Announcement

Collapse
No announcement yet.
X
Collapse
  • Page of 1
  • Filter
  • Time
  • Show
Clear All
new posts
  • #1

    Problem with bootstrapping stcrreg

    Dear Statalist,

    I am working on a survival model (long dataset format). The objective is to run a competing-risk regression for the composite outcome (variable 'comp =1 2 3 4 5'), with 'comp==9' as a competing risk.
    Variable 'event' represents the label for timepoints of 'date'.
    Code:
    * Example generated by -dataex-. For more info, type help dataex
clear
input int no byte(age rdm) float date byte(event ptdm) float comp byte acr
1 61 0 18272  1 0 0 .
1 61 0 18303  2 0 0 .
1 61 0 18394 18 1 0 .
1 61 0 19845  7 1 0 0
1 61 0 20137  8 1 0 0
1 61 0 20223 15 1 0 0
1 61 0 20420 19 1 2 0
1 61 0 20487  9 1 2 0
1 61 0 20489 16 1 2 0
1 61 0 22462 23 1 2 0
1 61 0 22691 24 1 2 0
6 52 0 18324  1 0 0 .
6 52 0 18343  2 0 0 .
6 52 0 18508  3 0 0 .
6 52 0 21208 11 0 0 0
6 52 0 21572 12 0 0 0
6 52 0 21943 13 0 0 0
6 52 0 22262 23 0 0 0
6 52 0 23142 24 0 0 0
7 41 0 18336  1 0 0 .
end
format %td date
label values event event
label def event 1 "KT", modify
label def event 2 "1mo", modify
label def event 3 "6mo", modify
label def event 7 "4y", modify
label def event 8 "5y", modify
label def event 9 "6y", modify
label def event 11 "8y", modify
    I can run the final model as:
    Code:
     stset date , id( no ) failure( comp ==1/5) origin(event==1) scale(365.25)

Survival-time data settings

           ID variable: no
         Failure event: comp==1 2 3 4 5
Observed time interval: (date[_n-1], date]
     Exit on or before: failure
     Time for analysis: (time-origin)/365.25
                Origin: event==1

--------------------------------------------------------------------------
      7,328  total observations
        558  observations end on or before enter()
        296  observations begin on or after (first) failure
--------------------------------------------------------------------------
      6,474  observations remaining, representing
        549  subjects
         55  failures in single-failure-per-subject data
  3,342.489  total analysis time at risk and under observation
                                                At risk from t =         0
                                     Earliest observed entry t =         0
                                          Last observed exit t =  13.21287

. stcrreg  age rdm ptdm acr  , compete(comp== 9)

        Failure _d: comp==1 2 3 4 5
  Analysis time _t: (date-origin)/365.25
            Origin: event==1
       ID variable: no

Iteration 0:   log pseudolikelihood = -269.83536  
Iteration 1:   log pseudolikelihood = -269.80709  
Iteration 2:   log pseudolikelihood = -269.80708  

Competing-risks regression                        No. of obs      =      5,923
                                                  No. of subjects =        541
Failure events:  comp == 1 2 3 4 5                No. failed      =         54
Competing event: comp == 9                        No. competing   =         17
                                                  No. censored    =        470

                                                  Wald chi2(4)    =     115.67
Log pseudolikelihood = -269.80708                 Prob > chi2     =     0.0000

                                   (Std. err. adjusted for 541 clusters in no)
------------------------------------------------------------------------------
             |               Robust
          _t |        SHR   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         age |   1.038975   .0125617     3.16   0.002     1.014644    1.063889
         rdm |   6.603938    2.20809     5.65   0.000     3.429222    12.71775
        ptdm |   2.388427   1.035484     2.01   0.045     1.021128    5.586551
         acr |   6.566593   2.450044     5.04   0.000     3.160463    13.64362
------------------------------------------------------------------------------
    However, when I ran the bootstrap command, it returned this:

    Code:
    bootstrap, noisily reps(10) seed(1): stcrreg  age rdm ptdm acr  , compete(com
> p== 9)
bootstrap: First call to stcrreg with data as is:

. stcrreg age rdm ptdm acr, compete(comp== 9)

        Failure _d: comp==1 2 3 4 5
  Analysis time _t: (date-origin)/365.25
            Origin: event==1
       ID variable: no

Iteration 0:   log pseudolikelihood = -269.83536  
Iteration 1:   log pseudolikelihood = -269.80709  
Iteration 2:   log pseudolikelihood = -269.80708  

Competing-risks regression                        No. of obs      =      5,923
                                                  No. of subjects =        541
Failure events:  comp == 1 2 3 4 5                No. failed      =         54
Competing event: comp == 9                        No. competing   =         17
                                                  No. censored    =        470

                                                  Wald chi2(4)    =     115.67
Log pseudolikelihood = -269.80708                 Prob > chi2     =     0.0000

                                   (Std. err. adjusted for 541 clusters in no)
------------------------------------------------------------------------------
             |               Robust
          _t |        SHR   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         age |   1.038975   .0125617     3.16   0.002     1.014644    1.063889
         rdm |   6.603938    2.20809     5.65   0.000     3.429222    12.71775
        ptdm |   2.388427   1.035484     2.01   0.045     1.021128    5.586551
         acr |   6.566593   2.450044     5.04   0.000     3.160463    13.64362
------------------------------------------------------------------------------

Bootstrap replications (10)
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
insufficient observations to compute bootstrap standard errors
no results will be saved
r(2000);
    I noticed that if I use stcox, the bootstrap worked fine:
    Code:
     bootstrap,  reps(10) seed(1): stcox  age rdm ptdm acr
(running stcox on estimation sample)

Bootstrap replications (10)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
..........

Cox regression with Breslow method for ties

Bootstrap results

No. of subjects =        541                            Number of obs =  5,923
No. of failures =         54
Time at risk    = 3,178.0561
                                                        Wald chi2(4)  = 522.77
Log likelihood = -267.79661                             Prob > chi2   = 0.0000

------------------------------------------------------------------------------
             |   Observed   Bootstrap                         Normal-based
          _t | haz. ratio   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         age |   1.041612   .0148074     2.87   0.004      1.01299    1.071042
         rdm |   6.674006   1.013341    12.50   0.000     4.956159    8.987272
        ptdm |   2.357624   .7253323     2.79   0.005     1.290027    4.308737
         acr |   6.452919   2.833782     4.25   0.000     2.728684    15.26016
------------------------------------------------------------------------------
    Do you have any ideas how to make bootstrapping work with stcrreg? Or did I miss any important options required in order to make boostrapping with stcrreg work?

    Thank you.
    Tags: None
  • #2
    Even if your original data set contains only one failure per subject, when you do bootstrap resampling, you are sampling with replacement so it is possible, even likely, that some subjects will end up with more than one failure in the bootstrap samples. -stcrreg- does not accept data sets with multiple failure per person. -stcox- allows them, although the way your data was -stset-, -stcox- will only consider the first failure per subject.

    I think you can get around this by adding the -cluster(no)- and -idcluster(pick_a_name_for_a_new_variable)- options to your -bootstrap:- prefix. That will cause Stata to treat re-samplings of the same no's observation as if they were different people and should avoid the problem.

    Comment

    • #3
      Dear Clyde,

      Thank you so much for your suggestion.
      However, it still could not overcome the error we've got.

      Code:
      bootstrap, noisily cluster(no)  idcluster(newno) reps(10) seed(1): stcrreg  ag
> e rdm ptdm acr abmr  ca a1c crcl, compete(comp== 9)
bootstrap: First call to stcrreg with data as is:

. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9)

        Failure _d: comp==1 2 3 4 5
  Analysis time _t: (date-origin)/365.25
            Origin: event==1
       ID variable: no

Iteration 0:   log pseudolikelihood = -198.67777  
Iteration 1:   log pseudolikelihood = -198.65779  
Iteration 2:   log pseudolikelihood = -198.65779  

Competing-risks regression                        No. of obs      =      3,822
                                                  No. of subjects =        507
Failure events:  comp == 1 2 3 4 5                No. failed      =         45
Competing event: comp == 9                        No. competing   =         15
                                                  No. censored    =        447

                                                  Wald chi2(8)    =     137.13
Log pseudolikelihood = -198.65779                 Prob > chi2     =     0.0000

                                   (Std. err. adjusted for 507 clusters in no)
------------------------------------------------------------------------------
             |               Robust
          _t |        SHR   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         age |   1.047996   .0157745     3.11   0.002      1.01753    1.079374
         rdm |   6.080825   2.332637     4.71   0.000     2.867061    12.89698
        ptdm |   2.697827    1.27997     2.09   0.036     1.064549    6.836949
         acr |   3.004378   1.375557     2.40   0.016     1.224704    7.370178
        abmr |   3.381073   1.885483     2.18   0.029     1.133379    10.08635
          ca |   .5943886   .1404099    -2.20   0.028      .374106    .9443789
         a1c |   1.155084   .0941302     1.77   0.077     .9845719    1.355126
        crcl |   .9782986   .0056848    -3.78   0.000     .9672199    .9895043
------------------------------------------------------------------------------

Bootstrap replications (10)
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
. stcrreg age rdm ptdm acr abmr ca a1c crcl, compete(comp== 9) 
data with multiple failures per subject not supported by stcrreg
an error occurred when bootstrap executed stcrreg, posting missing values
insufficient observations to compute bootstrap standard errors
no results will be saved
r(2000);

      Comment

      • #4
        I'm afraid I don't know what to tell you now. I thought that would solve the problem. Perhaps somebody else has some insight here.

        Comment

        Previous Next
        Working...
        X