Trial sequence meta-analysis can reject false-positive result calculated from conventional meta-analysis


  • Potential conflict of interest: Nothing to report.

To the Editor:

We read the article by Mauro Bernardi et al. with great interest.[1] The primary endpoints of the meta-analysis were postparacentesis circulatory dysfunction (PCD), hyponatremia, and mortality. This meta-analysis provides evidence that albumin reduces morbidity and mortality among patients with tense ascites undergoing large-volume paracentesis, as compared with alternative treatments investigated thus far. However, we would like to comment on the potential risk for false-positive results (i.e., report a treatment effect when in reality there is no effect) generated by conventional meta-analysis (relying on the conventional statistical criterion (two-sided α = 0.05) for deciding whether a treatment effect exists).

As we know, in a single trial, interim analyses increase the risk of type I error.[2, 3] Cumulative meta-analysis (i.e., also referred to as conventional meta-analysis) is one statistical method to combine available studies by time series, which are at risk of producing random errors because of repetitive testing on accumulating data.[4] Trial sequential analysis (i.e., a method adapted from interim monitoring boundaries in single randomized controlled trials applied to cumulative meta-analysis) was introduced to minimize random errors by calculating the required information size (i.e., the number of participants needed in a meta-analysis to detect or reject a certain intervention effect, analogous to the required sample size in a single randomized clinical trial). Usually, information size was based on the assumption of a plausible relative risk reduction (RRR) of 10%-30% or on the RRR observed in trials with low bias risk. On the basis of the required information size and risk for type I and type II error trial, sequential monitoring boundaries (i.e., a threshold for a statistically significant treatment effect) were constructed. In short, trial sequential analysis provides the necessary sample size for our meta-analysis and boundaries that determine whether the evidence in our meta-analysis is reliable and conclusive.[5] Based on this theory, we recalculated the data on the effect of albumin in PCD, hyponatremia, and mortality (Fig. 1). The results in PCD were in accordance with the claim of the traditional method, where albumin has an efficacy in ameliorating PCD, while the evidence that albumin could reduce the mortality rate and hyponatremia incidence rate was insufficient, as both the individual Z-curve failed cross-sequential monitoring boundaries.

Figure 1.

PCD, mortality, and hyponatremia in trials comparing albumin with alternative treatments. (A) PCD. (B) Mortality. (C) Hyponatremia. The horizontal line is set at ∼1.8, representing the traditional Z-test. In both mortality and hyponatremia, the cumulative scores are above the line, meaning that in traditional meta-analysis both could favor albumin's effect. But the real boundaries are higher (the black broken line above the traditional horizontal line), meaning in fact that the trials available are insufficient to draw the conclusion that favors albumin's effect.

In summary, meta-analytical inferences should not be based solely on the conventional statistical criterion of two-sided α = 0.05.[6] Trial sequential meta-analysis needs to be explored as a necessary supplement for the traditional meta-analysis.

  • Ren-Pin Chen, M.D.1 Chao Chen, M.D.1 Jie-Yu Yu, M.D.1 Xie-Lin Huang, J.R.2 Meng-tao Zhou, Ph.D.3* Zhi-Ming Huang, J.R.1*

  • 1Department of Gastroenterology First Affiliated Hospital Wenzhou Medical University, Wenzhou, China

  • 2RenJi Study, Wenzhou Medical University Wenzhou, China

  • 3Department of Hepatobiliary Surgery First Affiliated Hospital Wenzhou Medical University, Wenzhou, China