International Journal of Clinical Case Reports, 2025, Vol.15, No.6, 293-302 http://medscipublisher.com/index.php/ijccr 293 Feature Review Open Access Accuracy Analysis of AI Prediction Models in Early Screening for Diabetes JieZhang Institute of Life Science, Jiyang College of Zhejiang A&F University, Zhuji, 311800, Zhejiang, China Corresponding email: jie.zhang@jicat.org International Journal of Clinical Case Reports 2025, Vol.15, No.6 doi: 10.5376/ijccr.2025.15.0030 Received: 07 Oct., 2025 Accepted: 13 Nov., 2025 Published: 28 Dec., 2025 Copyright © 2025 Zhang, This is an open access article published under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Preferred citation for this article: Zhang J., 2025, Accuracy analysis of AI prediction models in early screening for diabetes, International Journal of Clinical Case Reports, 15(6): 293-302 (doi: 10.5376/ijccr.2025.15.0030) Abstract This study explores the performance of artificial intelligence prediction models in the early screening of diabetes and the main influencing factors, and summarizes the current prevalence of diabetes and prediabetes, the shortcomings of traditional screening methods, as well as the significance of defining high-risk populations and conducting early screening. This study analyzes; Common AI/ML models such as logistic regression, decision tree, random forest, support vector machine, deep neural network and convolutional neural network are used, and their performance in terms of discriminative ability such as AUC, sensitivity and specificity is compared with that of traditional risk scoring tools. Meanwhile, the key factors affecting the model's accuracy are analyzed from aspects such as data source type, sample representativeness, feature engineering, data quality, sample size and category imbalance. AI prediction models have high discriminative power and efficiency in the early screening of diabetes. However, due to limitations such as retrospective design, small sample size, and insufficient external validation, their universality and wide application still face challenges. In the future, it is necessary to carry out multi-center and large-sample prospective studies, promote the standardization of model development and reporting, enhance interpretability and cross-population transferability, and on the basis of improving supervision and multi-disciplinary collaboration, promote the safe, effective and fair application of artificial intelligence in the early screening of diabetes. Keywords Early diabetes screening; Artificial intelligence; Machine learning; Predictive models; Interpretability 1 Introduction Diabetes, including type 1 and type 2, has become a major global public health issue, with its prevalence and related complications continuing to rise. Recent epidemiological data indicate that more than 462 million people worldwide have type 2 diabetes, accounting for more than 6% of the global population, and it is expected to continue to increase in the coming decades (Khan et al., 2019). Under the influence of urbanization, sedentary lifestyle and dietary changes, the prevalence rate in low-and middle-income areas is also rising rapidly (Tinajero and Malik, 2021). Diabetes is one of the leading causes of death and significantly increases its incidence through complications such as cardiovascular diseases, renal failure, neuropathy and vision loss (Khan et al., 2019). The economic impact is equally significant, with direct medical costs and indirect losses due to productivity decline and premature death amounting to billions of dollars annually. Therefore, there is an urgent need to formulate effective strategies to curb the prevalence of diabetes and mitigate its long-term impacts on health and the social economy. Early detection and intervention are generally regarded as the key to the prevention and treatment of diabetes and its complications. Timely screening can identify high-risk populations in the early stage of the disease, promote lifestyle changes and medical management, thereby delaying or preventing disease progression and reducing serious consequences. However, traditional screening mostly relies on fasting blood glucose, oral glucose tolerance test or HbA1c determination, which has many limitations in practical application, including dependence on laboratory conditions and the need for patients to visit multiple times, especially hinders its wide promotion in resource-limited areas. Furthermore, a considerable proportion of diabetes cases remain undiagnosed. Research reports indicate that as many as 40% of patients are unaware of their condition, further increasing the burden on public health (Ali et al., 2025). The limitations of traditional methods highlight the necessity of innovative, scalable and easily accessible screening tools to enhance early detection rates and cover underserved populations.
RkJQdWJsaXNoZXIy MjQ4ODYzNA==