Intermediate markers as surrogate endpoints in cancer research

Because studies with surrogate cancer endpoints can be smaller, faster, and substantially less expensive than those with frank cancer outcomes, the use of surrogate endpoints is undeniably attractive. This attractiveness is likely to grow in coming years as the rapidly advancing discoveries in cell and molecular biology generate new therapies requiring testing and new markers that could plausibly serve as surrogates for cancer. Surrogate endpoint studies can certainly be suggestive. They continue to play a legitimate role in phase II studies, and they may give the right answers about intervention effects on or exposure associations with cancer. The problem is the uncertainty attached to most potential surrogates. Except for those few surrogates that are both necessary for and developmentally relatively close to cancer, the existence of plausible alternative pathways makes inferences about cancer from many surrogates problematic. Merely being on the causal pathway to cancer does not in itself constitute surrogate validity. It is the totality of causal connections that is critical. There is, unfortunately, a fairly extensive history of quite plausible surrogate markers giving the wrong answer about various chronic disease therapies. There is no reason to believe that cancer surrogacy is immune to such inferential difficulties. This article is, in part, an invitation, even a plea, for researchers to carry out the investigations necessary to evaluate potential surrogates, particularly surrogate-cancer studies and intervention or exposure-surrogate-cancer mediation analyses. Such studies are needed to generalize from surrogate endpoint findings to cancer. There is, however, an implicit and perhaps unavoidable irony here: the large, long, expensive studies required to evaluate potential surrogates fully are precisely the studies that surrogates were designed to replace. The exposure dependence alluded to earlier complicates matters further: establishing validity for a given surrogate for one intervention or exposure vis-à-vis cancer does not necessarily translate into validity for another intervention or exposure. One can enhance the inferential strength of surrogacy by using further "downstream" markers. Results of trials with CIN3 as an endpoint are arguably more persuasive than those from intervention studies with HPV infection endpoints. Similarly, one could consider only the advanced adenoma (> or = 1 cm, villous elements, or high-grade dysplasia) as the primary endpoint in adenoma recurrence trials. The inferential gain, however, comes with substantial costs: studies with CIN3 endpoints must be much larger than those with HPV infection endpoints; adenoma recurrence trials with sufficient rates of recurrence of advanced adenomas must be five or six times larger than trials with any recurrent adenomas as endpoints. A law emerges here: in using surrogate endpoints, inferential certainty is directly associated with study cost. In other words, one gets what one pays for. The problems inherent in using surrogate endpoints need not be regarded as a cause for pessimism in cancer research. If anything, the limitations of surrogacy are reminders of the complexity of cancer causation and affirm the continued importance of large clinical trials and observational epidemiologic studies with explicit cancer endpoints.