Figure 4
Effect of rosiglitazone on myocardial infarction (MI) in randomized controlled trials with at least 12 months of follow-up. Singh et al [9] included only "serious events" rather than "total events" for ADOPT [17], only events leading to hospitalization rather than all investigator-reported MIs for Dargie et al [19], and for the DREAM study [18] included an additional rosiglitazone patient reported to have an MI that was included in a recent presentation by the manufacturer to the Food and Drug Administration [7] but not included in the original publication. In contrast, Nissen and Wolski [1] included all MIs for ADOPT [17] and Dargie et al [19] and used the published values for DREAM [18]. While the decisions made by Singh et al [9] were reasonable, each increased the risk of MI for rosiglitazone relative to control and thus contributed to a statistically significant result (Panel A). Alternatively, using the same event inclusion decisions as Nissen and Wolski [1] and including only adjudicated events from the subsequently published RECORD trial [5], the increase in MI no longer retains significance at the p = 0.05 level (Panel B). If one limits the analysis to the two long-term trials in which the intervention and control groups differ only by rosiglitazone therapy [18, 19] the risk increases; however, again it loses its statistical significance when the data used by Singh et al (RR 2.27 [95%CI 1.06 to 4.87], p = 0.03) are replaced by those used by Nissen and Wolski (RR 1.83 [95%CI 0.88 to 3.81], p = 0.11). For this figure we have used the same effect measure as Singh et al [9] (Risk Ratio [RR]); the p-values are identical if Peto OR is used as the effect measure. Weight refers to the contribution of each study's RR to the overall pooled RR. The center of the diamond indicates the pooled RR, and the width of the diamond reflects the size of the 95% CI. Abbreviations: CI – confidence interval; I2 – I2 heterogeneity statistic; MI – myocardial infarction; n – number of patients with event in the intervention or control group; N – total number of patients in the intervention or control group; OR – odds ratio; RR – risk ratio; Z – Z test statistic.