Dead:1 is dead, 0 is alive, used as status variable.
DUR is survival time.
Normally I just use the last row of each ID(see below), and use data step inside PHREG to add time-dependent variable (Age in this case). Now I hope to model this data structure directly, but couldn't figure out how. Thank you so much if anyone could help !!!
This looks like a straightforward PHREG analysis. You may need to restructure your data to use age at onset as a covariate, but it is not time dependent.
I don't generally use the other two procedures (It's an age thing, I learned the Cox model before the other two SAS Procedures were available, way back in V 5.....).
BTW, DR Cox gave a great talk about the Student's t (William Gosset) and the relationships between Gosset, Fisher, and Pearson at the International Biometrics Society Conference in Dublin (home of Gosset's employer, Guiness). I'm afraid that we won't have too many more opportunities to hear some of these stories from first-hand observers. We added a plaque to him the Guiness Storehouse; look for it at your next visit.
I understand what you said of using last record of each person, and treat AGE as time independent. However, AGE is just one of the time-dependent variables in my dataset. I have over hundred of variables in my dataset, lots of them are time-dependent.
My data structure is one row for one person at each time point(see below). That's why I'm asking whether PHREG have a way to model this data structure directly, so that I don't need to restructure and aggregate my dataset into only one row for one person regardless of time points.
Thank you so much for your reply ! Hope you can help me with this question !!
I can't help you much further, I haven't done time-dependent covariates in a long time. One possibility is to look at Paul Allison's book on survival analysis (mine is out right now, or I'd look); he has good advice for solving problems; if you don't have one available locally, you can get it from the SAS bookstore.
Also, I will repeat, Age is NOT a time-dependent covariate. Age is a linear function of the age at time 0 and the difference from time 0 to time k. If you model it as time-dependent, you will have an over specified model and the results might be unreliable.