Reason for getting NA in a single category of a categorical variable in a Cox model

Question

I have a Cox model with time dependent variables as follows, where ibd is a categorical variable with 3 levels. However, I am not sure why there is a NA for regression coefficient in the last level of "ibd". When I reorder the levels, it is still the third level that is NA.

Call:
coxph(formula = Surv(adm_start, adm_end, adm_outcome == 1) ~ 
    adm + ibd + MontrealA + decade + prevadm, data = adm_cox, 
    cluster = ID)

  n= ..., number of events= 169 

                 coef exp(coef)  se(coef) robust se      z Pr(>|z|)    
adm1         2.492000 12.085419  0.732874  0.744455  3.347 0.000816 ***
ibdnon-IBD  -1.500172  0.223092  0.180067  0.185680 -8.079 6.51e-16 ***
ibdIBD             NA        NA  0.000000  0.000000     NA       NA    
MontrealAA2 -0.359413  0.698086  0.231540  0.239495 -1.501 0.133430    
MontrealAA3 -0.545712  0.579429  0.269156  0.274694 -1.987 0.046965 *  
decade1980+ -0.816816  0.441836  0.198816  0.212263 -3.848 0.000119 ***
decade1990+ -1.136621  0.320901  0.223769  0.233220 -4.874 1.10e-06 ***
decade2000+ -1.724058  0.178341  0.300258  0.325774 -5.292 1.21e-07 ***
decade2010+ -1.997544  0.135668  0.545032  0.563526 -3.545 0.000393 ***
prevadm     0.067405  1.069728  0.003267  0.003433 19.636  < 2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

On cross tabulation data is sparse on the first level for "ibd" predictor variable, but I still don't understand why the third level would be missing? Also, I am particularly interested in the coefficient of the third level, is there a way I could recode this variable so that it gives me a coefficient?

> table(adm_cox$ibd,adm_cox$adm_outcome,useNA="ifany")
         
               0      1      2
  None     14422     <3    269
  non-IBD 111869     57   4208
  IBD      47842    110    646

Answer 1

If you change the type of variable 'ibd' to numeric not as factor, then the output won't contain NA. And, only ibdIBD will shown (no ibdnon-IBD). Got the similar problem with a binary variable. Not sure if that's a bug from the print-codes. All the coefficients are correct and the model can call cox.zph as well.