0

I have 4 different parameters (HRTT, MRWT, HRWT and RNFLT) and am plotting sensitivity versus specificity curves for each. I want to:

  1. Find the partial area under the ROC curve (pAUC) for each parameter (for specificities in the range 90-100%) and the confidence interval around this pAUC for each parameter.
  2. Compare the pAUCs to evaluate which pAUC is statistically highest.

For 1., I calculated the pAUC estimate for HRTT:

auc(rocHRTT, partial.auc=c(100, 90), partial.auc.focus="spec",percent=TRUE, partial.auc.correct=TRUE, ci=TRUE)

I then calculated the CI around this by:

ci.auc(rocHRTT, partial.auc=c(100, 90), partial.auc.focus="spec",percent=TRUE, partial.auc.correct=TRUE, ci=TRUE)

I then repeated this for MRWT, HRTW and RNFLT.

For some reason, the estimate of pAUC often lies outside the CI for pAUC and I can't figure out why. For instance, MRWT the pAUC is 71%, but the CI is 77-91%!

For 2., I know that the function roc.test can be used to compare 2 pAUCs, but how can I compare 4 areas to evaluate which is highest?

Calimo
  • 7,510
  • 4
  • 39
  • 61
Riz Malik
  • 23
  • 5
  • It would be better if your question included a minimal, sufficient [reproducible example](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example). That will make it much easier to help you. – MrFlick Dec 03 '14 at 21:56
  • Bootstrap has some assumptions, in particular it's going to behave funny with less than 10 observations. Difficult to know if that's your case. – Calimo Dec 04 '14 at 10:02
  • Also if you have two questions, then please ask two separate questions (see your point 2). – Calimo Dec 04 '14 at 10:03
  • Verify if `partial.auc=c(100, 90)` is the problem. Do the analysis whit total area. – Robert Dec 04 '14 at 10:08

0 Answers0