log rank test for adjusted survival curve

[suganthiny kumarasamy]

Dear listers,

i created two survival curve adjusted by age
how can I compare this two curve. I am not sure I could use “sts test”
to compare two adjusted curve below is my command , curve is wider but
i got not significant , please advice me how can i compare adjusted curve


gen tpa_age=tpa*Age
sts test tpa_age
chi2(30) = 24.01
Pr>chi2 = 0.7714


xi:stcox Ageyears if tpa==0
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Age | .990738 .0357294 -0.26 0.796 .9231273 1.063301


xi:stcox Ageyears if tpa==1
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Age | 1.037272 .0244087 1.56 0.120 .9905185 1.086233

xi:stcox i.tpa Ageyears
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_Itpa_1 | .8745963 .2259029 -0.52 0.604 .527166 1.451001
Ageyears | 1.030515 .0202899 1.53 0.127 .9915047 1.07106




curve attach.

Appreciate your help.
Thank you

[Joseph Coveney]

For some reason, that graph doesn't seem as if it came from that fitted model. Try running the following code and show the output in a code block and attach the figure.

stset
stcox i.tpa c.AgeYears
stcurve , survival at1(tpa=1) at2(tpa=0)
graph export for_statalist.png

[Carlo Lazzaro]

Suganthiny:
like Joseph, I find difficult to link the graph you posted to the model you ran.
However, you can perform the -logrank- test with a hand-made interaction between age and drug:

use http://www.stata-press.com/data/r14/drugtr.dta
stcox  c.age##i.drug
g age_drug=age*drug
stcox  c.age##i.drug /// the result with the -fvvarlist- - made interaction and the hand-made one are the same for both -stcox- regressions
stcox  age drug age_drug
sts graph, by( age_drug ) /// the graph is difficult to read and probably unuseful to report
sts test age_drug , logrank

[Marcos Almeida]

Hello Sungathiny,

It seems your log-rank test has been done for the interaction term, when I guess you wished to perform the test for the group variable.

You may see that df = 30.

Below, we have a significant log-rank test for the group variable. Afterwards, we created an interaction term and perform a (non-significant) log-rank test.

. use "http://www.stata-press.com/data/r14/stan3.dta", clear
(Heart transplant data)

. sts test surgery

         failure _d:  died
   analysis time _t:  t1
                 id:  id


Log-rank test for equality of survivor functions

        |   Events         Events
surgery |  observed       expected
--------+-------------------------
0       |        69          60.34
1       |         6          14.66
--------+-------------------------
Total   |        75          75.00

              chi2(1) =       6.59
              Pr>chi2 =     0.0103

. gen surg_age = surgery*age

. tab surg_age

   surg_age |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |        148       86.05       86.05
         35 |          2        1.16       87.21
         36 |          2        1.16       88.37
         43 |          2        1.16       89.53
         45 |          4        2.33       91.86
         46 |          2        1.16       93.02
         47 |          2        1.16       94.19
         48 |          6        3.49       97.67
         52 |          4        2.33      100.00
------------+-----------------------------------
      Total |        172      100.00

. sts test surg_age

         failure _d:  died
   analysis time _t:  t1
                 id:  id


Log-rank test for equality of survivor functions

         |   Events         Events
surg_age |  observed       expected
---------+-------------------------
0        |        69          60.34
35       |         0           0.32
36       |         0           1.66
43       |         1           0.81
45       |         1           3.12
46       |         0           1.12
47       |         1           1.12
48       |         2           4.55
52       |         1           1.95
---------+-------------------------
Total    |        75          75.00

               chi2(8) =       8.09
               Pr>chi2 =     0.4250

Hopefully that helps!

Best,

Marcos

[suganthiny kumarasamy]

Thanks you so much Dr.Joseph, Dr.Carlo and Dr.Marcos your valuable comments.

this code result ican link with the stcox result
stcox i.tpa c.AgeYears
stcurve , survival at1(tpa=1) at2(tpa=0)



i have also difficult to link previous graph and the result i got it from
stcox all the model give non-significant when i this
graph command (previous graph)
sts graph, by(tpa) adj(Ageyears)
the curve separated widely
what does this graph tells from this command and when do we use this command



Thanks soo much

[Marcos Almeida]

It seems your graph conveys differences in survival according to the group (1 or 2), whereas your log-rank test was NOT done so as to account for differences between 2 treatment groups (please see the examples in #4), but for several levels of interaction between treatment group and age.

Had you "tested" the same way, for example:

.sts test tpa

I bet the graph and the output will go on the same verge!

Best,

Marcos

[suganthiny kumarasamy]

Thank you so much Dr.Marcos