Here is the abstract. The actual paper is behind a paywall.
"ROC analysis of [the test statistic], for cancers plus precancerous/suspect conditions vs. controls, cancer vs. precancerous/suspect conditions plus controls, and cancer vs. controls, gave areas under the curve of 0.87, 0.89, and 0.93, respectively (P<0.001). Optimization allowed test sensitivity or specificity to approach 100% with acceptable complementary measures."
The ROC curve has area under it of 1 for a perfect classifier and 0.5 for wild guessing. This is a more useful measurement than the p-value. (E.g. if I look at height vs sex for humans, it won't take too big a sample to get a great p-value for there being a difference, yet classifying people as male/female depending on whether they exceed some height threshold is a very poor diagnostic system.) I don't have much of a feel for how good ROC area of about 0.9 is for a medical test. I'd guess it is good enough to be useful, but you'd not want to rely on that test alone.