Surrogate endpoints in drug approval process examined, found wanting

Surrogate endpoints in clinical trials (ie, hemoglobin [Hb] A1c) on which drug approvals may be based only explain narrow aspects of complex conditions. For this reason, even when using a drug as labeled, there is always a level of uncertainty over outcomes with the drug, said Saul Malozowski, MD, PhD, MPH.

Even when outcomes trials with a particular drug are performed, the eventual population that is given the drug inevitably differs from the population enrolled in the clinical trials, which may lead to unexpected outcomes.

The impetus for his talk, said Dr. Maolozowski, was the meta-analysis of rosiglitazone by Nissen et al (N Engl J Med 2007;356:2457-2471), in which a review of randomized controlled trials uncovered an excess of myocardial infarction in rosiglitazone recipients relative to placebo or other drugs used to treat diabetes.

Rosiglitazone was approved on the basis of short-term studies using the surrogate endpoint of glycemic control. For a drug that activates a large set of genes, the overall risk : benefit balance is uncertain, he said, especially when the drug is meant to be used for a life-long condition. The uncertainty only increases when a drug is used off-label.

"It is naive to assume that one surrogate endpoint (HbA1c) alone will faithfully reflect all the alterations seen in a patient with diabetes," said Dr. Malozowski, Division of Diabetes, Endocrinology, and Metabolic Diseases National Institute of Diabetes and Digestive and Kidney Diseases. Diabetes is a diagnosis that encompasses multiple genetic variants as well as distinct patient populations, he noted. In addition, metabolic alterations are present even before the diagnosis is made.

"HbA1c is used to approve a drug because it reflects the level of glucose control," he said. However, the implications of using HbA1c on potential micro- or macrovascular outcomes is not part of the equation. The registration studies are not designed to answer these questions, he said.

Some critics have called for registration studies to include hard endpoints other than HbA1c, but such studies are costly, and finding somebody to pay for them is problematic. Another difficulty lies in designing studies of sufficient power to address all patient populations based on age, gender, and concurrent medications being taken.

Switching from surrogate endpoints would be a major undertaking. In a literature search, the terms "biomarker" or "surrogate endpoint" elicited 35,000 to 38,000 references. On the National Institutes of Health's home page, 13,500 documents use the term "biomarker." At the current meeting of the American Diabetes Association, achieved HbA1c levels were not able to predict cardiovascular complications in three major clinical trials (ADVANCE, ACCORD, VA Diabetes Trial). Is this reason enough to dismiss the use of HbA1c for drug development or approval, he asked.

In his 1999 evaluation of rosiglitazone, Dr. Malozowski addressed some of the recent concerns that have emerged regarding the risk : benefit profile with rosiglitazone. He believes that the major points in this evaluation remain relevant. He noted that reductions in HbA1c have been associated with a decrease in the risk for microvascular complications with other drugs for which outcomes trials were performed. Whether these benefits can also be attributed to rosiglitazone needed to be explored. Although the cardiac safety profile of rosiglitazone seemed to be benign in the population exposed, he wrote that it remained to be seen whether concerns are dispelled once the drug reaches the market and it is used in patients with profiles unlike the ones exposed in the registration trials, or if the time of exposure increases (the effects on HbA1c were examined over 12 months in clinical trials).

At the time, patients with New York Heart Association class III or IV had not yet been exposed to rosiglitazone. He noted that the worsening lipid profile observed with use of rosiglitazone appeared to be due to a direct pharmacologic effect of the compound, and this effect was not welcome in the treatment of type 2 diabetes.

In closing, he remarked, "analogies are great teaching instruments but should not be used to make prescribing decisions unless they are part of an ethically designed clinical study."