Response to Letter Regarding Article, “Effect of Celecoxib on Cardiovascular Events and Blood Pressure in Two Trials for the Prevention of Colorectal Adenomas”
We thank Dr Cen for his thoughtful comments about our work.1,2 He suggests combining covariates from the Adenoma Prevention With Celecoxib (APC) and Prevention of Colorectal Sporadic Adenomatous Polyps (PreSAP) trials into a single Cox model and using the resulting hazard ratio to estimate the overall effect of celecoxib on cardiovascular end points. This approach treats the variable trial (APC or PreSAP) as 1 ingredient in a stew. We believe in maintaining the integrity of the ingredients as long as possible to allow the flavor of each trial to retain its unique contribution. Using Cox models, we first tested whether the risk factors would mediate the effect of celecoxib on cardiovascular events. We reported several such analyses in this study1 (Table 4 of our original article) and in a preliminary analysis.2 Having shown that cardiovascular risk factors did not importantly mediate the effect of celecoxib on cardiovascular outcomes, we included only the stratifying variable, low-dose aspirin use, in our models. The lack of interaction between risk factors and the effects of celecoxib would lead one to predict that including the risk factors would not importantly influence the combined hazard ratio. In fact, for the end point of cardiovascular death, myocardial infarction, stroke, or congestive heart failure, tossing all the covariates into a single Cox-model pot yields a hazard ratio of 1.95, which becomes 1.94 if we limit the covariates to study and aspirin.
The more fundamental question is how best to combine the trials. As Cen rightly points out, Table 1 shows large differences in risk factors between the 2 trials. These observed differences are unlikely to reflect the only distinctions between the 2 populations. Hence, our analytic approach relies strongly on randomization, which controls for both measured and unmeasured variables. To maintain balance in unmeasured variables in each trial, we compare participants assigned to placebo treatment with those assigned to celecoxib. Cen’s proposed approach, which fails to preserve the studies’ structure, implicitly assumes that the only differences between the studies are those seen in the measured variables. Our method—estimating the effect within each trial and then combining those estimates by the inverses of their variances—yields a hazard ratio of 1.85, reported in Table 3 as 1.9. The degree of uncertainty inherent in the small number of end points and the different dose regimens leaves open the question of whether dose regimen influences the effect of celecoxib on cardiovascular outcomes, a question that further analyses of the data from these 2 trials cannot answer.
Drs Bertagnolli, Zauber, Levin, and Arber have received research grant support from Pfizer Inc, and Drs Levin and Arber have consulted for Pfizer. Drs Eagle and Lechuga are employees of Pfizer Inc. Drs Solomon, Finn, Pfeffer, McMurray, Wittes, and Hawk and R. Fowler report no conflicts relative to this article.
Solomon SD, Pfeffer MA, McMurray JJV, Fowler R, Finn P, Levin B, Eagle C, Hawk E, Lechuga M, Zauber AG, Bertagnolli MM, Arber N, Wittes J; APC and PreSAP Trial Investigators. Effect of celecoxib on cardiovascular events and blood pressure in two trials for the prevention of colorectal adenomas. Circulation. 2006; 114: 1028–1035.
Solomon SD, McMurray JJ, Pfeffer MA, Wittes J, Fowler R, Finn P, Anderson WF, Zauber A, Hawk E, Bertagnolli MM; Adenoma Prevention with Celecoxib (APC) Study Investigators. Cardiovascular risk associated with celecoxib in a clinical trial for colorectal adenoma prevention. N Engl J Med. 2005; 352: 1071–1080.