For some reason I want to know the Coef. of each variators in Cox proportional hazard model

[Tom Hsiung]

By default the output is hazard ratio. According to the logic behind Cox proportional hazard model, the point estimation of hazard ratio equals exp(b_i), where b_i is the predicted coef. I just want to find the point estimation and confidence interval for b_i. I tried the 'help stcox' syntax but no luck.

 stcox i.ttestforweightstatus i.ageforttest i.gendercode i.af i.hypertension i.chf i.hypoalbumi
> nemiaforttest i.akiforttest i.potentialamiodaroneddi i.ttestindication i.ttesttargetinr i.afan
> dwarfarinhistory i.non11ornon1non1 i.nonaaornonanona

         failure _d:  firsttherapeuticinrreached
   analysis time _t:  timetofirsttherapeuticinrd

Iteration 0:   log likelihood = -609.64264
Iteration 1:   log likelihood = -585.42388
Iteration 2:   log likelihood = -582.50792
Iteration 3:   log likelihood = -582.47598
Iteration 4:   log likelihood = -582.47597
Refining estimates:
Iteration 0:   log likelihood = -582.47597

Cox regression -- Breslow method for ties

No. of subjects =          174                  Number of obs    =         174
No. of failures =          135
Time at risk    =         1066
                                                LR chi2(14)      =       54.33
Log likelihood  =   -582.47597                  Prob > chi2      =      0.0000

-------------------------------------------------------------------------------------------
                       _t | Haz. Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------------------+----------------------------------------------------------------
   1.ttestforweightstatus |   1.020393   .2929401     0.07   0.944     .5812992    1.791163
            1.ageforttest |   1.498247   .3518103     1.72   0.085     .9456026    2.373875
             1.gendercode |   1.390171   .2653018     1.73   0.084     .9563653    2.020749
                     1.af |   .5818911   .1519922    -2.07   0.038     .3487422    .9709098
           1.hypertension |   .7258801   .1602377    -1.45   0.147     .4709367    1.118838
                    1.chf |   1.670382   .4451392     1.93   0.054     .9907856    2.816126
1.hypoalbuminemiaforttest |   2.158257   .6137253     2.71   0.007     1.236101    3.768358
            1.akiforttest |   1.268841   .3458445     0.87   0.382     .7436999    2.164796
 1.potentialamiodaroneddi |   .7580945   .1894779    -1.11   0.268     .4644878    1.237293
        1.ttestindication |   1.543941   .4843595     1.38   0.166     .8348241    2.855397
         1.ttesttargetinr |   1.465928   .5007634     1.12   0.263     .7504831    2.863416
   1.afandwarfarinhistory |   1.260178   .3985813     0.73   0.465     .6779624    2.342384
        1.non11ornon1non1 |   2.250942   .6506658     2.81   0.005     1.277359    3.966577
        1.nonaaornonanona |   .2831169   .0924991    -3.86   0.000     .1492326    .5371155
-------------------------------------------------------------------------------------------

[Carlo Lazzaro]

Tom:
you may want to try something along the following lines:

. use http://www.stata-press.com/data/r14/drugtr.dta
(Patient Survival in Drug Trial)

. stcox i.drug age, nohr

         failure _d:  died
   analysis time _t:  studytime

Iteration 0:   log likelihood = -99.911448
Iteration 1:   log likelihood = -83.551879
Iteration 2:   log likelihood = -83.324009
Iteration 3:   log likelihood = -83.323546
Refining estimates:
Iteration 0:   log likelihood = -83.323546

Cox regression -- Breslow method for ties

No. of subjects =           48                  Number of obs    =          48
No. of failures =           31
Time at risk    =          744
                                                LR chi2(2)       =       33.18
Log likelihood  =   -83.323546                  Prob > chi2      =      0.0000

------------------------------------------------------------------------------
          _t |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      1.drug |  -2.254965   .4548338    -4.96   0.000    -3.146423   -1.363507
         age |   .1136186   .0372848     3.05   0.002     .0405416    .1866955
------------------------------------------------------------------------------
.

[David Fisher]

Hi Tom,
The nohr option is what you want. It's definitely listed under help stcox!
Best wishes,
David.

[Marcos Almeida]

Indeed, the information is available in the example 1 of the Stata Manual, page 75, as disclosed here:

Once an stcox model has been fit, typing stcox without arguments redisplays the previous results. Options that affect the display, such as nohr—which requests that coefficients rather than hazard ratios be displayed—can be specified upon estimation or when results are redisplayed

[Tom Hsiung]

Thank you, all. And is there a syntax similar as 'estat vif' for multiple linear regress to detect interactions?

[Carlo Lazzaro]

Tom:
no, there's no a built-in Stata command similar to -estat vif- to detect quasi-extreme multicollinearity after -stcox-.
That said, you can use -estat vce, corr-:

. use http://www.stata-press.com/data/r14/drugtr.dta
(Patient Survival in Drug Trial)

. stcox i.drug age, nohr

         failure _d:  died
   analysis time _t:  studytime

Iteration 0:   log likelihood = -99.911448
Iteration 1:   log likelihood = -83.551879
Iteration 2:   log likelihood = -83.324009
Iteration 3:   log likelihood = -83.323546
Refining estimates:
Iteration 0:   log likelihood = -83.323546

Cox regression -- Breslow method for ties

No. of subjects =           48                  Number of obs    =          48
No. of failures =           31
Time at risk    =          744
                                                LR chi2(2)       =       33.18
Log likelihood  =   -83.323546                  Prob > chi2      =      0.0000

------------------------------------------------------------------------------
          _t |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      1.drug |  -2.254965   .4548338    -4.96   0.000    -3.146423   -1.363507
         age |   .1136186   .0372848     3.05   0.002     .0405416    .1866955
------------------------------------------------------------------------------

. estat vce, corr

Correlation matrix of coefficients of cox model

             |        1.         
        e(V) |     drug       age
-------------+--------------------
      1.drug |   1.0000          
         age |  -0.2281    1.0000

.

[Marcos Almeida]

Carlo gave insightful reply.

Just to avoid misunderstanding: as Carlo politely pointed out, neither -estat vif - nor - estat vce - will "detect interactions". There is no such thing. What is more, there would be no purpose on this, since it is up to us to select - or not - interaction terms. On such grounds, provided our command is correct, Stata will, well, "detect" that we included an interaction in the model so as to perform the estimations accordingly. But, even so, we know this for sure, well before "quick-as-lightning" Stata realizes.

[Tom Hsiung]

What does my output mean? @Cario

. estat vce, corr

Correlation matrix of coefficients of cox model

             |        1.        1.        1.        1.        1.        1.        1.        1.
        e(V) | ttestf~s  agefor~t  gender~e        af  hypert~n       chf  hypoal~t  akifor~t 
-------------+--------------------------------------------------------------------------------
1.ttestfor~s |   1.0000                                                                       
1.agefortt~t |   0.1665    1.0000                                                             
1.gendercode |   0.1467    0.1662    1.0000                                                   
        1.af |   0.0074   -0.2024   -0.2268    1.0000                                         
1.hyperten~n |   0.1587   -0.0474    0.1496   -0.0330    1.0000                               
       1.chf |  -0.1532   -0.1860    0.0415   -0.2974   -0.0074    1.0000                     
1.hypoalbu~t |  -0.0144    0.0198    0.0653   -0.1002    0.1000   -0.0650    1.0000           
1.akifortt~t |  -0.0720   -0.2801   -0.0226    0.1061   -0.1400   -0.0469   -0.1514    1.0000 
1.poten~eddi |   0.1392   -0.0869    0.0635   -0.0581   -0.0752    0.0835   -0.0135   -0.2020 
1.ttestind~n |  -0.0168   -0.4054   -0.2049   -0.0370   -0.0226    0.0246   -0.0273    0.0774 
1.ttesttar~r |   0.0828    0.0028   -0.0233   -0.1914    0.1527   -0.0551    0.0803   -0.2055 
1.afandwar~y |   0.0203   -0.0461    0.0016   -0.4261   -0.0059    0.0833    0.0247   -0.0247 
1.non11orn~1 |   0.1176    0.1104    0.0184   -0.1686   -0.0939   -0.1379   -0.0128   -0.0741 
1.nonaaorn~a |  -0.0444   -0.0803   -0.0533   -0.0590    0.0222   -0.0344    0.0291    0.0588 

             |        1.        1.        1.        1.        1.        1.
        e(V) | pot~eddi  ttesti~n  ttestt~r  afandw~y  non11o~1  nonaao~a 
-------------+------------------------------------------------------------
1.poten~eddi |   1.0000                                                   
1.ttestind~n |   0.0101    1.0000                                         
1.ttesttar~r |  -0.0134    0.6508    1.0000                               
1.afandwar~y |  -0.0191    0.0975    0.0576    1.0000                     
1.non11orn~1 |   0.1822    0.1101    0.0301    0.0696    1.0000           
1.nonaaorn~a |   0.0517    0.0041   -0.0484   -0.0514   -0.0273    1.0000

[Carlo Lazzaro]

Tom:
at a quick glance, it does not seem that your regression specification suffers from quasi-extreme multicollinearity.