Abstract 4944: Bayesian Analysis of TRITON-TIMI 38: Reassessing Prasugrel versus Clopidogrel in the Treatment of Patients With Acute Coronary Syndrome
Background: The application of Bayesian analytic techniques can provide clinically relevant information regarding efficacy of an intervention. Given the recent controversy surrounding prasgurel, as reported in the TRITON-TIMI 38 study, we sought to re-examine both the frequentist and Bayesian statistical methods which evaluated this study.
Methods: Magnitude of therapeutic benefit and a sensitivity analysis were performed, using posterior probabilities (Pr), in relation to probability of benefit versus harm, as well as probabilities of greater than 20% benefit (Pr>20%). The analysis was conducted with experiential probabilities obtained from the TRITON-TIMI 38 data, and a null prior probability with a log mean OR of 0, and a standard deviation of 2.
Results: Based on the study results, the Pr for the primary endpoint of death from cardiovascular causes, nonfatal MI, or nonfatal stroke is 99.9%. Calculated probability of at least a 20% benefit is 66.5%. This value is reasonably high enough to infer a clinically important benefit. All cause mortality was also found to have a clinically relevant benefit at Pr>20% of 99.1%. Analysis does indicate a clinically significant probability of harm pertaining to Non-CABG major and major plus minor bleeding, with a probability of at least 20% harm of 73.6 and 82.4% respectively.
Conclusions: On re-evaluation, the Bayesian analysis suggests the clinical relevance of prasugrel as a beneficial treatment in patients with ACS, which serves to support the implications of the conventional statistics reported in TRITON-TIMI 38. Of note, the clinically relevant benefit is juxtaposed against the harm related to bleeding complications.