Study On The Treatment Strategy Of Diabetic Foot Based On Random Forest

Objective:To screen out the clinical factors related to the decision of patients’ treatment mode through the analysis of the database of clinical sources,so as to provide help for clinical treatment.Methods:Clinical patients were classified according to the mode of treatment,including conservative treatment group,surgical intervention group and amputation foot.The data come from the data of 696 patients with diabetic foot and diabetic foot ulcer diagnosed by the first Hospital of Jilin University and the second Department of the first Hospital of Jilin University from March 2012 to June 2021.Exclusion criteria include:（1）patients with other systemic or local infectious diseases（2）patients with hematological diseases（3）patients with rheumatic immune diseases（4）patients with immunodeficient diseases（5）patients receiving immunosuppressive therapy（6）type 1 diabetes（7）gestational diabetes（8）other specific types of diabetes（9）malignant tumors.General data of patients,clinical data and hospitalization information were collected.Methodology:all the statistical analysis and random forest methods in the study are implemented by software Rversion4.0.4.Classified variables are represented by means,and numerical variables are represented by mean ±standard deviation.The mean values of numerical variables were analyzed by Kruskal-Wallis univariate ANOVA analysis,and the data of each group were compared.In the construction of the model,k sample sets are randomly selected by bootstrap method to build k decision trees.Then b operational features are extracted from the variable group,and the nodes are determined after calculating the information entropy of the variables.Pruning is carried out by reading the previous literature,and finally the training set data are classified by random forest,and the classification result is determined by each small classifier.By drawing the working characteristic curve of the subjects,the sensitivity and specificity and the maximum points were selected as cut-off points.The area under the subject working curve（ROC）（AUC）in the model was calculated as the evaluation index of the model effectiveness.Results:（1）Wagner grading standard,neutrophil percentage,absolute value of neutrophils,fibrinogen and hypersensitive C-reactive protein play a decisive role in the treatment of patients with diabetes.（2）The baseline data related to the timing of treatment,such as the course of disease,age and sex of patients,play only a weak role in decision-making.（3）In this model,we still retain the relevant factors that have been reported in the previous literature to affect the severity of diabetic foot and treatment decision-making,although it has a great impact on the model,but these clinical data and baseline data are also indispensable important factors in clinical decision-making.The model built in this paper can help doctors to make clinical decisions to some extent.