Evaluating cut-off values with time-dependent roc curves for Ki67

Abstract

Receiver operating characteristic (ROC) curve is one of the most active research areas in medical statistics. While classic ROC curve analysis is a well-established method to study the accuracies of clinical markers, it may be suboptimal for analyzing outcomes over time, such as prognosis. When the outcome variable of interest is an event that occurs not concurrent with the tests, time-dependent ROC curves should be used to accommodate this time lag.

A commonly used prognostic factor in breast cancer is proliferation rate of tumor cells. Ki67 monoclonal antibody is one of the agents that can be used to evaluate proliferation rate. The goal of this study is to evaluate the cut-off values of Ki67 as a diagnostic marker in breast cancer and to show the difference in terms of diagnostic accuracy using the classical ROC analysis and time-dependent ROC analysis for this marker.

In classical ROC analysis, no significant evidence was found that the area under the curve (AUC) was different from the null hypothesis value 0.5 for both Ki67 in lymph nodes and Ki67 in tumor cells. But in time-dependent ROC analysis that accommodates the event time outcomes, significant tests were found. Using bootstrapped variances, AUCs were significantly larger than 0.5 for 37-136(th) months and for lymph nodes AUCs were significantly larger than 0.5 for the 23-136(th) months.