I'm attempting to fit a GEE model and I have a question about using the AR(1) working correlation matrix. I've read some conflicting information about this correlation matrix. In some books and articles, I've seen that AR(1) should only be used when the measurements are made at equally spaced intervals, while in other articles, I've seen that this is not necessarily so. I have a population of patients with varying hospitals vists. Each of the patients in my study visit the hospital at various times and at different intervals. Given this, can it be appropriate to use an AR(1) working correlation matrix for this data? Does it really matter or make a big difference if I'm using the sandwich variance estimator and have about 90,000 patients/clusters?