A statistic for comparing two correlated markers which are prognostic for time to an event

A proliferation of biologic markers has led to increased interest in methods for comparing the prognostic value of markers in predicting time to recurrence of or death from disease. This paper proposes a non-parametric test statistic for comparing two continuous markers when the outcome of interest is time to an event and the outcome is subject to right censoring. Results from Monte Carlo studies show that the new testing procedure is reliable for practical use. We present an example in which we use the statistic successfully in a colon cancer study to select between two potentially useful markers.