Hello,
I have run a PROC PHREG for the "Longev" (longevity) of my experimental subjects. I have a predictor named "InfSta" (for infection status, coded as 0=uninfected, and 1=infected). I also have quantitative covariates "Size" and "TotFec" (for total fecundity).
I used the stepwise selection of variables, as seen in the code below:
PROC PHREG DATA=WORK.TMP0TempTableInput
PLOTS=SURVIVAL
PLOTS=CUMHAZ;
MODEL Longev = InfSta TotFec Size InfStaSize InfStaTotFec InfStaTotFecSize/
TIES=BRESLOW
CORRB
COVB
RISKLIMITS ALPHA=0.05
SELECTION=STEPWISE
SLENTRY=0.25
SLSTAY=0.15;
InfStaSize=InfSta*Size;
InfStaTotFec=InfSta*TotFec;
InfStaTotFecSize=InfSta*TotFec*Size;
RUN;TITLE;
See attachment for partial output.
While I can interpret the Hazard ratio and CI for InfSta, I have some questions about the other predictors' CIs. It is difficult to me to interpret the Hazard Ratio for TofFec and the interaction InfStaTotFec. They are both the same: 0.999 (and their 95% CIs are also the same 0.999 to 1.000)!!!!!
My question are:
1. Why are the Hazard Ratios de same for TofFec and InfStaTotFec?
2. Has anyone seen a CI so narrow (0.999 o 1.000)? How do you interpret such a narrow CI?
3. "TotFec" (Total Fecundity) and "Longev" (Longevity) are significantly correlated (n=74, Pearson r=0.665, p=0.000). What effect may this have in the Cox PH model?
4. If TotFec and Longev are correlated, is it sound to use TotFec as a continuous variable for Longev in PHREG?
Thank you for your help.
Regards,
Igforek