C-index not good for predictivity

[Abhaya Indrayan]

We have tried to show in the attached paper https://pubmed.ncbi.nlm.nih.gov/38668827/ that the area under the ROC curve (C-index) is good for assessing discrimination and classification of the KNOWN outcomes but is not adequate for assessing the performance for prediction of the UNKNOWN outcomes. It is erroneously used for assessing the performance of a prediction model. We have also proposed that for predictivity the best cutoff is not the Youden index but is the P-index where the (PPV + NPV) is maximum.

We would appreciate any feedback. Thanks.

~Abhaya Indrayan

--

Dr Abhaya Indrayan, MSc,MS,PhD(OhioState),FSMS,FAMS,FRSS,FASc

Personal website: http://indrayan.weebly.com