Researches And Applications Of The Methods For Multivariate Associated Meta-analysis

Meta-analysis refers to the statistical analysis method which collect a series of related studies and get a pooled statistic as the result. In traditional meta-analysis, it is common that only one or a few of the exposures of interest are focused, ignoring the association among the exposures. Meta-analysis combines the different effect values of an exposure in different studies to get a pooled-statistic from the narrow sense. This paper aims to research the methods for multivariate associated meta-analysis by using the idea of weight and big data.The main contents of this paper is to research three methods for multivariate associated meta-analysis and apply them to a group of esophageal cancer data:(1) Pooled statistic is essentially a weighted average of the corresponding effect values in studies. The variance is calculated according to the weights, which can be used to calculate the confidence interval. Because the weights are modified, the results of pooled statistics of exposures are correlated.(2) The multivariate associated meta-analysis oriented data can be viewed as an incomplete set of observations. In statistics, the data set with incomplete observations can be analyzed after imputation. The data is not monotone missing and hardly exists a complete variate. In that case, the Markov chain Monte Carlo method is the best choice.(3) The procedure of MCMC method is iteration to converge the estimates of the important parameters of the distribution under hypothesis. Meta-analysis can be viewed as a weighted average of observations in a narrow sense to estimate the parameters. Based on the characteristics of both, this paper comes up with a new method for multivariate associated meta-analysis.The main results of this paper are presented as follows:(1) The results of applications.From a professional point of view, E/S-C algorithm is the best, weight modification method is better than multiple imputation method. The confidence interval of pooled statistic calculated by multiple imputation method has the smallest scale, but it has no professional explanation and the variance is abnormal.(2) The result of simulations.In E/S-C algorithm, two simulations are designed to select the initial value of the variance-covariance matrix. In a non-informative prior, the correlation matrix with equal correlation coefficients is steadier. In an informative prior, the correlation matrix with equal correlation coefficients and priori matrix are almost the same.