Study on risk factors of diabetic peripheral neuropathy and development of a prediction model based on machine learning

Abstract

Abstract (1) Background: Diabetic peripheral neuropathy (DPN) stands as a prevalent complication in individuals with diabetes. This study aims to develop and validate a machine learning-based model to predict the probability of DPN in patients diagnosed with type 2 diabetes mellitus. (2) Methods: We conducted a retrospective analysis of data pertaining to 628 patients with type 2 diabetes mellitus who received treatment at the First Affiliated Hospital of Soochow University between 2022 and 2023. This dataset encompassed medical histories, physical examinations, and results from biochemical index tests. The cohort was divided into training and validation datasets at an 8:2 ratio randomly. Feature selection, parameter optimization, and model construction were carried out within the training set, while the validation set was employed to assess the predictive performance of the models. We utilized machine learning algorithms such as Gradient Boosting Machines (GBM), Random Forest (RF), Support Vector Machines (SVM), Naïve Bayes, Decision Trees (DT) and traditional logistic regression (LR). The performance of these models was evaluated through the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA). To interpret the best-performing model, we employed Shapley Additive exPlanation (SHAP) Plots and Local Interpretable Model Agnostic Explanations (LIME). (3) Results: The dataset, comprising 628 individuals from the First Affiliated Hospital of Soochow University, yielded significant variables following selection by the Boruta algorithm and logistic multivariate regression analysis. These significant variables included Age, HOMA-IR, Duration of diabetes, and (blood urea nitrogen) BUN. The GBM model outperformed the other models, demonstrating an accuracy of 0.9316, an F1-score of 0.9385, and an AUC of 0.9294. The validation set cohorts was further subdivided within the study, indicating that the GBM model remained an effective classifier in different subgroups. (4) Conclusions: The GBM model was composed of age, HOMA-IR, duration of diabetes and BUN may assist doctors with the early identification of DPN in patients with type 2 diabetes mellitus.