The Design And BAYES Analysis Of Repeated Measurement From Multi-centre Clinical Trials

In multi-center clinical trials, the same patient is observed and examined atdifferent times, so a subject can obtains repeated measurements data whenmulti-center clinical trials is finished, namely repeated measurement data fromMulti-center clinical trials. In multi-center clinical trials, the subjects in the samecenters have some common characters, there is some dependency betweenindividuals, but in different centers it is not the case. So analysis of multi-centertrials must consider the effects of the individual centers and possibility ofno-constancy of treatment effect differences among centers. if there isheterogeneity, then treatment-by-center interaction maybe exist. A characteristic of repeated measurement data is that there is intra-subjectcorrelation,and the closer the time point, the higher correlation. Now, under thecondition of homogeneity assumption, analysis of these data of any one timefrom a center usually is implemented by one-way anova, the compare betweenany two times is completed by t-test or one-way anova, then combining theseoutputs of different centers. But intra-subject correlation and treatment-by-center interaction are not be take into account, so the power of them is poorand easy to cause typeâ… error .Sometimes although considering the intra-subjectcorrelation and treatment-by-center interaction ,but ignoring the heterogeneity.If heterogeneity exists among centers, how to ascertain whether such interactionexists, if there is treat-by-center interaction, how to carry out the analysis hasbeen a focus of attention of many statisticians. In this paper, I advance several homogeneity test metheods implemented byBayes approach, then explore several method to identify and confirm interaction.Bayes mixed effects model is the main issue I discuss. Comparing the results ofBayes analyses with the result of fixed and mixed model effects applied to IIIrepeated measurement data from a multi-centre trial. With a view to exploit anew idea , to seek a analytic method easy to explain and applicable in morecases ,and to generalize the application of Bayes method in medical study. By reviewing the study of repeated measurement data from multi-centreclinical trials for more ten years , combining with progress in study of analyticmethod, I expound the related content of multi-centre clinical trials ,Bayesianbasic theory, Morkov Chain Monte Carlo and Gibbs sampling. Before analysisof models ,using the example of phase â…¡multi-centre clinical trials to illustratethe conditions of various statistical models, and compare the result of Bayesanalysis with the result of fixed and mixed model applied to repeatedmeasurement data from a multi-center trials ,the conclusion is as follow: General linear model is easy to understand, but it regards centre effect andtreatment-by-center interaction as fixed effect instead of random, theinformation is limited relately, the conclusion can't be generalized abroad.Mixedeffect model overcomes glm's shortcomings, and considers interaction andviews center effect and interaction as random, but it has some limit. Bayes methods provide attractive alternatives to conventional ANOVAs foranalyzing and reporting the findings of the repeated measurement data frommulti-center clinical trials and do not require more restrictive assumptions thenANOVAs approach,they can be applied to non-normolly distributed data suchas count data or binary data.Bayes approaches require regarding the centerseffects and treatment-by-cente as random instead of fixed ,a view which oftenwill reasonably describe outcomes of clinical trial in spite of the fact that theindividual centers certainly do not comprise a random sample of all possiblecenters. The Bayes approach makes the best of three sorts of information,adoptingWinbugs soft to model, implemented with Gibbs sampling,and provides aconvenient way to construct posterior and predictive di...