Comparative Diagnosis Test Accuracy of Five Weighting Methods for Type 2 Diabetes Mellitus with Dampness-heat Syndrome / 中医杂志

ABSTRACT

ObjectiveTo compare the diagnostic accuracy of five different weighting methods of Chinese medicine syndrome and then analyze their diagnostic efficacy and characteristics， by taking Diagnostic Standard for Type 2 Diabetes Mellitus （T2DM） with Dampeness-heat Syndrome （abbreviated as diagnostic standard） as an example. MethodsData from expert questionnaire on the diagnostic standard and a cross-sectional survey of 1021 patients were collected. The comparative diagnostic test accuracy （CDTA） method was used to calculate the area under the ROC curve （AUC）， area under the PR curve （AUPR）， accuracy （ACC）， sensitivity， and specificity of five commonly used weighting methods in two categories， including knowledge-driven weighting methods （expert scoring synthesis method， analytic hierarchy process， and precedence chart method） and data-driven weighting methods （logistic regression contribution method and entropy weighting method）. ResultsAmong 1021 patients with T2DM， 389 cases were diagnosed as dampness-heat syndrome. The expert scoring synthesis method， analytic hierarchy process method， and precedence chart method were basically consistent in the weight scores of each item. The expert scoring comprehensive method， analytic hierarchy process method， and entropy weighting method have a smaller difference in the weight scores of each item， while there was larger difference in the weight scores of each item of the precedence chart method and the logistic regression contribution method. The AUC （95% CI）， AUPR， ACC， sensitivity， and specifi-city of the expert scoring synthesis method were 0.913 （0.893， 0.932）， 0.851， 0.870， 0.868 and 0.875， respectively； while those of the analytic hierarchy process method were 0.910 （0.890， 0.930）， 0.838， 0.879， 0.848 and 0.896； of the precedence chart method were 0.919 （0.900， 0.937）， 0.858， 0.875， 0.871 and 0.875； of the logistic regression contribution method were 0.867 （0.842， 0.891）， 0.792， 0.853， 0.769 and 0.898； and of the entropy weighting method were 0.895 （0.873， 0.916）， 0.820， 0.869， 0.802 and 0.908. ConclusionThe knowledge-driven weighting methods are better than the data-driven weighting methods in terms of diagnostic efficacy and reflecting expert experience.