Elsevier

Knowledge-Based Systems

Volume 167, 1 March 2019, Pages 98-113
Knowledge-Based Systems

A lightweight acquisition of expert rules for interoperable clinical decision support systems

https://doi.org/10.1016/j.knosys.2019.01.007Get rights and content

Abstract

Background:

The process of adding new knowledge in the form of rules to already running Clinical Decision Support Systems (CDSSs) in hospitals is extremely costly and time consuming. There are two principal limitations: (1) the lack of a broad consensus regarding a uniform representation of clinical rules; and (2) the integration of new rule-based knowledge into hospital information systems.

Objective:

To provide a guideline with which to support knowledge acquisition for rule-based CDSSs and to facilitate the integration of that knowledge into hospital datasets using standard clinical terminologies and ontologies as reference elements.

Materials and Methods:

We have designed a straightforward 4-step methodology with which to incorporate the external knowledge sources and data integration required to run CDSSs in hospitals. This lightweight methodology is based on a reference ontology that integrates standard clinical terminologies and its objective is to effectively acquire procedural knowledge in the form of rules.

Results:

We have applied the methodology in the context of antimicrobial stewardship at a hospital. Recommendations from the European Committee on Antimicrobial Susceptibility Testing (EUCAST) were added to WASPSS, a CDSS running at the hospital. The reference ontology combines a subset of ATC terminologies for antibiotics and those of NCBI for microorganisms, including 584 and 1714 concepts, respectively. A total of 94 new rules were added to the CDSS so as to represent EUCAST knowledge. We also evaluated different implementations in order to study their scalability, during which time we analysed Drools 7.5 as a production rule engine, HermiT as an ontology reasoner and RuQAR as an integration tool. Our experiments show that the combination of a production rule engine and an ontology reasoner in runtime is more efficient than using a single rule engine with a knowledge base derived from the reference ontology (1.9 times faster than the next approach when executing 1000 expert rules on an ontology of 1000 concepts).

Discussion:

The methodology proposed helped to implement the knowledge acquisition process of EUCAST rules in a running CDSS. This methodology is applicable to other clinical domains when knowledge can be modelled with rules. Since it is a lightweight methodology, different implementation strategies are possible. The use of clinical standards also facilitates the future interoperation between CDSSs, particularly when using SNOMED as a reference ontology and employing future rule-sharing standards.

Introduction

Modern clinical decision support systems (CDSS) have the objective of providing healthcare professionals with relevant knowledge [1], [2]. The CDSSs running in clinical institutions are currently having a growing impact on patients’ healthcare and these systems are, therefore, under strict supervision and are strongly constrained. For example, if a CDSS is part of the clinical activity flow, it must be integrated into the hospital’s Health Information System and other databases following local regulations. These technical requirements signify that CDSS decisions are often computed using the production rules paradigm, which is considered an efficient, scalable and mature technology.

CDSSs are, at present, specifically adapted to the requirements of each hospital. However, the acquisition of new knowledge may have a positive effect on the quality of the system outcome. Indeed, physicians in daily practice share strategies and protocols published by high quality-tested recommenders, such as international healthcare institutions or national health systems. For example, European health institutions have recently published a catalogue of rules (EUCAST expert rules) in order to assist microbiologists during the tests carried out to evaluate the clinical success of an antibiotic against an infection [3]. Examples of these kinds of rules are (Rule 1.3) “IF the microorganism belongs to the Enterobacter cloacae species, THEN report as resistant to Amoxicillin-clavulanate,” or (Rule 13.5) “IF an Enterobacteriaceae is resistant to ciprofloxacin, THEN report as resistant to all fluoroquinolones.

We essentially identify two reasons why new knowledge is acquired by running CDSSs in hospitals: to interoperate with other CDSSs and to incorporate knowledge from specialised literature.

The interoperability between CDSSs, which is understood as the ability of systems to exchange interpretable data, is not a simple issue and, from the computational point of view, knowledge sharing is perhaps the most common bottleneck. In our opinion, there are two general strategies in literature by which to approach the mechanism of knowledge communication between systems: the top-down and the bottom-up strategies.

The first strategy in the process of CDSS interoperation (top-down) is to make all hospitals to share a single ontology, denominated as the global ontology, in order to define concepts and procedural knowledge (such as rules). This is a well-known approach in the Semantic Web community, in which the global ontology provides a homogeneous vocabulary with which to link terms, along with the axioms and relations between them [4]. However, when CDSSs are already in use in hospitals, the top-down approach implies a high-cost design. This approach requires the development of a global ontology in order to cover the terms employed at all hospitals and the redesign of each current knowledge base so as to adapt the rules to this ontology.

The objective of the bottom-up strategy is to minimise changes in the current representation of information in the CDSSs. From the point of view of medical informatics, this approach is considered to be a general interoperability problem. The common solution is to use a standard clinical model (reference model) to enable the use of communication mechanisms between the CDSSs. That is, each system maps its current terminology onto the reference model (e.g. SNOMED) to be shared.

Unlike CDSS interoperability, the acquisition of knowledge extracted from specialised literature does not require the development of technology for information interchange. Traditional approaches require a manual acquisition process carried out by knowledge engineers and validated by specialists. Human intervention is an advantage when different levels of granularity exist between the knowledge available in the sources (literature) and the rules and databases of the CDSS. This approach lacks standard mechanisms. However, the use of automatic or semi-automatic mechanisms could help to save costs and time.

To continue with the EUCAST example, the CDSS knowledge base stores specific treatments (e.g. ciprofloxacin, an antimicrobial) while the rule “IF an Enterobacteriaceae is resistant to ciprofloxacin, THEN report as resistant to all fluoroquinolones” is also defined over more general terms, such as Enterobacteriaceae, which is a large family of bacteria and fluoroquinolones, which is the group of antibiotics to which ciprofloxacin belongs.

Despite their differences, knowledge acquisition and CDSS interoperability approaches have to deal with similar problems: (1) the convenience of using ontologies to model new knowledge; (2) the need to extend the reasoning capacities of CDSS to support this new knowledge; and (3) the need to integrate CDSS reasoning into hospital information systems.

Various methodologies with which to extract and acquire knowledge have appeared in recent medical literature [5], [6] and some proposals suggest the use of theoretical models to interchange knowledge between rules. However, little attention is paid to approaching both problems from a holistic perspective in practical scenarios.

In this work, we propose a straightforward-4-step methodology with which to import knowledge from an external source in a rule-based CDSS. This methodology is suitable for acquiring the knowledge extracted from specialised literature but is also an essential step as regards enabling future CDSS interoperability.

The contributions of this paper are:

  • A review of the essential principles of clinical production rules and the categorisation of recent efforts concerning clinical knowledge representation in perspective (Sections 2 Background, 6 Related work).

  • A novel lightweight methodology based on a reference ontology (REO) to support the acquisition of new knowledge in rule-based CDSSs (Section 3).

  • The evaluation of our proposal, which was carried out by applying the methodology to a running CDSS for the antimicrobial testing problem and the realisation of scalability experiments. This is, to the best of our knowledge, the first methodological approach to incorporate EUCAST rules into a CDSS (Sections 4 Use case: Antibiotic susceptibility testing, 5 Performance experiments).

Section snippets

Background

In this section, we review the essential components of production rule systems, the principal approaches related to clinical rule representation, the technologies employed to integrate production rules and semantic technologies and, finally, current standard clinical terminologies.

Methodology proposed

In this work, we propose a straightforward lightweight methodology with which to acquire new clinical rules for CDSSs from clinical knowledge sources. This methodology is based in a set of assumptions: Firstly, several Health Information Systems (HIS)s are running in different hospitals using different terminologies for a similar clinical field. Secondly, a rule-based CDSS is available for each HIS, or at least it is possible to feed a rule-based system with HIS data to provide clinical

Clinical context: Susceptibility and EUCAST rules

When an infection is diagnosed in hospitals, a sample is taken from the patient and then analysed in the microbiology laboratory in order to determine to which species the microorganism causing the infection belongs. An antimicrobial susceptibility test is also performed: the microorganism is exposed to different concentrations of a set of antimicrobials so as to study its reaction and estimate the outcome of clinical therapies. Depending on the antimicrobial concentration in which the

Performance experiments

One key aspect for a successful acquisition of new knowledge in a CDSS is the scalability of the methodology. The number of rules to implement, the concepts in use and the implementation of the methodology are essential factors for scalability. We therefore analyse and discuss the differences in performance between the approaches described in Step 3 as regards the size of the REO to be queried and the number of expert rules to be executed.

Related work

The adoption of mechanisms with which to exchange knowledge among CDSSs has been widely analysed in recent years. In [52], [53], [54], the models used to acquire knowledge and grant the interoperability of CDSSs are reviewed. Based on such studies, we group the recent research available in literature as: terminologies, interpretable structures, logic specifications, semantic technologies and methodologies.

Clinical terminologies are the essential core of modern CDSSs and the first step required

Conclusion

In this paper, we have studied how knowledge is acquired and transferred in rule-based CDSSs already running in hospitals. We propose a straightforward 4-step methodology based on a reference ontology (REO). The objective of our methodology is to balance the importance placed on supporting the acquisition of new knowledge supported by ontologies and to preserve the current CDSS reasoning architecture and local hospital terminologies. We have successfully applied our approach in WASPSS, a

Acknowledgements

This work was partially funded by the Spanish Ministry of Economy and Competitiveness under the WASPSS project (Ref: TIN2013-45491-R) and its associated predoctoral grant (Ref: BES-2014-070682), and by the European Fund for Regional Development (EFRD, FEDER).

References (69)

  • Ben MessaoudM. et al.

    SemCaDo: A serendipitous strategy for causal discovery and ontology evolution

    Knowl.-Based Syst.

    (2015)
  • González-FerrerA. et al.

    Understanding requirements of clinical data standards for developing interoperable knowledge-based DSS: A case study

    Comput. Standards Interfaces

    (2015)
  • AhmadianL. et al.

    The role of standardized data and terminological systems in computerized clinical decision support systems: Literature review and survey

    Int. J. Med. Inform.

    (2011)
  • MarcosM. et al.

    Interoperability of clinical decision-support systems and electronic health records using archetypes: A case study in clinical trial eligibility

    J. Biomed. Inform.

    (2013)
  • PelegM.

    Computer-interpretable clinical guidelines: A methodological review

    J. Biomed. Inform.

    (2013)
  • SamwaldM. et al.

    The Arden Syntax standard for clinical decision support: Experiences and directions

    J. Biomed. Inform.

    (2012)
  • YeY. et al.

    An ontology-based hierarchical semantic modeling approach to clinical pathway workflows

    Comput. Biol. Med.

    (2009)
  • LezcanoL. et al.

    Integrating reasoning and clinical archetypes using OWL ontologies and SWRL rules

    J. Biomed. Inform.

    (2011)
  • JuarezJ.M. et al.

    Computing context-dependent temporal diagnosis in complex domains

    Expert Syst. Appl.

    (2008)
  • Marco-RuizL. et al.

    Publication, discovery and interoperability of clinical decision support systems: a linked data approach

    J. Biomed. Inform.

    (2016)
  • ChenY. et al.

    IBM Watson: how cognitive computing can be applied to big data challenges in life sciences research

    Clin. Ther.

    (2016)
  • GreenesR.A.

    Clinical Decision Support: The Road to Broad Adoption

    (2014)
  • GiarratanoJ.C. et al.

    Expert Systems: Principles and Programming

    (2005)
  • Berners-LeeT. et al.

    The semantic web

    Sci. Amer.

    (2001)
  • ZolhavariehS. et al.

    KQA: A knowledge quality assessment model for clinical decision support systems

    MEDINFO 2017: Precision Healthcare through Informatics - Proceedings of the 16th World Congress on Medical and Health Informatics, Hangzhou, China, 21-25 August 2017.

    (2017)
  • RileyG.

    Clips: an expert system building tool

    The Second National Technology Transfer Conference and Exposition, Vol. 2

    (1991)
  • Friedman-HillE.

    Jess in Action: Rule-based Systems in Java

    (2003)
  • . The JBoss Drools Team, Drools documentation - version 7.5.0.final,...
  • KahnM.G. et al.

    An expert system for culture-based infection control surveillance

    Proceedings of the Annual Symposium on Computer Applications in Medical Care.

    (1993)
  • DunsmuirD. et al.

    A knowledge authoring tool for clinical decision support

    J. Clin. Monit. Comput.

    (2008)
  • Van HilleP. et al.

    Comparing Drools and ontology reasoning approaches for telecardiology decision support

  • ClaytonP.D. et al.

    Issues and structures for sharing medical knowledge among decision-making systems: The 1989 Arden homestead retreat

    Proceedings of the Annual Symposium on Computer Application in Medical Care

    (1989)
  • PryorT.a. et al.

    The Arden Syntax for medical logic modules

    Int. J. Clin. Monit. Comput.

    (1993)
  • . W3C OWL Working Group, OWL 2 web ontology language document overview (Second Edition), W3C recommendation, W3C,...
  • Cited by (15)

    • Intelligent knowledge consolidation: From data to wisdom[Formula presented]

      2021, Knowledge-Based Systems
      Citation Excerpt :

      In the domain of healthcare, while the creation of Hospital Information Management Systems (HMIS), Clinical Decision Support Systems (CDSS), and other tools have seen massive improvements, the creation of an Artificial Medical Intelligence (AMI) is still far away. Literature points to a multitude of reasons behind this delay, including the heterogeneity of data sources [8], difficulty in expanding current solutions [9], the necessity for decision transparency [10], privacy of patient data [11], and many others. Resultantly, existing healthcare solutions suffer from a lack of quality and applicability, unless substantial effort, in terms of expert’s time and energy is spent on identifying and creating knowledge [12].

    • An interpretable knowledge-based decision support system and its applications in pregnancy diagnosis

      2021, Knowledge-Based Systems
      Citation Excerpt :

      In recent years, the fast development of eHealth systems has provided valuable resources to solve healthcare issues. As a specific application of eHealth, medical decision support system (MDSS) [5,6] utilizes patients’ data and some inference procedures to generate case specific advice and suggestions to the practitioner [7]. For example, Pal et al. [8] develop an MDSS to detect coronary artery disease at an early stage.

    • HaemoKBS: A knowledge-based system for real-time, continuous categorisation of adverse reactions in blood recipients

      2021, Neurocomputing
      Citation Excerpt :

      Also interesting, another review explored intrinsic weaknesses affecting the application of this methodology to the medical domain, notably difficulties in feature mining, presence of competence gaps affecting case base contents, challenges in the definition of a suitable adaptation strategy, and other complementary issues related to the increasing complexity of the domain [27]. On top of all these developments, it is also worth mentioning the development of rule-based systems that model clinical knowledge using sets of rules, in the form of IF-THEN statements [28]. That is, based on the available cases, the system selects one rule from the knowledge base to be fired in order to generate new knowledge or execute the corresponding actions.

    • Acquiring guideline-enabled data driven clinical knowledge model using formally verified refined knowledge acquisition method

      2020, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      Various knowledge acquisition approaches have been proposed to acquire clinical knowledge from these sources. For example, using machine learning and ontological approaches, knowledge models from patient data are created [3–5], and different cognitive approaches are used to develop knowledge models from CPGs and other medical resources [6–9]. Depending on the requirements, these knowledge models may need to be transformed into different model formats.

    View all citing articles on Scopus
    View full text