To view how well the model discriminates those with and those without the event, the user needs the area under the receiver operating characteristic curve (AUC). However, the options for the users are limited when the INMODEL= option is specified in the PROC LOGISTIC statement, with AUC only available when the user specifies both the SCORE statement and the FITSTAT option. But a big deficit in the process is the lack of confidence interval of AUC, which is automatically computed and displayed in the ROC statement but is not available when the INMODEL= option is specified in the PROC LOGISTIC statement. Therefore, when the user is validating a model, he/she cannot obtain it directly.
What is more, the formula of the standard error of the AUC is not reported in anywhere of SAS Help, so the user is unable to calculate the confidence interval on his/her own.
Therefore, I suggest that SAS (1) automatically compute the standard error and confidence interval of the AUC when the INMODEL= option is specified in the PROC LOGISTIC statement; (2) include the formula of the standard error of AUC in SAS Help.
