@f2harrell I agree that ROC curves makes things indirect. they are an alternative (and complicated) way of drawing risk distributions. nothing more, but imo nothing less. https://t.co/wO8n07pbMj https://t.co/FSBNKXFh4u
@f2harrell It's unfortunate that ROC got a threshold interpretation. The same risk distributions are also behind 2+ or 0 thresholds. ROC curve is nothing more than a transformation of risk distributions, nothing different
(am rewriting the article, but here is ref: https://t.co/wO8n07pbMj https://t.co/ycUCq5RbsQ
@GSCollins @f2harrell oh no! ROC plots show sample size much better than the CI of c-stat does. and skewness. and overlap of risk distributions. don't remove the plots but educate how to 'read' them. here's (again) one attempt to that: https://t.co/wO8n07pbMj
@DrMJoyner @stephensenn @f2harrell @VinayPrasad82 Allow me some promotion: if you want to know why the receiver operating curve is relevant, read this: https://t.co/wO8n07pbMj
ROC is just a different way of drawing risk distributions for patients and nonpatients, the essence of discrimination and prediction
Fast, open access, open for post-publication comments, copy-righted, and free. It's not indexed (pubmed, wos), which could be one reason to pay for formal open access publication, but are there more benefits? https://t.co/wO8n07pbMj https://t.co/1ynvSV2Vrd