OHDSI Home | Forums | Wiki | Github

What's the relationship between CohortMethod, Cyclops and FeatureExtraction?

Developers
#1

Hello.

I’m trying to understand the difference between the different HADES packages.

What’s the relationship between CohortMethod, Cyclops and FeatureExtraction (and other alternatives)?
It seems that some of their features overlap.
When should I use one or the other?

#2
  • Cyclops is our engine for doing regression (can be large-scale, with millions of subjects with millions of features).
  • FeatureExtraction is our generic approach to constructing a set of features for a cohort. The default set typically includes tens of thousands of features, such as multiple features for each drug, condition, demographic, etc. FeatureExtraction is used in multiple packages, for characterization, estimation, and prediction purposes.
  • CohortMethod can be used to perform comparative cohort analyses to estimate causal effects. CohortMethod uses FeatureExtraction to generate features for the exposure cohorts, and CohortMethod uses Cyclops to fit propensity models as well as outcome models using those features (although typically we keep the outcome model simple, because the outcome tends to be rare).

There is actually no overlap in functionality, since they depend on each other.

1 Like
#3

Then, can we just do everything from CohortMethod because internally it uses FeatureExtraction and Cyclops?
Or sometimes we may need to use FE and Cyclops individually?

#4

Yes, if all you want to do is perform comparative cohort studies then you only need to interact with CohortMethod (which will interact with FE and Cyclops for you).

1 Like
t