My latest discussion on this matter with the PHREG developer revealed the following sentiments: "Currently PROC PHREG does not estimate the survival function when there are time-dependent covariates. It is a bit controversial whether it is doable (for example, Fisher and Lin (1999)). My feeling is that it is doable in some situations (for example when all td-covariates are external). Such an implementation will require new syntax to input a profile (trajectory over time) for each time-dependent covariate. In the meantime, I will discuss ideas with my colleagues and our research into this topic is ongoing. If you have any references to share with us, by all means, please do so." I'm sorry to have to give you such bad news but there isn't really a way to plot survival curves for time-dependent covariates. ... View more

My latest discussion on this matter with the PHREG developer revealed the following sentiments: "Currently PROC PHREG does not estimate the survival function when there are time-dependent covariates. It is a bit controversial whether it is doable (for example, Fisher and Lin (1999)). My feeling is that it is doable in some situations (for example when all td-covariates are external). Such an implementation will require new syntax to input a profile (trajectory over time) for each time-dependent covariate. In the meantime, I will discuss ideas with my colleagues and our research into this topic is ongoing. If you have any references to share with us, by all means, please do so." I'm sorry to have to give you such bad news but there isn't really a way to plot survival curves for time-dependent covariates. ... View more

You don't get HR for interaction terms in the ParameterEstimates table. You need to use a HAZARDRATIO statement with the AT= option to specify the level of the interacting variable. Also, the COVS(AGGREGATE) does nothing unless you have an ID statement to identify the clusters. proc phreg data = CANIMPT.capstoneclean covs(aggregate) covm; class ironcat vitccat; model fup_t1d*case_t1d(0) = ironcat ironcat *vitccat /ties=discrete risklimits covb; hazardratio "iron unadjusted" ironcat / diff=ref cl=wald; title 'Ironcat consumption and T1D progression'; run; You will also need the PLOTS= option on the PROC PHREG statement in conjunction with the BASELINE statement in the code. See Example 8 Survival Curves in the PHREG documentation for the basics. ... View more

Your syntax looks OK to me but with only 2 observations per strata I'm not very surprised you are not getting the Quartiles to be computed. For this to happen the survival curve has to drop below the points of S(t) = .75, .50 and .25 for the first, second and third quartiles respectively. ... View more

You can use the TEST statement to test for the association of covariates with survival time. However, you cannot do a KM analysis or curves that take the covariates into account. See Example 1 in the LIFETEST documentation for a discussion of the TEST statement and it's results. ... View more

See Example 89.5 Conditional Logistic Regression for m:n Matching<> in the PHREG documentation. ... View more

The pointwise confidence limits for the survival function are in the OUTSURV= dataset.  This is an option on the PROC LIFETEST statement.  ... View more

You have referenced the correct SAS Note regarding the quartile upper confidence limits (quartile UCL). There was a Hot Fix in the SAS Note that applied to SAS9.4 (TS1M6) = SAS/STAT 15.1 and is now a part of SAS9.4 (TS1M7) = SAS/STAT 15.2. Typically, after the fix is applied we see that some of the quartile UCL do not actually exist and are missing in the output. I think your current results are correct. ... View more

I checked with the PHREG developer on this and "currently there are no methods to compare FG models in PHREG".  ... View more