This will seem very basic so my apologies.

Lets say I have 10 doctors who are inexperienced diagnosing cancer in 100 patients. 50 patients have cancer and the other 50 do not. The doctors provide a diagnosis for each patient which is a probability of cancer, summing to 100% i.e. cancer 90%, not cancer 10%.

I then construct ROCs for each doctor and calculate an AUC for each of them.

I then provide them with a computer assisted diagnosis tool and they repeat the exercise with the help of the computer.

I then repeat the analysis.

I want to show that the mean AUC has increased as a result of incorporating the computer assisted diagnosis tool into their evaluations.This to me is a comparison of AUCs between the two groups. Which test is most appropriate the show this? Would it be a comparison of the means on round 1 versus round 2, or test of paired data?

  • $\begingroup$ If you want to show that the AUC has increased and by how much (given that it is numeric and within 0 and 1) you can use a linear model. $\endgroup$ – user2974951 Oct 1 '18 at 6:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.