Clinical Trials—Multiple Treatments, Multiple End Points, and Multiple Lessons

In 1986, the GISSI group¹ published a report of a large randomized clinical trial demonstrating that intravenous streptokinase reduces the risk of death among patients with acute myocardial infarction. The design of the trial was simple, as was the published report. There was one treatment, intravenous streptokinase, whereas the control group received no streptokinase. There was no blinding and no placebo. The end point of interest was death from any cause during hospitalization.

Since the publication of the landmark GISSI-1 trial, many more randomized trials have been performed and their results published. However, as noted in 2 articles published in this issue of THE JOURNAL, the nature of clinical trials and their reporting has become more complex. Freemantle and colleagues² address the issue of trials that focus on composite end points that consist of multiple outcomes, rather than on a single end point such as all-cause mortality. The authors identified 167 randomized trials published in 9 journals since 1997 that included all-cause mortality as part of a composite end point. Among the 79 trials in which the composite end point yielded a statistically significant result, the mortality end point by itself was reported as statistically associated with the intervention in 19 (24%). In 22 trials, the impact of the intervention on mortality was not separately reported.

Freemantle et al identify a number of problems they consider of importance regarding composite end points. These include the inappropriate attribution of treatment effects on specific end points when only composite end points were found to yield significant results, the dilution of effect when individual end points might not all react the same way to treatment, and the excessively influential effects of end points that occur at the subjective discretion of clinicians, such as hospitalization, and that are not associated with irrevocable damage.

While Freemantle et al correctly highlight some of the complexities of using composite end points, a number of other issues deserve consideration. Among fatal end points, only all-cause mortality can be considered objective, unbiased, and clinically relevant.^{3 - 5} As previously reviewed in depth, the use of end points such as "cardiac death," "vascular death," and "arrhythmic death" are inherently subject to error due to biased assessment^{3 - 5} and to the biological complexities of disease, especially among elderly individuals.⁶ Failure to focus on all-cause mortality in the past has led to misleading impressions, as illustrated by the finding that amiodarone may be beneficial in survivors of myocardial infarction because it reduced the rate of arrhythmic death but had no impact on all-cause death.^{3 ,7}

Many contemporary trials add nonfatal end points to all-cause mortality, as pointed out by Freemantle et al, in part because this improves statistical efficiency.² A number of disease entities, such as acute coronary syndromes without ST-segment elevation, have low absolute mortality rates, meaning that sample sizes must be very large to demonstrate a significant treatment effect.⁸ Furthermore, because of advances in treatment, placebo or nontreatment controls are often no longer ethically permissible,⁹ meaning that the differences in outcomes between newer and older treatments will be relatively small, again mandating very large sample sizes. Nonetheless, more efficient trials with smaller sample sizes assembled to show a desired statistical effect do not necessarily compromise validity.

When considering composite end points, investigators designing clinical trials and clinicians reading the published results of clinical trials should ask a number of key questions. Are the end points themselves of clinical interest to patients and their physicians, or are they merely surrogates for clinically meaningful end points?⁵ End points that involve irrevocable tissue loss, such as myocardial infarction or stroke, have real clinical import. Similarly, adverse events directly related to therapy that themselves require specific treatment (eg, major bleeding) are legitimately of primary interest.¹⁰ Surrogate end points, on the other hand, are problematic, as was clearly illustrated by the trials in which antiarrhythmic agents successfully suppressed ventricular ectopy but led to a higher death rate.¹¹

A second important question is how nonfatal end points are measured. Any end point that requires a measurement involving human judgment is inherently subject to bias and hence mandates a blinded end points committee, a core laboratory, or both.^{12 - 15} For example, when myocardial infarction is considered, an electrocardiography core laboratory is often combined with a clinical events committee to determine whether or not an outcome event truly occurred.¹²

A third question is how many individual end points are considered part of a composite end point and how are they reported. When composite end points are analyzed, the individual end points are considered as being of equal clinical importance, something evidently not true whenever death is placed alongside a nonfatal event. Hence, it is clearly preferable to minimize the number of individual end points to no more than 3 or 4. As Freemantle et al correctly recommend, the results of the individual end points must be reported. Clinicians and patients need to be aware of treatments that primarily impact on only 1 element of a composite end point, such as late revascularization, as well as treatments in which individual end points behave differently. For example, in the recently reported TIME trial,¹⁶ aggressive revascularization in elderly patients had no impact on long-term quality of life, while it seemed to improve short-term quality of life—but this came at the expense of a trend toward higher early mortality and myocardial infarction. On the other hand, when each individual end point behaves in the same way, albeit not necessarily to a statistically significant degree, a more clinically robust effect of treatment might be reasonably inferred.

Moreover, it is important to ask how nonfatal events were analyzed. A composite end point that includes death as well as nonfatal events is subject to biases related to competing risks.^{17 - 18} Obviously, patients who die cannot later experience nonfatal myocardial infarction or be hospitalized. A treatment that leads to an increased risk of death may therefore appear to reduce the risk of nonfatal events. Although formal methods have been developed to analyze competing risks in an unbiased manner,¹⁸ the optimal approach to this problem is unclear.

In a second article in this issue of THE JOURNAL, McAlister and colleagues¹⁹ systematically reviewed trials that used a factorial design, in which randomization to more than 1 treatment was performed in the same study cohort. Whereas nearly all of the 44 trials they considered did correctly compare all patients receiving 1 treatment with all patients receiving its control, only 26 formally reported testing for interactions among treatments and only 29 reported data for each individual treatment cell. McAlister et al propose a method of analyzing factorial trials that makes it possible to determine the individual effects of each treatment as well as how the treatments might interact with each other.

As McAlister et al conclude, factorial designs are a highly efficient means to study the effects of multiple treatments by means of randomized trials. Factorial designs should therefore be encouraged and fostered. Yet factorial designs present other opportunities of critical importance in modern medicine. Because a number of chronic diseases, such as coronary artery disease, heart failure, and AIDS, can successfully be treated using multiple agents, it is critically important to know whether and how different treatments interact. For example, one recent trial examined the impact of statins and vitamins on the progression of coronary disease; not only did vitamins fail to produce a beneficial effect, they also blocked the beneficial effects of statins.²⁰ Only a factorial design trial can reliably lead to the conclusion that such a clinically important interaction occurs. As another example, a controversy exists as to whether or not aspirin and angiotensin-converting enzyme inhibitors may interfere with each other's salutary effects.²¹ Because a factorial trial has not been done, a definitive answer to this question remains elusive.

Future factorial trials will be needed to identify the benefits and pitfalls of polypharmacy in different patient groups. Factorial trials need not necessarily be limited to 2 × 2 designs; one can envision 2 × 3 or even more complex groupings that mirror trends in contemporary practice where 3 or more drugs are frequently combined. Furthermore, factorial designs make it possible to test different dosing strategies on outcome.

Most major high-impact peer-reviewed journals have adopted the CONSORT guidelines, ensuring standardized precise presentation of trial results.^{22 - 23} The reports by Freemantle et al and McAlister et al take these guidelines to a higher level and are important additions to the literature on the design and reporting of clinical trials. When composite end points are used, the individual components must be appropriately chosen, objectively measured in an unbiased manner, and individually reported. Factorial designs also should be analyzed and reported in such a way that individual treatment effects as well as treatment interactions can be assessed.

The lessons learned from randomized trials go beyond published reports of trial results. Because randomized trials are labor intensive, time consuming, and expensive; because their results may be difficult to generalize; and because some phenomena cannot be subject to randomization, much contemporary clinical investigation is observational, rather than experimental.^{24 - 25} Yet the issues explored by Freemantle et al and McAlister et al are applicable to the design, analysis, and reporting of observational studies as well. When mortality is considered, only all-cause mortality is a valid end point, while end points such as "cardiac death" and "arrhythmic death" should be actively discouraged.⁴ Composite end points may be appropriate, but may well be limited by biases related to differential associations of exposures and individual outcomes and related to competing risks. Nonfatal end points, such as myocardial infarction, should only be considered valid and reliable if assessed by a blinded clinical events committee or a core laboratory; otherwise, they should only be reported as secondary or supplementary. In addition, exposures, including treatments, should not be considered as isolated entities but rather as risk factors that may well interact with one another in a clinically important way. Thus, the multiple lessons learned from clinical trials should be carefully considered throughout the entire sphere of patient-oriented research.