ABSTRACT
In this chapter, two types of input data for the meta-analysis are considered. The first type consists of survival proportions at one or more times, for one or more groups, with their standard errors. The survival proportions are jointly analyzed, using a multivariate common-effect or random-effects meta-analysis model. A complexity in this type of analysis is that the within-study correlations between the survival proportions of the same group must be estimated. This is solved by fitting a linear (mixed) model iteratively. The second type of input data is based on the reconstruction of a life table for a chosen uniform partition of the time axis per group and study. To reconstruct the life tables, all relevant information given in a publication is used, including Kaplan–Meier graphs and information on patient accrual and drop-outs. Assuming constant hazards per interval, the life tables are analyzed in a two-stage or one-stage fashion. Two methods are presented for a comparative analysis of two groups. Under a proportional hazards assumption, a simple two-stage analysis is described. Avoiding a proportional hazards assumption, a Poisson correlated gamma frailty model is described.
