Developing A Diagnostic Tool For Chinese Medicine Syndrome In Patients With Chronic Kidney Disease Based On PROs And Decision Tree Method

ObjectiveIn order to promote the diagnostic accuracy of Chinese medicine (CM) syndromes and rational use of Chinese patent products for the western medicine doctors, it is necessary to develop a method to construct a simple and reliable tool. This approach also provides a new thinking and method for the CM syndrome diagnosis study, and it is a demonstration of the construction of other CM syndrome diagnosis.MethodWe collected data from patients with chronic kidney disease who were hospitalized in the nephropathy department wards of Guangdong Province Hospital of Chinese Medicine and the First Affiliate Hospital of Guangzhou University of Chinese Medicine using the PROs information collection tool from October 2015 to March 2016. After data preprocessing, we establish the syndrome diagnosis model using the Clementine SPSS decision tree C5.0 algorithm and CART algorithm. Three aspects were used to evaluate the decision tree model: the accuracy of judgment, the gains chart, the diagnostic test evaluation index of test set (setsensitivity, specificity, positive predictive value, positive likelihood ratio, and Jorden index).ResultsThere were 300 patients were included analysis in the syndrome diagnosis model, which involve three CM Syndrome types,57 variable index. We establish the syndrome diagnosis model using the Clementine SPSS decision tree C5.0 algorithm and CART algorithm. The accuracy rates of C5.0 syndrome diagnosis model for the training set was 75.11%, the accuracy rates of C5.0 syndrome diagnosis model for the test set set was 68%, the accuracy rates of CART syndrome diagnosis model for the training set was 75.11%, the accuracy rates of CART syndrome diagnosis model for the test set set was 62.67%. The accuracy of the two decision tree models were not so good. The gains charts indicated that the two models failed to well provide information. For the C5.0 decision tree model, the sensitivity was 40%, specificity was 70%, positive predictive value was 8.7%, positive likelihood ratio was 1.33, Jorden index was 0.1. For the CART decision tree model, the sensitivity was 27.3%, the specificity was 68.8%, the positive predictive value was 13.0%, the positive likelihood ratio was 0.87 and Jorden index was -0.04. Thus, the test results of the two decision tree model were not so good for CM syndromes classification.ConclusionThe results of this study indicated that the accuracy rates of syndrome diagnosis model were not so good. The gains charts of the decision tree models showed that the model failed to provide a good information. The diagnostic test evaluation index showed that the test results of the two decision tree models were not good for CM syndromes classification.