ROC curve (receiver operating characteristics) is a powerful tool for evaluation of diagnostic accuracy. In its most simplest form (empirical ROC curve), true positive rates (TPF, sensitivity) are drawn above false positive rate (FPF, 1-specificity) using every measured value as a cut-off.
ROC curve for evaluation of diagnostic accuracy
Main issue is not the calculation of the curve, but the planning of the referring diagnostic study on the one hand, and its interpretation on the other hand. Especially biases due to selection and golden standard could lead to severe deviation from reality.
ROC curve for determination of cut-off
ROC curve helps to understand optimal usage of a diagnostic test. Cut-Off can be choosen due to clinical requirements (e.g. high specificity).
Alternative approach to ROC analysis: DAC analysis
DAC analysis (Keller et al. (2005), Clin Chem 51, 532-539) is an alternative approach when diagnostic accuracy of high correltaing tests should be compared.
Links for ROC curves
--> ROC curve MS Excel tool
--> statistical background and application of ROC curves (written in german)
--> ACOMED statistics: analysis of ROC curves
--> Addstat: CRO specialized on diagnostic studies