optimism in flexible parametric survival analysis by stpm2, please

[Kullathorn Thephamongkhol]

Dear Professor Lambert, Professor Royston and other statalists,

I am doing prognostic modelling and I would like to do internal validation and get Dxy, R2, slope, D (as example below in R from function validate from rms package)
So I can use new c statistics, shrunk model and new intercept.


index.orig training test optimism index.corrected n
Dxy 0.4393 0.4676 0.4066 0.0610 0.3783 200
R2 0.0614 0.0742 0.0529 0.0213 0.0400 200
Slope 1.0000 1.0000 0.8613 0.1387 0.8613 200
D 0.0518 0.0632 0.0444 0.0189 0.0329 200
U -0.0034 -0.0036 0.0037 -0.0073 0.0038 200
Q 0.0552 0.0668 0.0406 0.0262 0.0291 200
g 0.8751 0.9585 0.8046 0.1538 0.7213 200
How I get these information after I use stpm2 command, please?

Also, if I use MICE or ICE package how can I use with stpm2 and do internal validation, please?

P.S I cannot get these values from R with function validate because it doesn't support rstpm2 package
P.S I have tried boostrap command in stata but it can display only bootstrp standard error. How can I get optimism and index.corrected?

[Kullathorn Thephamongkhol]

I have tried this command below: to calculate the bias-corrected C statistics in stcox first (and I will use in stpm2 later)

////
stset fuptime, failure(hipfrac )

capture program drop b_conc
program define b_conc, rclass
stcox sex smoking prevfrac age10 weight10, nohr
estat concordance
return scalar c = r(C)
end
bootstrap c=r(c), reps(500): b_conc
estat bootstrap, all

////

I got this results:

| Observed Bootstrap
| Coef. Bias Std. Err. [95% Conf. Interval]
-------------+----------------------------------------------------------------
c | .71963748 .0055138 .04872778 .6241328 .8151422 (N)
| .6304039 .8140494 (P)
| .6130046 .7974204 (BC)

I interpreted that the corrected c statistics is .71963748 - .0055138 = 0.71412368


Unfortunately, I have tried this in R software as below (Dxy = 0.4393, corrected Dxy = 0.3920) then corrected c statistics = (0.3920+1)/2 = 0.696


index.orig training test optimism index.corrected n Dxy 0.4393 0.4568 0.4096 0.0472 0.3920 500

I think it's too different from two software. (stata is too good) I have checked that coefficient in stata and in R are exactly the same.

Please correct me if I am wrong.

[Kullathorn Thephamongkhol]

So when I use this below for stpm2, I am not sure I can believe the results: (because I check with stcox in R and in stata , C statistics is very different)

capture program drop b_conc

. program define b_conc, rclass
1. stpm2 sex smoking prevfrac age10 weight10, df(3) scale(haz)
2. stcstat2
3. return scalar c = r(C)
4. end

. bootstrap c=r(c), reps(500): b_conc


------------------------------------------------------------------------------
| Observed Bootstrap Normal-based
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
c | .7193295 .0468472 15.35 0.000 .6275106 .8111484
------------------------------------------------------------------------------



Any suggestions will be appreciated.