Choice of vocabularies for use in annotating Cochrane content 

Our aims in chosing vocabularies have included the following:

• Acceptable for use across Cochrane
•  Structured in a way that allows them to be linked to other parts of the linked-data universe.
• Compatible with vocabularies already in use within Cochrane or by our close  collaborators.

Vocabularies we are currently using

Population

• For age and sex - MeSH
• For other aspects of Population  - SNOMED CT and MedDRA

Intervention and Comparison 

For drugs - RxNorm and ATC 

For non-drug interventions - SNOMED CT and MedDRA

The OHDSI vocabularies and common data model

In choosing specific controlled vocabularies we have built on the work of OHDSI (Observational Health Data Sciences and Informatics).  This (mainly US based) consortium is primarily interested in using large observational datasets (e.g. computerized medical records, or adverse drug reaction reports submitted to regulators) to identify drug risks and benefits. They are building on and extending prior  work done by The Observational Medical Outcomes Partnership (OMOP).  

OMOP has done a lot of work on relevant vocabularies (http://omop.org/Vocabularies) and has produced a Vocabulary Specification Document that includes rich detail about SNOMED CT, RxNorm, MeSH, LOINC ATC and others, along with a conceptual basis for linking terms across vocabularies and a set of mappings from one vocabulary to another within this set.  Ongoing development and maintenance of these vocabularies has now been taken over by the IMEDS program of the Reagan-Udall Foundation for the FDA.

Many of their domain concepts (see this table for a list of domains with relevant vocabularies) fit well with our PICO concepts.  E.g. their Demographic and Condition would be part of P (Observation - which could include specific findings on a physical examination, abnormal lab or xray finding etc - would also fit with P), Drug & Procedure would  fit in I or C etc.   They have identified RxNorm as their primary vocabulary for drugs and have mapped the WHO's ATC vocabulary & others to it (including extending RxNorm where there are non-US drugs that have been omitted) They are using SNOMED CT in a similar primary role for Condition and Procedure and SNOMED CT plus LOINC  for Observation.

Higher-level categories for interventions and outcomes

In addition to precise annotations using terms from controlled vocabularies, we are annotating interventions, comparison interventions and outcomes using broad categories that have been derived from studies of the contents of Cochrane Reviews.

Intervention and Comparison

We are using a 15 item classification of interventions  from an article by Davey et al that classified every meta-analysis with at least two studies in every review in the January 2008 issue of the Cochrane Database of Systematic Reviews.  This classification is  based in turn on the the Health Research Classification System developed by the UK Clinical Research Collaboration and contains the following categories:  

• Pharmacological
• Vaccines
• Surgical
• Medical devices
• Cellular and gene
• Radiotherapy
• Physical
• Educational
• Behavioural
• Screening
• Complementary
• Psychological
• Resources and infrastructure
• Complex
• Other

Outcomes

We are working closely with the COMET (Core Outcome Measures in Effectiveness Trials) Initiative to build their Core Outcome Sets into our model as they become available.  In the interim, we have collaborated with COMET to develop the following high-level set of outcome categories to use in our initial annotation of Cochrane Reviews.  Eventually, specific components of individual Core Outcome Sets will be mapped to the appropriate category on our high-level list.

Our categories are very similar to the list of 23 outcome categories from Table 3 in an article by Davey et al cited above.

Components and characteristics of an intervention.  The Template for Intervention Description and Replication (TIDieR) checklist has been developed as an attempt to improve the quality of description of interventions in publications.  We have incorporated some  of the components of TIDieR into our model and will explore their usefulness in annotating our reviews.