| Objective:This study has three main objectives. The first one is to explore the application of second-order confirmatory factor analysis in TCM (Traditional Chinese Medicine) syndrome elements. Next, in practical clinical applications, is to explore different experts’recognition to the classification of syndromes and syndrome elements and the degree of consistency to each index. The third objective expects to compare whether the two EFA (Exploratory Factor Analysis) models based on covariance matrices and polychoric correlation coefficient have differences. In a word, the third objective is to draw application conditions of the two models in TCM clinical dialectical process.Method:Firstly, second-order confirmatory factor analysis based on covariance matrices and polychoric correlation coefficient matrices was used to extract syndrome elements by Mplus, and to find therelationships among syndrome elements, syndromes and four diagnostic information. Secondly, according to the results of confirmatory factor analysis, expert questionnaire was designed and experts of hypertension were invited to give subjective ratings to justify the importance of the syndrome elements optional index, the rationality of the name of the syndrome’s expression, the precision of the name of the syndrome elements’ expression, and by using the method of experts’ consultation to evaluate experts’ scoring consistency. In the end, Monte Calor simulations were excuted by Mplus. The original datum of several different conditions in terms of different sample size, different factor loading, different number of factors were simulated. Then the datum were transformed into ordered categorical data, covariance coefficient EFA and polychoric correlation coefficient EFA were applied to fit the data, then the pros and cons of each EFA was evaluated in terms of identification and fitting.Result:The results of second-order confirmatory factor analysis show that five hypertension syndrome, respectively, can be divided into two or three syndrome elements. This study names these elements from two aspects:the nature of disease and location of disease. Among these elements, the locations of disease contain liver, kidney, spleen, etc. The natures of diease contain qi deficiency, qi depression, Yin deficiency, virtual, yang hyperactivity, excess heat, wetness hyperactivity, etc. Comparing the results of two EFA models, polychoric correlation coefficient EFA is slightly better than covariance coefficient EFA. Comparing the fit indexs’ results of two CFA methods, polychoric correlation coefficient CFA is similar to covariance coefficient CFA. The result of Delphi method indicates that six syndromes consistency evaluation coefficient is range from 0.171 to 0.271, and P<0.05. So this result shows that experts have high degree of consistency. The result of Monte carlo simulations show that when the sample size and factor loading is under 250 and 0.5 respectively, the identifying ability and model index of polychoric correlation coefficient EFA is better than that of covariance coefficient EFA. But with the increase of sample size and factor loading, the two EFA models tend to be similar. And no matter what kind of method is used, factor loading and sample size is larger, the number of factors is less, the easier it is to build a model.Conclusion:The results of second-order confirmatory factor analysis can be considered that polychoric correlation coefficient EFA is slightly better than traditional covariance coefficient EFA. Polychoric correlation coefficient CFA is similar to covariance coefficient CFA. And the study of Monte Carlo simulation indicates that when we address the binary or ordered categorical variables, on the condition of large sample size, low factor loading and large number of factor, polychoric correlation coefficient EFA is slightly better than traditional covariance coefficient EFA.Other factors need to be studied further which can influence the model by these two methods. |
PDF Full Text Request