FeatureExtraction 2 - Demographics - Is there a way to calculate distinct counts by subject_id? about featureextraction

[schuemie]

Could you help me: I'm having a hard time imagining a use case where this is appropriate. Cohort characterization is a clearly defined task, but you seem to want to characterize the population that is participating in a cohort. If that is the case, why not construct a new cohort that is specifically the population you wish to characterize? (For example, de-duping subjects by picking the first cohort entry per subject.)

[gowthamrao]

Thank you @schuemie . You are correct, one approach may be to construct a new cohort which is constructed using the first occurrence of the subject_id.

Use case - we want to calculate a measure - records per subject_id for a defined cohort. This measure is an indicator of churn/stability of the population - is the population going in and out of the cohort. We need to look at the populations distribution of the churn. An extremely squewed distribution indicates that only a few people are churning a lot, while a uniform distribution indicates that everyone is uniformly churning.

e.g. we may have a cohort where few subject_ids in the population churning 5 times, while vast majority are not churning. The subject_id that churned most is an outlier, that we may want to take out - during a sensitivity analysis of models sensitivity to the churn.

I felt this is easier to handle in the package, then build another cohort for the same population - with the de-duping.