Background: Results from apparently conclusive meta-analyses may be false. A limited number of events from a few small trials and the associated random error may be under-recognized sources of spurious findings. The information size (IS, i.e. number of participants) required for a reliable and conclusive meta-analysis should be no less rigorous than the sample size of a single, optimally powered randomized clinical trial. If a meta-analysis is conducted before a sufficient IS is reached, it should be evaluated in a manner that accounts for the increased risk that the result might represent a chance finding (i.e. applying trial sequential monitoring boundaries).
Methods: We analysed 33 meta-analyses with a sufficient IS to detect a treatment effect of 15% relative risk reduction (RRR). We successively monitored the results of the meta-analyses by generating interim cumulative meta-analyses after each included trial and evaluated their results using a conventional statistical criterion (alpha = 0.05) and two-sided Lan-DeMets monitoring boundaries. We examined the proportion of false positive results and important inaccuracies in estimates of treatment effects that resulted from the two approaches.
Results: Using the random-effects model and final data, 12 of the meta-analyses yielded P > alpha = 0.05, and 21 yielded P </= alpha = 0.05. False positive interim results were observed in 3 out of 12 meta-analyses with P > alpha = 0.05. The monitoring boundaries eliminated all false positives. Important inaccuracies in estimates were observed in 6 out of 21 meta-analyses using the conventional P </= alpha = 0.05 and 0 out of 21 using the monitoring boundaries.
Conclusions: Evaluating statistical inference with trial sequential monitoring boundaries when meta-analyses fall short of a required IS may reduce the risk of false positive results and important inaccurate effect estimates.