GraphPad Statistics Guide

Comparing ROC curves

Comparing ROC curves

Previous topic Next topic No expanding text in this topic  

Comparing ROC curves

Previous topic Next topic JavaScript is required for expanding text JavaScript is required for the print function Mail us feedback on this topic!  

Prism does not compare ROC curves. It is, however, quite easy to manually compare two ROC curves created with data from two different (unpaired) sets of patients and controls.

1.Separately use Prism to create two ROC curves by separately analyzing your two data sets.

2.For each ROC curve, note the area under the curve and standard error (SE) of the area.

3.Combine these results using this equation:

4.If you investigated many pairs of methods with indistinguishable ROC curves, you would expect the distribution of z to be centered at zero with a standard deviation of 1.0. To calculate a two-tail P value, therefore, use the following Microsoft Excel function:

=2*(1-NORMSDIST(z))

The method described above is appropriate when you compare two ROC curves with data collected from different subjects. A different method is needed to compare ROC curves when both laboratory tests were evaluated in the same group of patients and controls. To account for the correlation between areas under your two curves, use the method described by Hanley and McNeil (1).

 

1. Hanley, J.A., and McNeil, B. J. (1983). Radiology 148:839-843. Accounting for the correlation leads to a larger z value and, thus, a smaller P value.