I've seen it detailed that a concordance of < 0.5 can be viewed as a measure of better than random predictive power, similarily to the AUC. When a classifier obtains an AUC of 0.2, you know that it's predicting the negativily labelled class higher than the positive one, and an accurate prediction can be made by simply reversing every future prediction.

How would the concordance be used in a similar fashion? I've got a boosted Cox model with concordances of < 0.1, which as far as I'm aware means it's got the same diagnostic power as a model with a concordance >= 0.9.

When predicting survival curves would I simply subtract the probability of survival from 1 to reflect this?


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.