st: Cox model with a segmented time-dependent covariate

[Marta Garcia-Granero <mgarciagranero@gmail.com>]

I am trying to replicate some SPSS commands in Stata. This data were 
modified from David Collett (2003) Modelling Survival Data in Medical 
Research (Chapman & Hall/CRC Texts in Statistical Science), and show a 
hypothetical clinical trial on 12 patients. I want to test if treatment 
is significant, adjusted for age and ln(bilirrubin). Ln(bilirrubin) were 
measured 3 times: at baseline (lbr0); approximately at 3 months -with 
the exact day recorded in t1 (lbr1); and approximately at 6 months - 
with exact day recorded in t2 (lbr2).

+----------------------------------------------------------------------------+
     | patient   time   status   treatment   age   lbr0    t1 lbr1    
t2   lbr2 |
|----------------------------------------------------------------------------|
  1. |       1    281     Died     Placebo    46    3.2    47 3.8   
184    4.9 |
  2. |       2    604    Alive     Placebo    57    3.1    94 2.9   
187    3.1 |
  3. |       3    457     Died     Placebo    56    2.2    61 2.8    
97    2.9 |
  4. |       4    384     Died     Placebo    65    3.9    92 4.7   
194    4.9 |
  5. |       5    341    Alive     Placebo    73    2.8    87 2.6   
192    2.9 |
|----------------------------------------------------------------------------|
  6. |       6    842     Died     Placebo    64    2.4    94 2.3   
197    2.8 |
  7. |       7   1514     Died   Treatment    69    2.4    74 2.9   
202      3 |
  8. |       8    182    Alive   Treatment    62    2.4    90 2.5   
182    2.9 |
  9. |       9   1121     Died   Treatment    71    2.5   101 2.5   
410    2.7 |
 10. |      10   1411    Alive   Treatment    69    2.3   182 2.2   
847    2.8 |
|----------------------------------------------------------------------------|
 11. |      11    814     Died   Treatment    77    3.8   167 3.9   
498    4.3 |
 12. |      12   1071     Died   Treatment    58    3.1   108 2.8   
187    3.4 |
+----------------------------------------------------------------------------+

With SPSS, I had to define a time varying covariate (T_COV_) and add it 
to the model:

TIME PROGRAM.
COMPUTE T_COV_ = (T_<t1)*lbr0 + (T_>=t1&T_<t2)*lbr1 + (T_>=t2)*lbr2.
COXREG   time
  /STATUS=status(1)
  /CONTRAST (treatment)=Indicator
  /METHOD=ENTER T_COV_ treatment age.

In SPSS, T_ is the same as Stata's _t.

This is the output:
                     coef      se         HR    ----95%CI------ 
LR        de     sig
T_COV_       4.533    2.488    93.05    0.71    12192.54 10.865    1    
0.001
Treatment    4.749    2.860    115.50    0.42    31421.66 5.814    1    
0.016
Age            -0.078     0.094    0.93     0.77    1.11    0.808 1    0.369

With Stata, I suspect I have to use, somehow, -stsplit- and -stjoin- , 
but I can't figure how (even though I have used help stsplit, and read 
the whole PDF on stcox).

Any hints?

Thanks,
Marta Garcia-Granero

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/