How to calculate concordance for Cox proportional hazards model for recurrent events

[SeaMoon_168]

I want to calculate the concordance for a Cox proportional hazards model with the Andersen-Gill extension for recurrent events. However, I got the error message below:

NOTE: Neither concordance nor ROC analysis is performed when the COVS option in the PROC PHREG
statement is specified.

I found some similar topics from the forum but none of them have been resolved, instead, they advised using R to solve this issue. I wonder if any newly updates on this issue using SAS are available. Many thanks.

[quickbluefish]

It would be helpful to see your code.

[SeaMoon_168]

Thank you for your response. The data and code can be found from https://documentation.sas.com/doc/en/statug/15.2/statug_phreg_examples10.htm

data Bladder;
   keep ID TStart TStop Status Trt Number Size Visit;
   retain ID TStart 0;
   array tt[4] T1-T4;
   infile datalines missover;
   input Trt Time Number Size T1-T4;
   ID + 1;
   TStart=0;
   do Visit=1 to 4;
      if tt[Visit] = . then do;
         TStop=Time;
         Status=0;
      end;
      else do;
         TStop=tt[Visit];
         Status=1;
      end;
      output;
      TStart=TStop;
   end;
   if (TStart < Time) then delete;
   datalines;
1       0       1     1
1       1       1     3
1       4       2     1
1       7       1     1
1       10      5     1
1       10      4     1   6
1       14      1     1
1       18      1     1
1       18      1     3   5
1       18      1     1   12  16
1       23      3     3
1       23      1     3   10  15
1       23      1     1   3   16  23
1       23      3     1   3   9   21
1       24      2     3   7   10  16  24
1       25      1     1   3   15  25
1       26      1     2
1       26      8     1   1
1       26      1     4   2   26
1       28      1     2   25
1       29      1     4
1       29      1     2
1       29      4     1
1       30      1     6   28  30
1       30      1     5   2   17  22
1       30      2     1   3   6   8   12
1       31      1     3   12  15  24
1       32      1     2
1       34      2     1
1       36      2     1
1       36      3     1   29
1       37      1     2
1       40      4     1   9   17  22  24
1       40      5     1   16  19  23  29
1       41      1     2
1       43      1     1   3
1       43      2     6   6
1       44      2     1   3   6   9
1       45      1     1   9   11  20  26
1       48      1     1   18
1       49      1     3
1       51      3     1   35
1       53      1     7   17
1       53      3     1   3   15  46  51
1       59      1     1
1       61      3     2   2   15  24  30
1       64      1     3   5   14  19  27
1       64      2     3   2   8   12  13
2       1       1     3
2       1       1     1
2       5       8     1   5
2       9       1     2
2       10      1     1
2       13      1     1
2       14      2     6   3
2       17      5     3   1   3   5   7
2       18      5     1
2       18      1     3   17
2       19      5     1   2
2       21      1     1   17  19
2       22      1     1
2       25      1     3
2       25      1     5
2       25      1     1
2       26      1     1   6   12  13
2       27      1     1   6
2       29      2     1   2
2       36      8     3   26  35
2       38      1     1
2       39      1     1   22  23  27  32
2       39      6     1   4   16  23  27
2       40      3     1   24  26  29  40
2       41      3     2
2       41      1     1
2       43      1     1   1   27
2       44      1     1
2       44      6     1   2   20  23  27
2       45      1     2
2       46      1     4   2
2       46      1     4
2       49      3     3
2       50      1     1
2       50      4     1   4   24  47
2       54      3     4
2       54      2     1   38
2       59      1     3
;

title 'Intensity Model and Proportional Means Model';
proc phreg data=Bladder concordance covm covs(aggregate);
   model (TStart, TStop) * Status(0) = Trt Number Size;
   id id;
   where TStart < TStop;
run;

[quickbluefish]

I suspect it's something to do with the fact that when you specify the COVS option, it's calculating a robust SE, and conversely, when you calculate concordance stats, it seems that that option is calculating its own SE and the two aren't compatible.  I imagine some R person has worked out whatever crazy, nearly impossible to verify method for aligning the two methods, but the people at SAS have avoided this.  You could of course run PHREG twice, once for concordance and once with COVS specified (using the confidence intervals from the version with COVS specified), but I imagine that's not ideal.  

I only play a statistician on TV (took one look at the documentation for the concordance option and immediately closed that page), but I have a fair amount of PHREG experience, including with longitudinal data.  Hopefully some actual statisticians will reply.  Good luck.  

[Ksharp]

According to the LOG

NOTE: Neither concordance nor ROC analysis is performed when the COVS option in the PROC PHREG statement is specified.

I think you need to get rid of option "covs(aggregate)".

But after removing this option ,you would get this NOTE.

NOTE: Neither concordance nor ROC analysis is performed with the START/STOP specification of time.

a.k.a  you can't perform this test by statements like:

 model (TStart, TStop)

therefore, you need to change your data structure .