The Study Of Missing Data Processing Method In Clinical Research Of Traditional Chinese Medicine

ObjectiveClinical research of Chinese medicine usually involves three links: subject, study factor, experimental effect, these links usually influence the quality of the research. In the research, as a result of a variety of reasons, we could find some missing values from the data of the research.these missing data may cause bias in the result of evaluation process, thus to reduce the accuracy of the research conclusion, and even may come to a wrong conclusion.This article will analyze the missing data in the process of clinical study, and summarize the influence of missing data in the analysis of clinical research, then compare the advantage of different processing methods with the missing data, and select the appropriate analysis method in the process of clinical study, in order to ensure the accuracy of the evaluation conclusion.MethodsThis article will discuss the related concepts and the types of missing data, analyze the influence of missing data in the clinical study. First, we simulate a group of complete data about clinical efficacy evaluation, and analyze this data. Then we make some different levels of incomplete data from this data, and analyze these incomplete data with listwise deletion and mean substitution(Mean) and last observation carried forward (LOCF) and expectation maximization (EM) and multiple imputation (MI), evaluate and revise these proper imputations of missing data. Finally, compare the difference of the data before and after the imputation and the stabilization of analytic result.ResultsIn the simulated data, we compare the advantages of the different methods to handle missing data. Results suggest that:If the missing rate was below10%, the outcomes of several methods were stable. When the missing rate was above30%, the outcomes of listwise deletion and Mean were not so well. The outcomes of LOCF was quite well in this study; but the outcome of EM was not so well; it is worth to be mentioned that the effect of the imputation with MI was better then other methods. Thus, this article will analyze one clinical research data, then analyze this data with different methods, and analyze the differences of the results.ConclusionIn the analysis of the simulated data, we could find that MI was much better than other methods in different levels of missing rate, it seems closer to complete data on the results of statistical analysis, that reflect irreplaceable superiority in the high missing rate of data after MI. At the same time, it could not be denied that the stability of the statistical results is fairly well after listwise deletion and Mean. Moreover, in clinical study, these two methods were much more simple and practical. Generally speaking, when the missing rate is high, the stability of the results was worse no matter what methods were adopted. Sometimes, its result deviated from the complete data.Therefore, in the analysis of clinical study, we should consider MI for the analysis of the missing data, in order to reduce the probability of the bias. In clinical study, as a precaution, we should use two or more methods to compare, in order to prevent making some biased conclusions.