Abstract:
In this essay we will consider the meta-analysis of summary statistics versus individual participant-level data of trials with binary outcomes. Numerous techniques will be considered to essentially determine, both theoretically as well as numerically, by the use of real data analysis, whether or not, in certain instances, analyzing individual participant data (IPD) from all investigations undoubtedly gains efficiency over combining summary statistics from each study in a binary setting. We will be considering both the one-step as well as two-step meta-analysis, with the focus being mainly on the standard fixed-effects model and partially on the random-effects model. An application of two different clinical studies will be considered. The first study is called the Cannon (Cannon et al., (2006)[13]) Cardiovascular Clinical Trials which consists of four clinical studies. The second study is called the Bacillus Calmette-Guerin Vaccine Study which consists of a total of 13 clinical studies. We observed that a meta-analysis of summary data is more similar to a meta-analysis of IPD data for the first study, but different for the second study. This is due to the fact that the number of studies
are small in both studies. Therefore the asymptotic equivalence should be cautious when the sample size is small and the number of studies is small. In this case, the estimation of the between study variance will be very unstable which will lead to a different conclusion between a meta-analysis of summary statistics (SS) and a meta-analysis of individual participant-level (IPD) data.
