Concordance index for Competing risks (PROC PHREG)

[ks17]

I am fitting a Fine and Gray regression model to competing risks data. I am trying to compute the time-dependent AUC for this model, but the "plots=auc rocoptions(method=ipcw(cl seed=1234))" option does not work. I get the note "Neither concordance nor ROC analysis is performed when the EVENTCODE option in the MODEL statement is specified". However, I need that option to specify fitting a Fine and Gray regression. The data is coded as having Status 0 if censored, 1 if experience the event of interest, 2 if they experience the competing event. Here is my SAS code: 

proc phreg data=train plots=auc rocoptions(method=ipcw(cl seed=1234));
class X1 X2;
model Time*Status(0)=X1 X2/eventcode=1;
run;

Any suggestions for how I can obtain the concordance index from a Fine and Gray regression model? Thanks in advance! 

[k-san]

Did you ever find a solution to this?  We are having the same issue.

[ks17]

I emailed one of the authors of "Analyzing Survival Data with Competing Risks using SAS Software" in February 2018. They were quick to respond and said that the AUC and time dependent ROC method (using the IPCW) could be carried over to the Fine-Gray model, but that it was not yet available in PHREG and that it would be a future project. 

For now, I am using the FGR function in the riskRegression package in R (https://cran.r-project.org/web/packages/riskRegression/riskRegression.pdf) for the Fine-Gray model, and the timeROC package in R to get the time-dependent AUC. 

[Bowen2006]

I used cause specific model, and created AUC at different time point, it works.

Try this:

proc phreg data=train plots=auc rocoptions(method=ipcw(cl seed=1234));
class X1 X2;
model Time*Status(0, 2)=X1 X2;
run;

[FP_Italy]

It is not a Fine and Gray model, so it works

[Season]

There is a paper on the concordance statistic for competing risks models: Concordance for prognostic models with competing risks | Biostatistics | Oxford Academic. The authors themselves provided R code in the supplemental material of the paper: Supplementary material to “Concordance for prognostic models with competing risks”. Another person has also developed an R package for this calculation based upon the paper mentioned above: R: Concordance index in the Presence of Competing Risks. An issue to note is that independence of observations was assumed in the paper. Therefore, this method is not applicable in the analysis of clustered data with competing risks.