NLM-DDI CD corpus
DailyMed Cardiovascular Product
Labels Annotated
with Drug-drug Interactions
developed by the National Library of
Medicine
1. Entities
Drug classes are
assigned based on different characteristics: chemical ingredients,
administration method, mechanism of action, and the target organ or target
anatomical system. Any of these categories is a drug class and can include one
or many individual drugs or substances.
A drug is a chemical
substance administered for medicinal purposes. Drug names can be generic
(chemical name of the substance) or brand names.
Substances refer to
any material entities that are not drug or drug classes. These include foods,
nutritional supplements and other things that can be found in the environment.
Span refers to a span
of text that:
*provides
information about an interaction (context for interaction, severity of
interaction, strength of evidence)
*is
a fragment of text associated with an entity
*is
an antecedent to an anaphor that refers to an event
An expression is a
span of text containing an anaphor. The anaphora may refer to an entity or an
event represented by a span.
2.
Relations
Drugs
are linked to classes with a hasDrugMember relation. Drug
classes are linked to parent classes with a hasClassMember
relation. Substances are linked to classes with a hasSubstanceMember
relation.
Expressions
containing an anaphora are linked to their antecedent with a Coreference relation. The antecedent must be an entity
(drug, drug class, substance, Span, or Expression).
Fragmented
parts of an entity are connected by the hasFragment
relation.
3.
Drug Interaction Events
Interactions are biochemical or
physiological changes on the human body caused by the concomitant action of two
or more chemical components. Interactions can occur between drug classes,
drugs, and substances. Interactions should not be confused with side effects or
drug actions.
Caution interactions are general
precautions about the use of two entities together without specific mention of
an effect.
Increase interactions indicate higher levels or increased effects of the object drug in the system as a result of the precipitant drug.
Decrease interactions indicate
lower levels or decreased effects of the object drug in the system as a result
of the precipitant drug.
Specific interactions indicate
specific effects resulting from the interaction.
All drug interaction event types contain the Interaction
Indicator as well as entities in the following roles:
The object of interaction is a drug
class, drug or a substance whose effect is altered by the precipitating entity.
The object entity is linked to the interaction indicator with a hasObject-Argument relation.Precipitant of interaction (mandatory)
The precipitant of interaction is a drug class, drug, or substance that alters the pharmacological and/or other action of the object entity. The precipitant entity is linked to the interaction indicator with a hasPrecipitant-Argument relation.
The context is a population characteristic or event for which the drug interaction was observed. Context is a property of the interaction and qualifies the patient, not the drug. The Context for Interaction Span is linked to the interaction indicator with a hasContext-Argument relation.
The
strength refers to the scientific research evidence for the interaction. The
Strength of Evidence Span is linked to the interaction indicator with a hasStrength-Argument relation.
The severity is reserved for contraindicated interactions. The Severity of Interaction Span is linked to the interaction indicator with a hasSeverity-Argument relation.
Drug-drug
Interactions are encoded with the same XML schema as the SemEval
Task 9 DDI DrugBank and DailyMed
corpora.
A drug interaction is
encoded in the following way:
Sentence
containing the DDI
Annotated
entities within the sentence with a corresponding normalization if available
List of
annotated relations
List of
interacting drug pairs within the sentence and attributes of the interaction
The complete
annotation guidelines are available here.
The corpus is
available for download here.
The corpus with the original annotations can be downloaded here.
Contact
Johann Stan, johann.stan@nih.gov
Olivier Bodenreider, Olivier.bodenreider@nih.gov