OHDSI Home | Forums | Wiki | Github

Measurement data included for Propensity Score generation?

Hi everyone!

Can i confirm whether any of the measurements or observation variables are used in propensity score generation? So for example BMI, smoking status

If there are included - how is the missingness of these variables dealt with when the propensity scores are generated?

Thanks for your help

The default set of covariates used to construct propensity scores does not contain verbatim measurement values. Instead, two types of binary covariates are constructed for measurements:

  1. Was the measurement observed during the lookback window? (1 binary variable per measurement concept ID)
  2. Was a measurement value observed below, within, or above the normal range during the lookback window? (3 binary variables per measurement concept ID)

In both cases missingness doesn’t really apply: either the measurement was observed (within a range), or it was not.

I think smoking status is an observation, not a measurement?

Thank you for the quick reply! Thanks for the clear explanation :slight_smile: Sorry you are totally right smoking is an observation - newbie error! Is there be a way to include observations such as smoking status like you described for measurements? If this info is in the book or any of the vignettes let me know in case i missed it!