proper sample size and number of estimation for comparing two predictors' AUCs

#1
Hello,

I am trying to do classification of patients by using different predictors. I want to do bootstrapping for a number of times so that I can come up with a set of Area Under Curve's (AUCs) of ROC curve for each predictor and then compare their effectiveness in classification by performing a two-sample MWU test on their AUCs.

I have a question on how to set the sampling size and number of estimations of AUCs properly. By the following command in R,

aucs = replicate(A,mean(sample(pos.scores,B,replace=T) > sample(neg.scores,B,replace=T))),

I come up with A number of AUCs by using B number of samples from the positive class and B number of samples from the negative class. But then how can I set A and B properly so that I can compare the effectiveness of classification by two different predictors? I have around 50 samples in the positive class and around 380 in the negative class.

Thank you for your help in advance!
 
#2
I hope I am communicating my question well.

If I increase B, the number of samples from each class, then the standard deviation of aucs will decrease. If I increase A, the number of replicates, then the p-value will be smaller when I perform two-sample MWU test on the two set of AUCs came up with by two different predictors. But how can I set A and B properly based on the number of samples in the positive class and the number of samples in the negative class?