We are conducting a study focuses on the associations of the neuropsychiatric symptoms and the diagnosis of the cognitive impairment. We are using an observational datasets, and want to compare people with the similar age, gender and the educational level. Thus we want to use Propensity score matching to ensure that our datasets are balanced. propensity score matching produces a weight.
Then, we also have participants died before diagnosis of cognitive impairment, so we want to use fine and gray model which treats death as a competing risk. Do you happen to know if there is a way that we can add propensity score weights to the Fine and Gray model? or if there is some other ways which we can use propensity score weighting which also take the competing risks into consideration?
It’s not clear why you would use propensity score adjustment here as opposed to plain covariate adjustment. More on this at BBR .
With a competing risk analysis you would be estimating the probability of cognitive improvement that preceeds death. “That preceeds death” is a loaded clause and I never really know how to interpret this. I would think that a state transition model with death as an absorbing state would be far more appropriate. Some thoughts about this are at Longitudinal Ordinal Models as a General Framework for Medical Outcomes | Statistical Thinking . With a state transition model you can compute state occupancy probabilities such as P(death), P(cognitive level > y and alive), P(dead or cognitive level < y), etc.
Dear Dr. Harrell,
Thank you so much for your reply. I will read your discussion on the propensity score matching .
Yiqi