GS: "Before I fully understood what AUC really meant, I erroneously thought a .95 meant it was a good test in all respects. It says nothing about the sensitivity of the test but only that the positive hits were 95% accurate (specificity)."

GS,

You misunderstand AUC. A receiver operating characteristic (ROC) plots sensitivity against 1-specificity. The ROC curve gives you the tradeoff between sensitivity and specificity.

No real-world test is perfect. You would like perfect results: 100% sensitivity and 100% specificity. But you inevitably have to trade off sensitivity against specificity. You could decide you want 90% sensitivity (90% of cancers detected) in which case you might get e.g., 80% specificity (20% non-cancers are reported as cancer).

Alternatively, you might decide you want to boost the sensitivity to 95% (95% of cancers detected), but the consequence will be that specificity falls (e.g. to 70%, so 30% non-cancers are reported as cancer). These are make-believe numbers, simply to illustrate there is a tradeoff between sensitivity and specificity. Biocurex ultimately needs to pick a level of recaf that they think works best. This will give some combination of sensitivity and specificity.

The area under the curve (AUC) is exactly what is says: it is the area under the ROC curve. It is not the specificity or the sensitivity. Rather, it condenses the whole ROC curve to one number. Note that you cannot calculate the AUC given one particular combination of sens. and spec. In order to calculate the AUC you need the whole ROC curve.

AUC does not change as you vary the recaf level to find the best tradeoff between sensitivity and specificity. It is a single number that represents the overall performance of the test, independent of exactly what combination of sens. and spec. you choose. IMO, for a real-world test, AUCs above 0.9 are IMO very good, and AUCs above 0.95 are exceptional.

Bear in mind, however, that biocurex' reported AUCs are for small datasets and need to be confirmed by larger studies performed independently of biocurex. Also, doesn't matter how good your AUC is if your competitor (e.g., epca-2) has a significantly better one.

I think you are doing everyone a dis-service by claiming that bocx' reported results are bad, and are the reason for Abbott rejecting the test. In fact, IMO if bocx delivers on their reported AUCs in clinical trials shareholders should be extremely satisfied.

QB