Stcomlist - stcompet : Difference between cif curve and table values

[Robin Schaffar]

Dear Statalisters,
I have issues with stcomlist and stcompet since I found differences between curves (from stcompet) and tables values (from stcomlist)
Here the code I am using to get the tables:

. stset dEv, fail(Ev==1) scale(365.25) origin(ddiag) exit(time mdy(12,31,2015))

     failure event:  Ev == 1
obs. time interval:  (origin, dEv]
 exit on or before:  time mdy(12,31,2015)
    t for analysis:  (time-origin)/365.25
            origin:  time ddiag

------------------------------------------------------------------------------
        776  total observations
          2  observations end on or before enter()
------------------------------------------------------------------------------
        774  observations remaining, representing
         67  failures in single-record/single-failure data
  6,305.235  total analysis time at risk and under observation
                                                at risk from t =         0
                                     earliest observed entry t =         0
                                          last observed exit t =  20.98836

. stcomlist,compet1(2) compet2(3) compet3(4) at(1 5 10 15) saving("Tables\CR_AllPatients.dta",replace)

            failure:  Ev == 1
 competing failures:  Ev == 2 3 4

    Time       CIF         SE     [95% Conf. Int.]
--------------------------------------------------
       1    0.0104     0.0037     0.0050    0.0198
       5    0.0541     0.0085     0.0392    0.0724
      10    0.0928     0.0120     0.0711    0.1180
      15    0.1327     0.0181     0.0996    0.1705


            failure:  Ev == 2
 competing failures:  Ev == 1 3 4

    Time       CIF         SE     [95% Conf. Int.]
--------------------------------------------------
       1    0.0078     0.0032     0.0033    0.0163
       5    0.0640     0.0092     0.0477    0.0836
      10    0.1079     0.0129     0.0843    0.1347
      15    0.1357     0.0170     0.1044    0.1710


            failure:  Ev == 3
 competing failures:  Ev == 1 2 4

    Time       CIF         SE     [95% Conf. Int.]
--------------------------------------------------
       1    0.0091     0.0034     0.0041    0.0179
       5    0.0175     0.0048     0.0098    0.0290
      10    0.0449     0.0089     0.0297    0.0647
      15    0.0635     0.0130     0.0413    0.0921


            failure:  Ev == 4
 competing failures:  Ev == 1 2 3

    Time       CIF         SE     [95% Conf. Int.]
--------------------------------------------------
       1    0.0026     0.0018     0.0005    0.0089
       5    0.0068     0.0030     0.0026    0.0151
      10    0.0128     0.0046     0.0060    0.0245
      15    0.0256     0.0091     0.0119    0.0484

Then I derive my graph and add lines corresponding to the values found in the table above.

stset dEv, fail(Ev==1) scale(365.25) origin(ddiag) exit(time mdy(12,31,2015))
cap drop CIFEv1
stcompet CIFEv1=ci,compet1(2) compet2(3) compet3(4)
cap drop _tEv1
rename _t _tEv1

stset dEv, fail(Ev==2) scale(365.25) origin(ddiag) exit(time mdy(12,31,2015))
cap drop CIFEv2
stcompet CIFEv2=ci,compet1(1) compet2(3) compet3(4)
cap drop _tEv2
rename _t _tEv2

stset dEv, fail(Ev==3) scale(365.25) origin(ddiag) exit(time mdy(12,31,2015))
cap drop CIFEv3
stcompet CIFEv3=ci,compet1(2) compet2(1) compet3(4)
cap drop _tEv3
rename _t _tEv3

stset dEv, fail(Ev==4) scale(365.25) origin(ddiag) exit(time mdy(12,31,2015))
cap drop CIFEv4
stcompet CIFEv4=ci,compet1(2) compet2(3) compet3(1)
cap drop _tEv4
rename _t _tEv4

twoway (line CIFEv1 _tEv1 if Ev==1&_tEv1<=30,sort lcolor(gs0) connect(step)) ///
(line CIFEv2 _tEv2 if Ev==2&_tEv2<=30,sort lcolor(gs3) lp(dash) connect(step)) ///
(line CIFEv3 _tEv3 if Ev==3&_tEv3<=30,sort lcolor(gs6) lp(dot) connect(step)) ///
(line CIFEv4 _tEv4 if Ev==4&_tEv4<=30,sort lcolor(gs9) connect(step)), ///
xline(10) yline(0.0449 0.1079 0.0928 0.0128) ///
ytitle("Cumulative incidence") xtitle("Time since diagnosis (years)") ///
xscale(r(0 20)) ///
title("Competing risk analysis.") ///
legend(rows(2) symxsize(5) subtitle("Competing events") ///
label(1 "Local relapse") label(2 "Distant relapse") label(3 "Second BC") label(4 "Deaths")) graphregion(fcolor(white)) ///
saving("Figures\3_Competingk_Risk_AllPatients",replace)

We can see that there is a discordance since the curves do not pass through the points. With some other data, the gap is even more important.
Would someone have an explanation?

Many thanks!

[Ange Lee]

Hi- were you able to find a resolution to this discrepancy? I am having a similar issue, but am using stcomlist and stcurve.

. stset duration_ac, fail(vte_event=1)

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

------------------------------------------------------------------------------
111 total observations
0 exclusions
------------------------------------------------------------------------------
111 observations remaining, representing
6 failures in single-record/single-failure data
1,018.74 total analysis time at risk and under observation
at risk from t = 0
earliest observed entry t = 0
last observed exit t = 56.40604

. stcrreg doacorlmwh, compete(vte_event=2)

failure _d: vte_event == 1
analysis time _t: duration_ac

Iteration 0: log pseudolikelihood = -25.355546
Iteration 1: log pseudolikelihood = -25.355026
Iteration 2: log pseudolikelihood = -25.355026

Competing-risks regression No. of obs = 111
No. of subjects = 111
Failure event : vte_event == 1 No. failed = 6
Competing event: vte_event == 2 No. competing = 13
No. censored = 92

Wald chi2(1) = 0.00
Log pseudolikelihood = -25.355026 Prob > chi2 = 0.9895

------------------------------------------------------------------------------
| Robust
_t | SHR Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
doacorlmwh | .9895018 .7964889 -0.01 0.990 .2042918 4.792721
------------------------------------------------------------------------------

. stcomlist, compet1(2) by(doacorlmwh) at(6)

failure: vte_event == 1
competing failures: vte_event == 2

Time CIF SE [95% Conf. Int.]
--------------------------------------------------
doacorlmwh=0
6 0.0471 0.0325 0.0086 0.1407
doacorlmwh=DOAC
6 0.0563 0.0316 0.0148 0.1405

. stcurve, cif at1(doacorlmwh=0) at2(doacorlmwh=1) range (0 6) title("Cumulative Incidence of VTE Events")

The curves for LMWH and DOAC do not intersect at 5.6% and 4.7% respectively that stcomlist expresses. Instead it seems like its displaying cumulative incidence w/o competing risk (see below, 5.5% and 5.4% respectively)

. cs vte_event doacorlmwh

| Exposed Unexposed | Total
-----------------+------------------------+------------
Cases | 3 3 | 6
Noncases | 52 53 | 105
-----------------+------------------------+------------
Total | 55 56 | 111
| |
Risk | .0545455 .0535714 | .0540541
| |
| Point estimate | [95% Conf. Interval]
|------------------------+------------------------
Risk difference | .000974 | -.0831682 .0851162
Risk ratio | 1.018182 | .2147181 4.828165
Attr. frac. ex. | .0178571 | -3.65727 .792882
Attr. frac. pop | .0089286 |
+-------------------------------------------------
1-sided Fisher's exact P = 0.6520
2-sided Fisher's exact P = 1.0000

Would appreciate any insight to this issue! Thanks