Landmark Survival Analysis - STS Graph not accounting for landmark time of entry

[Winta Mehtsun]

Hello Statalist Community,

I am performing a landmark survival analysis to investigate whether the addition of chemotherapy to surgery for older patients (age>=80) with pancreatic cancer is associated with improved overall survival. My dependent variable, overall survival, was calculated as time from 90 days after diagnosis until death or last known follow-up, I used this landmark time point of 90 days from diagnosis until death or last know follow-up and excluded patients who died within 90 days of diagnosis to account for immortal time bias. My exposure was receipt of chemotherapy after surgery.

I have two variables that define follow up time (months):

1. Real follow-up time (t=postopfu) *
2. landmarked time (postopfu90=t-3 mo)

I tried to use the origin and entry options on stset so that I don’t have to redefine my follow up time to account for the landmark entry (Approach A vs. Approach B in code below), and so I can plot the survival curves with real follow up time. But the issue I am having is that the survival curves I have generated already show separation of the survival curves at the landmark point of entry when I graph them, where as when I redefine my follow up time in option A the curves start together as you would expect. Is there something I missing in my code with the way sts graph pulls data to graph - it appears the origin and entry terms do not apply when graphing sts? Is there a way around this?

** Follow up time**
summarize postopfu, detail
generate postopfu90 = postopfu - 3
summarize postopfu90, detail
** A: Overall survival - landmark time accounted for **
stset postopfu90, failure(laststatus==1)
stci
by anychemo: stci, median
sts test anychemo, logrank
stset postopfu90, failure(laststatus==1)
sts graph, by(anychemo) tmin(3) tmax(60) xtitle(Time (months)) xlabel(0 12 24 36 48 60) risktable(, size(3) order(1 "No Chemotherapy" 2 "Chemotherapy")) legend(label(1 "No Chemotherapy") label (2 "Chemotherapy") title(, size(1))) title("") xline(3, lcolor(blue) lpattern(dash))

** B: Overall survival - real follow up time **
stset postopfu, failure(laststatus==1) enter(time 3) origin(time 3)
stci
by anychemo: stci, median
sts test anychemo, logrank
sts graph, by(anychemo) tmin(3) tmax(60) xtitle(Time (months)) xlabel(0 12 24 36 48 60) risktable(, size(3) order(1 "No Chemotherapy" 2 "Chemotherapy")) legend(label(1 "No Chemotherapy") label (2 "Chemotherapy") title(, size(1))) title("") xline(3, lcolor(blue) lpattern(dash))

Graph A:




Graph B:

[Paul Dickman]

Can you show the output?

[Winta Mehtsun]

. summarize postopfu, detail 

                          postopfu
-------------------------------------------------------------
      Percentiles      Smallest
 1%         3.52           3.12
 5%         4.53           3.12
10%         5.75           3.12       Obs               1,968
25%          9.3           3.22       Sum of Wgt.       1,968

50%         17.1                      Mean           24.06058
                        Largest       Std. Dev.      21.48896
75%       31.375         132.37
90%        51.52         135.69       Variance       461.7753
95%        66.99          139.1       Skewness       2.024202
99%       109.73         142.23       Kurtosis       8.090972

. generate postopfu90 = postopfu - 3

. summarize postopfu90, detail 

                         postopfu90
-------------------------------------------------------------
      Percentiles      Smallest
 1%          .52       .1199999
 5%         1.53       .1199999
10%         2.75       .1199999       Obs               1,968
25%          6.3            .22       Sum of Wgt.       1,968

50%         14.1                      Mean           21.06058
                        Largest       Std. Dev.      21.48896
75%       28.375         129.37
90%        48.52         132.69       Variance       461.7753
95%        63.99          136.1       Skewness       2.024202
99%       106.73         139.23       Kurtosis       8.090972

. 
. ** Overall survival - landmark time accounted for **
. stset postopfu90, failure(laststatus==1) 

     failure event:  laststatus == 1
obs. time interval:  (0, postopfu90]
 exit on or before:  failure

------------------------------------------------------------------------------
      1,968  total observations
          0  exclusions
------------------------------------------------------------------------------
      1,968  observations remaining, representing
      1,655  failures in single-record/single-failure data
  41,447.22  total analysis time at risk and under observation
                                                at risk from t =         0
                                     earliest observed entry t =         0
                                          last observed exit t =    139.23

. stci

         failure _d:  laststatus == 1
   analysis time _t:  postopfu90

             | Number of 
             |  subjects         50%      Std. Err.    [95% Conf. Interval]
-------------+-------------------------------------------------------------
       Total |      1968       14.87      .4628914        13.72       15.6

. by anychemo: stci, median

-------------------------------------------------------------------------------------------------------------------------------
-> anychemo = 0

         failure _d:  laststatus == 1
   analysis time _t:  postopfu90

             | Number of 
             |  subjects         50%      Std. Err.    [95% Conf. Interval]
-------------+-------------------------------------------------------------
       Total |      1097        12.7      .6159029        11.69      14.02

-------------------------------------------------------------------------------------------------------------------------------
-> anychemo = 1

         failure _d:  laststatus == 1
   analysis time _t:  postopfu90

             | Number of 
             |  subjects         50%      Std. Err.    [95% Conf. Interval]
-------------+-------------------------------------------------------------
       Total |       871       16.48      .7465297        15.43      18.45

. 
. ** KS Curve - landmark time accounted for **
. sts test anychemo, logrank 

         failure _d:  laststatus == 1
   analysis time _t:  postopfu90


Log-rank test for equality of survivor functions

         |   Events         Events
anychemo |  observed       expected
---------+-------------------------
0        |       937         844.06
1        |       718         810.94
---------+-------------------------
Total    |      1655        1655.00

               chi2(1) =      20.94
               Pr>chi2 =     0.0000

. sts graph, by(anychemo) tmin(0) tmax(60) xtitle(Time (months)) xlabel(0 12 24 36 48 60) risktable(, size(3) order(1 "No Chemo
> therapy" 2 "Chemotherapy")) legend(label(1 "No Chemotherapy") label (2 "Chemotherapy") title(, size(1))) title("") xline(0, l
> color(blue) lpattern(dash)) 

         failure _d:  laststatus == 1
   analysis time _t:  postopfu90

. 
end of do-file

. drop _st _d _t _t0

. do "/var/folders/k6/6scj6cy11w95g95zr82lg6p00000gp/T//SD64501.000000"

. ** Overall survival - real follow up time **
. stset postopfu, failure(laststatus==1) enter(time 3) origin(time 3)

     failure event:  laststatus == 1
obs. time interval:  (origin, postopfu]
 enter on or after:  time 3
 exit on or before:  failure
    t for analysis:  (time-origin)
            origin:  time 3

------------------------------------------------------------------------------
      1,968  total observations
          0  exclusions
------------------------------------------------------------------------------
      1,968  observations remaining, representing
      1,655  failures in single-record/single-failure data
  41,447.22  total analysis time at risk and under observation
                                                at risk from t =         0
                                     earliest observed entry t =         0
                                          last observed exit t =    139.23

. stci

         failure _d:  laststatus == 1
   analysis time _t:  (postopfu-origin)
             origin:  time 3
  enter on or after:  time 3

             | Number of 
             |  subjects         50%      Std. Err.    [95% Conf. Interval]
-------------+-------------------------------------------------------------
       Total |      1968       14.87      .4628914        13.72       15.6

. by anychemo: stci, median

----------------------------------------------------------------------------------------------------------------
-> anychemo = 0

         failure _d:  laststatus == 1
   analysis time _t:  (postopfu-origin)
             origin:  time 3
  enter on or after:  time 3

             | Number of 
             |  subjects         50%      Std. Err.    [95% Conf. Interval]
-------------+-------------------------------------------------------------
       Total |      1097        12.7      .6159029        11.69      14.02

----------------------------------------------------------------------------------------------------------------
-> anychemo = 1

         failure _d:  laststatus == 1
   analysis time _t:  (postopfu-origin)
             origin:  time 3
  enter on or after:  time 3

             | Number of 
             |  subjects         50%      Std. Err.    [95% Conf. Interval]
-------------+-------------------------------------------------------------
       Total |       871       16.48      .7465297        15.43      18.45

. 
. ** KS Curve - real follow up time **
. sts test anychemo, logrank 

         failure _d:  laststatus == 1
   analysis time _t:  (postopfu-origin)
             origin:  time 3
  enter on or after:  time 3


Log-rank test for equality of survivor functions

         |   Events         Events
anychemo |  observed       expected
---------+-------------------------
0        |       937         844.06
1        |       718         810.94
---------+-------------------------
Total    |      1655        1655.00

               chi2(1) =      20.94
               Pr>chi2 =     0.0000

. sts graph, by(anychemo) tmin(3) tmax(60) xtitle(Time (months)) xlabel(0 12 24 36 48 60) risktable(, size(3) or
> der(1 "No Chemotherapy" 2 "Chemotherapy")) legend(label(1 "No Chemotherapy") label (2 "Chemotherapy") title(, 
> size(1))) title("") xline(3, lcolor(blue) lpattern(dash)) 

         failure _d:  laststatus == 1
   analysis time _t:  (postopfu-origin)
             origin:  time 3
  enter on or after:  time 3

. graph save "/Users/winta/Documents/Research /Wang/NCDB/KS_landmark_overall_cohort_2021", replace  
(file /Users/winta/Documents/Research /Wang/NCDB/KS_landmark_overall_cohort_2021.gph saved)

[Paul Dickman]

It appears to me as the two analytic approaches give identical results, as they should. I can't see the two graphs in your original post. I tried to reproduce the problem (see code below) but couldn't (the two graphs A and B are identical).

Can you provide us with some data, or reproduce the problem using the brcancer data?

An obersvation: In #1 you have tmin(3) and xline(3) in both graphs. In #3 you have tmin(0) and xline(0) in one graph but tmin(3) and xline(3) in the other. Therefore, you should get different graphs in #3 since you have specified different graphs. I don't know your research question, but it's not obvious why you would want to use tmin(3) and xline(3) given how you have set up the data.

webuse brcancer, clear
generate rectime300=rectime-300

// A:
stset rectime300, f(censrec==1)
sts graph, by(hormon) tmin(300) tmax(2700) risktable xlabel(0(300)2700) ///
           xline(300, lcolor(blue) lpattern(dash)) name(A, replace)
           
// B:
stset rectime, f(censrec==1) enter(time 300) origin(time 300)
sts graph, by(hormon) tmin(300) tmax(2700) risktable xlabel(0(300)2700) ///
           xline(300, lcolor(blue) lpattern(dash)) name(B, replace)