There is no direct way to compute the statistical significance of the difference between the ROC areas of two riskfactors. This program implements the Monte Carlo method to sample each riskfactor and return an estimation of the spread of the ROC area distributtion by sampling with replacement (i.e. it could also be called a
bootstrapping method since it is not done analytically, but rather emperically).
For cross-validation with Matlab's perfcurve, this program uses: function AROC=AROCforOptimization(observedoutcome,fittedoutcome)
for comparison. The results are quite close.
It save the Area ROC using methods in separate files for statistical test
for difference if needed.
A unit test script is provided. Please contact me if you have any thoughts.
Rex Cheung (2019). SCOPE: bootstrapping the ROC areas and errors in two ways (https://www.mathworks.com/matlabcentral/fileexchange/37366-scope-bootstrapping-the-roc-areas-and-errors-in-two-ways), MATLAB Central File Exchange. Retrieved .