چكيده به لاتين
Diabetes is a chronic disease associated with an increase in blood glucose levels, posing serious health risks such as diabetic nephropathy, which describes kidney damage. Diabetic nephropathy is a significant complication of diabetes, impacting kidney function from mild to severe levels and requiring specialized interventions and care. The presence of microalbuminuria, defined as the excretion of more than 30 milligrams of protein in 24 hours, is a hallmark of this disease. Diagnosing such conditions often incurs substantial costs due to the necessity of numerous tests. Nowadays, abundant data on various diseases are collected, and data mining serves as a mechanized system to aid in discovering existing patterns and predicting future occurrences, contributing significantly to medical advancements. In this research, eight machine learning algorithms, including k-Nearest Neighbors, Decision Tree, Random Forest, Logistic Regression, Support Vector Machine, Extreme Gradient Boosting, Light Gradient Boosting, and AdaBoost Classifier, were employed to predict diabetic nephropathy in a dataset of 6403 patients with type 2 diabetes. Furthermore, the study evaluated the roles of TyG, AIP, AVI, and the HDL/LDL ratio in the occurrence of diabetic nephropathy. Given the importance of early screening in diabetic patients to control microalbuminuria, a composite index of anthropometric features was developed to determine the patients' need for key tests for the diagnosis of the disease. As a result, the study achieved an accuracy of over 94% using the Catboost algorithm. Direct associations between the TyG, AIP, and AVI indices and an inverse relationship between the HDL/LDL ratio and diabetic nephropathy were identified. Finally, three effective indices for the early prediction of patients at risk of nephropathy, with a high sensitivity of 80%, were identified.
