Hello!
I am running PHREG with repeated covariate measures and I cannot understand how SAS deals with missing data. I specifically created a data set with full repeated covariate data (i.e. there are no missing values in any of the variables). Nevertheless, I still get notes in the LOG such as "5 event observations have missing values in the time-dependent covariates. They are not counted as event observations, but they may still be used in the risk sets of other event times." Of course the outcome and follow-up time are also complete. Does anyone knows why this may be happening? The code I use is:
proc phreg data=check;
where cohort>1 and Years~=.;
class sex(ref="1") cohort(ref="2");
model Years*Outcome(0)=cohort sex bodymassindex / ties=breslow risklimits selection=backward slstay=0.08 include=1;
array adur{*} dur1-dur6 dur8 dur10 dur14;
array abmi{*} mbmi1-mbmi6 mbmi8 mbmi10 mbmi14;
do i=1 to 9;
if (Years ge adur{i}) and (adur{i}~=.) then do;
bodymassindex=abmi{i};
end;
end;
run;
Thank you!
Tina.
P.S. The reason I am concerned about this is that when I use the same exact data set with PROC LIFETEST, I get very different results. When covariates outside the fixed cohort and sex are not included in PHREG the effect of cohort is not significant, just like in LIFETEST. However, when bmi is included as a covariate with repeated measures, the effect of cohort becomes very significant, though bmi increases over time across all cohorts (though it is lower at study entry in one compared to the other two cohorts, bmi becomes similar across cohorts by year 3). I was thus wondering whether the 5 events not counted as such may be the answer, although I am not sure which 5 are excluded as cases - all have data on bmi throughout follow-up.
