Generalized Linear Mixed Models In The Center Of The Multiple Application Of Traditional Chinese Medicine (tcm) Clinical Trial

Backgrounds:The center effect usually exists in data from the multi-center clinical trials of Traditional Chinese medicine. And also the data is not independent. In order to ensure the quality of the clinical trial and obtain authentic research results, to ensure the specification of the clinical trials is necessary, and the choice of statistical methods is more important. At present, China use traditional statistical methods for statistical analysis of multi-center clinical trials. However, these methods are more demanding requirements for data analysis, so when the data type does not meet the conditions, the conclusions tend to be inaccurate. Therefore, to select the appropriate statistical methods is of great importance.Objective:According to the theoretical research of generalized linear mixed model, we will analyze the qualitative data of multi-center clinical trials of Traditional Chinese medicine by generalized linear mixed models and explore its applying in multi-center clinical trials of Traditional Chinese medicine.Methods:1. The research expounds the theory of generalized linear mixed model systematically.2. Case study: data is from a traditional Chinese medicine clinical trial of the randomized, double-blind, double-dummy designed. The trial selected seven centers, four visit viewpoints. Statistical methods are generalized linear mixed model and traditional statistical methods.Results:The results of traditional statistical methods:the results of Wilcoxon rank test is that there is statistical significance between the experimental group and the control group (P<0.05), there is statistical significance between the experimental group and the control group (P<0.05) in the2nd center, the4th center, the6th center (P<0.05). The results of Chi-square test is that TCM symptom classification efficacy between the experimental group and the control group has statistically significance (P<0.05), the difference of TCM syndrome classification efficacy of the6th center is statistically significant (P<0.05). The difference of TCM syndrome efficacy between two groups of3visits viewpoints is statistically significant (P<0.05). From the efficiency, the treatment time is longer, the effective rate is higher. The results of generalized linear mixed model as following:central effect has statistically significance (P<0.05), center and treatment interaction effect has not statistical significance (P>0.05). The results show that the difference of the efficacy between the different treatment groups is statistically significant (P<0.05), and the ranked data efficacy of the experimental group is2.669times better than the control one, and the dichotomous data efficacy of the experimental group is2.114times better than the control one. The efficacy among the different centers is different, of which the difference in the3rd center, the4th center and the7th center is statistically significant (P<0.05), other centers have not significant difference (P>0.05). The results to analyze repeated measured data by generalized linear mixed model as following:the difference of efficacy between two groups is statistically significant (P<0.05), the experimental group is better than the control one. The efficacy difference among different visits viewpoints is statistically significant (P<0.05), and with the longer medication time, the efficacy is better. Different individuals have different influence to efficacy. Some individuals’efficacy differences at different time points are statistically significant (P<0.05).Conclusion:There are some difference between traditional statistical analysis methods and generalized linear mixed model in field of application and purpose. The results of the two methods are basically the same. When evaluatiing multi-center clinical trials of traditional Chinese medicine, if the data does not exist center effect and the correlation, it is relatively simple to analyze it by traditional statistical methods, if there are center effect and the correlation in data, generalized linear mixed models is a better choice.