Early Detection of Diabetic Nephropathy Using Machine Learning Models and Retrospective Clinical Data

Abstract

In this work, we aimed to develop machine learning models for the identification of diabetic nephropathy (DN) during early stages using retrospective clinical data. Using a variety of patient data such as demographic information, lab test results, and medical history; through the training of various models such as Random Forest, SVM and Gradient Boosting, we hoped to predict the onset of DN at earlier stages. Model performance was evaluated through accuracy, precision, recall, and area under the receiver operating characteristic curve (AUC). The Gradient Boosting model had an AUC of 0.92, the best of all models, indicating superior discrimination between patients who would go on to develop DN and those who would not. We also found that the addition of renal biomarkers including serum creatinine concentrations and albuminuria significantly increased the predictive capability of the model. This study highlights the use of machine learning methods that can timely screen and identify individuals at high risk for DN and intervene to prevent progressive DN progression. With implications for clinical practice, this research adds to the prospect on AI based diagnostic modalities to facilitate early identification and renal preservation, thus preventing the long-term complications of diabetic kidney disease.