Ibuprofen May Not Increase Bleeding Risk in Plastic Surgery: A Systematic Review and Meta-Analysis
First, our group searched correlative articles about the risk of bleeding when taking nonsteroidal antiinflammatory drugs. The increased risk of bleeding and cardiovascular events was seen with all antithrombotic treatment regimens and occurred with selective cyclo-oxygenase2 nonsteroidal antiinflammatory drugs (i.e., rofecoxib and celecoxib) and nonselective agents (i.e., ibuprofen and diclofenac). Their results showed that the risk of bleeding in patients taking nonsteroidal antiinflammatory drugs was more than double that in patients not taking these drugs (hazard ratio, 2.02; 95 percent CI, 1.81 to 2.26).2 The current meta-analysis included limited studies with small sample size and concluded that ibuprofen was not associated with an increased risk of bleeding, which was usually accompanied by the possibility of being a false-positive result. Thus, the conclusion should be interpreted with caution. In addition, we were not sure whether the bleeding events were restricted in the surgical areas. Therefore, the definition of bleeding events should be clear.
Second, the meta-analysis eventually included only four studies, and the authors used the funnel plot to assess for publication bias, concluding that there is a lack of significant publication bias. However, this conclusion may be incorrect. Because of the poor statistical property influenced by “small-study effects” and the bad performance of the funnel plot in detecting publication bias,3 an expert panel composed of statisticians, epidemiologists, and methodologist recommended that publication bias contributed as only one possible explanation for the asymmetry of the funnel plot, and tests for funnel plot asymmetry should not have been used when there were fewer than 10 studies in the meta-analysis because test power was usually too low to be distinguished.4
Third, the investigation of heterogeneity is a crucial part of any meta-analysis. The power increases with the total information available rather than simply the number of studies, and it is substantially lowered if, as is quite common in practice, one study constitutes a large proportion of the total information.5 When meta-analyses included limited studies with small sample size, the pooled results were usually accompanied by possible random error and false-positivity, subsequently causing overestimation of intervention effect. Therefore, the authors should discuss heterogeneity.
Fourth, the Jadad scale is not recommended for bias assessment, and the tools of Cochrane are recommended. In addition, one minor mistake might have occurred because of carelessness in the review. The risk bias analysis in Figure 2 was performed by RevMan review software not the Grades of Recommendation Assessment, Development and Evaluation system. Finally, we totally agree with the authors that additional and more comprehensive meta-analyses should be conducted to achieve a more accurate and prudent conclusion.