• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Wednesday, November 18, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Health

Seventy Systematic Reviews Reveal Artificial Intelligence in Diabetes Faces Implementation Barriers

by
November 18, 2026
in Health
Reading Time: 6 mins read
0
Seventy Systematic Reviews Reveal Artificial Intelligence in Diabetes Faces Implementation Barriers

Seventy Systematic Reviews Reveal Artificial Intelligence in Diabetes Faces Implementation Barriers

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

A comprehensive tertiary review examining 70 systematic reviews of artificial intelligence in diabetes mellitus reveals that while computational research has accelerated rapidly, the overall evidence base remains clinically fragmented and heavily skewed toward specific disease subtypes and modalities. Led by Sander M. W. Janssen and colleagues, the synthesis evaluates secondary studies published between 2014 and October 2025 across multiple clinical domains. Rather than demonstrating immediate translational readiness, the secondary literature indicates that AI development is at a transitional stage characterized by uneven methodological rigor and significant barriers to routine clinical deployment. Across the analyzed literature, researchers identified sharp disparities between technically mature experimental algorithms and their practical integration into real-world healthcare settings. By examining reviews spanning type 1, type 2, and gestational diabetes, as well as critical complications, the investigation offers the first high-level systematic overview assessing the structural maturity and reliability of artificial intelligence across the entire diabetes continuum.

The urgency surrounding computational tools stems from the immense and expanding global burden of diabetes mellitus and its complex multi-system complications. Source data indicate that diabetes prevalence reached approximately 9.3 percent of the global population in 2019, with projections indicating an increase to 10.2 percent, amounting to 578 million individuals, by 2030 and 10.9 percent, or 700 million people, by 2045. More recent estimates from the Global Burden of Disease Study documented 529 million affected individuals in 2021, with numbers anticipated to escalate to 1.31 billion by 2050. The condition manifests predominantly as type 2 diabetes, which represents more than 90 percent of all cases and had an estimated adult prevalence of 10.5 percent in 2021. Meanwhile, autoimmune type 1 diabetes affects roughly 9 million people worldwide, including more than 1.5 million children, and gestational diabetes complicates an estimated 14 percent of pregnancies globally. Managing these conditions requires continuous monitoring, personalized behavioral strategies, and timely interventions to avert severe microvascular and macrovascular sequelae.

The clinical burden of diabetes is compounded by debilitating complications that provide distinct targets for automated diagnostic and predictive technologies. Diabetic kidney disease, also termed diabetic nephropathy, affects up to 40 percent of individuals with diabetes and represents the foremost cause of chronic kidney failure globally, progressing through metabolic, hemodynamic, and inflammatory pathways that frequently remain clinically silent until advanced organ damage occurs. Diabetic foot ulcers present another severe manifestation, exhibiting a worldwide prevalence of 6.3 percent among individuals with diabetes; between 50 and 60 percent of these ulcerations progress to active infections, 15 percent lead to lower-limb amputations, and patients face a five-year mortality rate 2.5 times higher than those without ulcers. In parallel, diabetic retinopathy constitutes the leading cause of preventable adult blindness, with an estimated 22 percent prevalence affecting approximately 103 million people in 2020 and projected to reach 160 million by 2045. These disparate pathologies generate unique diagnostic imaging, physiological sensor feeds, and clinical records that demand targeted computational approaches.

To rigorously evaluate the existing secondary evidence base, the investigators implemented a tertiary review protocol based on guidelines formulated by Kitchenham and Charters for software engineering and adapted from medical syntheses. The study executed comprehensive multi-database queries across PubMed, Scopus, Web of Science, and IEEE Explore covering literature published from January 1, 2014, through October 31, 2025. An initial retrieval yielded 912 citations, which narrowed to 595 records following the removal of duplicate entries across platforms. Title and abstract screening subsequently identified 155 candidate publications for full-text eligibility appraisal against explicit inclusion criteria. A total of 63 studies were excluded during this stage, consisting of 44 papers that lacked systematic methodology, 14 that did not align with specific clinical research questions, and five non-journal items. The remaining 92 secondary reviews underwent formal quality assessment using the four-item DARE criteria established by York University Centre for Reviews and Dissemination, with 22 low-scoring studies eliminated.

The final curated corpus comprised 70 rigorous secondary studies, encompassing 60 systematic reviews and 10 formal meta-analyses distributed across prominent international biomedical publications. The Journal of Medical Internet Research hosted five of these reviews, followed by Artificial Intelligence in Medicine with four, though most contributing journals published only one or two eligible syntheses. Geographically, institutional affiliations were concentrated in the United Kingdom with 13 publications, the United States with 10, and China with nine, collectively accounting for 32 of the 70 included works. Publication trends revealed a pronounced acceleration after 2020, reaching an annual peak of 16 secondary reviews in 2025 alone. Across all 70 syntheses, the authors recovered complete inclusion lists representing 2,382 citations and 1,988 unique primary investigations, with only 292 primary studies appearing in multiple reviews. Corrected covered area analysis demonstrated a slight overall overlap of 0.29 percent, confirming that the analyzed secondary literature generally drew from distinct primary research pools.

Mapping the distribution of computational research revealed substantial structural imbalances across clinical categories, with primary attention centered on type 2 diabetes across 50 reviews, closely followed by type 1 diabetes. In contrast, gestational diabetes and diabetic kidney disease remained the least researched clinical categories in the synthesized corpus. A co-occurrence network analysis of titles and abstracts delineated five primary interconnected research domains: clinical and behavioral evaluations assessing digital tools like chatbots and lifestyle interventions against biomarkers such as glycated hemoglobin; supervised learning and predictive modeling using traditional algorithms; decision-support and continuous monitoring utilizing deep learning and continuous glucose monitor streams; precision and reliability studies focusing on false positives, sensitivity limits, hypoglycemia, and neuropathy; and systemic integration examining interpretability and data governance. Dense conceptual clustering showed that these domains evolve simultaneously, yet their close interaction highlights how clinical utility remains contingent on addressing technical and methodological vulnerabilities.

Detailed examination of analytical pipelines revealed distinct pairings between data modalities and machine learning paradigms across different clinical challenges. Tabular datasets extracted from electronic medical records and registries heavily dominated research, appearing in 31 type 2 diabetes reviews and 21 type 1 diabetes reviews. Conversely, image-based inputs were concentrated in diabetic foot ulcer screening across 12 reviews and diabetic retinopathy across 10 reviews, where automated computer vision algorithms were predominantly deployed. Wearables and continuous glucose monitoring data featured primarily in type 1 diabetes across 14 reviews and type 2 diabetes across 17 reviews, feeding predictive time-series models and clinical decision-support architectures for insulin titration. However, advanced computational paradigms remained conspicuously rare: natural language processing appeared in only seven reviews, reinforcement learning in just five reviews, and genomics or omics in a negligible fraction. This pronounced concentration highlights that algorithmic exploration has largely adhered to conventional tabular machine learning rather than adaptive, multimodal approaches.

Through bibliometric clustering of 73 frequent limitation terms, the tertiary review categorized widespread weaknesses that undermine the validity and generalizability of published machine learning models. The dominant limitation cluster centered on data bias, stemming from an over-reliance on a narrow collection of public, homogeneous, non-Western datasets that fail to capture ethnic and demographic variations in diabetes pathogenesis. Another prominent challenge involved analytical opacity and model interpretability, where complex neural architectures operate without clear feature attribution, leaving clinicians unable to discern underlying predictive reasoning. Furthermore, evaluation methodologies frequently conflated internal statistical validation with true external reproducibility. A majority of primary studies assessed performance using internal data splits from single institutions rather than prospective, external cohorts in diverse clinical environments, which often inflates reported discrimination accuracy while masking poor generalizability, calibration deficits, and vulnerabilities to operational dataset shifts in real-world practice.

Beyond algorithmic design, the synthesized evidence exposed systemic organizational bottlenecks that impede the routine integration of artificial intelligence into day-to-day healthcare workflows. Practical deployment is frequently obstructed by technical incompatibilities with legacy electronic health record systems, creating substantial friction and adding to clinician administrative burdens rather than alleviating them. Furthermore, healthcare organizations encounter resistance among clinical staff due to limited digital literacy, unresolved ethical dilemmas, and ambiguities regarding regulatory compliance and patient data privacy. Crucially, the secondary literature rarely evaluates the ongoing maintenance demands of deployed clinical models, overlooking operational hazards such as algorithmic drift over time, software updating expenses, and the periodic recalibration of embedded clinical decision rules. Implementation appears most feasible where clinical tasks are tightly circumscribed and supported by standardized existing workflows, such as automated retinal screening and targeted decision support for insulin dosing, whereas expansive multi-task systems face persistent friction.

To transition from experimental algorithmic validation to trustworthy clinical utility, the tertiary review outlines explicit structural and methodological requirements for future diabetes research. Investigators must expand beyond homogeneous cohorts by curating multi-institutional datasets and leveraging federated learning to preserve privacy while capturing diverse patient demographics. Future inquiry should integrate underrepresented modalities, incorporating natural language processing for unstructured clinical narratives, multi-omics biomarkers, and reinforcement learning for closed-loop adaptive insulin delivery. Surprisingly, despite searches extending through late 2025, no included systematic review directly examined large language models, exposing a temporal lag between primary exploratory studies and secondary syntheses. Finally, advancing the field demands explainable artificial intelligence frameworks, such as SHAP and LIME, alongside standardized prospective trials that evaluate hard clinical outcomes, equitable access, and patient safety rather than isolated retrospective accuracy metrics.

Subject of Research: Tertiary review of artificial intelligence applications across diabetes mellitus care and complications

Article Title: Artificial intelligence in diabetes mellitus: a tertiary review of the current landscape and future research directions

Article References: Janssen, S. M. W., Bouzembrak, Y., Cuypers, M., Tekinerdogan, B., & Leusink, G. (2026). Artificial intelligence in diabetes mellitus: a tertiary review of the current landscape and future research directions. Frontiers in Endocrinology, 17, Article 1933129. https://doi.org/10.3389/fendo.2026.1933129

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1933129

Keywords: artificial intelligence, diabetes mellitus, type 1 diabetes, type 2 diabetes, gestational diabetes, diabetic retinopathy, diabetic foot ulcer, diabetic nephropathy, tertiary review, clinical decision support, machine learning, continuous glucose monitoring, computer vision, methodological quality, external validation, data bias

News Source: Blake Davidson. (November 18, 2026). Seventy Systematic Reviews Reveal Artificial Intelligence in Diabetes Faces Implementation Barriers. Scienmag.

Tags: Artificial IntelligenceClinical decision supportComputer VisionContinuous Glucose Monitoringdata biasDiabetes mellitusdiabetic foot ulcerDiabetic nephropathydiabetic retinopathyexternal validationgestational diabetesMachine Learningmethodological qualitytertiary reviewType 1 diabetesType 2 diabetes
Share12Tweet7Share2ShareShareShare1

Related Posts

Italian Medical Amendment Reinstating Sanctioned Physicians Sparks Ethical and Public Health Concerns

Italian Medical Amendment Reinstating Sanctioned Physicians Sparks Ethical and Public Health Concerns

November 18, 2026
Elective Case Shows Automated Pressure Monitoring For Broviac Catheter Transfusion

Elective Case Shows Automated Pressure Monitoring For Broviac Catheter Transfusion

November 18, 2026

Restraint Use and Consultations Associated With Longer Psychiatric Emergency Stays

November 18, 2026

Israeli Study Links Water Fluoridation Halt to Increased Tooth Decay in Children

November 18, 2026

POPULAR NEWS

  • Colonial Legacies and Environmental Shifts Drive Mesoamerican Zoonotic Disease Vulnerabilities

    Colonial Legacies and Environmental Shifts Drive Mesoamerican Zoonotic Disease Vulnerabilities

    29 shares
    Share 12 Tweet 7
  • Motor Imagery Vividness Shows No Link to Parkinson’s Motor Symptom Severity

    29 shares
    Share 12 Tweet 7
  • Brain Tumour Review Identifies Ferroptosis Checkpoints and Spatial Therapeutic Vulnerabilities

    29 shares
    Share 12 Tweet 7
  • Hepatitis C Vesicles Maintain Immunosuppressive Signals Despite Antiviral Cure in Cirrhosis

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Sequencing Links Tubulin Gene Variants to Cases Diagnosed With Dandy-Walker Malformation

Italian Medical Amendment Reinstating Sanctioned Physicians Sparks Ethical and Public Health Concerns

Elective Case Shows Automated Pressure Monitoring For Broviac Catheter Transfusion

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 88 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.