GA4GH Computable Cohort Representation – Scope Statement

Definitions:

  • Cohort is “A set of persons who satisfies one or more inclusion criteria for duration of time”. ~Ref OHDSI
  • Phenotype is “A specification of an observable, potentially changing state of an organism”. With respect to a Cohort, a Phenotypes are typically used to define one or more inclusion criteria using its code
  • Phenotyping Algorithms are heuristic or probabilistic algorithms that characterise phenotypes
  • Computable Cohort Representation is “a schema to represent a Cohort along with its associated Phenotypes as inclusion Criteria that can be executed against a resource to obtain results that match that definition”.
  • Cohort Summary is “a resource to describe the statistics of a group of entities, including associations of measured variables within the cohort”. ~ Ref FIHR
  • Cohort Study is “a pre-defined collection of cohorts conducted by an organisation with a specific theme”

Functional Representation

What is a Computable Cohort Representation?

  • A Computable Cohort Representation aims to represent a collection of subjects that satisfy a defined criteria represented by a Phenotype.
  • Each criterion is represented by combining Cohorts or Phenotypes using set (UNION, INTERSECT, EXCEPT) and logical (AND, OR, NOT) operators optionally within time interval (INTERVAL, FIRST, LAST)
  • Each Cohort inclusion criteria can be defined using recursively transcluded into Cohort representations
  • A Computable Cohort Representation query resource that accepts an inclusion criteria within a specified observation date range and an optional list of data sources.
  • A Cohort Summary is a resource that represents the summary statistics (mean, median and proportions) and association statistics (correlation coefficient)
  • Based on the configuration of the Cohort Resource, the response may contain an existence response (Yes/No), or Counts, or Cohort Summary statistics or full extracts of subjects that match the Cohort Definition
  • MH: Provenance Information needs to be captured
  • Capture mapping rules for terminologies

Use cases

  • Matching queries: retrieve the individual records matching (asthma=yes AND diabetes=no)
  • Provide shared computational representation of (asthma=yes AND diabetes=no) to enable nested queries eg (asthma=yes AND diabetes=no) OR (high blood pressure=yes)
  • Cohort studies would provide a query interface to enable matching queries to be run
  • Provide a summary statistics of the content of a cohort/cohort study - eg 30% of participants match the query (asthma=yes AND diabetes=no)
  • Enable differential queries based on different guidelines - eg criteria for asthma definition may be different between countries. Severity may also matter.
  • Enable mappings between terminologies used eg snomed vs hpo
  • Query: Patients with fanconi anemia patients
  • Patients with Bifid Thumb; OR
  • ...patients that do not have bifid thumbs but have A, B and C

=> need to be able to fetch the union of those 2 phenotypes in absence of an asserted fanconi anemia diagnosis

  • NLp to database Query
  • Crosswalk of single phenotypes, e.g. Diabetes across different phenotype libraries

Next steps

  1. Identify a simple query, and apply it to one source using Beacon 2.0 or Data Connect API.
  1. Jonathan will share links to Jupyter notebook used for Connect API: Data Connect queries over the “384 Phenopackets” collection https://colab.research.google.com/drive/1jhq5ONPj59jrQ4bmZGJtjakdur2tCFGd#scrollTo=u2WwrfudIbLp and Data Connect over KidsFirst FHIR Observations https://colab.research.google.com/drive/1Fr6l02-1clmouxVuhUgtsRVe7cVphjQA#scrollTo=qO_rYcj7glbS
  2. Melissa mentions we can query N3C - the largest clinical cohort in the US. https://ncats.nih.gov/n3c 
  1. Run same query across different sources and integrate results (summary statistics)

What is it not?

  • It is not a replacement for representing existing terminologies, vocabularies and ontologies. It aims to reuse existing terminologies within its definition
  • It is not a replacement of the GA4GH Phenopacket schema and aims to leverage the Phenopacket schema as part of its transmission protocol
  • It is not a replacement for the Beacon v2 schema and aims to leverage and augment Beacon protocol to query a resource for the existence of Cohort

Cohort Representation Landscape Analysis