سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / اولین همایش بین المللی تحول خدمات سلامت: تلفیق مراقبت، درمان، بازتوانی و امیدبخشی

An Integrated Framework for Chronic Disease Prediction and Management Based on Machine Learning and Clinical Data Analysis: A Systematic Review of Applications, Challenges, and Future Directions

نویسندگان :

Alireza Talebi Tadi¹ Alireza Ostadzadeh² Shohreh Ajoudanian³

1- دانشگاه شهید اشرفی اصفهانی 2- دانشگاه شهید اشرفی اصفهانی 3- دانشگاه شهید اشرفی اصفهانی

کلمات کلیدی :

Machine Learning،Chronic Diseases،Prediction،Clinical Data،Systematic Review،Explainable AI،Artificial Intelligence

چکیده :

Background: Chronic non-communicable diseases (NCDs), including diabetes, cardiovascular disease, cancer, and chronic respiratory disease, remain the leading cause of death worldwide. Machine learning (ML) and deep learning (DL) have emerged as powerful tools for early prediction, diagnosis, and management of these conditions, with demonstrated applications spanning structured clinical data, medical imaging, wearable sensor streams, and unstructured electronic health record (EHR) text. Methods: This systematic review was conducted using the Integrated Flexible Systematic Review (IFSR) framework (Daraio, Di Leo et al. 2025) combined with the PRISMA 2020 guideline (Page, McKenzie et al. 2021). A systematic search was performed across Scopus, PubMed, Web of Science, and IEEE Xplore for records published between January 2019 and July 2026. Sixty-six references were synthesized in total: 59 core studies identified through the systematic search and forward/backward snowballing, and 7 foundational or methodological references used to ground the review's methods and discussion. Findings: The included literature clusters around three broad disease domains diabetes (11 studies), cardiovascular disease including hypertension and stroke (8 studies), and cancer (4 studies) alongside chronic kidney/liver disease, chronic respiratory disease, and mental-health/hormonal comorbidities. Random Forest, XGBoost, and Logistic Regression remain the most frequently reported classical algorithms, while LSTM, CNN, and attention-based deep architectures report the strongest performance ranges in imaging- and sequence-based tasks. Seven persistent barriers to clinical adoption were identified, most frequently low model interpretability, data scarcity/quality, and privacy/security constraints. Most strikingly, a Technology Readiness Level (TRL) assessment of the 59 core studies found that 64% remain at the conceptual or review-synthesis stage, 36% reach single-center validation, and 0% report prospective, multi-site clinical deployment indicating that, within this review's sample, no reviewed system has yet reached routine clinical use. Conclusion: A proposed four-layer integrated framework Data Collection, Machine Learning Analysis, Clinical Decision-Making, and Security & Privacy is offered as a roadmap for translating this literature into deployable chronic-disease prediction and management systems. Realizing this potential will require sustained investment in explainable AI, federated and privacy-preserving learning architectures, standardized external-validation protocols, and context-specific pilot studies, including in under-represented healthcare systems such as Iran's.