I was surprised by the recent meta-analysis by Johnson and colleagues [1] on nonsteroidal anti-inflammatory drugs (NSAIDs) and blood pressure. Several important aspects of the validity and generalizability of the study need serious attention before readers can apply the information.

First, the "methodologic rigor" for individual studies was not very rigorous. For example, the study by Radack and colleagues [2] included 41 patients, not 30. The patients were predominantly black women (21 of 41 patients), not white (their Appendix Table 1), and the male-to-female ratio was 15:26, not 7:19 (their Appendix Table 1). We highlighted these differences by discussing the potential effect on both the pharmacoepidemiology and generalizability of our findings. Second, in the first published methodologic review of NSAIDs and blood pressure control [3], my colleagues and I reached different conclusions on the quality of the clinical trials (those done before 1987). In fact, the methodologic heterogeneity was so great that we omitted a pooled summary estimate to avoid a meaningless conclusion. Third, Johnson and colleagues [1] reported their results as mean arterial pressure, a relatively useless outcome variable for the practicing clinician. Although all articles reviewed up to 1986 (and after) have reported both systolic and diastolic blood pressure values, the latter was omitted from Johnson and colleagues' meta-analysis. Given the well-established lack of consistent and quantitatively important changes in many of the laboratory measures reported in their meta-analysis, information on systolic and diastolic blood pressure would have been far more useful and clinically sensible to practitioners than some of the data provided in their Table 1.

Finally, as an advocate and producer of meta-analyses, I strongly urge more careful and thoughtful assessment by reviewers and editors if the credibility of meta-analyses is to grow [4].