OHDSI Home | Forums | Wiki | Github

Balancing Covariates in a PLE Package


(Matt Spotnitz) #1

Hi Everyone,

Here is a suggested 5 step method for balancing covariates in a PLE package.

  1. In ATLAS, create a concept set that includes the defining exposure variables of the target and comparator cohorts.
  2. In the ATLAS estimation tool, select the exposure variable concept set for exclusion both in the “Comparative Cohort Settings” and in the “Effect Estimation Analysis Settings.” If the PLE package has multiple analyses, select the exposure variable concept for exclusion in each analysis. It may be helpful to exclude descendant concepts, which is an option in the “Effect Estimation Analysis Settings.”
  3. Download the PLE package and run it. An error message may list the covariates that are correlated with the exposure variables. Those covariates can also be found in the error log.
  4. Update the exposure variable concept set with the new covariates.
  5. Update the PLE package with the new exposure variable concept set in the “Comparative Cohort Settings” and “Effect Estimation Analysis Settings.”

For some PLE packages, steps 3-5 are unnecessary. For other packages, those steps may need to be performed more than once. It is important to know the counts of the target and comparator cohorts. If both are small, then covariate balance may be difficult to achieve. I hope these suggestions are helpful.

Thanks,
Matt


t