OHDSI Home | Forums | Wiki | Github

Basic questions: "cohort vs era" and "cohort entry vs inclusion criteria"?


I’m starting to learn the basics of OMOP, Atlas… and I have some simple questions.

What is the difference between cohort and era?

The book of OHDSI defines…
COHORT as a set of persons who satisfy one or more inclusion criteria for a duration of time. (Often interchanged with the term phenotype).
A CONDITION ERA is defined as a span of time when the Person is assumed to have a given condition.

I find both definitions quite similar.

Is it just that cohort refers to the persons and era is the time?

I’m also confused between “cohort entry events” and “inclusion criteria” in ATLAS. May they be interchangeable?

For example if I want to predict or estimate death among people who (take aspirin & have had a stroke).

I could use…
the date of the first aspirin intake as cohort entry and the date of stroke as inclusion criteria.
or reverse the order and use…
stroke as cohort entry and aspirin intake as inclusion criteria.

Would it be the same?
Everything could be considered both as an event and as a inclusion criteria?

There is a tutorial where “ace inhibitors” is used as cohort entry event and “diagnosis hipertension” as inclusion criteria. Why not the opposite?

In this context, yes, but for more reasons what you think: both the cohort and condition era are people with durations of time (cohort_id = condition_concept_id, cohort_start_date = condition_start_date, etc). Moreover, these periods are non-overlapping for a person within the given cohort/condition_concept. This is a critical feature of cohorts, and so condition era could serve as a cohort table if you re-mapped the columns to the correct names. The main difference between these, tho, is a cohort represents criteria in the form of inclusion criteria, while the condition era is built from condition occurrence coding, with a 30d gap between same-conditions are grouped together.

No, a cohort is both the persons and the time.

Entry events represent ‘potential entry’ into a cohort. Anyone may enter the cohort with condition X (an entry event) but only if they are being exposed to drung Y within 365d prior to the index condition X (the inclusion rule).

We sometimes think of entry events as determining the ‘sensitivity’ of a cohort definition, and the inclusions control ‘specificity’…but that’s a story for another day.

It would be the same in the sense you have identified people with asprin and stroke. But the difference is that the cohort_start date in the former is the asprin exposure, and the latter will have the stroke as cohort_start_date. You also need to specify relative time of one or the other…is aspirin before stroke or stroke before asprin?

Because they want to study efficacy or safety of ace inhibitors for a specific outcome (like GI bleeding) so the time-at-risk starts at the exposure, not the hypertension. We use hypertension because we are interested in the sub-population of ace inhibitor users. The opposite might be used on a predition study: you have patients taking aspirin, but experience a new diagnosis of hypertension and you want to predict GI bleeds.

Hope this helps!


Thank you for your thorough explanation.

Once you fulfill a entry event you will be tagged with it forever, but entry conditions just tag you for a short time?

I’m not sure what you mean ‘for forever’ vs. ‘for a short time’…do you mean 10,000 years (forever) vs. 30 days (a short time) or do you mean that the process will consider you in the cohort briefly until the entry events confirm your entry? If the latter, you can have a cohort definition without any inclusion rules, meaning all entry events will qualify you for the cohort.

As far as ‘how long will you remain in the cohort’, that is controlled in 3 ways:

  1. Until end of continuous observation. This is the default.
  2. For a fixed number of days
  3. Until the end of a continuous drug expoure (typically used in drug exposure cohorts)
1 Like

Another example from here:

“Persons hospitalized during the 2008-2009 influenza season with a diagnosis of influenza 21 days prior or during the hospital stay, with no prior continuous enrollment required and with no influenza hospitalization in the 6 months prior to hospital admission.”

Here it seems they are using as entry event three things together: the hospitalization date and Influenza (diagnosis or measurement).

Why aren’t they using the diagnosis or hospitalization as as inclusion criteria?

In this example, the patient is eligible for entry in the cohort at an In Patient visit (assumed to be hospitalization). So inclusion criteria for this doesn’t make sense because they want the cohort entry to be based on the IP Visit start date, and inclusion criteria filter out entry events, not establish entry into a cohort.

The ‘Restrict Initial Events’ is like an inclusion criteria, but instead of putting it udner ‘inclusion rules’ there is an option to ‘qualify’ the entry events with inclusion criteria before proceeding to inclusion rules. There’s reasons for making this choice, but they could have made an inclusion rule with the criteria for diagnosis and measurement as an inclusion rule and gotten the same result.