Hello all,
I am working with a dataset in long format, where each row represents a specific month for a patient (e.g., row 1 = month 1, row 2 = month 2, etc.). Patients become censored (censor=0 -> censor=1) at the month they deviate from their assigned treatment strategy (or by month=12 if they do not start their assigned treatment).
I am performing an inverse probability of censoring weighting (IPCW) analysis to estimate the probability of remaining uncensored. For the denominator model, I have included the variables month, baseline covariates (cov1, cov2) and time-varying covariates (cov3, cov4).
However, I am unsure about the appropriate specification for the numerator model. Should it include only baseline covariates (cov1, cov2) (option 1) or should it just be the month variable (option 2)?
numerator - option 1
proc genmod data=example descending;
class id month (ref='12');
model censor = month / dist=binomial link=logit;
repeated subject = id;
output out=censor_num p=num;
run;
numerator - option 2
proc genmod data=example descending;
class id cov1 cov2;
model censor = cov1 cov2 / dist=binomial link=logit;
repeated subject = id;
output out=censor_num p=num;
run;
denominator
proc genmod data=example descending;
by exposure_final;
class id month (ref='12') cov1 cov2;
model censor = month cov1 cov2 / dist=binomial link=logit;
repeated subject = id;
output out=censor_den p=den;
run;
