When may k-medians be preferable to k-means?

A. When every feature is a class label B. When numeric data contain substantial outliers C. When cluster centers must be arithmetic means D. When frequent itemsets are needed