Abstract
RuleEd is a web-based editing environment which enables clinical trial eligibilty rules entered as free text to be represented as a series of terms mapped to unique concepts in a controlled vocabulary. RuleEd provides interfaces for creating and refining concept mappings for terms within rules and disambiguating multiply-mapped terms. A combination of interactive and non-interactive methods enables authors to specify eligibilty rule representations with a fine level of control.
Background
The Trial Bank Project has demonstrated that data from randomized clinical trials (RCTs) can be captured as computable representations in specialized knowledge models [1]. Eligibility rules are essential components of RCTs that have not been represented in a generic, consistently computable way. Representation problems include mapping terms within individual eligibility rules to concepts in a controlled clinical vocabulary and capturing intended relationships between concepts and their modifiers.
Previous approaches for parsing comparable segments of medical text, such as MetaMap [2], SemRep [3] and others, have largely focused on non-interactive techniques such as grammar analysis, part-of-speech tagging, context-based concept disambiguation and combinations thereof. None have so far consistently achieved levels of precision that are acceptable for representing eligibility rules in RCTs.
Method
When RuleEd is opened (rctbank.ucsf.edu:8081/BaT/RuleEd.html), the user is presented with a text-input box to enter an eligibility rule as free text. RuleEd then seeks out the longest phrases in the submitted rule that can be mapped to UMLS concepts. The mapped and unmapped rule elements are internally represented as an S-expression, which is displayed as a series of HTML input objects: mapped terms are designated by labelled HTML submit-buttons, unmapped phrases are designated by HTML text-input boxes, while semantic relations and boolean connectors are designated by HTML select-choice lists (Figure 1).
Figure 1.
An HTML representation of a parsed rule
Clicking on a submit-button that designates a uniquely mapped term displays that term’s semantic neighbors in the UMLS hierarchy (i.e. parent, child, sibling, etc). Selecting a semantic neighbor from this interface immediately updates the parsed eligibilty rule and shifts the local context view of the network to the semantic neighborhood of the selected concept.
Ambiguous terms in the parsed rule are represented as labelled submit-buttons with an appended (n) notation, where n is the number of concepts literally mapped to that term string (Figure 1). Clicking an ambiguously mapped term invokes a concept dismbiguation interface that presents a list of the UMLS preferred terms for each concept that shares a literal mapping to the ambiguous term. Selecting a concept from this list causes the ambiguous term to be replaced with the selected term and then redisplayed without an (n) notation.
Discussion
RuleEd provides interactive methods for those parsing subtasks at which humans excel, such as concept disambiguation, thus enabling precise control over the specification of concept mappings for terms within rules, including their level of generality.
A concept may be specified more broadly by selecting one of its parent concepts, which causes the local context view to be moved upward in the UMLS hierarchy. Selecting a child concept gives a narrower specification and moves the local context view downward in the hierarchy.
RuleEd provides a powerful set of tools for capturing free text eligibility rules in computable form.
References
- 1.Sim I, Olasov B, Carini S. An ontology of randomized controlled trials for evidence-based practice: content specification and evaluation using the competency decomposition method. J Biomed Inform. 2004;37(2):108–19. doi: 10.1016/j.jbi.2004.03.001. [DOI] [PubMed] [Google Scholar]
- 2.Aronson AR. The MetaMap Program. 2001. Effective Mapping of Biomedical Text to the UMLS Metathesaurus. [PMC free article] [PubMed] [Google Scholar]
- 3.Tringali M, Rindflesch TC, Kilicoglu H, Fiszman M, Bodenreider O. Strategies for Mapping Concepts in Gastrointestinal Endoscopy Reports to the UMLS Metathesaurus. 2004. [Google Scholar]