There are several issues. First the Cox model is not for a binary outcome. It is for a time-to-event outcome whose status (event vs. censored) is binary. Second, split-sample validation does not work well with n < 20000 in the combined sample. Third, reclassification tables are not used to validate models; they are used to compare two models. Fourth, the rms package has several methods for truly validating Cox models. Frank Petergodsk wrote
Thank you very much to Prof. Harrell for the comment. I have fitted a Cox model on one data set and need to validate it on another dataset with a binary outcome. I can't find a way to make the reclassification (or the PredRisk) function in the PredictABEL package to accept my Cox model. I have used predictSurvProb to calculate predicted survival which is accepted by the ImproveProb function, but - as mentioned - not by the reclassification function in the PredictABEL package. Does anyone have a solution? Thanks, Peter Godsk
----- Frank Harrell Department of Biostatistics, Vanderbilt University -- View this message in context: http://r.789695.n4.nabble.com/NRI-reclassification-table-improveProb-Cox-tp4653768p4653878.html Sent from the R help mailing list archive at Nabble.com.