OHDSI Home | Forums | Wiki | Github

Cohort Exit in ATLAS

@Chris_Knoll or @anthonysena, in ATLAS there is a Cohort Exit Criteria and Censoring Events section. However the “Censoring Events” section sounds like it is technically a “Cohort Exit Criteria”. Would it make sense if they both sections fell under “Cohort Exit Criteria”?

There is a difference between the two:

Cohort exit generally applies to cohort_end_date

Censor applies to either or both cohort_start_date or cohort_end_date. It’s like a date based chopping block. Censor start chops cohort_staet_date, censor end chops cohort end date

Actually cohort exit explicitly applies to the end date :smile:

@gowtham: Erica is pointing to censoring events, which also is explicitly for setting the end date. So censor events is a bit different than the censor window

Censoring all applies to cohort_start_date correct?

@Chris_Knoll why this came up is I’m thinking about reporting the study design (aka Print Friendly) and it felt like these were both describing aspects of “Cohort Exit” and I thought they should be grouped as such (basically I want the stuff below to be in the “Cohort Exit Criteria” box too).

Does it make sense for me to open a ticket on GitHub for this recommendation for change and let the design team debate it / nix it / take it from there?

Yes, Erica, that sounds good.

Not sure what you mean by ‘censoring all’ but I think you mean ‘specifying a censor window start and end date’. Yes, censoring all trims the cohort periods by both the start and end date.

However, ‘Censoring Events’ are a way to pick random events from a patient history that can ‘end’ your cohort presence. This is different then censoring window which is used to ‘bound’ a cohort to a calendar period of time. Maybe censoring events is the wrong term or censoring window is the wrong term, but, to Erica’s point: censoring events and cohort exit criteria are specifically for defining the cohort end dates.

I think it makes sense to keep ‘Censor Window’ it’s own separate thing.

-Chris

Done:

t