Researchers of the study noted that using RR as an end point reduced study time by a mean of just over 1.5 years, while using PFS as an end point reduced mean study time by just 11 months. “The benefits of using surrogate end points are that they are faster to measure, and hence, help trials to be completed early, thereby leading to faster approval and access. However, as [the] recent JAMA Internal Medicine paper showed — the use of surrogate end points, in fact, saves study time by only around a year at most.”
Other, frequently used surrogate end points not evaluated in this analysis include disease-free survival (DFS), cancer-specific mortality, and recurrence-free survival (RFS). These are compared by authors Kemp and Prasad in BMC Medicine with “hard” end points in oncology, OS and quality of life (QoL), which they refer to as being “intrinsically valuable to patients.”5
Replacing these hard end points with any surrogate should then be seen as an attempt to use another metric, which may be found more quickly than a hard end point like OS, to predict whether use of the drug in question will meet a hard end point in the future.
According to Dr Gyawali, the biggest concerns are, however, “approving and exposing patients to potentially toxic drugs that may be inefficacious — or at worst, even harmful — on the basis of surrogates. And [that] we as a society [are] paying huge amounts for access to these drugs that are of no benefit.”
It’s unclear, then, what the PFS or RR threshold should be to say, with certainty, that a novel therapy is likely to improve OS.
Research published in the BMJ Open in 2013 may give some insight. The authors of this paper found, based on analysis of PFS and OS data from 2331 patients with advanced non-small cell lung cancer (NSCLC) treated with chemotherapy, only “modest support for considering PFS as an acceptable surrogate for OS in patients with advanced NSCLC.”6 The authors noted, in particular, that only those treatments with a “major” impact on PFS — corresponding to a risk reduction of at least 50% — were likely to also improve OS.