OHDSI Home | Forums | Wiki | Github

Effect of No Covariates on Propensity Score (PLE)


As part of developing a study, I have created some study packages with different levels of covariates. One contains the full OHDSI-recommended large-scale propensity score, while others specify either no covariates, covariates only as Age + Charlson Comorbidity Index, or covariates only Age + Sex.

The attrition tables show that when the study package is run with no covariates, even when there are quite a few subjects in the cohorts, the population drops to 0 at the stage “Matched on propensity score”. Of course we are not planning on running a study without any covariates, but are curious why this might be the case (hopefully, the answer might help with fixing errors in other versions of the package). Does anyone have any ideas?

I’m pretty sure this is just an edge case that wasn’t covered by the software. It assumes that if there is no covariate data, then something must be wrong.

Note that even when covariates are created, you can decide not to use them (e.g. decide not to fit a PS model). That is actually more efficient when you have multiple analyses in a single study, some doing adjustments and some not, since then CohortMethod will re-use the same data object for both.

Thank you! And to not use the PS, you would choose “None” in the analysis when the selection is “Do you want to perform matching or stratification?”, the options are “Matching”, “Stratification” or “None”?

That’s right!