Statistically significant AUC improvement

[Jennifer Xu]

Hi,

I’ve been using the logit command to combine biomarkers, then postestimation ROC analysis to generate a curve and AUC. There was significant improvement in AUC’s from just using single biomarkers. My question is how do I statistically compare the AUC value from single biomarkers with my combinations to see if there is a statistically significiant difference in AUC’s.

Thanks! Much appreciated answers.

[Rich Goldstein]

the p-value for the added coefficient is the p-value for the change in AUC; see Pepe, et al., (2013), "Testing for improvement in prediction model performance," Statistics in Medicine, 32: 1467-1482, or Seshan, et al. (2013), "Comparing ROC curves derived from regression models," Statistics in Medicine, 32: 1483-1493

[Jennifer Xu]

Would you mind explaining what you mean by the "p-value of the added coefficient"? Thanks.

[Rich Goldstein]

did you read either of the articles I cited - this is explained there