Is the Underlying Biology of the Component End Points Similar Enough That One Would Expect Similar Relative Risk Reductions?
Comfort with using a composite end point of a study as the basis of clinical decision making rests in part on confidence that similar RRRs apply to the more important and less important components. Investigators should therefore construct composite end points in which the biology would lead us to expect similar effects across components.
For example, the Irbesartan Diabetic Nephropathy Trial12 randomized 1715 patients with hypertension and nephropathy and type 2 diabetes to irbesartan, amlodipine, or placebo. The primary end point was the composite of a doubling of the baseline serum creatinine concentration, the onset of end-stage renal disease (serum creatinine >6.0 mg/dL, initiation of dialysis, or transplantation), or death from any cause. It is extremely plausible that, for 2 of these 3 components—doubling of creatinine and crossing the creatinine threshold of 6.0 mg/dL—the treatment effects would be similar. Indeed, one would be very surprised if results indicated otherwise. On the other hand, there are many contributors to all-cause mortality aside from renal failure (including, for instance, cardiac disease), and it might well be that treatments have different effects on these contributors. Thus, the biologic rationale that the treatments would have similar effects on all 3 components is weak. The relatively weak biologic rationale increases our reluctance to base treatment decisions on the composite end points, as opposed to its components. Indeed, in this instance, irbesartan lowered the incidence of both doubling of creatinine and end-stage renal disease but without apparent effect on all-cause mortality (Figure 12.4-2).
In contrast, the authors of the Clopidogrel versus Aspirin in Patients at Risk of Ischemic Events (CAPRIE) study, an RCT of aspirin vs clopidogrel in patients with a variety of manifestations of atherosclerosis, argued explicitly for the biologic rationale of their composite end point.13 Citing results of prior trials of antiplatelet agents vs placebo, they note the similar biologic determinants of ischemic stroke, MI, and vascular death: “A meta-analysis of 142 trials … shows clearly that antiplatelet drugs reduce the incidence of a CEP [composite end point] of ischemic stroke, myocardial infarction, and vascular death, the odds reduction being 27%, which is consistent over a wide range of clinical manifestations.”13 Their argument strengthens the case for assuming, until evidence suggests otherwise, that RRRs are consistent across components of the trials' composite end point.
Are the Point Estimates of the Relative Risk Reductions Similar and Confidence Intervals Sufficiently Narrow?
No matter how compelling the investigators' biologic rationale, only similar RRRs can strongly increase our comfort with a composite end point.
For example, in the Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition with Prasugrel–Thrombolysis in Myocardial Infarction (TRITON-TIMI) 38 trial,14 investigators randomized 13 608 patients with moderately high-risk acute coronary syndrome and scheduled PCI to receive either prasugrel or clopidogrel for 6 to 15 months and recorded the effect of these medications on a primary composite end point of cardiovascular mortality, MI, and stroke. A reasonable biologic rationale suggests that if there is a difference between these 2 antiplatelet agents, the difference should be similar across the 3 outcomes. Figure 12.4-3, however, shows that this is not to the case. The RRR for the composite end point of 19% (95% CI, 10%-27%) does not apply to the individual components: RRRs of −2% for stroke, 24% for MI, and 11% for cardiovascular death. This variability suggests that clinicians should focus on individual end points. The TRITON-TIMI 38 trial suggests that a prasugrel-based regimen, compared with a regimen based on clopidogrel, may reduce the risk for MI but has uncertain effects on cardiovascular mortality and stroke in patients with acute coronary syndrome and scheduled PCI.14
The UK Prospective Diabetes Study (UKPDS) trial of intensive glycemic control vs conventional control in patients with type 2 diabetes provides another example.15 This study reported that the primary end point of the trial was time to first diabetes-related end point (sudden death, death from hyperglycemia or hypoglycemia, fatal or nonfatal MI, angina, heart failure, stroke, renal failure, amputation, vitreous hemorrhage, retinal photocoagulation, blindness in one eye, or cataract extraction); diabetes-related death (death from MI, stroke, peripheral vascular disease, renal disease, hyperglycemia or hypoglycemia, and sudden death); or all-cause mortality. Although the investigators reported a significant 12% reduction in the RR in the composite end point (95% CI, 1%-21%), the results do not exclude a harmful effect on diabetes-related deaths (RR, 0.90; 95% CI, 0.73-1.11) and all-cause mortality (RR, 0.94; 95% CI, 0.80-1.10).15 Moreover, most of the apparent effect was a reduction (2.7% of the 3.2%, or 80% of the absolute reduction in risk of microvascular complications) in retinal photocoagulation.15,16
Reviewers typically summarize the results as showing a reduction in any of 21 diabetes-related end points with intensive glycemic control, and only 1 in 35 reviews of the UKPDS results highlighted the dominance on the overall effect of the reduction in the risk of photocoagulation.17
These results contrast with those of the HOPE trial of ramipril vs placebo in patients at high risk of vascular events that we described earlier.11 Here, the RRRs in the same 3 end points were 26% (95% CI, 13%-36%) for cardiovascular death, 20% (95% CI, 10%-30%) for MI, and 32% (95% CI, 16%-44%) for stroke. For each of the 3 components of the composite end point, the clinician can be confident that treatment affects the component favorably.
Finally, consider the results of the Clopido-grel in Unstable Angina to Prevent Recurrent Events (CURE) trial, in which investigators randomized 12 562 patients with acute coronary syndrome to clopidogrel or placebo and examined the effect on the same composite end point as reported in the trial reviewed in our opening clinical scenario: cardiovascular death, MI, or stroke.18 In this trial, there was a reduction in the RR of the composite outcome of 20%, with a reduction in the RR of MI, cardiovascular death, and stroke of 23%, 7%, and 14%, respectively. Here, although one could interpret the point estimates of the RRR as consistent with one another and with the composite, the range of the CIs should give us pause. Although the point estimate and 95% CI on the RRR leave us reasonably confident of an important treatment effect on MI (23%; 95% CI, 11%-33%), the same is not true of either cardiovascular death (7%; 95% CI, –8% to 21%), or stroke (14%; 95% CI, –18% to 37%). As a result, the statement that clopidogrel reduced a composite end point of cardiovascular death, stroke, and MI by 20% is potentially misleading, and use of the composite end point as a basis for clinical decision making is problematic.
TRITON-TIMI 38 Trial14
Abbreviations: CI, confidence interval; RRR, relative risk reduction; TRITON-TIMI, Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition with Prasugrel–Thrombolysis in Myocardial Infarction.
Many of the examples we have presented highlight the typical situation. The number of events and magnitude of effect are typically greater (and often considerably greater) for the less important end points than for the more important. We have found that this is the case in randomized trials in cardiovascular interventions, in which components of greater importance to patients were associated with smaller treatment effects than less important ones (RRR of 8% for death and 33% for components of minor importance to patients).5 In individual trials, wide CIs may leave us uncertain of the relative effect of the intervention on more and less important end points. Sometimes, however, when data accumulate from many trials, it becomes clear that skepticism about treatment effect on the most important outcomes, even in the presence of convincing evidence of impact on outcomes of lesser seriousness, was well warranted.
For example, consider that trials that compare drug-eluting stents with bare-metal stents indicate conclusively that the drug-eluting stent reduces the composite end point MACE (major adverse cardiac events—a composite of death, MI, and target lesion revascularization) compared with the bare-metal stent. The first RCT of drug-eluting stents found a nonconclusive effect at 1 year on the most important end points of death (RRR, 2%; 95% CI, –680% to 76%) and MI (RRR, 51%; 95% CI, –530% to 96%), whereas the drug-eluting stent had a large effect on reducing revascularization (RRR, 97%; 95% CI, 75%-99%).19 These early results suggest a substantial difference in effects on the more important and less important end points, but the wide CIs around the former leave the issue in doubt. Subsequent systematic reviews have found that drug-eluting stents have no benefit in survival or Q-wave MI (in fact, there is an unfavorable trend for Q-wave MI), whereas there is a large benefit in reducing the need for revascularization.20
CLINICAL SCENARIO RESOLUTION
Let us return to the scenario with which our discussion began, that of the patient reluctant to undergo cardiac surgery to improve his prognosis and to control his angina. Is it reasonable to use the composite end point from the SYNTAX trial1—death, MI, stroke, and subsequent revascularization—to guide the decision, or should we focus on individual results of the 3 components?
To address this issue, we can ask the 3 questions in our Users' Guides (Box 12.4-1). Most patients will find death, stroke, and large MI with subsequent disability far more important than the need of a new revascularization. Subsequent revascularization occurred far more frequently than the 3 more important events (Table 12.4-1). Biologic rationale may support a presumption that the surgery strategy will have similar effects on cardiac end points; however, one could also argue the opposite and make the case for a positive effect on MI and the need for new coronary revascularization but not on stroke. The relative effects on the 3 outcomes proved very different (Table 12.4-1). In the SYNTAX trial, the composite end point fails all 3 criteria, and the trial therefore requires focus on the component end points. In this case, the patient proved extremely stroke averse. “Doctor,” he says, “that PCI is pretty simple, so the possibility that I might need a repeat is not a big deal to me. You tell me that although the CABG might reduce my risk of dying or having an MI, you are not sure. But it is very likely the CABG will increase my risk of stroke. To me, a stroke would be awful. I'll go with the PCI.”