Replies to post #109103 on Biotech Values

[iwfal]

Just some more color on calcium scans:

The ROC AUC for calcium scores is about 0.80 or a little under. Whereas the full Framingham ROC AUC is somewhere below 0.70. (Note that the AUC for calcium scores appears to be even better in high risk patients - in the high 0.80's.. Implying that it is lower in the lower risk patients - which should be no surprise since in the general population 1/2 the population has a score of zero and so cannot be used as predictive.)

That difference of 0.1 in AUC doesn't sound like much but it should be kept in mind that the AUC tops out at 1.0. And measuring predictive accuracy by a different measure makes the benefit clearer - the relative risk of the highest quartile compared to the lowest quartile for Calcium score is somewhere in the neighborhood of 10, but for the best individual test (LDL/HDL) the relative risk is around 4.


PS My guess is that other measurement techniques (e.g. CIMT) could be very substantially improved if they could get their act together. The calcium score community has collaborated on finding the most predictive ways of calculating scores, running really big trials (needed to sort among different possible predictors) ... in a way that the proponents of other techniques have not.

[genisi]

The CT scan is a more sensitive test than hs-CRP but in addition to the radiation issue, is more expensive and I believe it has a high false positive rate. Also, I remember reading that using both hs-CRP and Lp-PLA2 gives better diagnostic accuracy than either alone. So perhaps (you tell me if you know), this combined test might not be so far from the scan in terms of predicting of heart disease.