Estimate propensity score - random effect?

[bollibompa]

Hi,

I am performing a PS-matched comparative effectiveness study of two drugs (A and B). I have data for 3 years and I want to estimate PS based on data for these years. PS are estimated based on covarites before initiating treatment A or B.

During this period a subject initially treated with drug A may have switched to drug B during this time-period of three years. This subject may thefore exist in the data both treated with drug A and with drug B (but with different covarites before before initiation of A and B). They are probably not independed.

Have anyone experience of estimating propensity score (PS) where the same subject may occur two times?

Is it possible? Mixed-effect logistic regression?

(I am just curious about the estimation of PS, the estimated treatment effect I will handle at a later stage)

Thanks in advance!

Thomas