The PROforma guideline specification language: progress and prospects

Bury, J., Fox, J., Sutton, D.

Advanced Computation Laboratory, Imperial Cancer Research Fund, London, UK.

{jb,jf,ds}@acl.icnet.uk

Abstract. Medical guidelines are constructed with the aim of assisting clinicians in making decisions that are informed by the best available medical evidence. In order to achieve this aim, they must be disseminated in a form that makes them easy for clinicians to use and easy for domain experts to critique. Furthermore, the language

in which they are expressed must facilitate their transfer between institutions and their adaptation to local conditions. This paper describes PROforma, a knowledge representation language that attempts to meet these desiderata. PROforma is formal knowledge representation language designed to capture the content and structure of a clinical guideline in a form that can be interpreted by a computer. The language embodies many contemporary themes in machine interpretable guideline representation schemas whilst retaining some distinctive features. This paper describes the key features of the PROforma language in the context of recent trends and developments in the field of electronic guideline representation formats. We describe our experiences in applying PROforma to a range of clinical decision making and workflow problems, and the benefits and limitations of the current language specification are discussed. Finally, we outline plans for the further refinement of PROforma.

Keywords :

Decision support systems, Clinical guidelines, Knowledge engineering, PROforma.

1 Introduction

This paper describes the key features of the PROforma language in the context of recent trends and developments in the field of electronic guideline representation formats. The structure of the paper is as follows: Section 2 outlines the problems that clinical guidelines address and section 3 explains the paradigms that have arisen for the expression of such guidelines. Section 4 describes the theoretical basis of PROforma.

2 Guidelines and workflow

The explosion of interest in developing and publishing clinical guidelines, both using conventional media and on the World Wide Web, is an indicator of the growing importance attached to clinical guidelines in medicine. Kohn et al.[1] describe the recent estimate from the US Institute of Medicine that there may be as many as 98,000 unnecessary deaths per annum resulting from avoidable clinical error, which suggests there is a genuine need for such guidelines. The vast majority of clinical practice guidelines are published as text, and typically include criteria describing their applicability to particular groups of patients, the recommended processes of care, appropriate use of materials and procedures and so on, as well as providing ancillary information such as supporting evidence. Clinical guidelines are generally concerned with two main areas that directly affect the quality and effectiveness of patient care: the quality of clinical decision-making, and the correct and timely management of clinical tasks. A clinical guideline thus incorporates information relating to both decision-making and workflow management.

2.1 Decision-making: What to believe and what to do

Decision theory describes two main classes of decision: those concerning what to believe (e.g. what is wrong with a patient, how serious a disease is, the possible prognosis) and those concerning what to do (e.g. whether or not a patient should be referred for specialist care, what treatment or treatment strategy is appropriate, whether tests or investigations are required). Both of these classes of decision are vulnerable to many kinds of error. Errors in diagnosis or prognosis decisions can result from the absence of critical information about a patient, or difficulties in assessing relative strengths of evidence for example. Errors in treatment and other management actions can result from incorrect decisions about eligibility or failures to allow for possible drug interactions, etc.

2.2 Task management: Who does what, when, how and where

Even assuming that all clinical decisions are taken correctly, there are still countless ways in which errors can be made, and which can be costly in terms of the efficacy or efficiency of patient care. Tasks may be inadvertently omitted, carried out too late or adverse events may be missed. More complex clinical procedures such as care pathways or chemotherapy protocols may be carried out over an extended timescale by multiple healthcare providers, resulting in further potential hazards e.g. through a failure to keep sufficiently complete clinical records or ensure that staff in the care team are kept informed about actions that have been carried out or are planned.

2.3 Delivering Clinical Guidelines

While the systematic preparation and publication of clinical guidelines is of great importance, it is only a first step towards ensuring consistent compliance with the standards of care represented in those guidelines [2]. A guideline may set out current best-practice in great detail, but if clinicians do not have time to access the guidelines, or fully absorb their content and apply it correctly in their decision-making, or the clinical organisation fails to ensure that all the required tasks are carried out in a timely manner, then the objectives of the guideline may not be reflected in clinical outcomes.

From an AI perspective there is an obvious alternative to publishing guidelines solely in human readable form such as text, tables or flow diagrams, which is to formalize the medical knowledge contained in these guidelines in a form that a computer can apply to support clinicians in their routine work. The PROforma guideline specification language and its associated software attempt to provide such a formal framework. In this paper we describe the PROforma language in the context of recent trends in guideline representation, review the experience we have gained in applying PROforma technology in a variety of clinical domains, and present our plans for the future development of PROforma.

3 Paradigms for representing clinical guidelines and workflow

3.1 The procedural approach

Early computerised clinical support systems, including decision support systems such as statistical decision aids and clinical algorithms, whether expressed as conventional programs or in visual notations such as flow diagrams, did not separate the clinical knowledge from the computation details of how that knowledge was to be delivered.

Although such systems can be satisfactory, many limitations of the procedural approach have been identified in the AI and knowledge engineering literature. First of all users, particularly non-programmers, find it difficult to comprehend the underlying clinical logic by simple inspection of a computer program. Second, maintaining the guideline software as medical knowledge increases is notoriously difficult, and such systems lack reusability - the knowledge they embody cannot be readily exported to other systems in a modular form. As a consequence it is increasingly recognised that a better way of representing medical knowledge is in declarative form, which allows knowledge of medical concepts such as diseases, drugs and tests to be clearly separated from the reasoning and problem-solving processes that use that knowledge in particular medical contexts.

3.2 Rule based systems

Rule based expert systems such as MYCIN [3] and Oncocin [4] reflected the recognition of the importance of declarative knowledge representation and pioneered a significant change in the development of clinical decision support systems in the 1970s and ‘80s. Such systems clearly separated domain-specific knowledge expressed as sets of rules from the generalised inference engine, typically a backward-chaining or forward-chaining system, which would apply those rules. This represented a paradigm shift in the construction of decision support systems, and promised improved readability, modularity and reusability of the knowledge bases. However, despite the popularity and technical elegance of rule based systems, and the enhanced readability they have brought, they have been less successful in achieving modularity and reusability of knowledge. In practice rules can be written in a procedural way and frequently depend upon being carefully crafted to ensure that they are only applied in specific situations making their transfer to other applications difficult [5] and potentially hazardous, as demonstrated in studies with the Arden syntax [6], which can be viewed as a hybrid of simple rule-based and more traditional procedural methods.

3.3 Task based systems

The limitations of rule-based systems have led to the recent emergence of a further paradigm shift in the development of clinical knowledge systems. Here, rule-based and task-based formalisms are combined to represent clinical processes. How a particular logical condition should be interpreted and acted upon depends, for example, on whether that rule represents part of a diagnostic process, a precondition for a particular clinical intervention, or a contraindication to a particular drug therapy. Task-based systems therefore attempt to contextualise rules within explicit and intuitive models of the clinical process with the aim of combining the programmatic richness of a procedural representation with the logical clarity of declaratively expressed knowledge.

Use of an intuitive representation of the clinical process has at least two potential advantages. Firstly, building the knowledge base should be an easier process for domain experts to participate in, given that tasks provide a higher-level and more intuitive set of design primitives than conventional programming languages or languages based on rules. Secondly, systems built around this approach have the potential to integrate into the clinical environment and workflow more easily, based as they are around primitives which correlate with real world concepts and processes familiar to users. PROforma [7] is one example of a task-based guideline representation format. Other schemas that embody some or all of the features of the paradigm include the Guideline Interchange Format (GLIF), developed by the Intermed Collaboratory [8], SMART [9], Eon [10], and the Asgaard Project [11].

4 PROforma – a task based approach to guideline representation

The term “Clinical Guidelines” describes a heterogeneous range of textual material designed to support practitioners in their decision-making. Individual guidelines may describe interventions concerning populations or individuals. These interventions may be simple “atomic” actions or complex plans of therapy to be conducted over time. The decisions they describe include diagnostic, prognostic, therapeutic and risk assessment decisions. Unlike clinical protocols, which are intended to be followed rigidly, usually within special contexts such as clinical trials, clinical guidelines often outline general principles of management, to be considered by experienced staff as they exercise their clinical judgement. Guidelines therefore frequently contain ancillary information such as the evidence and literature on which their recommendations are based, guideline authorship, and the statements of the applicability of the material presented to particular patient groups.

When work began on PROforma in the early 1990s, the main requirements for the language were that it should:

·  Be sufficiently expressive to fully represent a range of clinical processes

·  Be sufficiently general to describe processes in any clinical specialty, from routine care to clinical research

·  Use concepts that are intuitive for clinical users whilst ensuring that

·  Processes specified in the language can be enacted by machine

·  The semantics of the language are demonstrably sound

·  Applications can be automatically checked for consistency and other properties PROforma combines the features of a formal specification language as developed in software engineering with the features of knowledge representation languages as developed in artificial intelligence. The PROforma language structure is based on a simple but versatile clinical process model referred to as the domino model. This model was abstracted from a variety of empirical studies of clinical decision-making and the development of aids to support patient management. The logical form of the domino model is shown in figure 1.

As described earlier, decision-making is a core function of any guideline representation and enactment schema. Statistical decision theory is the most formal decision model but in many respects has proved difficult to use and unattractive to clinicians. Decision theory based on mathematical probabilities is sound and well understood, but decision support systems based on statistical decision theory are often inflexible and difficult to use, and of limited applicability to knowledge-rich domains such as healthcare where precise parameters like quantitative probabilities are seldom known.

Discussion and conclusions

We believe that the task based paradigm represents a promising solution to the problem of developing guideline representation formats that maintain a declarative knowledge base whilst reflecting the constraints of the clinical workflow and process within which that knowledge is to be applied. We have found the particular task model adopted in PROforma to be both expressive and intuitive, with its simplicity and formal foundations being significant benefits. Although the development of the PROforma task model appears to have been productive, we must now address a second set of challenges. The technical challenges of providing interfaces between decision support systems and other clinical information systems such as electronic patient records and laboratory systems are well known. Within the context of the PROforma 2000 project, the definition of a clear API for PROforma technologies will be a first step towards supporting communication between PROforma and other systems. We also hope that the adoption of techniques from the agent-based paradigm will help to address the difficulties of communication and negotiation between guidelines, and facilitate the development of clearer behavioural semantics for goal-based guideline representation.

The development of more intelligent functions, such as plan repair and the dynamic modification of plans represents a longer term objective for the PROforma format. These features will require a deeper understanding of general cognitive functions and how thesecan be modelled in formal systems. A further challenge in providing decision support at the point of care is in integrating computer systems into the dynamic of the clinical consultation. We are exploring the notion of “use models”, in which we consider the impact of a computer on the complex discourse between patient and clinician. Introducing a computer as a “third party” in this discourse may have significant effects on the psychological aspects of the consultation, both for clinicians and patients. If the decision support systems we wish to provide are to be integrated into clinical practice, their design must support the subtleties of the human relationship that exists between patients and their clinicians so that they may enhance, rather than detract from, this relationship.

Acknowledgements

We thank Richard Thompson, Ali Rahmanzadeh, and Christian Blum for their help in preparation of this manuscript.

[1] Kohn, L., Corrigan, J., Donaldson, M. (Eds). To Err is Human: Building a Safer Health System. National Academy Press, Washington, D. C., 1999.
[2] Shiffman RN, Brandt CA, Liaw Y, Corb GJ, 1999. A design model for computer-based guideline implementation based on information management services. J Am Med Inform Assoc 1999; 6(2):99-103
[3] Wraith SM, Aikins JS, Buchanan BG, Clancey WJ, Davis R, Fagan LM, Hannigan JF, Scott AC, Shortliffe EH, van Melle WJ, Yu VL, Axline SG, Cohen SN. Computerized consultation system for selection of antimicrobial therapy. Am J Hosp Pharm 1976; 33(12):1304-8
[4] Shortliffe EH. Update on ONCOCIN: a chemotherapy advisor for clinical oncology. Med. Inform. 1986; 11(1):19-21
[5] Shwe M, Sujansky W, Middleton B. Reuse of knowledge represented in the Arden syntax. Proc Annu Symp Comput Appl Med Care 1992; 47-51
[6] Hripcsak G, Wigertz OB, Kahn MG, Clayton PD, Pryor TA. ASTM E31.15 on health knowledge representation: the Arden Syntax. Stud Health Technol Inform 1993;6:105-12
[7] Fox J, Johns N, Rahmanzadeh A, Thompson R. PROforma: A method and language for specifying clinical guidelines and protocols. In: J Brender, J P Christensen, J-R Scherrer P McNair (eds) Medical Informatics Europe '96 IOS Press, Ams terdam 1996 pp. 516-520.
[8] Ohno-Machado L, Gennari JH, Murphy SN, Jain NL, Tu SW, Oliver DE, Pattison-Gordon E, Greenes RA, Shortliffe EH, Barnett GO. The guideline interchange format: a model for representing guidelines. J Am Med Inform Assoc 1998; 5(4):357-72
[9] Pisanelli DM, Consorti F, Merialdo P , SMART: a system supporting medical activities in real-time. Stud Health Technol Inform 1997; 43 Pt A:343-7
[10] Tu SW, Musen MA. A Flexible Approach to Guideline Modeling. AMIA 1999 Annual Symposium, Washington D.C., 420-424. 1999.
[11] 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(1-2):29-51