stintcox, tvc() with vs without lrphtest

[MaryEllen Mackesy-Amiti]

I ran a model with stintcox and tvc to test the proportional hazards assumption with lrphtest and it produced output, but when I left out lrphtest it fails with the message "coefficient estimates cannot be computed because matrix is close to singular or badly scaled". Why does including lrphtest allow the model to run but it fails without it? Using Stata MP 18 for Windows.

. stintcox i.bl_res_chicago , id(idnum) time(studytime) status(hcvresult) tvc(i.bl_res_chicago) nohr lrphtest
note: recurrent events detected; recurrent events ignored.
note: observations after events detected and ignored.
note: using adaptive step size to compute derivatives.

Performing EM optimization (showing every 100 iterations):

Fitting main model:
Iteration 0:   Log likelihood = -77.790501
Iteration 100: Log likelihood = -49.582331
Iteration 124: Log likelihood = -49.582156

Fitting full model:
Iteration 0:   Log likelihood = -49.582156
Iteration 30:  Log likelihood =  -43.84777

Computing standard errors: ...... done

Interval-censored Cox regression                  Number of obs      =     294
Baseline hazard: Reduced intervals                Number of subjects =      92
                                                          Uncensored =       0
ID variable: idnum                                     Left-censored =       7
Examination time: studytime                           Right-censored =      75
Status indicator: hcvresult                           Interval-cens. =      10

                                                  Wald chi2(2)       = 1623.94
Log likelihood = -43.84777                        Prob > chi2        =  0.0000

----------------------------------------------------------------------------------
                 |                 OPG
       studytime | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-----------------+----------------------------------------------------------------
main             |
1.bl_res_chicago |   82.66612   5.573796    14.83   0.000     71.74168    93.59056
-----------------+----------------------------------------------------------------
tvc              |
1.bl_res_chicago |   -3.67076   1.827981    -2.01   0.045    -7.253537   -.0879834
----------------------------------------------------------------------------------
Notes: Standard error estimates may be more variable for small datasets and
       datasets with low proportions of interval-censored observations.
       Variables in tvc equation interacted with _t.

LR test that [tvc] = 0: chi2(1) = 11.4688                 Prob > chi2 = 0.0007

.
. ** Wald test
. stintcox i.bl_res_chicago , id(idnum) time(studytime) status(hcvresult) tvc(i.bl_res_chicago) nohr
note: recurrent events detected; recurrent events ignored.
note: observations after events detected and ignored.
note: using adaptive step size to compute derivatives.

Performing EM optimization (showing every 100 iterations):
Iteration 0:   Log likelihood = -77.790501
coefficient estimates cannot be computed because matrix is close to singular or badly scaled
r(198);

end of do-file

r(198);