Multiple Hypothesis Testing And Its Applications Based On FDR Control

With the continuous advancement and development of science and technology,high-dimensional data frequently comes to us together with the steps of modern medicine,for which many classical statistical methods are not applicable any more.At the moment,the multiple hypothesis testing gets extensive attention since it provides an important theoretical basis for the analysis of high-dimensional data.Based on the practical problems in the detection of drug efficacy,this paper analyzes the influencing factors of the therapeutic effect of Xuebijing drug on patients with severe sepsis,by applying multiple hypothesis testing to the data analysis model.One of the key tasks of multiple hypothesis testing is to develop a reasonable and effective error measure to improve the effectiveness of the overall test under the premise of controlling the probability of category I errors.First of all,we reviewed the traditional method(family-wise error rate,also known as FWER for short)and how it works.What’s more,we discussed the false discovery rate(known as FDR for short)of multiple hypothesis testing.For FDR content,our study mainly focused on improving the testing efficiency.For the analysis of efficacy of Xuebijing drugs to those suffering serious sepsis,we used the multiple hypothesis testing after checking p-values of regression coefficients derived from either principal component regression or classical linear regression.With the help of involved methods in this article to control FWER or FDR,we tried to control the p-values and give the significant impact indicators in the controlling process.After comparing the testing results of classical statistical methods and multiple hypothesis testing,factors discovered by multiple hypothesis testing were more reliable since it had adjusted p-values in the processes.This indicates that the multiple hypothesis testing works better than the traditional statistical methods in the examination of drug efficacy with high-dimensional data involved.