Pencina et al. respond to 'The incremental value of new markers' and 'Clinically relevant measures? A note of caution'.

摘要

We thank Drs. Cook, Kerr, Bansal, and Pepe for then-careful review of our work and insightful critiques. Several of the points they raise require further highlighting, discussion, or rebuttal.In her commentary, Dr. Cook presents 2 interesting examples that shed additional light on some of the properties of the 3 measures of interest (1). On the basis of the developments presented by us (2), as well as those presented by Kerr et al. (3), the increments between the measures of interest must go in the same direction. Cook's example with diabetes shows that this need not be the case for binary variables. She explains the reason for this anomaly: "The problem is that counterintuitively, there are more cases among people without diabetes than among those with because nondiabetic participants comprise the majority of the cohort" (1, p. 488) and concludes that "[t]he new model may look worse because more are moving in the wrong direction, but the correct changes are larger and the incorrect changes are much smaller" (1, p. 488).