PROC PHREG Standardized CIF estimations

[Rasmus_G]

Hi all,

I hope you can help me or point me in the right direction!

I'm analysing 30-day mortality for a group of patients after discharge. They are divided into different clinical groups. For each group, i report age (mean), sex, M3 comorbidity score (mean), length of stay and 30-day mortality (cumulative risk at 30 days).

Since the groups differ in age, sex, comorbidity and length of stay, I also want to present "standardized 30-day mortality", standardized to mean covariate values and most frequent sex.

However, I get some results that are off what i expect - for instance the standardized risk is lower for a group despite standardizing to a higher age and comorbidity (which does increase risk in my data as well).

I have used the PROC PHREG BASELINE statement - but I'm in doubt if it in fact does, what i want it to do. Is there another way I should go about it?

I have tried to make the estimations with a model for each group and a model for all groups, with group as a covariate - both versions give the same problem.

It might also be, that there are too few events for the model to make reliable standardized estimates.

I don't have statistical or similar education - so bear with me, if it is rubbish 🙂

Thanks in advance!

Best regards,

Rasmus

Data covar_1;
Age=58; sex='F'; M3Score=0.62; Los_H=16.2; output;
Run;

Proc phreg data=xxx plots=cif;
Where unspec_diag_gruppe = ’Abdominal pain’;
Class koen (ref=’F’);
Model mort_30_event * mort_30_days (0) = age koen M3Score / eventcode=1;
Baseline out=mort_standard covariates=covar_1 timelist=30 CIF=_ALL_;
Run;

[OsoGris]

 I don't see anything that your are doing wrong necessarily.  Your code looks OK to me.  The BASELINE statement should be estimating the CIF at the covariate values and time you specified. We don't have a benchmark for the number of events but, as always, the more the better. 

[Rasmus_G]

Okay, thank you for the reply and clarification. It's great with an extra set of eyes on the code. I will "blame it" on relatively few events across the groups.

Enjoy your day.