When we shouldn't borrow information from an existing network of trials for planning a new trial.

ABSTRACT

Introduction: To achieve higher power or increased precision for a new trial, methods based on updating network meta-analysis (NMA) have been proposed by researchers. However, this approach could potentially lead to misinterpreted results and misstated conclusions. This work aims to investigate the potential inflation of type I error risk when a new trial is conducted only when, based on a p-value of the comparison in the existing network, a "promising" difference between two treatments is noticed. Methods: We use simulations to evaluate the scenarios of interest. In particular, a new trial is to be conducted independently or depending on the results from previous NMA in various scenarios. Three analysis methods are applied to each simulation scenario with the existing network, sequential analysis and without the existing network. Results: For the scenario that the new trial will be conducted only when a promising finding (p-value <5%) is indicated by the existing network, the type I error risk increased dramatically (38.5% in our example data) when analyzed with the existing network and sequential analysis. The type I error is controlled at 5% when analyzing the new trial without the existing network. Conclusion: If the intention is to combine a trial result with an existing network of evidence, or if it is expected that the trial will eventually be included in a network meta-analysis, then the decision that a new trial is performed should not depend on a statistically "promising" finding indicated by the existing network.