# ROC

Hi folks,

I'm trying to compare different biomarkers' ability to detect a disease using ROC curves.

Does anyone know if there's a way to combine two different markers to calculate if the combination performs even better? (using mainly GraphPad5 and SPSS)

Thanks a lot in advance.

I'm trying to compare different biomarkers' ability to detect a disease using ROC curves.

Does anyone know if there's a way to combine two different markers to calculate if the combination performs even better? (using mainly GraphPad5 and SPSS)

Thanks a lot in advance.

## All Answers (9)

Mohammad Salehi-Marzijarani· Shiraz University of Medical Sciencesyou can combine information of biomarkers using regression. then use the results as a new test.

DeletedMohammad Salehi-Marzijarani· Shiraz University of Medical SciencesAlthough interaction effect and its contribution in the model can improve the prediction, but I think we need to ask this question that which terms( biomarkers ) have to enter the model and why? In other words, Do we use all the biomarkers in hand or subset of those terms? If we want to use a subset of terms, how we can do it?

Do We have to use statistical algorithms or do the selection based on theory?

Mahmoud ElbannaNorman Bravo· Independent ResearcherMatthew Marler· Carnegie Mellon UniversitySlightly better is MANOVA followed by canonical variates; retain the canonical variate values and choose cutoffs based on them. I say it is slightly better, but it depends on your design; Hastie, Tibshirani and Friedman (The Elements of Statistical Learning) make a case for logistic regression.

Norman Bravo· Independent ResearcherCan you help by adding an answer?