Statistical validation of intermediate markers of precancer for use as endpoints in chemoprevention trials.

Using an intermediate marker of precancer as an endpoint for evaluating agents that may prevent cancer involves a presumption that the modification of the marker will be accompanied by a modification of cancer incidence. This presumption can hold only if the marker is on or very closely linked to a causal pathway. Epidemiologists have discussed the nature of evidence required to infer causal relationships, and we briefly survey their work. Studies relating exposure (E) to marker (M) provide only indirect evidence for causality. Those relating marker (M) to disease (D) are more relevant. We propose a new validation criterion based on an analysis of the three-way relationship of exposure (E), marker (M) and disease (D). We discuss the level of evidence required for using intermediate markers as endpoints for Phase II and Phase III trials, and propose very stringent criteria for Phase III trials. For Phase II trials, we propose less stringent criteria, but still recommend that the marker (M) should have been shown to have a strong association with disease (D).