Are You Sure About That Surrogate Endpoint?
Even if it makes sense, it might be misleading
The goal of healthcare is twofold: to help patients live longer, and to help patients live better in terms of their quality of life. As clinicians, our job is to implement interventions (like prescribing medications or performing procedures) with the aim of achieving these ends. But first, we need to know if an intervention is safe and effective. Clinical trials are specifically designed to help provide this information.
Clinical trials have pre-specified endpoints, which are the measurable outcomes used to assess whether or not the trial has met its objectives. There are many different kinds of endpoints, but for now, we’ll focus on clinical endpoints and surrogate endpoints. Clinical endpoints are the outcomes that definitely matter. Did patients who received the intervention live longer? Did they have fewer heart attacks and strokes? Did they have fewer fractures? Clinical endpoints provide invaluable information, but measuring them is often difficult, impractical, and expensive.
Therefore, researchers often use surrogate endpoints that are much easier to measure as stand-ins for the clinical endpoints we’re most interested in. Surrogate endpoints can be lab values like HbA1c and cholesterol levels, or other measurable variables like weight, blood pressure, bone density, or tumor size. Vinay Prasad, a hematologist-oncologist and associate professor at the University of California San Francisco, adds, “A surrogate endpoint is an endpoint a patient didn’t know mattered until the doctor said it mattered.”
The problem is, although surrogate endpoints are thought to predict the clinical endpoints we care about most, these relationships aren’t guaranteed. Does lowering blood pressure actually lower the chance a patient will have a stroke? Does improving bone density actually decrease the risk of a patient getting a fracture? Unfortunately, in many cases, conclusions have been drawn from clinical trials based on surrogate endpoints that turned out to be misleading. Consider the following example about the surrogate endpoint of premature ventricular contractions (PVCs).
PVCs are extra heartbeats that are strongly associated with sudden death, and flecainide is a medication that was widely used to suppress these extra heartbeats. Then, the results of the Cardiac Arrhythmia Suppression Trial (CAST) were published. Contrary to clinicians’ expectations, flecainide was shown to increase patients’ risk of death (the clinical endpoint) even though it decreased PVCs (the surrogate endpoint). In Ending Medical Reversal, Prasad and coauthor/physician Adam Cifu note, “CAST taught us that even the most careful reasoning and the best scientific models do not guarantee an effective clinical treatment. What works in the lab, or on a computer, or in the head of the smartest researcher does not always work in a patient.” Although we should have learned this lesson many times over by now, unfortunately, we’re still learning it the hard way.
The point is not that surrogate endpoints are always misleading and should never be used; in some cases, they’re both necessary and useful. But surrogate endpoints warrant skepticism. We can’t know for sure that a surrogate endpoint is going to provide an accurate indication of the clinical endpoint we’re really interested in, and we need to do a better job of keeping this in mind.