Results of meta-analysis should be treated critically

Rong-Qiang Liu, Yi Shao

Abstract Proton pump inhibitors use increases hepatic encephalopathy risk in patients with liver disease.

Key words: Proton pump inhibitor; Hepatic encephalopathy; Liver disease

TO THE EDITOR

We read the recently published meta-analysis on the association between proton pump inhibitor (PPI) and hepatic encephalopathy (HE)[1]. The authors found that there was a significant association between PPI use and HE risk with low heterogeneity (I2= 14.2%). In fact, a correct analysis showed an even stronger association. First, the author made an error in extracting the data. The study by Tsai et al[2]provided three dose-related odds ratios (ORs) of 1.41 (1.09-1.84), 1.51 (1.11-2.06)and 3.01 (1.78-5.10), and the meta-analysis by Ma et al[1]incorrectly used only the first OR. According to the study of Hamling et al[3], we recalculated that the correct OR was 1.59 (1.30-1.95). The results of the meta-analysis revealed that PPI use was significantly associated with the risk of HE with a pooled OR of 1.97 (95% confidence interval [CI]: 1.51-2.58) by using a random effects model (I2= 57.1%). The forest plot is shown in Figure 1. In addition, we used the data which were given by Ma et al[1]to perform comprehensive analysis. The pooled OR was 1.97 (1.47, 2.65) instead of 1.50(1.25, 1.75) and the I2was 61.2% instead of 14.2%. The results are shown in Figure 2.The incorrect operations of Ma et al[1]from using software may be the cause of inconsistencies. Due to the obvious heterogeneity, subgroup analysis and metaregression should be performed to explore the sources of heterogeneity.

The fixed effects model assumes that all the included studies have the same true effect size, while the true effect size in the random effects model varies with different studies. When the observed effect size of each study approaches or equals its true effect size, the heterogeneity is not obvious and the fixed effect model should be used.Otherwise, the random effects model is used. The Cochrane Handbook has stated that when I2is less than 40%, a fixed effects model should be used for meta-analysis.Otherwise, a random effect model should be used. In general, the conclusions obtained by random effects models tend to be conservative, and thus can be used in any case. However, when the heterogeneity is significant, only the random effect model can be used, and further analysis is needed to find the sources of heterogeneity.

In conclusion, our results demonstrated that PPI use increased HE risk, consistent with the study of Ma et al[1]. However, the meta-analysis by Ma et al[1]is flawed. More accurate statistical analysis is necessary to improve the quality of this meta-analysis.

Figure 1 Forest plot of the association between proton pump inhibitor use and hepatic encephalopathy.

Figure 2 Forest plot of proton pump inhibitor use and risk of hepatic encephalopathy.