We read with interest the report by Spector et al titled “Meta-Analysis of Ablation of Atrial Flutter and Supraventricular Tachycardia.” Although the report provides extensive information about the topic, the study was an observational meta-analysis, which presents some methodologic drawbacks potentially flawing the results presented in the report.

The first issue of concern is “publication bias,” which was not investigated in Spector et al’s study. Studies with negative results can take longer to be published, and results not conforming to the desired outcome may not even be reported. An investigation of this bias can be performed using a graphical test, such as a funnel plot, which is created by plotting the estimated treatment effect against the study size. An indication of publication bias would be an absence of small studies with small effects in the funnel plot’s lower left-hand corner. Several methods have also been suggested that translate the graphical approach of the funnel plot into a statistical model. Egger et al introduced a linear regression approach in which the standardized effect size is regressed into a measure of precision. The greater the value of the regression coefficient, the greater the evidence for small study effects.

The second issue is that a graphical illustration of the results of the meta-analysis was not provided. A useful graphical tool to summarize the results of such quantitative analyses is the Forrest plot. Unfortunately, SAS, the statistical package used in this work, cannot perform this function. We suggest taking full advantage of Review Manager version 5 (Cochrane Collaboration, Copenhagen, Denmark), freeware that is able to display such a graphical approach.

Finally, Spector et al extracted data from single studies without taking account of the simple unadjusted estimates. The Meta-Analysis of Observational Studies in Epidemiology (MOOSE) group recommended the formal assessment and reporting of confounders in reviews of nonrandomized studies. Reporting “crude” estimates without considering potential confounders can lead to biased results. Thus, reporting adjusted estimates in such a analysis is a “key point,” and if they are not available, investigators should clearly declare so.