Question about my sample

[iuri_leite]

Dear members of SAS community,

I do not have a SAS specific question. In fact, I would like to argue about the need to give enough time of follow-up to analyze data with logistic regressions instead of any survival model that incorporates censoring.

Suppose I am interested in identifying factors associated with loss of follow-up of participants from a cohort study. Suppose further that the loss of follow up is defined as failure to return for a period of six months.

I could use a cox proportional hazards model to consider patients who were enrolled less than six months before the ending date of the study.

If I use a logistic regression instead of a survival model that incorporates censoring, shouldn’t I exclude all the patients who were enrolled less than six months before the ending date of the study? I ask that because I did not follow them for enough time to observer the event.

It seems to be that a worse decision is to exclude those who did not return in the six months before the end of the study and keep in the analysis those who return in the six months. Is that correct?

Best regards,

[HB · Accepted solution]

If you don't have six months of data on some subjects to be able to determine if they had proper follow-up then yes, it would seem reasonable to exclude those subjects. 

[StatDave]

I don't see one model as being preferable over the other but rather as the two models answering different questions. If your interest is in modeling the time to the event, then use the survival model and your censored data. If the interest is in modeling the probability of the event occurring within six months, then use the logistic model and those that don't experience the event are simply nonevent observations to be used along with the event observations.