OsoGris Tracker

Re: How to compute survival probabilities from PROC PHREG fitted model?

OsoGris — Thu, 14 Nov 2024 11:12:26 GMT

The formulas for the survival probabilities are given in the Details => Survivor Function Estimators section of the PROC PHREG documentation. These formulas are not easily replicated outside of PHREG and we rarely, if ever, get asked about them. The only one that seems tractable is for METHOD=PL (product limit). This estimate is S(t) = So(t)**exp(XBETA) where So(t) is the baseline survival function (more on this later) and XBETA is the linear predictor XBETA = b1*x1 + b2*x2 + . . . . + bk*xk. The bi are the coefficients and xi are the covariate values for a particular observation.

The baseline survival function So(t) can be had using the BASELINE statement and OUT= data set where the COVARIATES= data set/option has a zero value for all of the covariates. Then you can manually put together S(t) = So(t)**exp(XBETA).

Note that the BASELINE statement is typically used in conjunction with the PLOTS= option to produce survival curves. These curves are produced at the settings given in the COVARIATES= data set. The COVARIATES= data set typically only has a few observations used for plotting particular survival curves. See Example xx.8 Survival Curves in the PHREG documentation (Examples section) for an illustration. The BASELINE statement produces estimates at the event times in the data.

Also note that if you are using the model (t1, t2) counting-process syntax then the BASELINE statement enforces METHOD=Breslow regardless of what METHOD= option you specify. I think a NOTE is written to the log in this case.

Re: How to plot cumulative incidence with number at risk from survival analysis with competing risk

OsoGris — Fri, 08 Nov 2024 15:13:24 GMT

When using the EVENTCODE= option for a competing risks analysis PROC LIFETEST does not support the ATRISK plot option.

Re: The Weibull accelerated failure time model parameters estimated but how to use them for predicti

OsoGris — Thu, 09 May 2024 13:56:39 GMT

Tom: You can get the survival probability in this manner: use the OUTPUT statement to request the CDF= keyword and post-process this as Survival = 1 - CDF in a subsequent DATA step. I'm not sure how to get a cumulative incidence value since PROC LIFEREG doesn't model any competing risks scenario for which the CIF is appropriate.

Re: Proc Phreg programming statement to create Time-Varying Covariates

OsoGris — Thu, 11 Jan 2024 17:14:26 GMT

Please note that when using programming statements to create time-dependent covariates they must appear inside the PROC PHREG step. You cannot create them in a prior DATA step unless you are using the model (t1, t2) counting-process syntax (instead of programming statements). Please see the following SAS NOTE:

Usage Note 24554: Why do I get different results when I form time-dependent covariates in a DATA step as opposed to forming them inside a PROC PHREG step?

https://support.sas.com/kb/24/554.html

Re: Propensity Match in 1:1:1 fashion

OsoGris — Fri, 22 Dec 2023 16:50:18 GMT

You are correct that PROC PSMATCH does not match with more than two groups (treatment and control). I am not aware of a procedure that does matching of 3 groups.

Re: Strange result of Proc lifetest in one strata

OsoGris — Wed, 13 Dec 2023 17:25:29 GMT

Shouldn't you be specifying a FAILCODE= value in your code?

proc lifetest data=Phreg12121   
plots=cif(test cl)  outcif=survival_data;
time T_FI1YR*outcome(0)/failcode; 			
strata cohort ;
*WEIGHT weight0_SD;
run;

Re: How to generate compact letter display from multiple comparisons in PROC LIFETEST?

OsoGris — Mon, 11 Dec 2023 15:52:16 GMT

There is nothing similar to the "pdiff lines" options for the ADJUST= option in PROC LIFEEST.

Re: PROC PHREG with IPTW weighted cohort: robust standard error

OsoGris — Mon, 11 Dec 2023 11:26:16 GMT

Robust standard errors are produced by the COVSANDWICH(AGGREGATE) option on the PROC PHREG statement. You will also need an ID statement to list the variable that defines your clusters.

Re: Interaction term in the PHREG procedure

OsoGris — Thu, 30 Nov 2023 14:50:50 GMT

You may need to put your interaction terms in programming statements inside the procedure if they involve the time variable. Then you can list them in the MODEL statement as well. This would be for so-called "time-dependent interactions with time" variables.

Re: Including log rank test with proc phreg

OsoGris — Tue, 07 Nov 2023 06:26:54 GMT

There is no log-rank test with PROC PHREG. You can get a log-rank test in PROC LIFETEST with the STRATA statement.

Re: procedure for TWANG

OsoGris — Wed, 01 Nov 2023 10:53:54 GMT

I don't know if PROC PSMATCH was based on TWANG or not but it does propensity score matching of treatment and control groups.

Re: procedure for TWANG

OsoGris — Tue, 31 Oct 2023 07:09:54 GMT

You might want to look into the PROC PSMATCH procedure. It does propensity score matching of treatment and control groups.

Re: Proc Phreg for time varying exposure variable- different output

OsoGris — Wed, 18 Oct 2023 05:36:15 GMT

I think you should go with the start, stop counting-process syntax. Your data are already in that form.

Re: Harrell's C indexes in competing risk model

OsoGris — Mon, 25 Sep 2023 10:17:23 GMT

You can try using the CONCORDANCE option on the PROC PHREG statement. I don't know if that's enabled, necessarily, for the competing risks analysis.

Re: Cox regression plot - plot two lines in the same graph

OsoGris — Thu, 21 Sep 2023 10:40:31 GMT

Because you don't have a COVARIATES= option/data set you are getting the default behavior in the BASELINE statement of producing only the reference level of group. You need to define a COVARIATES= option/data set where you specify both groups. See the example in the documentation titled "Survival Curves" for an example of using the COVARIATES= option. I think you would also need to use the CUMHAZ= keyword on the BASELINE statement.

Re: Log-rank test at a specific time point

OsoGris — Fri, 25 Aug 2023 06:14:45 GMT

The only way to get a log-rank test at 12 months and 24 months would be to subset your data to those time intervals. That is not really something we would recommend. I don't see any other way to get a log-rank test at a specific time instead of over all the data.

Re: Estimating Hazards for Time-Variant Treatment Groups in PROC PHREG

OsoGris — Mon, 07 Aug 2023 10:28:19 GMT

PHREG does not produce hazard estimates. The underlying model is h(t) = ho(t) * exp(xbeta) where ho(t) is an unknown and unspecified baseline hazard. Cox's method of partial likelihood does away with ho(t) but still estimates the coefficients. These coefficients are exponentiated to give hazard ratios which is the main output of PHREG. It you have time-dependent covariates (td-covars) defined by programming statements inside the PHREG step then, as you noticed, the OUTPUT and BASELINE statements are not available. If you could use the model (t1, t2) counting-process style of input (cp-syntax) then the BASELINE statement could be used with the CUMHAZ= keyword to get cumulative (integrated) hazard estimates.

Re: How to conduct competing-risk based on proc surveyphreg using sas 9.4

OsoGris — Fri, 04 Aug 2023 06:08:52 GMT

unfortunately, PROC SURVEYPHREG does not have the EVENTCODE= option (like PROC PHREG does) to enable a competing risks analysis. You could try using PHREG with a WEIGHT statement; the parameter estimates would be the same but not the standard errors as from SURVEYPHREG.

Re: Proc lifetest with multiple imputation

OsoGris — Wed, 02 Aug 2023 13:05:42 GMT

If you want curves for treated (events?) and untreated (censored?) separately then it seems you would have to run two PROC LIFETEST steps for each imputed data set. That would mean 10 PROC LIFETEST steps. I don't see any other way around this.

Re: Proc lifetest with multiple imputation

OsoGris — Mon, 31 Jul 2023 15:50:14 GMT

Have you tried "strata _imputation_" ?