Cohorts

Nathan Hall

2024-10-22

Introduction

Cohort identification within real-world evidence (RWE) analyses of data adhering to the Observational Medical Outcomes Partnership Common Data Model (OMOP CDM) is a foundational step in understanding healthcare outcomes and treatment effects. Observational health data, sourced from diverse sources including Electronic Health Records (EHRs), health insurance claims, registries, and patient-generated data, offer valuable insights into patient health status and healthcare delivery. However, these data were not originally collected for research purposes, leading to the need for sophisticated methods to infer relevant clinical information for research objectives. This summary explores the importance of cohort identification, the methods endorsed by the Observational Health Data Sciences and Informatics (OHDSI) community, and the tools available for creating cohorts within OMOP CDM datasets.

Importance of Cohort Identification

Cohort identification is crucial for conducting meaningful analyses in healthcare research. A cohort, defined as a set of individuals meeting one or more inclusion criteria over a duration of time, serves as the basis for studying healthcare outcomes, treatment effects, and disease epidemiology. In the context of OHDSI research, cohorts are fundamental building blocks used for executing research questions, developing phenotypes, and conducting comparative effectiveness studies. OHDSI’s approach emphasizes the independence and reusability of cohort definitions, allowing researchers to define cohorts tailored to specific research questions while ensuring consistency and reproducibility across studies.

Methods for Cohort Identification

OHDSI endorses standardized methods for cohort definition, ensuring transparency, and reproducibility in research. Two main approaches are employed for constructing cohorts:

Rule-based Cohort Definitions

Rule-based cohort definitions rely on explicitly stated inclusion criteria to define cohort membership. These criteria are typically based on domain expertise and clinical knowledge, allowing researchers to specify cohort attributes such as clinical conditions, procedures, medications, and temporal relationships. OHDSI provides standardized components for assembling these criteria, including domains, concept sets, domain-specific attributes, and temporal logic.

Probabilistic Cohort Definitions

Probabilistic cohort definitions leverage machine learning techniques to compute the probability of cohort inclusion based on patient characteristics and clinical events. These models are trained on example data to automatically identify relevant patient characteristics predictive of cohort membership. The resulting probabilities can be used to classify patients into cohorts or as inputs for certain study designs.

Tools for Cohort Creation & Identification

OHDSI offers a suite of open-source tools to support cohort identification within OMOP CDM datasets, such as:

  1. ATLAS: A web-based tool that enables researchers to define, execute, and share cohort definitions using standardized terminologies and criteria. ATLAS streamlines the cohort creation process by providing an intuitive interface for specifying cohort criteria and visualizing cohort characteristics.
  2. SQL: Structured Query Language (SQL) provides a powerful means for defining cohorts through custom queries. OHDSI encourages the use of SQL for advanced cohort definition tasks and complex analyses.
  3. CohortGenerator: The CohortGenerator R package is a tool within the Observational Health Data Sciences and Informatics (OHDSI) ecosystem designed to facilitate the creation of cohorts from observational healthcare data stored in databases adhering to the OMOP Common Data Model (CDM). It is an R Package that streamlines the cohort creation process, allowing for efficient and reproducible cohort identification across different datasets.

The Cohorts Module

In the Cohorts tab of the OHDSI Analysis Viewer, there are 3 main sections, each with their own tab, the user can explore:

  1. Cohort Counts: Gives the counts of both the number of subjects and the number of records for each cohort and each database.
  2. Cohort Generation: Gives the cohort generation information for each cohort and each database, including an indicator if the cohort was generated (or not), the generation start and end time, and the duration (in minutes) of the cohort generation.
  3. Inclusion Rules & Attrition: Gives both a tabular and graphical representation of cohort attrition statistics for each cohort and each database. The user may select whether they want to view the results at the subject or record-level.