Skip to content

Day 3 · Cohort Definition and Characterization

Objectives

By the end of Day 3 you will be able to:

  1. Define an OHDSI cohort and explain how it differs from a simple code list.
  2. Identify the four parts of a cohort definition: entry event, concept sets, inclusion criteria, and exit.
  3. Apply temporal logic to inclusion rules.
  4. Build and generate a cohort in ATLAS.
  5. Characterize a generated cohort and read the output.

What a cohort is

In OHDSI, a cohort is a set of persons who satisfy one or more criteria for a period of time. This is an important distinction: a cohort is not just a list of people with a diagnosis code. It is a rule set that says who qualifies, under what conditions, and for how long. The same definition, run on any OMOP CDM, should produce a comparable cohort, which is what makes studies reproducible across sites.

The four parts of a cohort definition

ATLAS and the Book of OHDSI (chapter 10) present a cohort definition as cohort entry events, inclusion criteria, and cohort exit, all built on concept sets. This program teaches those as four parts so that the Day 2 concept set work stays in view:

  1. Cohort entry event (initial event): what event qualifies a person to enter, for example the first exposure to riluzole? This is built on a concept set.
  2. Concept set: the reusable set of standard concepts that the entry event and rules reference. (This is the Day 2 building block.)
  3. Inclusion criteria: additional conditions applied to the entry events, for example "at least 365 days of prior observation" or "no prior edaravone." In ATLAS there is no separate exclusion list; an exclusion is written as an inclusion rule that requires exactly zero occurrences of the excluded event.
  4. Cohort exit: when and how a person stops being in the cohort, for example at the end of continuous drug exposure or end of observation.

Temporal logic

Inclusion rules are usually time-relative. "No prior edaravone" really means "no edaravone exposure in some window before the entry event." Getting the window right (for example 365 days before, anytime prior, or during a fixed period) is where most cohort logic errors live, so it deserves explicit attention.

Building a cohort in ATLAS

ATLAS is the graphical interface for cohort building. The typical flow:

  1. Start a new cohort definition.
  2. Define the entry event using a concept set (for example new use of riluzole).
  3. Set the required prior observation (ATLAS offers this in the entry event section) and add inclusion criteria, each as its own named rule with its temporal window.
  4. Define the exit.
  5. Save, generate against a CDM, and review the counts and attrition at each inclusion step.

The attrition report shows how many people each rule removed, which helps you find a rule that is stricter or looser than you intended.

Characterization

Once a cohort is generated, characterization summarizes who is in it: demographics, conditions, drugs, and procedures before and after entry. It is the first sanity check that your definition captured the population you intended.

Slides and materials

File Description
Instructor Deck Full slide deck with speaker notes
Participant Workbook Workbook with fill-in exercises
Kahoot Quiz Cohort definition quiz
Tutorial: Cohort Definitions Basics and Atlas Written walkthrough of a riluzole new-user example

The hands-on lab is on the Day 3 exercise page.

Further reading

  • The Book of OHDSI, Chapter 10 (Defining Cohorts): https://ohdsi.github.io/TheBookOfOhdsi/