Suppose I aim to examine if treatment A is better than treatment B. My dataset consists of 10,000 individuals with a total of 100,000 observations. Thus each individual is observed 10 times and his covariates (blood pressure, blood lipids etc) are updated each visit. To provide more robust estimates, I use all observations in the Cox regression, by counting process syntax. I would like to adjust for propensity score as well but the problem is:

(1) Should the propensity score be calculated for each observation (i.e 10 times for each individual)?

(2) Should the propensity score be calculated once (per individual) and then held fixed?

My second question

I have read R D'Agostinos review (http://www.stat.ubc.ca/~john/papers/DAgostinoSIM1998.pdf) on use of propensity score but after reading a paper (which investigated differences in treatment A and treatment B on risk of death), where authors stated that their covariance adjustment was done only by stratification with a propensity score; is it not "better" to include the covariates in the Cox regression as well, despite being included in the prop score?

I did not attach a data set because this is a methodological discussion.


  • $\begingroup$ When is the decision to assign treatment A or B is made? If at the beginning, then you want to calculate the propensity only based on covariates present at the beginning. $\endgroup$ – Aniko Aug 29 '14 at 13:25
  • $\begingroup$ Yes, decision is made at the beginning. So you suggest i calculate the prop score at inclusion and then hold it fixed throughout observations. Seems logical. $\endgroup$ – Adam Robinsson Aug 29 '14 at 14:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.