Ontology Outreach Advisory

General Description
The Protocure II project (www.protocure.org, [Protocure-II]) has been a multidisciplinary project (2004 – 2006) bringing together medical doctors and computer scientists and aimed at improving the quality of computerized guidelines and funded by the European Commission under contract number IST-FP6-508794. The project addressed the topic of quality improvement of clinical guidelines and protocols by integrating formal methods of software engineering in the life cycle of guidelines development and maintenance.

The problem
Medical guidelines and protocols describe the optimal care for a specific group of patients and are meant to improve the quality of patient care by providing better patient informedness and lower inter-practician variability. It has been proved that adherence to guidelines and protocols may reduce health-care costs up to a 25%. The effort spent on developing and disseminating the rather high number of medical guidelines and protocols developed within the last decade far outweighs the efforts on guaranteeing their quality. Indeed, anomalies like ambiguity and incompleteness are frequent in medical guidelines and protocols. Recent efforts have tried to address the problem of quality improvement. These approaches are not sufficient since they rely on informal processes and notations. As a result, many guidelines and protocols in practical use are still ambiguous or incomplete. A different approach, grounded on a formal representation, can answer these needs, as demonstrated in the Protocure I project. The approach taken in Protocure I views medical protocols as computer programs, and formal methods as means to assess and improve their quality. The Protocure II project extends this metaphor to guidelines and makes the analogy between “medical protocol development” and Software Engineering. Thus, Protocure II is aimed at integrating formal methods in the life cycle of guidelines, by developing techniques & tools to support the whole guideline development process. An important part of this effort is dedicated to defining re-usable relations and concepts to build a part of a model of a medical guideline, which can be applied to several guidelines.

The Solution
The research methods used to achieve the project goals are as follows:
1) Study the current guideline development process and augment it to support the use of formal methods and tools for assisting high-quality and up-to-date guidelines (living guidelines);
2) Develop a methodology and tools for supporting the transition from narrative to formal representation of guidelines;
3) Define a library of guideline components, in the form of re-usable patterns
4) Adapt and apply theorem proving and model checking techniques to the formal representation.

To address steps 2 and 3 mentioned above, the problem of linguistic guideline templates and the role of medical ontologies in identifying and maintaining these templates has been studied.

The text of medical guidelines often has a modular structure that is suitable for automatic translation into a formal representation. Patterns include:
•In the event of [MedicalContext] the treatment of choice is [Treatment].
•In the event of [MedicalContext], [Treatment] is recommended.
•[MedicalTargetGroup] [recommendation_op] receive [Treatment] with [MedicalGoal].
Such linguistic templates can be generated and instantiated with the use of an ontology of the medical domain. Producing meaningful linguistic pattern templates and translating them into a formal representation cannot be fully automated, but partial automatable steps can reduce greatly the effort spent in building and maintaining a model of a medical guideline.

To identify compositional linguistic patterns, one can learn an ontology of guideline terms by extracting key phrases from guidelines and assigning them to vocabulary categories (selected as a subset of semantic categories from existing medical thesauri). With the help of a guideline domain ontology that takes into account the guideline components (e.g., action, condition, goal, effect), medical knowledge, and relationships, the terminology specific to a specific disease can be decoupled from the terminology specific to the guideline.

The guidelines contain basically two kinds of pattern templates:
•templates that describe medical background knowledge;
•templates that describe control knowledge.
For the objectives of the Protocure II project, aimed at producing an operational model of a guideline, we have identified action-centric patterns which belong to the following classes:
•Effects of actions
•Associations action-goal
•Recommendations for specific actions for target groups
•Preferences for a specific medical intervention
•Background knowledge-centered patterns
•Associations disease-treatment.
•Associations disease-target group
•Combinations of actions, focused on hierarchical decomposition, sequencing and temporal relations between actions
•Action Sequencing
•Patterns for more complex action sequencing
•Patterns for high-level coordination of actions
•Search for the right dosage of medication
We have evaluated the usefulness of these patterns in guideline formalization, by performing the following steps: (1) a rough comparison (quantitative) of the amount of knowledge (automatically) identified by using patterns with respect to the knowledge modelled by (manual) knowledge acquisition; (2) an analysis (qualitative) of the utility of identified pattern instances, performed on specific fragments of the guideline; and (3) in connection with the two previous steps, an identification of the essential knowledge elements that current patterns overlook.

The conclusion was that the most useful patterns are the control patterns, which are the closest to the implementation and include:
•decomposition, ordering, and repetition, or describe constraints such as the fact that an action may be associated with a time frame, intention and medical effect;
•patterns produced by knowledge engineers to describe relations between medical terms.

Key Benefits of Using Semantic Web Technology
Existing medical ontologies/thesauri, such as [MESH], [UMLS], [NCIOntology], already contain a lot of background medical information that can be used when building a domain model for the medical guideline and for verifying the consistencies of the relations defined in a clinical guideline or protocol.
More than 15 semantic categories used in the UMLS thesaurus are included in a mini-ontology for the medical domain used in the linguistic template retrieval application used for the Protocure II experiments. By establishing mappings between existing medical thesauri and a custom-built guideline ontology, more medical background information can be extracted from existing medical knowledge sources and incorporated in a guideline model, reducing the risk of inconsistencies and the effort of (re)defining relations between medical terms ([AIIMJ2006, FCTC2006]).

References
[Protocure-II]Protocure 2 Project. Integrating formal methods in the development process of medical guidelines and protocols. URL: http://www.protocure.org. Accessed: 1 June 2006.
[AIIMJ2006]R. Serban, A. ten Teije, F. van Harmelen, M. Marcos, C. Polo-Conde. Extraction and Use of Linguistic Patterns for Modelling Medical Guidelines. Artificial Intelligence in Medicine Journal, Volume 39, Issue 2, Pp. 137-149, 2006.
[FCTC2006]Radu Serban, Annette ten Teije., Formalization of medical guidelines exploiting medical thesauri. Proceedings of the Workshop on Foundations of Clinical Terminologies and Classifications (FCTC 2006), Timisoara, Romania.
[MESH]Mesh (Medical Subject Headings (MESH)). URL:http://www.nlm.nih.gov/mesh/meshhome.html. Accessed: 1 June 2006.
[UMLS]Unified Medical Language System (UMLS). URL: http://www.nlm.nih.gov/research/umls/. Accessed: 1 June 2006.
[NCIOntology]National Cancer Institute (NCI) Ontology. URL: http://www.mindswap.org/2003/CancerOntology/. Accessed: 1 June 2006.

See Other Use Cases