Problems with using biomarkers as surrogate end points for cancer: a cautionary tale.

Abstract

Investigations employing surrogate cancer end points are especially attractive because they may be smaller, shorter, and cheaper than comparable studies with explicit cancer outcomes. For many potential surrogate end points--epithelial cell proliferation will be taken as an example--inferences are problematic because of the existence of alternative causal pathways to cancer that bypass the surrogate end point. Evaluating potential surrogates requires information on the following three questions: (1) What is the relation of the surrogate end point to cancer? (2) What is the relation of the intervention (or exposure) to the surrogate? (3) To what extent does the surrogate end point mediate the relation between intervention (exposure) and cancer? Data for these questions may derive from animal experiments, human metabolic studies, observational epidemiologic investigations (including ecologic studies), and randomized trials. Inferences to cancer from such downstream markers as colorectal adenomatous polyps and persistent human papillomavirus infection of the cervix are strong, though not absolutely unassailable. For all but these very-close-to-cancer markers, considerable caution is warranted in extrapolating from surrogate effects or associations to cancer.