OHDSI Home | Forums | Wiki | Github

Care Sites in PLE

Dear all,
Does any of you have any experience adding care_site as a covariate in a propensity score model within the PLE package? I see it’s not included in Feature Extraction, so any advice would be very much appreciated!

Thank you!

@wallislau @mattspotnitz @schuemie @Adam_Black @edburn @Chris_Knoll

Hi @CarmenOT. Adding such covariates makes sense, but is actually not easy. I just created an issue in FeatureExtraction detailing the request, but it could be a while before it is implemented.

How much time do you have?

Hi @schuemie, thank you for raising the issue. We have 10 days roughly, do you think it’ll be feasible?

In theory, but I don’t think we have developers available right now. A developer who knows FeatureExtraction inside out could do it in a day. Other developers may need two weeks or more.

1 Like

Hi @schuemie Can I hope to see it is being implemented in a month or two? We may need it in the future.

@Adam_Black , @Paul_Nagy : Could we enlist the Khieron’s for this? This is a fairly high complexity issue to start with, but nicely contained and with clear objective.

Because of the urgency expressed by the researchers (in private communications), I decided to prioritize this and do it myself. I have a working version (tested with unit tests) of FeatureExtraction here. To install this version, use

remotes::install_github("ohdsi/FeatureExtraction", ref = "careSiteId")

To generate the features, you’ll need to set useCareSiteId = TRUE when calling the createCovariateSettings() function in R.

@wallislau: since you appear to be using a settings JSON file, you could achieve this behavior by adding

"CareSiteId": true,

to the covariateSettings entries in your StudySpecification.json file. Don’t forget to rebuild your study package afterwards, and to rerun the study in an empty folder!

1 Like

I agree and I did post it in the OSC workgroup but I don’t think anyone was able to jump on it quickly.