topic Re: How do I do a sample size calculation based on AUC in SAS? in Statistical Procedures

How do I do a sample size calculation based on AUC in SAS?

Tpham — Tue, 26 Apr 2016 02:43:05 GMT

So I am trying to replicate this statement in SAS (basically to come up with the 300 patients part), but I can't seem to figure out how:

60% of the patients will be considered adherent, 300 patients will allow an AUC of 0.80 to be estimated with an accuracy (i.e. half width of the 95% confidence interval) of 0.049

Re: How do I do a sample size calculation based on AUC in SAS?

SteveDenham — Tue, 26 Apr 2016 13:57:46 GMT

Some things to consider. AUC is likely lognormally distributed. Iget something like this from:

proc power;
   OneSampleMeans test = t Dist = LogNormal
      Alpha = 0.05
      Sides = 2
      Mean = 0.8
      NullMean = 0.79
      CV = 0.06125
      Power = 0.8
      NTotal = .
   ;

Where I got the CV as .049/0.8.

This may just be coincidental, though.

Steve Denham

Re: How do I do a sample size calculation based on AUC in SAS?

acozzilepri — Sun, 14 Jun 2020 10:18:24 GMT

This simple macro below gives you the figures using the formula in Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982 Apr;143(1):29-36.

N=302

%macro power_auc(a=, k=, j=, out=);

data a;

call streaminit(11);
do i=1 to 1;

AUC=&a.;
Q1=(AUC / (2 - AUC));
Q2=(2*AUC**2)/(1+AUC);
N1=&k.;
N2=&j.;
SE=sqrt(((AUC*(1-AUC))+(N1-1)*(Q1 - AUC**2)+(N2-1)*(Q2-AUC**2))/(N1*N2));

output;
end;

proc print;
run;

%mend;

%power_auc(a=.8, j=151, k=151, out=a8_n302);