Composite Endpoints: Proceed with Caution

The choice of a valid primary endpoint is crucial for a randomized controlled trial (RCT) designed to demonstrate efficacy of a new pharmaceutical drug. Recent improvements in medical management, however, have led to a continuous decline in mortality and morbidity for several common disorders, i.e., also in the incidence of clinically relevant outcomes that may be addressed in a drug trial as an endpoint. For example, secondary prevention after myocardial infarction with aspirin, β-blockers, ACE inhibitors, and statins has remarkably lowered the rate of a subsequent cardiovascular event,¹ and combination antiretroviral therapy for HIV infection has made mortality even unfeasible as a clinical endpoint.²

Moreover, for many diseases a standard of care has been established, and it is often not possible anymore to compare new therapeutic options with no treatment or placebo for ethical reasons.^3,4 As a consequence, the effect sizes observed with add-on treatments when compared to standard treatment are generally lower today. Both the reduced incidence of disease-related events and the limited chance to provide significant differences in treatment effects require trials with large sample sizes.⁵

As a solution to maintain feasibility of RCTs, particularly those facing low event rates, the investigation of composite endpoints has been introduced and become fashionable in the last few years.

Why to use a composite endpoint

A composite endpoint in a RCT consists of multiple single endpoints that are combined in order to confront an investigational drug with a higher number of events expected during the trial. For instance, the primary composite endpoint may include mortality along with nonfatal endpoints such as hospitalization and cardiac arrest in chronic heart failure patients,⁶ or along with myocardial infarction and stroke in hypertensives.⁷

The major advantages in using a composite endpoint are: statistical precision and efficiency will be increased⁸ ; trials become smaller, less costly; and the results of promising new treatments will be available earlier. If more than one outcome is important for efficacy evaluation, a composite endpoint can efficiently deal with the issue of multiplicity.⁹ A summary measure for drug efficacy can be defined¹⁰ (e.g., the assessment of skeletal-related events [SREs] in trials for prevention or treatment of bone metastases).

The selection of composite endpoints has been regarded as "a useful strategy" for registration of pharmaceuticals in the ICH E9 guideline.⁹ A Points-to-consider document released by the European Agency for the Evaluation of Medicinal Products (EMEA) specifies regulatory requirements.¹¹

Clinical and regulatory requirements

The use of a composite endpoint in a clinical trial is usually justified if the following assumptions are respected:

• The individual components of the composite are clinically meaningful and of similar importance to the patient.⁸

• The expected effects on each component are similar, based on biological plausibility¹² (which is, in the end, the rationale for using a composite endpoint). Accordingly, regulatory guidelines also require components for which it can be assumed treatment will beneficially influence in a similar way.

• The clinically more important components of composite endpoints should at least not be affected negatively.¹¹

As a consequence of the third assumption, regulatory authorities require all components of a composite endpoint to be analyzed separately.^9,11 One needs to be aware whether a treatment affects all components or just a single outcome.⁵

Limitations of composite endpoints

The selected individual components of a composite endpoint, as reported in the biomedical literature, are not always clinically meaningful. The suitability of a combination of all-cause mortality, the need for intubation and mechanical ventilation together with intensification of steroid therapy in a trial of patients with chronic obstructive lung disease¹³ may be questioned, as patients are likely to consider the need for steroids of trivial importance.¹²

Moreover, the problems with nonvalidated surrogate endpoints need to be kept in mind when selecting outcome variables for the composite, particularly when a relationship between the surrogate and mortality has not been established. Although strongly effective in suppressing ventricular arrhythmia, class IC anti-arrhythmic agents were shown to significantly increase mortality after myocardial infarction.¹⁴ In contrast, carvedilol treatment demonstrated favorable effects on survival in patients with moderate to severe heart failure; however, there was little benefit on exercise tolerance or quality-of-life scores.¹⁵

If a composite endpoint consists of death, myocardial infarction, and stroke, for example, the distribution of these single outcomes is of lower importance. The situation may be different if the less important outcomes account for the majority of events. This is particularly the case when "soft" clinical endpoints like refractory angina or the associated need for revascularization are combined with the risk for death or myocardial infarction.

The TIME study,¹⁶ for instance, compared invasive and medical treatment in an elderly population with chronic angina. The frequency of the composite endpoint (death, nonfatal myocardial infarction, and hospital admission for ACS) was much lower with revascularization; however, this was due to a marked difference in hospital admissions, which accounted for 75% of the events in the medical treatment group. In contrast, there were twice as many deaths in the invasive treatment group. The question remains how to interpret the results and inform a patient who has to decide between conservative or surgical therapy.¹²

The Irbesartan Diabetic Nephropathy Trial¹⁷ compared the renoprotective effect of the angiotensin blocker irbesartan with amlodipine in type 2 diabetics with nephropathy. The primary composite outcome consisted of all-cause mortality, end-stage renal disease (ESRD), and doubling of the serum creatinine concentration. In the definition of ESRD, the serum creatinine concentration (increase > 6.0 mg/dl) was one component. The authors did not mention how ESRD diagnosis was assessed according to the underlying definitions; a significant contribution of creatinine values to ESRD diagnosis can not be excluded, which would mean that a laboratory value had determined two of the three components of the composite.

But even independent of this, the majority of events concerned doubling of the serum creatinine concentration and were the main basis for the overall beneficial effect of irbesartan compared to amlodipin. Reduction of ESRD with irbesartan failed to be statistically significant (when adjusted for blood pressure), and mortality—the most important endpoint—was not different between both treatment groups. The question remains: What at all is the value of serum enzyme concentrations as part of a composite endpoint, particularly if the risk of death is unchanged in the investigated population?

Adding a component that foreseeably is insensitive to treatment increases variability and, as a direct consequence, makes it more difficult to provide clear results, particularly for interpreting noninferiority trials.¹¹ Time-to-treatment failure is therefore seldom useful for regulatory purposes, as treatment discontinuation because of toxicity or voluntary withdrawal has no direct relevance to the underlying disease process or the effectiveness of a therapy.^8,18

Indication and bias awareness

One needs to be aware that several drugs (or new indications, respectively) have been approved based on composite endpoints although evidence of mortality reduction was lacking. As an example, glycoprotein IIb/IIIa inhibitors have shown to significantly reduce a composite outcome of death, myocardial infarction, or refractory angina; however, all-cause mortality, the most important component, was not changed.⁸

To assess the real clinical value of a drug, it is necessary to look in detail at the wording of the approved indication and the effects on single, clinically relevant endpoints. According to the systematic review by N. Freemantle et al.,⁸ more than one third of published trials using a primary composite outcome with a mortality component showed an overall significant result, although the single mortality endpoint did not.

Finally, one should be aware of a potential bias. There may be competing risks between endpoints, and one needs to particularly look whether a positive composite endpoint camouflages a negative individual outcome or dilutes the effect of the treatment on mortality. Myocardial infarctions or hospitalizations may be reduced by a treatment only because mortality is increased. Studying nonfatal events without including deaths is methodologically invalid, as those with the worst outcome are censored.¹⁹ Endpoints that require subjective judgement (e.g., ECG findings or X-rays) are also subject to bias. Therefore, it is important to standardize endpoint assessment, preferably by a blinded endpoint adjudication committee, as recommended by an EMEA guideline.²⁰

The following example demonstrates how important an endpoint committee may be. In the so-called TRIM trial, the endpoint committee reduced the number of patients with the endpoint myocardial infarction or refractory angina by 13.6% and reclassified study-related events for 28% of the patients.²¹

Some words of caution

There is no guarantee for success when using a composite endpoint. Despite an underlying biologic rationale, the effects on each component of the composite endpoint may be unexpectedly different, as shown in the LIFE trial⁷ : compared to atenolol, losartan significantly reduced the incidence of stroke, but, on the other hand, it increased the incidence of myocardial infarction.

A "worst case scenario" happened in the CAPRICORN trial,²² which investigated carvedilol in patients with left-ventricular dysfunction after myocardial infarction. The original endpoint, all-cause mortality, was changed during the trial to a new composite consisting of mortality and hospital admissions. Whereas the "old" endpoint achieved a p-value of 0.03, the "new" combined endpoint, in contrast, failed to show statistical significance.

A further problem may arise if the decision to stop a trial early is based on the monitoring of a composite endpoint, particularly if this is driven by the least patient-important outcome.²³ Such an approach may lead to overestimation of the benefit and underestimation of the risk.

Questions instead of answers

The use of a composite endpoint in a clinical trial is comprised of chances and risks. Composite endpoints can be of value if both requirements and limitations are respected. When planning a trial or reading about a trial with a composite endpoint, the following key questions should be answered:

• Does the composite endpoint really measure a disease?

• Does the use of a composite endpoint solve a medical problem or is it just for statistical convenience?

• Are the individual components of the composite endpoint valid, biologically plausible, and of importance for patients?

• Are the results clear and clinically meaningful? Do they provide a basis for therapeutic decisions? Does each single endpoint support the overall result?

• Is the statistical analysis adequate?

And a last word addressed to journal editors: Trials with composite endpoints should be reported unequivocally, providing all information necessary for judgement of the trial.

Peter Kleist, MD, FFPM, a Swiss specialist in pharmaceutical medicine, is medical director with PFC Pharma Focus AG, Chriesbaumstrasse 2, CH-8604 Volketswil, Switzerland, email: peter.kleist@pfc.ch

References

1. S. Orn and K. Dickstein, "Pharmacotherapy Following Myocardial Infarction—A Review of Current Treatment Practices," Expert Opinion on Pharmacotherapy, 1, 1105–1116 (2000).

2. F.J. Palella, K.M. Delaney, A.C. Moorman et al., "Declining Morbidity and Mortality among Patients with Advanced Human Immunodeficiency Virus Infection," New England Journal of Medicine, 338, 853–860 (1998).

3. World Medical Association, "Declaration of Helsinki," www.wma.net/e/policy/b3.htm (accessed 16 December 2005).

4. K.J. Rothman and K.B. Michels, "The Continuing Unethical Use of Placebo Controls," New England Journal of Medicine, 331, 394–398 (1994).

5. M.S. Lauer and E.J. Topol, "Clinical Trials—Multiple Treatments, Multiple End Points, and Multiple Lessons," Journal of the American Medical Association, 289, 2575–2577 (2003).

6. J.N. Cohn and G. Tognoni, for the Valsartan Heart Failure Trial Investigators, "A Randomized Trial of the Angiotensin-Receptor Blocker Valsartan in Chronic Heart Failure," New England Journal of Medicine, 345, 1667–1675 (2001).

7. B. Dahlöf, R.B. Devereux, S.E. Kjeldsen et al., "Cardiovascular Morbidity and Mortality in the Losartan Intervention for Endpoint Reduction in Hypertension Study (LIFE): A Randomised Trial against Atenolol," Lancet, 359, 995–1003 (2002).

8. N. Freemantle, M. Calvert, J. Wood, J. Eastaugh, C. Griffin, "Composite Outcomes in Randomized Trials. Greater Precision but with Greater Uncertainty?" Journal of the American Medical Association, 289, 2554–2559 (2003).

9. International Conference on Harmonization, Guideline E9, "Statistical Principles for Clinical Trials," www.ich.org/LOB/media/MEDIA485.pdf (accessed 16 December 2005).

10. F. van Leth and J.M.A. Lange, "Use of Composite End Points to Measure Clinical Events," Journal of the American Medical Association, 290, 1456–1457 (2003).

11. Committee For Proprietary Medicinal Products (CPMP), "Points to Consider on Multiplicity Issues in Clinical Trials," www.emea.eu.int/pdfs/human/ewp/090899en.pdf (accessed 16 December 2005).

12. V.M. Montori, G. Permanyer-Miralda, I. Ferreira-Gonzalez et al., "Validity of Composite End Points in Clinical Trials," British Medical Journal, 330, 594–596 (2005).

13. D.E. Niewoehner, M.L. Erbland, R.H. Deupree et al., "Effect of Systemic Glucocorticoids on Exacerbations of Chronic Obstructive Pulmonary Diease," New England Journal of Medicine, 340, 1941–1947 (1999).

14. D.S. Echt, P.R. Liebson, I.B. Mitchell et al., "Mortality and Morbidity in Patients Receiving Encainide, Flecainide, or Placebo: The Cardiac Arrhythmia Suppression Trial," New England Journal of Medicine, 324, 781–788 (1991).

15. M. Packer, W.S. Colucci, J.D. Sackner-Bernstein et al., "Double-Blind, Placebo-Controlled Study of the Effects of Carvedilol in Patients with Moderate to Severe Heart Failure. The PRECISE Trial," Circulation, 94, 2793–2799 (1996).

16. The TIME Investigators, "Trial of Invasive Versus Medical Therapy in Elderly Patients with Chronic Symptomatic Coronary-Artery Disease (TIME): A Randomised Trial," Lancet, 358, 951–957 (2001).

17. E.J. Lewis, L.G. Hunsicker, W.R. Clarke, T. Berl, M.A. Pohl, J.B. Lewis et al., "Renoprotective Effect of the Angiotensin-Receptor Antagonist Irbesartan in Patients with Nephropathy Due to Type 2 Diabetes," New England Journal of Medicine, 345, 851–860 (2001).

18. J.R. Johnson, G. Williams, R. Pazdur, "End points and United States Food and Drug Administration Approval of Oncology Drugs," Journal of Clinical Oncology, 21, 1404–1411 (2003).

19. H. Skali, S.D. Solomon, M.A. Pfeffer, "Are We Asking Too Much of Our Trials?" American Heart Journal 143, 1–3 (2002).

20. Committee for Medicinal Products for Human Use (CHMP), "Guideline on Data Monitoring Committees," www.emea.eu.int/pdfs/human/ewp/587203en.pdf (accessed 19 December 2005).

21. U. Näslund, L. Grip, J. Fischer-Hansen, T. Gundersen, S. Lehto, L. Wallentin, "The Impact of an End-Point Committee in a Large Multicentre, Randomized, Placebo-Controlled Clinical Trial," European Heart Journal, 20, 771–777 (1999).

22. The CAPRICORN Investigators, "Effect of Carvedilol on Outcome After Myocardial Infarction in Patients with Left-Ventricular Dysfunction: The CAPRICORN Randomised Trial," Lancet, 357, 1385–1390 (2001).

23. V.M. Montori, P.J. Deveraux, N.K.J. Adhikari et al., "Randomized Trials Stopped Early for Benefit. A Systematic Review," Journal of the American Medical Association, 294, 2203–2209 (2005).