Author + information
- Ruchira Glaser, MD, FACC⁎ (, )
- Henry A. Glick, PhD,
- Howard C. Herrmann, MD, FACC and
- Stephen E. Kimmel, MD, MSCE, FACC
- ↵⁎Hospital of the University of Pennsylvania, Division of Cardiovascular Medicine, 9 Gates Pavilion, Philadelphia, Pennsylvania 19104
We thank Dr. Boersma for his comments on our work (1) and we appreciate the opportunity to respond. We agree that the many therapeutic options now available for acute coronary syndrome (ACS) patients create a need to implement these options rationally. It was because of this need that we used data from randomized clinical trials in a formal decision analysis to provide estimates of the probability of outcomes with each of the therapeutic strategies modeled.
Concern is raised about three specific estimates used and their effect on the results of the principal analysis. First, Dr. Boersma comments on the value of zero for the lower limit of the confidence interval (CI) for death in TARGET (Do Tirofiban and ReoPro Give Similar Efficacy Outcomes?) trial ACS (2). This was a typographical error; as indicated in Table 3, the range that was used was 1.0 to 2.5. Thus, our sensitivity analysis ranges from being neutral to detrimental for the upstream strategy. Second, we did not use the differing rates of major bleed reported in the TARGET trial for our baseline analysis, because these were not statistically significantly different (p = 0.43); thus, there was insufficient evidence of a true difference in bleeding. Nonetheless, had we used the TARGET trial rates, the number of bleeds saved with abciximab would have increased from 500 to 660 per 100,000 patients treated, and the cost per year of life saved with the upstream strategy would have changed from $18,000 to $18,350, which is still economically favorable. Third, the principal estimate for the effect of small-molecule glycoprotein (GP) IIb/IIIa inhibitors versus placebo during medical management was obtained from a meta-analysis of relevant randomized clinical trials (3). The range we tested in sensitivity analysis encompasses the 95% CI from that same meta-analysis, but it was also widened to explore the effect of utilizing other relative risks (RRs) cited in the literature, as listed in our Table 1 (1). However, because we agree that post hoc analyses may be biased, these other estimates were not used for our principal analysis; thus, they do not affect the presented principal results (4,5). In all of these cases, regardless of the ranges that are used, it is the principal estimate that is the basis of the results, and thus differing ranges cannot bias the main result of the study.
We did not report a probabilistic sensitivity analysis—assessed by second-order Monte Carlo simulation. We cannot rule out that had we performed such an analysis, the resulting CIs might have widened. But readers should not overinterpret the possibility of such wide CIs. When the point estimate ($18,000) is (substantially) less than our willingness to pay, wide CIs may suggest that we cannot be confident that 2 strategies differ in their value for the cost, but they do not support an interpretation that upstream use is bad value compared to selective use.
We agree that decision analysis techniques occasionally may be mistrusted and misunderstood. We believe that the presentation of our data was transparent in its choices and limitations. A clinical trial addressing the question of upstream versus selective use of GP IIb/IIIa inhibitors is needed. In the absence of a clinical trial, though, a decision analysis provides valuable insight. In this regard, we believe that the importance of decision models is their ability to oblige clinicians and researchers alike to formalize how each piece of evidence or assumption at key decision points impacts our belief of what constitutes an optimal management strategy.
- American College of Cardiology Foundation
- Glaser R.,
- Glick H.A.,
- Herrmann H.C.,
- Kimmel S.E.
- Stone G.W.,
- Moliterno D.J.,
- Bertrand M.,
- et al.
- Pieper K.S.,
- Tsiatis A.A.,
- Davidian M.,
- et al.