An Overview For Meta-analysis: From Statistical Analysis Of Original Research To Proposal For Standard Reporting Of Meta-analysis

[Objective] In the background of process of conducting of Meta-analysis and current recognized international reporting standards,to explore the basic principles of statistics,operation norms,quality control strategies and standard procedure in the research process of Meta-analysis in both China and abroad;to represent the development profile,status quo of quality,and to recommend further specifications for reporting the results of a Meta-analysis;eventually,by these measures,to improve the quality of evidences,reach a high standard in scientific reporting so as to contribute evidences which can stand the test of time and practice.[Methods] A total of 814 Meta-analysis literatures and 693 primary studies,both published in Chinese and English,were included as research object in this thesis.I review and evaluate these literatures,from conducting a meta-analysis to standard reporting,from the following aspects,1)From the view of study design and conducting of a trial,this thesis focuses on the following key steps,such as ethical issues,experimental design,the “Three-Essential-Factors” and “Four-Principles” underlying a study,data management and statistical design,3 special comparative type in clinical trial of new drug,quality control.like the experimental study type,statistical design type,comparison of clinical trials of new drugs,etc.;2)Focusing on source studies included in a Meta-analysis,this thesis exploring the features of databases,language limits and search strategies;3)This thesis compare the difference between the meta-analysis studies and their primary studies analyzed by using 2 schools of statistical inference,the Bayesian school and the frequency school;4)The analysis of difference between net-work meta-analysis and traditional meta-analysis,and between meta-regression and traditional meta-analysis;5)The analysis of how to evaluate the Meta evidence quality by using the Grading of Recommendations Assessment,Development and Evaluation(GRADE)system.The existing problems were summarized,excellent Meta-analysis reports were interpreted and recommended and the standardized report templates were recommended according to the existing problems.[Results] 1)The quality assesssment criteria and standards of reporting of different research types are different.Therefore it’s not recommended to merge the results from different research and conduct Meta-analysis directly.The GRADE quality rating is heavily affected by the research type.The level of abidance of the“Three-Essential-Factors” and “Four-Principles”also affects the correctness and quality of the conclusion.The choices of the research design type and comparison type of the original research are closely related to the choice of statistic methods.These considerations need to be taken when assessing the quality of the original studies.If the meta-analysis is on clinical trial of a new medicine(often needs efficacy and safety evaluation),one must distinguish the statistical assumptions of non-inferiority study,superior study and equivalence study,and the assumptions of difference test,and in general should not simply merge the results of original studies with different comparison type,as such merge may result in heterogeneity.2)In the search process of original publications for meta-analysis,one should consider the possible bias of inclusion due to the rules of inclusion of the database of publications.In the inclusion process,the languages the author(s)can use may limit the completeness of the studies.The pre-condition for a complete publication set is the inclusion of the publication from advantageous countries and institutes;one could extend on such list.3)Bayesian method is a statistical method which estimates unknown parameters based on posterior distribution of unknown parameters obtained on the basis of Bias’ s theorem by integrating the general information,prior information and sample information.The advantage of the method is that the posterior probability is used to sort the intervention measures,which is fully utilized in network Meta-analysis.4)In comparison with traditional Meta-analysis,the conducting of a network meta-analysis should focus on the following points,a)Evaluation of transitivity and consistency between direct and indirect comparison studies,which,in a large extent,determines the decision of how to combine the effect size estimates,and how to conduct a sub-group analysis and sensitivity analysis;b)Carefully manage and interpret the single standing nodes on the evidence network.c)Regarding the reporting of a network meta-analysis,a network figure showing all the eligible comparisons should be given with explanations to each nodes.d)The surface under the cumulative ranking curve(SUCRAs)is clear and particularly useful for clinical work because it recommendation sequence for multiple interventions compared within this present network,which is apparently the advantage of the network meta-analysis.As thus,for conducting and reporting a network Meta-analysis,it is necessary to individualize the aforesaid sections on the basis of PRISMA 2009 checklist.5)GRADE evidence quality evaluation system is to rate and recommend the evidence obtained through main and secondary outcome indicators.Quality evaluation of original literature and GRADE evidence quality evaluation complements each other,and can contribute to the production of high quality Meta-analysis results.However,the application of GRADE system in the new Meta-analysis research is not mature and needs more practice to improve.[Conclusions] In this study,meta-analysis papers(n=814)in both Chinese and English and some of t he included papers of original studies(n=693)are randomly sampled,scrutinized,and analyzed.The standards for the structure,procedure,and reporting style have formed,but improvements are still in demand in the evidence reliance and reporting standards.In the original papers in the meta-analysis,significant differences exist between the r esearch design,procedure,and conclusion aspects,as well as the “scientific,complete,systematic,comprehensive” and “high quality and standardized” aspects.This study is based on the thorough understanding of the meta-analysis and generalized overvie w evaluation method(evaluation of the research design,implementation,analysis,rep orting,and conclusion).It gives scientifically rigorous and hands-on references and te mplates for meta-analysis research and provides a critical evaluation of the “input evi dence” and the “output evidence”.