Research On Predictive Models Of Amputation After Diabetic Foot Ulcer:Based On Machine Learning

Objective: To build up predictive models of amputation with diabetic foot ulcer using logistic regression and machine learning algorithms,and to evaluate the predictive effectiveness of models and the ability to filter variables.Methods: Clinical data of all diabetic foot ulcer(DFU)patients which were in hospital from Jan.2015 to Dec.2020 were retrospectively collected.The risk factors of amputation in DFU patients were evaluated,and predictive models were built up using logistic regression,decision tree,support vector machine,random forest and XGBoost algorithms.Then the predictive effectiveness and clinical significance of models were compared by sensitivity,specificity,accuracy,precision,areas under curve(AUC)of receiver operating characteristic(ROC)and F1 score.Results: The clinical data of 364 DFU patients were collected,including213 amputated patients and 151 non-amputated patients.Multivariate logistic regression analysis showed that cerebrovascular disease,Wagner>2,fibrinogen,hemoglobin and average platelet distribution width were independent influencing factors of amputation in DFU patients.The sensitivity,specificity and the AUC of this model were 0.861、0.642 and 0.821,respectively,while 78.18% in accuracy and 87.37% in precision,0.867 in F1 score,which had the best overall prediction performance.The prediction efficiency of each model was evaluated by F1 score,and the order is logistic regression,XGBoost,decision tree after pruning,decision tree before pruning,random forest and SVM,while by AUC was as follows: XGBoost,logistic regression,SVM,random forest,decision tree after pruning and decision tree before pruning.Models using machine learning algorithms screened variables from different perspectives,and the results obtained were more clinically valuable.Conclusions: In this study,the prediction efficiency of machine learning model is not better than that of traditional statistical model.Different predictive models are used to evaluate the amputation risk of patients with DFU from different algorithmic,which can provide complementary basis for DFU treatment decision.