Abstract 21109: Bayesian Approach to Moderate Implausible Treatment Effects Observed in Trials Stopped Early: Lessons Learnt from JUPITER Trial.
Background: The JUPITER trial was halted early after interim analysis showed a 44% risk reduction in the primary endpoint in primary-prevention patients randomly allocated to treatment with rosuvastatin 20 mg compared to placebo. It has been argued that premature stopping may have contributed to the unexpectedly large (implausible) and rapid treatment benefit in JUPITER.
Objectives: To moderate the implausibly large treatment benefit in JUPITER by using the Bayesian approach.
Methods: The prior information from CORONA, a placebo-controlled rosuvastatin trial in ischemic congestive heart failure (published results available when JUPITER was stopped early) was integrated with the empirical evidence based on JUPITER to derive posterior estimates via the Bayes’ theorem.
Results: LDL (45% vs. 47% in JUPITER) and hsCRP (32% vs. 38% in JUPITER) lowering with rosuvastatin was similar in the 2 trials. The results shown in Table demonstrate that treatment benefit is moderated down from 20% to 8% with respect to mortality, from 47% to 13% with respect to the combined endpoint of CV death, MI or stroke, and from 54% to 27% with respect to MI, estimates that even though different from zero, are substantially less than and almost excluding the observed effects. The moderated estimates are in alignment with the results of subsequent rosuvastatin trials (GISSI-HF and AURORA). The corresponding absolute risk difference in CV death, MI or stroke endpoint is reduced from a 0.83% to 0.23% and the NNT increased from 119 to 434.
Conclusions: A Bayesian integration of data from the CORONA and JUPITER trials provides a more tempered estimate of the true difference and formalizes the notion that surprising effects are probably overestimations. Thus, the treatment benefit in JUPITER, although real, is likely to be overestimated and the risk is likely to be underestimated. This has important implications for cost effectiveness assessment, guideline and policy development, and clinical practice.
- © 2010 by American Heart Association, Inc.