Figure 2From: Rosiglitazone: can meta-analysis accurately estimate excess cardiovascular risk given the available data? Re-analysis of randomized trials using various methodologic approachesEffect of Rosiglitazone on Myocardial Infarction and Cardiovascular Mortality. For each adverse clinical outcome, the pooled effect of rosiglitazone in different trial groupings is expressed as odds ratios (OR) and 95% confidence intervals (CI) using different statistical methods. For the Bayesian OR, CI denotes credible interval. Statistical methods with "Potential Negative Bias" may show reduced treatment effects. (1) This analysis includes only trials in which the sole difference between the intervention and control groups is rosiglitazone therapy. Compared to the original analysis, this excludes trials AVM100264, 49653/020, 49653/080, 49653/097, 49653/137, SB-712753/009, and ADOPT. It excludes one arm in each of the 3-armed trials 49653/093, 49653/079, and SB-712753/007. It also excludes RECORD. (2) Peto Odds Ratios as reported by Nissen and Wolski [1]. (3) Inverse Variance (IV) Odds Ratios using standard 0.5 continuity corrections in trials with no events in one group (Review Manager Software using a Random Effects Model with no heterogeneity [I^{2} = 0]). (4) Mantel-Haenszel (MH) Odds Ratios using standard 0.5 continuity corrections (CC) in trials with no events in one group (Review Manager Software using a Fixed Effects Model). (5) Mantel-Haenszel (MH) Odds Ratios using treatment arm continuity corrections (CC) in trials with no events in one group as proposed by Sweeting et al [11] to minimize bias. (6) Mantel-Haenszel (MH) Odds Ratios using empiric continuity corrections (CC) in trials with no events in one group as proposed by Sweeting et al [11] to minimize bias. (7) Odds Ratios calculated using exact statistical methods with no requirement for continuity corrections (StatXact 8 [Cytel Inc, Cambridge, MA, USA]). (8) Odds Ratios along with 95% credible intervals calculated using fixed effects Bayesian method and non-informative prior distributions with no requirement for continuity corrections (WinBUGS 1.4.1 [available at http://www.mrc-bsu.cam.ac.uk/bugs/]). Abbreviations: CC – continuity corrections; CI – Confidence or Credible Interval; I^{2} – I^{2} heterogeneity statistic; IV – inverse variance; MH – Mantel Haenszel; OR – Odds Ratio; RSG – rosiglitazone.Back to article page