OHDSI Home | Forums | Wiki | Github

Prediction in the atlas: what are included in the covariate settings?

(panpan) #1

dear all:
when I use the atlas to do prediction research, I am a little confused about the covariates in the domain. For example, if I choose the observation domain, then the included covariates are only the observation_concept_id , the different observations my cohort’s members had prior to the cohort start date, or they include all concept ids, such as observation_ type_concept_id and value_as_concept_id,qualifier_concept_id,unit_concept_id and so on, which are also observed prior to the cohort start date

thank you very much

(Martijn Schuemie) #2

There is some information in the documentation, although I agree not very clear, e.g.:

useObservationLongTerm One covariate per observation in the observation table in the long term window. (analysis ID 802)

What this setting will do if set to TRUE is create one covariate per observation_concept_id in the long term window, which by default starts 365 days before cohort entry, and ends on the cohort start date (including the cohort start date). All other concept IDsm such as unit_concept_id are ignored.

(panpan) #3

thank you very much @schuemie, so is it possible that we include the covariates such as value_as_concept_id in the observation domain when developing atlas because this domain contains so much important information which can’t be included in other domains.

thank you very much

(Martijn Schuemie) #4

We did briefly discuss this in the past, but there is a silly technical issue we need to overcome.

(Vojtech Huser) #5

I replied to the github issue. Indeed, knowing the inner guts of the package helps a lot in using it. What you describe in discussion of the issue is very useful.