Abstract 4016: Does Prasugrel Provide a Clinically Important Treatment Benefit Compared with Clopidogrel? A Bayesian Analysis of TRITON-TIMI 38
Background: In contrast to the conventional frequentist analysis, the Bayesian method is ideally suited to provide a quantitative estimate of a given magnitude of treatment benefit or harm.
Objective: A reappraisal, with emphasis on both statistical significance and clinical importance, of the recently reported TRITON-TIMI 38 trial that compared prasugrel with clopidogrel in ACS patients scheduled for PCI.
Methods: Posterior probabilities (Pr) that the relative difference (d) in benefit and risk is greater than the putative threshold value for clinical importance (d>0% to 20%) were derived from a Bayesian analysis using an uninformative or “agnostic” prior (log mean RR, = 0, SD = 2) and empirical evidence from TRITON-TIMI 38.
Results (Table⇓): Although the Pr(d>0%) and Pr(d>10%) difference in the primary endpoint are 100% and 97.5%, respectively, the Pr(d>20%), the “minimum clinically important difference” that the study was powered to detect, is only 40%. Similar patterns are observed with all-cause death/MI/CVA. The Pr(d>10%) harm (for either TIMI major or TIMI major plus minor bleeding) exceeds 90%, and Pr (d>20%) harm exceeds 65%. The Pr(d>20%) harm exceeds the Pr(d>20%) benefit. Combining ischemic endpoints and TIMI major bleeding into net clinical benefit, the Pr(d>10%) benefit is <80% and Pr(d>20%) benefit is <5%. If TIMI major plus minor bleeding is included, the probability of net clinical benefit with prasugrel therapy is reduced even further (<5% for d>10% and <1% for d>20%).
Conclusions: The probability of clinically important benefit associated with prasugrel treatment in TRITON-TIMI 38 is far less than that implied by conventional statistical significance. Bayesian analysis provides a useful tool in quantifying clinically meaningful difference. Ultimately, the clarity of the balance between “clinically important”, not just “statistically significant”, benefit or harm should influence guideline recommendations and treatment decisions.