st: RE: Test of proportional-hazards assumption using estat phtest after stcox

["Kieran McCaul" <Kieran.McCaul@uwa.edu.au>]

...

When I'm testing a PH model, I would rarely rely solely on the results
from -phtest-.  Basically the number of events occurring within each
category of a variable are going to determine the power of each
individual test and also how problematic any apparent deviation from
proportionality is likely to be. 

I usually have a look at the Schoenfeld residuals with a lowess smoother
run over them.  If there is no proportionality problem, I should get a
flat line. Any deviation away from that indicates a potential problem
with lack of proportionality and when it's occurring during the course
of follow-up.  Based on what I see in this situation, I might run some
sensitivity analyses to get an idea of how influential the problem might
be.

I note that you only have three age-groups.  If these are broad
groupings, then it is possible that there is significant heterogeneity
in risk within the age-groups and this could cause some proportionality
problems if follow-up is long.  You could try increasing the number of
age-groups or modelling age as a continuous variable, preferably using
restricted cubic splines.

Also I usually -stplit- my data on birthday so I can allow age to
reflect each participants accrued age during follow-up.  That way I'm
modelling age-at-event rather than age-at-recruitment.  The latter is
really just defining a birth cohort.


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Weichle,
Thomas
Sent: Friday, 21 May 2010 1:09 AM
To: statalist@hsphsun2.harvard.edu
Subject: st: Test of proportional-hazards assumption using estat phtest
after stcox

Hi Statalisters,
I'm wondering if there is a valid proportional hazards test for a
categorical variable with > 2 levels in the case where not all of the
levels reject the null hypothesis.  For instance, _Iagegrp_2 is rejected
but _Iagegrp_3 is not rejected.  Similarly, not all levels of the
_Idivision are rejected.  I'm thinking along the lines of a global test
for all of the levels of a particular categorical variable with > 2
levels kind of like a likelihood ratio test does.


estat phtest, rank detail

      Test of proportional-hazards assumption

      Time:  Rank(t)
      ----------------------------------------------------------------
                  |       rho            chi2       df       Prob>chi2
      ------------+---------------------------------------------------
      _Iagegrp_2  |     -0.12386         7.42        1         0.0064
      _Iagegrp_3  |     -0.04730         1.10        1         0.2950
      black       |     -0.02018         0.20        1         0.6567
      married     |      0.05330         1.42        1         0.2334
      _Ichrlson_~1|     -0.03836         0.70        1         0.4024
      _Ichrlson_~2|     -0.10741         5.62        1         0.0178
      _Ichrlson_~3|     -0.16425        13.24        1         0.0003
      chemo_dx_6~h|     -0.01827         0.16        1         0.6857
      surgery     |      0.10080         5.08        1         0.0242
      outpt_even~4|      0.11063         6.08        1         0.0137
      va_rept2    |      0.09753         4.51        1         0.0338
      dual~c_36mth|      0.09944         4.93        1         0.0265
      hs_edu10    |      0.04712         1.16        1         0.2818
      _Idivision_1|      0.06742         2.24        1         0.1344
      _Idivision_2|      0.00182         0.00        1         0.9686
      _Idivision_3|      0.09970         4.72        1         0.0299
      _Idivision_4|      0.04497         0.99        1         0.3204
      _Idivision_5|     -0.01216         0.07        1         0.7887
      _Idivision_7|      0.08346         3.32        1         0.0682
      _Idivision_8|      0.09293         4.20        1         0.0405
      _Idivision_9|      0.06551         2.12        1         0.1450
      ------------+---------------------------------------------------
      global test |                     59.16       21         0.0000
      ----------------------------------------------------------------

Tom Weichle
Math Statistician
Center for Management of Complex Chronic Care (CMC3)
Hines VA Hospital, Bldg 1, C202
708-202-8387 ext. 24261
Thomas.Weichle@va.gov 

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

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