Abstract 3291: Does Clopidogrel Provide Clinically Important Benefit in Atherothrombotic Cardiovascular Disease? A Bayesian Perspective
Background: The conventional P value and confidence interval (CI) provide information about statistical significance. However, estimation of “clinically important difference” (CID) remains a challenge.
Objective: To quantify probability of CID with respect to benefit (reduction in death, MI or stroke), harm (major bleeding) and net clinical effect (harm + benefit) associated with clopidogrel and aspirin vs. aspirin alone.
Methods: Posterior probabilities (Pr) that the relative risk difference is greater than the putative threshold value for clinical importance (δ>0%, >10%, >20%) are derived from a Bayesian analysis utilizing an uninformative prior (log odds mean = 0, standard deviation = 100) and the empirical evidence obtained from 5 large RCTs. All trials were powered to detect a CID of >15% risk reduction (RR) except for COMMIT (10%).
Results (Table⇓): Pr for any outcome fell as the threshold for CID increased. For any (>0%) benefit, Pr was 0.95 in all trials except CHARISMA; for >10% benefit, Pr was >0.90 in CURE & CREDO; for >20% benefit, Pr ranged from 0 – 0.74. The Pr was >0.95 for >0% harm in all 3 longer follow-up trials, and >0.90 for both >10% & 20% harm in CURE & CHARISMA. Pr of net clinical effect δ>0% was >0.95 in 2 trials (CURE & COMMIT); for δ>10%, it ranged from 0.03– 0.66; and for δ>20%, from 0 – 0.19.
Conclusions: Despite published conclusions to the contrary, probability of clinically important difference (typically 10–20% RR for vascular events) with clopidogrel is far less than that implied by conventional statistical significance. Thus, estimation of clinical importance using the Bayesian approach provides useful & complimentary information in weighing the evidence.