type 2 diabetes comorbidities. Sci Rep. 2020;10(1):1776. https://doi.org/10.1038/s41598-020-58601-7
36. Akhtar S, Aftab S, Ali O, Ahmad M, Khan MA, Abbas S, et al. A deep learning based model for diabetic retinopathy grading. Sci Rep. 2025;15(1):3763. https://doi.org/10.1038/s41598-025-87171-9
37. Daramola O, Nyasulu P, Mashamba-Thompson T, Moser T, Broomhead S, Hamid A, et al. Towards AI-enabled multimodal diagnostics and management of COVID-19 and comorbidities in resource-limited settings. Informatics. 2021;8(4):63. https://doi.org/10.3390/informatics8040063
38. Ait-Aissa N. Can digital technology revolutionize continuous education in diabetes care? J Diabetes Sci Technol. 2024: 19322968241298000. https://doi.org/10.1177/19322968241298000
39. Wang Y, Liu C, Hu W, Luo L, Shi D, Zhang J, et al. Economic evaluation for medical artificial intelligence: accuracy vs. cost-effectiveness in a diabetic retinopathy screening case. NPJ Digit Med. 2024;7(1):43. https://doi.org/10.1038/s41746-024-01032-9
40. Pierucci N, Laviola D, Mariani MV, Nardini A, Adamo F, Mahfouz K, et al. Remote monitoring and heart failure. Eur Hear J Suppl. 2025;27(Supplement_1):i126–31. https://doi.org/10.1093/eurheartjsupp/suae116
41. Khanna NN, Maindarkar MA, Viswanathan V, Fernandes JFE, Paul S, Bhagawati M, et al. Economics of artificial intelligence in healthcare: diagnosis vs. treatment. Healthcare. 2022;10(12):2493. https://doi.org/10.3390/healthcare10122493
42. Muzammil MA, Javid S, Afridi AK, Siddineni R, Shahabi M, Haseeb M, et al. Artificial intelligence-enhanced electrocardiography for accurate diagnosis and management of cardiovascular diseases. J Electrocardiol. 2024;83:30–40. https://doi.org/10.1016/j.jelectrocard.2024.01.006
43. Chi EA, Chi G, Tsui CT, Jiang Y, Jarr K, Kulkarni CV, et al. Development and validation of an artificial intelligence system to optimize clinician review of patient records. JAMA Netw Open. 2021;4(7):e2117391. https://doi.org/10.1001/jamanetworkopen.2021.17391
44. Ramoutar RR. An economic analysis for the use of artificial intelligence in screening for diabetic retinopathy in Trinidad and Tobago. Cureus. 2024;16(3):e55745. https://doi.org/10.7759/cureus.55745
45. Bennett DJ, Feng J, Goldman S, Kothari A, Gottlieb LM, Durstenfeld MS, et al. Reducing readmissions in the safety net through AI and automation. Am J Manag Care. 2025;31(3):142–8. https://doi.org/10.37765/ajmc.2025.89697
46. Wang L, Li G, Ezeana CF, Ogunti R, Puppala M, He T, et al. An AI-driven clinical care pathway to reduce 30-day readmission for chronic obstructive pulmonary disease (COPD) patients. Sci Rep. 2022;12(1):20633. https://doi.org/10.1038/s41598-022-22434-3
47. Gala D, Behl H, Shah M, Makaryus AN. The role of artificial intelligence in improving patient outcomes and future of healthcare delivery in cardiology: a narrative review of the literature. Healthcare. 2024;12(4):481. https://doi.org/10.3390/healthcare12040481
48. Hossain S, Hasan MK, Faruk MO, Aktar N, Hossain R, Hossain K. Machine learning approach for predicting cardiovascular disease in Bangladesh: evidence from a cross-sectional study in 2023. BMC Cardiovasc Disord. 2024;24(1):214. https://doi.org/10.1186/s12872-024-03883-2
49. Chengyu Z, Xueyan H, Ying F. Research on disease management of chronic disease patients based on digital therapeutics: a scoping review. Digit Heal. 2024;10:20552076241297064. https://doi.org/10.1177/20552076241297064
50. Selvaprasanth P, Ramasamy S. AI-Powered prediction of chronic kidney disease : a machine learning perspective. J Sci Technol Res. 2025;1(6):1–18.
51. Subrata SA. The role of artificial intelligence in chronic illness care: navigating challenges in clinical nursing practice. Pacific Rim Int J Nurs Res. 2025;29(2):213–8. https://doi.org/10.60099/prijnr.2025.272061
52. Budi Susilo YK, Rahman SA, Amgain K, Yuliana D. Artificial intelligence-powered precision medicine for cardiovascular disease prevention and management. medrxiv. 2025. https://doi.org/10.1101/2025.03.23.25324474
53. Provost C, Broughan J, McCombe G, Kelly MO, Ledwidge M, Cullen W, et al. Artificial intelligence (AI) Models for cardiovascular disease risk prediction in primary and ambulatory care: a scoping review. medrxiv. 2025. https://doi.org/10.1101/2025.03.21.25324379
54. Mihan A, Pandey A, Van Spall HGC. Artificial intelligence bias in the prediction and detection of cardiovascular disease. NPJ Cardiovasc Heal. 2024;1(1):31. https://doi.org/10.1038/s44325-024-00031-9
55. Juhn YJ, Malik MM, Ryu E, Wi CI, Halamka JD. Socioeconomic bias in applying artificial intelligence models to health care. In: Artificial intelligence in clinical practice. Amsterdam: Elsevier; 2024. pp. 413–35. https://doi.org/10.1016/B978-0-443-15688-5.00044-9
56. Khosravi M, Zare Z, Mojtabaeian SM, Izadi R. Artificial intelligence and decision-making in healthcare: a thematic analysis of a systematic review of reviews. Heal Serv Res Manag Epidemiol. 2024;11:23333928241234863. https://doi.org/10.1177/23333928241234863
57. Reddy S, Shaikh S. The long road ahead: navigating obstacles and building bridges for clinical integration of artificial intelligence technologies. J Med Artif Intell. 2025;8:7. https://doi.org/10.21037/jmai-24-148
58. Swinckels L, Bennis FC, Ziesemer KA, Scheerman JFM, Bijwaard H, de Keijzer A, et al. The use of deep learning and machine learning on longitudinal electronic health records for the early detection and prevention of diseases: scoping review. J Med Internet Res. 2024;26:e48320. https://doi.org/10.2196/48320
59. Ampavathi A, Tamilselvi P, Kumar Lingamallu R, Sethy A, Gogula S, Sharath MN, et al. Integrative approach for modern health risk modeling and predicting in patients through artificial intelligence method. In: Satyanarayana K, Bobba PB, Perveen A, Debnath S, editors. MATEC Web Conf. Vol. 392; 2024. p. 01144. https://doi.org/10.1051/matecconf/202439201144
60. Chopra H, Annu, Shin DK, Munjal K, Priyanka, Dhama K, et al. Revolutionizing clinical trials: the role of ai in accelerating medical breakthroughs. Int J Surg. 2023;109(12):4211–20. https://doi.org/10.1097/JS9.0000000000000705
61. Panda M, Mahanta SR. Explainable artificial intelligence for healthcare applications using random forest classifier with LIME and SHAP. arXiv:2311.05665. 2023. https://doi.org/10.1201/9781003442509-6
62. Metta C, Beretta A, Pellungrini R, Rinzivillo S, Giannotti F. Towards transparent healthcare: advancing local explanation methods in explainable artificial intelligence. Bioengineering. 2024;11(4):369. https://doi.org/10.3390/bioengineering11040369
63. Carrasco-Ribelles LA, Llanes-Jurado J, Gallego-Moll C, Cabrera-Bean M, Monteagudo-Zaragoza M, Violán C, et al. Prediction models using artificial intelligence and longitudinal data from electronic health records: a systematic methodological review. J Am Med Inform Assoc. 2023;30(12):2072–82. https://doi.org/10.1093/jamia/ocad168
64. You JG, Hernandez-Boussard T, Pfeffer MA, Landman A, Mishuris RG. Clinical trials informed framework for real world clinical implementation and deployment of artificial intelligence applications. NPJ Digit Med. 2025;8(1):107. https://doi.org/10.1038/s41746-025-01506-4
65. Moneer O, Miller JE, Shah ND, Ross JS. Direct-to-consumer personal genomic tests need better regulation. Nat Med. 2021;27(6):940–3. https://doi.org/10.1038/s41591-021-01368-9
66. Teplytska O, Ernst M, Koltermann LM, Valderrama D, Trunz E, Vaisband M, et al. Machine learning methods for precision dosing in anticancer drug therapy: a scoping review. Clin Pharmacokinet. 2024;63(9):1221–37. https://doi.org/10.1007/s40262-024-01409-9
67. Thomas A, Heinemann L. Algorithms for automated insulin delivery: an overview. J Diabetes Sci Technol. 2022;16(5):1228–38. https://doi.org/10.1177/19322968211008442
68. Babel A, Taneja R, Mondello Malvestiti F, Monaco A, Donde S. Artificial intelligence solutions to increase medication adherence in patients with non-communicable diseases. Front Digit Heal. 2021;3:669869. https://doi.org/10.3389/fdgth.2021.669869
69. Chaturvedi A, Kaur J. Leveraging advanced AI agents for unified electronic health records: a novel approach to healthcare interoperability. J Artif Intell Mach Learn Data Sci. 2023;1(4):1481–6. https://doi.org/10.51219/JAIMLD/akhil-chaturvedi/334
70. Sharma V. AI and RPA in healthcare: transforming administrative operations for better outcomes. Int J Multidiscip Res (IJFMR). 2021;3(2):1–18. https://doi.org/10.36948/ijfmr.2021.v03i02.36362
71. Ciecierski-Holmes T, Singh R, Axt M, Brenner S, Barteit S. Artificial intelligence for strengthening healthcare systems in low- and middle-income countries: a systematic scoping review. NPJ Digit Med. 2022;5(1):162. https://doi.org/10.1038/s41746-022-00700-y
72. Odeh VA, Chen Y, Wang W, Ding X. Recent advances in the wearable devices for monitoring and management of heart failure. Rev Cardiovasc Med. 2024;25(10):386. https://doi.org/10.31083/j.rcm2510386
73. Covi E, Donati E, Liang X, Kappel D, Heidari H, Payvand M, et al. Adaptive extreme edge computing for wearable devices. Front Neurosci. 2021;15:611300. https://doi.org/10.3389/fnins.2021.611300
74. Cheng W, Zhang Z, Hoelzer S, Tang W, Liang Y, Du Y, et al. Evaluation of a village-based digital health kiosks program: a protocol for a cluster randomized clinical trial. Digit Heal. 2022;8:205520762211291. https://doi.org/10.1177/20552076221129100
75. Ali MS, Ahsan MM, Tasnim L, Afrin S, Biswas K, Hossain MM, et al. Federated learning in healthcare: model misconducts, security, challenges, applications, and future research directions – a systematic review. arXiv:2405.13832. 2024.
76. World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance. Geneva: WHO; 2021.

Cite this article as:
Fahmy AM. Exploring the Role of AI in Predicting Chronic Disease Progression: Diabetes and Cardiovascular Diseases. Premier Journal of Public Health 2025;4:100021

Export to EndNote Export to Mendeley