Establishment and external verification of an early warning model of diabetic peripheral neuropathy based on random forest logistic regression

The aim of this study was to explore the risk factors of diabetic peripheral neuropathy (DPN) based on clinical data and biochemical indicators, and to build an early diagnosis and prediction model. Retrospective data of 1446 patients with diabetes in the First Affiliated Hospital of Anhui Province were collected and divided into training set and internal validation set according to 7:3 ratio, and 360 patients with diabetes in the second affiliated hospital were used as external validation cohort. The single variable logistic regression feature selection was carried out on the training set, and the variables with p value < 0.05 were determined and the logistic regression model was established. At the same time, the random forest algorithm is applied to the training set, and the 10 most important features are identified through grid search combined with cross-validation to optimize the hyperparameters. ROC curve, decision curve analysis and correction curve were used to evaluate the model performance, Hosmer-Lemeshow test evaluated the model, Brier score evaluated the random forest model, and further verified the model through ten fold cross-validation and SHAP analysis to enhance the model interpretibility. Logistic regression was used to establish a neagram model with limb numbness, limb pain, diabetic retinopathy, diabetic nephropathy, urinary protein, diastolic blood pressure, white blood cell count, HbA1c, HDL cholesterol as the main features. The AUC of the training set, validation set and test set of the model are 0.91, 0.88 and 0.88, respectively, and the average AUC of the ten-fold cross-validation is 0.902. The AUC of the training set, the validation set and the test set of the random forest model were 0.95, 0.88 and 0.88, respectively, with an average AUC of 0.88 and a 10-fold cross-validation AUC of 0.886. The results show that the random forest model has good practical application effect.