Surrogate Endpoints

Note: this is a short primer on surrogate endpoints. I plan to flesh out this post at a later date.

Surrogate endpoint is the most important, and the most dangerous, tool in the clinical trialist’s arsenal.

Surrogate endpoints are endpoints that are not clinically meaningful in itself but correlate with and can predict the clinically important endpoint. One classic example is hyperglycemia. High level of blood sugar is a surrogate endpoint that correlate with, and (hopefully and presumably, but perhaps not quite as effectively as we hope) predict renal failure, cardiovascular events, diabetic retinopathy, and other sequelae of metabolic syndrome. In and of itself, high levels of glucose (unless high enough to cause polyuria) is not symptomatically or clinically detrimental. The reason we treat diabetes is to prevent the long-term sequelae, with the assumption that lowering glucose level will lower the risk of the sequelae.

Examples of Surrogate Endpoints

·  Pharmacokinetic/pharmacodynamic Measures

·  In-vivo biomarkers (e.g., CD4 count, viral Load, Glucose level, cholesterol level)

·  Clinical Surrogates (e.g., Blood Pressure)

·  Ex-vivo measures

·  MIC (minimal inhibitory concentration) of an anti-bacterial agent

·  ADP-induced platelet aggregation inhibition

·  Non-clinical measures (e.g., FEV1, radiographic findings)

Surrogate endpoints are used when it is difficult or impractical to use real clinical outcome as endpoints.  It may take too long or require too many patients to see a real clinical outcome endpoint.   Or it may be too costly or cause too much discomfort to measure a real clinical outcome. For example, bone mineral density can be used in place of hip fractures.

Since surrogate endpoints commonly guide treatment decisions in clinical practice (e.g., a 95% stenosis in a coronary artery may lead to a percutaneous coronary intervention, high glycosylated hemoglobin levels might lead to increase in the insulin dose, and active urine sediment might precipitate aggressive immunosuppression in a lupus patient), many clinical trials use surrogate endpoints, and their results often can drive clinical practice.

However, a surrogate endpoint is never as informative as the clinical endpoint, and in many instances, surrogate endpoints have turned out not to be predictive of clinical response at all.  For example, antiarrhythmics that prevent premature ventricular contractions (PVC’s) have actually increased mortality. Some drugs that lower blood pressure do not lower the risk of cardiovascular problems.  Moreover, a surrogate endpoint that works for one drug may not for another drug with a different mechanism of action.

As a result, regulatory authorities and many clinicians insist on clinical rather than surrogate endpoints. Because regulatory agencies rarely accept surrogate endpoints for approval of new therapies, surrogates are typically used more often in Phase II rather than Phase III studies.

An Example of a Failed Surrogate Endpoint: Cardiac Arrhythmia Suppression Trial (CAST) Study

High rates of premature ventricular contractions (PVCs) are predictive of sudden death after myocardial infarction.  Several drugs were developed (and approved) for suppression of PVCs, with the ultimate goal of reducing death after myocardial infarction.  The CAST study was initiated in 1987 with flecainide, moricizine, and encainide, which had been shown to be highly effective at reducing PVCs.  At initiation of the trial, there was debate over whether it was ethical to randomize patients to placebo when the drugs had been demonstrated to reduce PVCs.  In all, 2309 patients were randomized.

The Data Safety Monitoring Board (DSMB) stopped the study early because the patients receiving antiarrhythmic therapy had an unacceptably high mortality.  The relative risk (RR) of death and nonfatal events at 10 months was 4.6 in favor of placebo.

As an aside, the rate of mortality seen in the antiarrhythmic group was lower than historical controls.  If it had been deemed unethical to conduct a placebo-controlled trial, then we may still be using these drugs in post-myocardial infarction patients.

This post is excerpted from Principles and Practice of Clinical Trial Medicine.