Cumulative meta -analysis of clinical trials: From a retrospective exercise to a dynamic procedure

Meta-analysis is an increasingly important method in clinical research. In cumulative meta-analysis (CMA) an updated meta-analysis is performed every time the results of a new trial appear. The goal of this more fluid approach is to decide when the evidence provided by the accumulated data about a new treatment is statistically significant and clinically convincing enough to adopt or reject it.;The current practice of CMA gives rise to the multiple looks problem. The adaptation of the spending function method (Lan and DeMets 1983) that was proposed by Pogue and Yusuf (1997) as a solution is further investigated. A theoretical justification to such an adaption is given, as well as sensitivity analysis to the choice of spending function and model (fixed or random effects). Simulation studies show that in switching to the proposed adaptation there is a relatively small loss in power, and gain in coverage.;Another feature of the current practice of CMA of clinical trials is that it is used retrospectively, making any conclusions about the effectiveness of the treatment have little timely practical value. A new methodology for conducting CMA prospectively in introduced, utilizing a family of delayed alpha-spending functions. Practical considerations regarding the choice of key parameters are discussed, as well as guidelines for the researcher on how to proceed once these choices are made. A complete and detailed example accompanies the new methodology.;In the final chapter properties of spending functions are further investigated. In particular, the effect of the frequency and configuration of monitoring on the magnitude of the critical values is discussed, as well as the "deceptive intuition" of the spending function method.