How to obtain covariate-adjusted median survival time in PROC PHREG

[JEPSANTOS1]

I have a question that I hope you can help me with.

I often use the Cox’s hazard regression with the PHREG procedure of SAS fitting multivariable models with several covariates. Below is an example of a model I used to analyze time to pregnancy in cows according to several covariates (all categorical covariates)

DPREG= Days to pregnancy

PEND = Pregnant at the end of study (0 or 1)

Covariates in the model are 4, all categories, and each with 2 levels: PTACCR, EXPN, PARPRE, and AIGRP

I obtain covariate adjusted survival functions with the BASELINE COVARIATES function in SAS.

My question to you is how can I obtain the covariate-adjusted median survival times and the respective 95% CI for each level of each covariate in the statistical model?

I request the predicted days when 50% of the population survives (OUT=PRED), but it does not give me the 95% CI.

If I use PROC LIFETEST, I obtain the median days to event and respective 95% CI, but those are from univariable models without multiple covariate adjustments; however, I would like to obtain the median days to event and respective 95% CI from the multivariable model in PHREG.

Is there a function or code in SAS PHREG that allows me to obtain the median days to event and respective 95% CI from the multivariable model (median days adjusted for covariates)?

ODS HTML;

ODS GRAPHICS On;

PROC PHREG DATA=GENONEW PLOTS (OVERLAY)=SURVIVAL;

CLASS PTACCR EXPN PARPRE AIGRPN;

MODEL DPREG*PEND(0) = PTACCR EXPN PARPRE AIGRP/ RISKLIMITS TIES=EFRON;

BASELINE COVARIATES=GENONEW OUTDIFF=DIFFSURV OUT=PRED  SURVIVAL=_ALL_/DIRADJ GROUP=PTACCR;

OUTPUT OUT=OUTPUTFILE XBETA=Xb STDXBETA=SE RESMART=Mart RESDEV=Dev;

HAZARDRATIO PTACCR;

ASSESS PH/RESAMPLE;

RUN;

[sbxkoenk]

---

@JEPSANTOS1 wrote:

Is there a function or code in SAS PHREG that allows me to obtain the median days to event and respective 95% CI from the multivariable model (median days adjusted for covariates)?

I don't think so, but I may be wrong.

If you analyze time-to-event data (including covariates), you can try -- by means of a model (incl. target + regressors) -- to find out about :

• mean survival time
• median survival time
• mean survival rate at a specified time
• median survival rate at a specified time
• restricted mean survival time = RMST (sometimes called the restricted mean event time)

The RMST is an alternative measure that is more often reliably estimable than the mean and median survival time in certain situations. Also, it provides a summary of the whole survival curve up to a time horizon, in contrast to the survival rate at a specified time.

RMST can be fit with the RMSTREG Procedure.
Have a look ... maybe that RMST is what you are looking for.

https://support.sas.com/documentation/onlinedoc/stat/151/rmstreg.pdf

Koen

[JEPSANTOS1]

Koen,

Thank you! I was not aware of RMSTREG as a method to estimate adjusted means for time to event analysis with censored data. I will try it.

Best,