ROC curve - AUC related question

[health_economist2904]

Hi guys!

I am trying to determine the best cut off value for a single diagnostic test by performing the ROC curve analysis. The diagnostic test has AUC of 0.99, so the test has a high discriminating potential, which is great, and I found the best cut off value by choosing the optimal cut off value with sens and spec that suits well to his intended clinical utility. However, I got a question from my peer how come that I don't have different AUC values for different cut off points, showing me the right cut off value. Based on my understanding, there cannot be different AUC for different cut off values for a single diagnostic test - am I right ? If there is a way to calculate the AUC in different cut off values, could you please share with me some recommended papers/ any literature? Thanks!

 

[Dason]

Do you know how a ROC curve is generated?

 

[health_economist2904]

Hi Dason!
Thank you very much for taking the time to look at my question. To be honest, I thought I did, but now, I'm just uncertain. I did understand that the AUC is a summary of overall performance of the test, accounting for all cut off values and their corresponding true negative and false positive rates.

Could it be possible to shed some light on this? Thanks!

 

[hlsmith]

There is a true prevalence for the outcome of interest. Can you post the AUC fpormula?

 

[health_economist2904]

Yes - there is a true prevalence for the outcome of interest, and it is indeed a slightly different than the one I have in my study sample. I was using SPSS for the analysis, is there a way to see where is this formula, so I can share it with you?

 

[health_economist2904]

Just a small note - I believe AUC was calculated by trapezoidal approach. Are you thinking about partial AUC maybe?