Guideline Representation Formalism and Electronic Decision Support Systems: Addressing the Guideline – Implementation Gap - Part 2

• Tri-Model Guideline Representation Architecture
  • The Structured Document Model
  • The Guideline Knowledge Element Model
  • The Guideline Execution Model
• From Narrative to Executable Guidelines
• Conclusion
• References

Click here for Part 1 of this paper

Tri-Model Guideline Representation Architecture
The preferred format/medium for guideline representation by domain experts appears to be natural language. The document-centric formalism also allows the capturing of useful additional information about the guidelines including versioning, authoring, sponsoring organisation and quality of evidence, etc.

However, narrative guidelines tend to be semantically complex and inherently ambiguous by nature of the varying levels of knowledge abstraction by experts during their external representation processes. The semantic complexity, ambiguity and low retrievability of guideline knowledge components contribute to their low utilisation by practitioners. One approach to improve retrievability of guideline knowledge components is to format narrative guideline documents into a more structured model. The three-column, one-page guideline structure used by the New Zealand Guideline Group (NZGG) and the XML mark-up document structure used by GEM and CPG-RA represent examples of such efforts. These strategies can improve the readability of narrative guidelines and, hence, knowledge component retrieval. However, they do not necessarily address the semantic complexity and ambiguity problems inherent in this approach or the issue of "guideline representation – comprehension mismatch". For electronic guidelines to be useful, the knowledge and logic components must be structured so that the semantic relationships between the knowledge components and the execution logic/workflow are clearly defined.

There is a clear need for both formalisms to co-exist so that these guidelines, which are very expensive to develop, can be made fit for both human and machine consumption. This author proposes a tri-model architecture for clinical guideline representation that aims at accommodating document-centric and improved (execution) model-centric formalisms.

The Structured Document Model
A structured document format aims to provide a standard architecture for structuring narrative clinical guidelines as standard XML documents to support clinical guideline authoring, review, update and access via standard XML editors and web browser tools. The greatest benefit is the human readability of the guideline contents and computer usability of the structured contents. The lack of agreement between the GEM and CPG-RA project teams makes it unlikely that a standard narrative guideline representation model will emerge soon. The international community may be forced to make a difficult decision. Given the track record and the openness of GEM, it may be most productive to resolve GEM’s technical issues. The usability and technical merits (of GEM) will ultimately help select a winner between the two competing architectures.

The Guideline Knowledge Element Model
The knowledge elements embedded in narrative guidelines consist of a set of decision variables (or critical determinants) and actions that can be organised into plans and subplans that unfold over time. The guideline knowledge element components are concerned with clear definition of the decision variables and the actions triggered. They also define the linkage structures that organise the action variables in sequence, in parallel, in iterative or cyclic and nesting structures.

The decision variables and actions should be mapped to standard terminology with the meaning and data type clearly defined. This is necessary to ensure that data from electronic health or medical records can be accurately evaluated against the guideline knowledge element concepts by decision support systems.

Most of the model-centric formalisms reported in international literatures have similar formats. A degree of cross-format standardisation may be possible, especially in those areas (eg, action plans) with common components and component structures.

The Guideline Execution Model
The guideline execution model defines the decision rules and flow control rules that control the entry points (and re-entry) into guidelines, especially in multi-encounter care processes characteristic of chronic disease management. These rules also govern the action plans execution (eg, sequence, parallel, iterations) based on workflow management logic, patient status (pre-conditions as decision criteria) and outcome goals (post condition evaluations).

The decoupling of guideline knowledge elements from the execution rules has the advantage of improving the flexibility, reusability and maintainability of the knowledge element and execution logic components of computer executable clinical guidelines.

From Narrative to Executable Guidelines
Most clinical guidelines exist in narrative format. To improve their maintainability and accessibility, they need to be transformed into structured XML documents. GEM is a reasonable architecture. The transformation processes involve conceptualising and organising knowledge concepts contained in narrative guidelines into a hierarchy of elements. Based on a guideline mark-up architecture such as the GEM, the hierarchy of knowledge concepts and guideline texts are marked up with XML tags and converted into a structured XML document format. The NZGG and many professional colleges have developed many narrative practice guidelines that can benefit from such transformations.

For narrative guidelines to be transformed into computable structures, the semantic complexity and, in particularly, the ambiguity problems must be adequately resolved. A number of steps are required:

• conceptual analysis and extraction of the structured knowledge concepts/elements from the structured guideline documents
• classification of the knowledge elements as decision variables and actions:
• definitions of each knowledge element identified, together with their relationships; also development of definitions of the data type for each element (eg, according to the HL7 data types)
• mapping of these to standard vocabularies
• mapping of the guideline knowledge elements and decision rules to knowledge architectures (eg, knowledge element model and guideline execution model) that can be used by guideline-based decision support systems.

The guideline elements extraction, definition and mapping processes force guideline developers to examine their recommendations from a fresh (stepwise reasoning) perspective and validate the recommendations for internal corrections (ie, errors in logical flows such as incompleteness and ambiguity). They are useful processes for resolving guideline ambiguity and converting semantically complex propositions into a set of simple and clearly stated propositions that minimise the need for generating inferences by users (eg, by "filling" the gaps of missing information in text-based guidelines), hence minimising variation in interpretation.

Clinical use cases are extremely useful tools in the disambiguation and validation processes. Each use case consists of a set of clinical management decision scenarios for a specific clinical problem. The scenarios capture the pre-conditions (decision variables), the management events triggered, the post-conditions (desired goals) and the flow of the events. These use cases are developed and refined by clinical domain experts. They can be used to validate the narrative guidelines as they are being developed. By comparing the events described in the clinical scenarios with the propositions (knowledge elements and their semantic relations), domain experts can resolve any ambiguity or missing components during and after the guideline transformation processes. The use cases can also serve as the precursors and catalysts to scenario-based authoring of executable guideline by domain experts. With the support of appropriate authoring tools, scenario-based guideline authoring enables domain experts to integrate clinical experiences appropriate to specific clinical problems (context-specific) with evidence-based knowledge during executable guideline authoring processes. The transformation and validation procedures are executed directly by domain experts, thus minimising the risks of inaccuracy and knowledge loss during the transformation processes. Scenario-based guideline authoring methodology will be reported on in a separate paper.

It will take considerable time for the informatics community to complete the cross-format harmonisation and standardisation. Until a standardised guideline representation architecture is available, the PREDICT^TM rule-based architecture developed by Enigma Publishing Ltd (a private provider of online health knowledge systems) can provide a useful output platform for knowledge elements extracted from structured guideline documents like those marked up according to the GEM architecture.

Conclusion
Evidence-based clinical guidelines are extremely useful and important tools for enabling the delivery of quality health care services and in disease management programs. They are expensive to develop and require regular updating as large amounts of new biomedical knowledge and evidence are discovered each day. Electronic clinical guidelines provide the flexibility of allowing easy review, validation, updates by domain experts and distribution to clinicians. However, for these guidelines to exert optimal positive impact on clinical practices, they need to be integrated seamlessly into clinical workflow and decision support systems. Such needs have become the drivers of numerous projects that produce many electronic guideline representation architectures. Products of these efforts include Arden Syntax, PROforma, GLIF, Asbru, Asgaard, EON, CommanKADS, OCML, Protégé, UOML, GEM, CPG-RA and many more.

The informatics community faces a number of problems and challenges in clinical guideline development and representation. Inexact mapping from experts’ internal knowledge to external representations as narratives and then to machine-executable formats means that the external representations can be plagued with ambiguities. Further distortions can arise as clinical users attempt to "fill in the gaps" with their own interpretations of the guidelines. Significant variations exist in all guideline representation (narrative and executable) formalisms. These variations create problems for sharing of guideline knowledge and also for developers of evidence-based decision support systems.

This author proposes a tri-model approach to guideline representation architecture. Decoupling guideline knowledge elements from the execution rules allows better flexibility, reusability and maintainability of the knowledge element and execution logic components of computer executable clinical guidelines. Direct scenario-based guideline authoring by clinical domain experts presents an excellent methodology to resolve the ambiguity and "knowledge loss" problems in guideline representation processes. The HL7 clinical guideline and decision support technical workgroup needs to initiate standardisation of multiple guideline representation formalisms as a matter of high priority and urgency.

References

1. Kohn LT, Corrigan JM, Donaldson MS (eds). To err is human: building a safer health system. Washington, DC: National Academy Press; 2000. 
2. Institute of Medicine. Crossing the quality chasm: a new health system for the 21st century. Washington, DC: National Academy Press; 2001.
3. Allen VG, Arocha JF, Patel VL. Evaluating evidence against diagnostic hypothesis in clinical decision making by students, residents and physicians. Int J Med Inform. 1998;51:91-105.
4. Ohno-Machado L, Gennari JH, Murphy SN, Jain NL, Tu SW, Oliver DE, Pattison-Gordon E, Greenes RA, Shortliffe EH, Barnett ;. The guideline interchange format: A model for representing guideline. J Am Med Inform Assoc. 1998;5(4):357-372.
5. Grimshaw JM, Russell IT. Effect of clinical guidelines on medical practice: a systematic review of rigorous evaluation. Lancet. 1993;342(8883):1317-1322.
6. Lomas J, Anderson GM, Dominic-Pierre K, Vaida D, Enkin MW. Do practice guidelines guide practice? The effect of a consensus statement on the practice of physicians. N Engl J Med. 1989;321;1306-1011.
7. McAlister FA, Teo KK, Lewanczuk RZ, Wells G, Montague TJ. Contemporary practice patterns on the management of newly diagnosed hypertension. Can Med Assoc J. 1997;157(1):23-30.
8. Tunis SR, Hayward RSA, Wilson MC. Internists’ attitudes about clinical practice guidelines. Ann Intern Med. 1994;120(11):956-963.
9. Peleg M, Tu S, Bury J, Ciccarese P, Fox J, Greenes RA, Hall R, et al. Comparing computer-interpretable guideline models: a case-study approach. J Am Med Inform Assoc. 2003;10(1): 52-68.
10. Shiffman RN, Karras BT, Agrwal A, Chen R, Marenco L, Nath S. GEM: a proposal for a more comprehensive guideline document model using XML. J Am Med Inform Assoc. 2000;7(5):488-498.
11. Patel VL, Arocha JF, Diermeier M, Greenes RA, Shortliffe EH. Methods of cognitive analysis to support the design and evaluation of biomedical systems: the case of clinical practice guidelines. J Biomed Inform. 2001:1-15.
12. Patel VL, Arocha JF, Diermeier M, How J, Mottur-Pilson C. Cognitive psychological studies of representation and use of clinical practice guidelines. Int J Med Inform. 2001;63:147-167.
13. Quaglini S, Stefanelli M, Lanzsola G, Caporusso, V, Panzarasa S. Flexible guideline-based patient careflow systems. Artif Intell Med. 2001;21:65-80.
14. de Clercq PA, Hasman A, Blom JA, Korsten H. The application of ontologies and problem-solving methods for the development of sharable guidelines. Artif Intell Med. 2001;22: 1-11.
15. Svatek V, Ruzicka M. Step-by-step mark-up of medical guideline documents. Int J Med Inform. 2003;70:329-335.
16. Sutton DR, Fox J. The syntax and semantics of the PROforma guideline modelling language. J Am Med Inform Assoc. 2003;10(5):433-443.
17. Musen MA. Modern architectures for intelligent systems: reusable ontologies and problem solving methods. Proceedings, AMIA Annual Symposium; 1998:46-52.
18. Peleg M, Boxwala AA, Bernstam E, Tu S, Greenes RA, Shortliffe EH. Sharable representation of clinical guidelines in GLIF: relationship to the Arden Syntax. J Biomed Inform. 2001:34:3170-3181.
19. Fox J, Johns N, Rahmanzadeh A. Disseminating medical knowledge: the PROforma approach. Artif Intell Med. 1998;14:157-181.
20. Shahar Y, Miksch S, Johnson P. The ASGAARD project: a task-specific framework for the application and critiquing of time-oriented clinical guidelines. Artif Intell Med. 1998;14:29-51.
21. Musen MA, Gennari JH, Erikson H, Tu SW, Puerta AR. Protégé II: computer support for development of intelligent systems from libraries of components. Proceedings, MedInfo 95: The 8th World Congress on Medical Informatics; 1995:766-770.
22. Gruber TR. A translation approach to portable ontologies. Knowledge Acquisition 1993;5(2): 199-220.
23. Schreiber AT, Wielings BJ, de Hoog R, Akkermans H, van der Velder W, Anjewierden A. A CommanKADS: a comprehensive methodology for KBS development. IEEE Expert 1994:28-37.
24. Motta E . Reusable components for knowledge modelling. In: Case studies in parametric design problem solving. Amsterdam: IOS Press; 1999.
25. Fensel D, Benjamins VR, Motta E, Wielinga B. UPML: a framework for knowledge system reuse. IJCAI-99;1999: 66-79 .
26. de Clercq PA. An ontological approach for the development of shareable guidelines. Proceedings, AMIA Symposium; 2000:231-235.
27. Agrawal A, Shiffman R. Using GEM-encoded guidelines to generate medical logic modules. J Am Med Inform Assoc. 2001; 8:489-499.

Click here for Part 1 of this paper