Prediction of 30-day risk of acute exacerbation of readmission in elderly patients with COPD based on support vector machine model

Background: Acute exacerbation of chronic obstructive pulmonary disease (COPD) is an important event in the process of disease management. Early identification of high-risk groups for readmission and appropriate measures can avoid readmission in some groups, but there is still a lack of specific prediction tools.The predictive performance of the model built by support vector Machine（SVM） has been gradually recognized by the medical field. This study intends to predict the risk of acute exacerbation of readmission in elderly COPD patients within 30 days by SVM, in order to provide scientific basis for screening and prevention of high-risk patients with readmission. Methods: : A total of 1058 elderly COPD patients from the respiratory department of 13 general hospitals in Ningxia region of China from April 2019 to August 2020 were selected as the study subjects by convenience sampling method, and were followed up to 30 days after discharge. Discusse the influencing factors of patient readmission , and built four kernel function models of Linear-SVM, Polynomial-SVM, Sigmoid-SVM and RBF-SVM based on the influencing factors.According to the ratio of training set and test set 7:3, they are divided into training set samples and test set samples, Analyze and Compare the prediction efficiency of the four kernel functions by the precision, recall, accuracy, F1 index and Area under curve(AUC). Results: : Education level, smoking status, coronary heart disease, hospitalization times of acute exacerbation of COPD in the past 1 year, whether Long-term home oxygen therapy, whether regular medication, nutritional status and seasonal factors were the influencing factors for readmission. The training set shows that Linear-SVM、Polynomial-SVM、Sigmoid-SVM and RBF-SVM precision respectively were 69.89, 78.07, 79.37 and 84.21; Recall respectively were 50.78、69.53、78.74 and 88.19; Accuracy respectively were 83.92、88.69、90.81 and 93.82; F1 index respectively were 0.59、0.74、0.79 and 0.86; AUC were 0.722、0.819、0.866 and 0.918。Test set precision respectively were86.36、87.50、80.77 and 88.24; Recall respectively were51.35、75.68、56.76 and 81.08; Accuracy respectively were 85.11、90.78、85.11 and 92.20; F1 index respectively were 0.64、0.81、0.67 and 0.85; AUC respectively were 0.742、0.858、0.759 and 0.885. Conclusions: : The overall prediction performance of SVM model is good, and the Prediction performance of RBF-SVM model is the best.