The Application Of Different Statistical Methods In The Study Of Chinese Medicine Multi-center Clinical Trial "center Effect"

Backgrounds:Statistical analysis is a very important part of clinical trials, and plays a vital role of the whole clinical research. Standard statistical analysis is not only the foundation and basis of evaluating the safety and effectiveness of drugs correctly, but also an important symbol of standardized and high-quality clinical trial. It is mostly used randomized, blinded, controlled form and also carried out in all centers in the field of clinical trials of traditional Chinese medicine(TCM). So it is big concerned about which statistical methods should be used and how to control the central effects and comprehensively consider about the specificity of TCM.Objectives:Comprehend the circumstances of clinical evaluation of new drugs at home and abroad, combined with the specific circumstances of evaluation of multi-center clinical trials of TCM and on account of the specificity of TCM treated the disease which is called "treatment with syndrome differentiation", and discuss the rationality and feasibility of the different statistical methods applied in clinical trials. Through comparing of four statistical methods and making relevant suggestions, in order to provide methodological reference for the statistical analysis of clinical evaluation of new drugs.Methods:By consulting relevant literature and books, make a deep understanding of four statistical methods' theory and using conditions which include CMH test, Logistic regression, covariance analysis and META analysis. The data was obtained by participating and observing aⅢclinical trial which is some TCM treats insomnia. Discuss the application of the four statistical methods through making study of the data. Use EpiDATA3.1 for data entry, and SAS8.2 and Revman5 Software for statistical analysis.Results:TCM syndrome's effect and TCM syndrome scores both are analyzed in the instance. The data whose response variable is the secondry or the fourth can be analyzed by CMH test after deduction of central effect. We can test central effect by Breslow-Day test when the response variable is the sencondary. And logistic regression should meet the prerequisite assumption which is that the response variable is parallel lines when the response variable is ranked. But in the instance, group and center both are categorical variable, so we don't need to meet the prerequisite. When the response variable is binary variable, it does not need a parallel line hypothesis testing, so logistic regression is applicable. We can obtain regression model and the value of OR, and we also can test central effect by likelihood ratio test. For the TCM syndromes, we use covariance analysis, considering the impact of center, group and the baseline. Because there is interaction between the centre and the group, which does not meet the analysis of covariance conditions, we can just only analyze from each centre. The META analysis can analyze both TCM syndrome's effect and TCM syndromes scores. The two syndromes categorical's results of TCM syndrome's effect are the same as CMH test's results. But because of using Revman5 software, only for dichotomous response variable can be analyzed. Because the difference between TCM syndromes scores has heterogeneity, we adopted the random effects model. We can get those from the instance. First, for statistical methods of count data, we can use CMH test, logistic regression and META analysis of the binary variables. But for the logistic regression analysis of multi-valued variables, prerequisite is required. META analysis is limited by the software limit, so it's lack of analysis of multi-valued variables. Second, for statistical methods of the measurement data, analysis of covariance needs each group are from the same normal population whose variance is equal and the overall linear regression coefficient is equal. META analysis can select model through the results of the heterogeneity test.Conclusions: Central effect is very important problem in the multi-center clinical trials of TCM. For count data, we can take CMH test, logistic regression and META analysis for reduction central effect. For the measurement data, the analysis of covariance need to meet the conditions of its application, META analysis need to consider the generating reasons center effects, then take the appropriate model. It is suggested to approach the evidence-based medicine and constantly update statistical theory and methods, then find the statistical methods which are considered the various effects.