Is there a sound way to infer a ROC curve (a (smoothed?) approximation of the real ROC curve) from a given AUC? The authors did not supply a figure of the ROC curve, only the AUC and corresponding confidence intervals.
Vasilis,
It is not possible as several different curves may have the same area under the curve. Why are you interested in curves? did not they provide cut-off points analysis?
Hi Vasilis,
I will direct you to this link:
gim.unmc.edu/dxtests/roc3.htm
Also you can read evidence-based-diagnosis (Thomas B. Newman; Micheal A. Kohn) (Chapter 4 - Multilevel continuous tests test - ROC curves, LRs for multilevel tests and optimal cut-off between positive and negative for a multilevel test).
Dear Diego,
I understand this issue. I am looking for a sensible way to get pairs of sensitivity and specificity as the curve is not supplied nor are the cutoff points. The results will be pooled with results of other studies for the determination of diagnostic accuracy of an index test. Of course there would be some loss of information.
Dear Oyekola,
Thanks for the suggestions but unfortunately they do not address my question.
Vasilis,
Did you try to contact the authors?
Evidence synthesis is an interesting area, but it is frustrating when people do not publish all the requiere data. Particularly, when it is explained by selective publication bias
- Badges
- Science method