Ebook: Data and Knowledge for Medical Decision Support

This volume contains the proceedings of the Thirteenth EFMI Special Topic Conference, held in Prague, Czech Republic, from 17–19 April 2013. The EFMI STC 2013 is Europe's leading forum for presenting the results of current scientific work in health-informatics processes, systems, and technologies this year.

EFMI STC 2013 has been organized by the European Federation for Medical Informatics (EFMI) in cooperation with the Czech Society of Biomedical Engineering and Medical Informatics and the Czech Medical Society J.E. Purkyne. It follows previous conferences in Bucharest, Romania (2001), Nicosia, Cyprus (2002), Rome, Italy (2003), Munich, Germany (2004), Athens, Greece (2005), Timisoara, Romania (2006), Brijuni Island, Croatia (2007), London, UK (2008), Antalya, Turkey (2009), Reykjavik, Iceland (2010), Laško, Slovenia (2011) and Moscow, Russia (2012).

Ensuring patient safety and providing high quality health services are the dominant challenges faced by healthcare systems around the world. The sharing of advanced knowledge and best practice in diagnosis, therapy, process optimization and prevention, the inclusion of public health and social sciences, as well as the deployment of any relevant information, is of vital importance. This includes enhanced networking, socially and technologically, based on advanced interoperability.

EFMI STC 2013 is entitled “Data and Knowledge for Medical Decision Support”, and addresses this important field, linking traditional and translational medicine with natural sciences and technology with a view to the design, implementation and deployment of intelligent systems. Medical decision support is an important part of this strategy. It has spawned research in the areas of knowledge discovery, formalization and distribution of knowledge, different ways of reasoning based on that knowledge depending on the nature of facts and processes, establishment and exchange of clinical guidelines, and also the inclusion of decisions not based on knowledge, such as neuronal networks and genetic algorithms. Data and knowledge sharing also confronts aspects of concept representation and languages, i.e. terminologies and ontologies. Most of the developed decision support systems can be integrated more or less easily into clinical information systems, both as part of those systems connected through standardized interfaces or as services to be remotely accessed. Knowledge representation must be appropriate for the different stakeholder groups for decision support systems to be accepted. Such systems must meet the expectations of developers, of users, such as physicians, nurses and other health professionals, but also those of patients. The usability of the system and the comprehensibility of the knowledge and decision support offered are critical in this regard. The standardization of principles, methodologies and means, and also the availability of specifications and tools such as open source products and artefacts are crucial to ensure wide and harmonized use. Another important factor for success is the careful evaluation and certification of systems as regards quality and functionality. EFMI STC 2013 is the latest in a series of events in Prague dedicated to the subject of STC, including the IMIA International Working Conference on Computer-Aided Medical Decision Making in 1985 and the Symposium on Computerized Guidelines and Protocols in 2004, which augurs well for a successful conference. The conference will be introduced by two Keynotes: Prof. Marion Ball (Baltimore, USA) presents Social and Cognitive Computing for Patient Engagement and Decision Support, in cooperation with Dr. Joseph M. Jasinski, and Prof. Jan van Bemmel (Rotterdam, The Netherlands) addresses The Future of Computer-Assisted Medical Decision Making: Can We Learn from the Past? Additionally, two invited speeches will frame basics, principles, methods, and advanced results relevant for the main streams of the conference. A rigorous review process has selected the best of the submissions, resulting in a scientific programme of 49 oral presentations, 31 poster presentations, 1 panel and 7 workshops, bringing active participants from 31 countries to the conference.

EFMI STC 2013 is complemented with panels on special topics, plenary poster sessions, and workshops. Most of the workshops are organized by EFMI Working Groups, such as “Sharing Knowledge and Tools for Decision Support in Biomedicine and Health Care”, realized by the EFMI WG Education in Health Informatics, “Decision Support and Decision Making Enabled by Personal Portable Devices”, organized by the EFMI WG Personal Portable Devices, “Using Information to Improve the Quality of Care in Type 2 Diabetes in Primary Care”, performed by the EFMI WG Primary Care Informatics, “Socio-economic Features of Traceability, ePrescription and Pharmacovigilance”, managed by the EFMI WG Traceability, and “Health Information Management for Europe-Ways and Perspectives”, established by the EFMI Project Group Health Information Management Europe. Another workshop addresses “Clinical Decision Support – From Research to Practice”. Finally, HL7 International performs a special workshop on “Standardization of knowledge management and innovations in science – Prerequisite or conflict?”

The editors would like to thank all the contributing authors for their excellent work, and the reviewers for lending their expertise to the conference, thereby enabling the final achievement. Furthermore, they are indebted to HL7 International, GS1 Europe, and HL7 Germany for sponsoring the printing of the proceedings. Last but not least, they would also like to thank Thomas Schabetsberger (Innsbruck, Austria), who collaboratively responded to all requests related to the Online-Submission System, and Roman Muška and Tereza Jeníkova from AIM Group International for managing the EFMI STC 2013 Website, the registration process and a number of communications.

Bernd Blobel, Arie Hasman and Jana Zvárová

Editors

Based on the paradigm changes for health, health services and underlying technologies as well as the need for at best comprehensive and increasingly automated interoperability, the paper addresses the challenge of knowledge representation and management for medical decision support. After introducing related definitions, a system-theoretical, architecture-centric approach to decision support systems (DSSs) and appropriate ways for representing them using systems of ontologies is given. Finally, existing and emerging knowledge representation and management standards are presented. The paper focuses on the knowledge representation and management part of DSSs, excluding the reasoning part from consideration.

The clinical worksite constitutes a naturally clustered environment, posing challenges in the statistical analysis of quality improvement interventions such as computerized decision support. Ignoring clustering in the analysis may lead to biased effect estimates, underestimating the variance and hence type I errors. This paper presents a secondary analysis on data from a previously published, cluster randomized trial in cardiac rehabilitation. We compared six different statistical analysis methods (weighted and unweighted t-test; adjusted χ² test; normal and multilevel logistic regression analysis; and generalized estimation equations). There were considerable differences in both point estimates and p-values derived by the methods, and differences were larger with increasing intracluster correlation.

Decision Support Systems (DSS) have a vital role to play in today's scenario for Patient Care. They can embody a vast knowledge not normally found in one individual where diagnosis and treatment are involved. This paper highlights the training in minute details and precise mathematics needed in a successful DSS and indicates how such attention-to-detail was instilled into the writer as a result of working with Alan Turing and Emil Wolf who have both since achieved world-wide recognition in their own fields as a result of international publicity by the current writer. The article discusses four Decision Support Systems written by the present writer all of which have been shown to improve patient treatment and care, and which are of such complexity that, without their use, patient care would fall short of optimum. The Systems considered are those for Intensive Care Units, Cardiovascular Surgery, a Programmed Investigation Unit, and Diagnosis of Congenital Abnormalities. All these Systems have performed better than the human alternatives and have shown their value in the improvement of patient care.

Variety of prognostic models can be designed on the basis of learning sets by using the principle of linear separability. The degree of linear separability of two learning sets can be evaluated on the basis of the minimal value of the perceptron criterion function, which belongs to a larger family of the convex and piecewise linear (CPL) criterion functions. Parameters constituting the minimal value of a given CPL criterion function can define particular prognostic model. Prognostic models have been designed this way, for example, on the basis of genetic data sets.

Advances in genomics and human genetics have enabled a more detailed understanding of the impact of genetics in a disease and its treatment. In addition to a patient's clinical signs and symptoms, physicians can now or in near future consider genetic data for their diagnosis and treatment decisions. This new information source based on genome and gene expression analysis makes clinical decision processes even more complex. Beyond, behavioral and environmental aspects should also be considered in order to realize personalized medicine. Given these additional information sources, the need for support in decision making is increasing. In this paper, we introduce a vision how knowledge-based systems or decision support systems can help to realize personalized medicine and we explore the upcoming challenges for clinical decision support in that context.

Clinical decision support systems (CDSSs) are gaining popularity as tools that assist physicians in optimizing medical care. These systems typically comply with evidence-based medicine and are designed with input from domain experts. Nonetheless, deviations from CDSS recommendations are abundant across a broad spectrum of disorders, raising the question as to why this phenomenon exists. Here, we analyze this gap in adherence to a clinical guidelines-based CDSS by examining the physician treatment decisions for 1329 adult soft tissue sarcoma patients in northern Italy using patient-specific parameters. Dubbing this analysis “CareGap”, we find that deviations correlate strongly with certain disease features such as local versus metastatic clinical presentation. We also notice that deviations from the guideline-based CDSS suggestions occur more frequently for patients with shorter survival time. Such observations can direct physicians' attention to distinct patient cohorts that are prone to higher deviation levels from clinical practice guidelines. This illustrates the value of CareGap analysis in assessing quality of care for subsets of patients within a larger pathology.

Web ontology language (OWL), used in combination with the Protégé visual interface, is a modern standard for development and maintenance of ontologies and a powerful tool for knowledge presentation. In this work, we describe a novel possibility to use OWL also for the conceptualization of knowledge presented by a set of rules. In this approach, rules are represented as a hierarchy of actionable classes with necessary and sufficient conditions defined by the description logic formalism. The advantages are that: the set of the rules is not an unordered set anymore, the concepts defined in descriptive ontologies can be used directly in the bodies of rules, and Protégé presents an intuitive tool for editing the set of rules. Standard ontology reasoning processes are not applicable in this framework, but experiments conducted on the rule sets have demonstrated that the reasoning problems can be successfully solved.

Dutch general practices have a high adoption rate for computerized patient records and clinical decision support. We sought to measure the attitudes and experience of Dutch general practitioners towards clinical decision support. Methods: A preliminary survey was created based on questions from published surveys, modified with the results of interviews. The final web-based survey was administered to 43 general practitioners in a practice area where a decision support implementation is planned. Results: Thirty general practitioners (70%) completed the survey. Most felt that decision support is a good idea (23/30), although fewer reported positive experience with decision support (10/30). Participants were supportive of rules and guidelines, but commonly had the sense that there were too many alerts. Conclusion: Dutch clinicians are positive about decision support, but future efforts should try to reduce the perception of overload, for example by ensuring that alerts are relevant and choosing less interruptive forms of notification for less severe alerts.

Online access to records is part of the process of empowering patients. National health services in both France and England have introduced systems to provide online access to summary health data. The English system was called the “Summary Care Record (SCR),” made accessible to patients through “HealthSpace”. The French system Dossier Médical Personnel (DMP) is a patient controlled record clinicians enter data into. The objective was to compare the programmes and lessons from the introduction of patient access. We carried out a literature review. The English system has been progressively de-scoped, with HeathSpace due to close in 2013, only 0.01% of the population signing up for “advanced accounts”. The French system slowly grows as more documents are added; though only 0.31% of the population have opened a DMP. The English SCR has an opt-out consent model, whereas the French DMP is patient controlled opt-in consent model. The SCR sits within an NHS intranet while the DMP sits on the Internet. Both systems have costs of around 200million Euro. Providing patients online access to their medical records is potentially empowering. However, the English HealthSpace and SCR have failed to deliver and are due to be withdrawn as methods of providing patients online access. The French system is still in operation but much criticized for its high costs and low uptake. The design of these systems does not appear to have met patients' needs or been readily integrated into physicians workflow.

Electronic Patient Records can be interfaced with medical decision support systems and quality of care assessment tools. An easy way of measuring the quality of EPR data is therefore essential. This study identified a number of global quality indicators (tracers) that could be easily calculated and validated them by correlating them with the Sensitivity and Positive Predictive Value (PPV) of data extracted from the EPR. Sensitivity and PPV of automatically extracted data were calculated using a gold standard constructed using answers to questions GPs were asked at the end of each contact with a patient. These properties were measured for extracted diagnoses, drug prescriptions, and certain parameters. Tracers were defined as drug-disease pairs (e.g. insulin-diabetes) with the assumption that if the patient is taking the drug, then the patient is suffering from the disease. Four tracers were identified that could be used for the ResoPrim primary care research database, which includes data from 43 practices, 10,307 patients, and 13,372 contacts. Moderately positive correlations were found between the 4 tracers and between the tracers and the sensitivity of automatically extracted diagnoses. For some purposes, these results may support the potential use of tracers for monitoring the quality of information systems such as EPRs.

Healthcare providers are facing an enormous cost pressure, wherefore the assurance of an efficient care on a high level of quality is of decisive importance. Clinical guidelines and clinical pathways have been established for that purpose. Clinical guidelines offer abstract recommendations for diagnostic and therapeutic issues, while clinical pathways are a road map of patient management. The consideration of clinical guidelines during pathway development is highly recommended. But the transfer of evident knowledge (clinical guidelines) to care processes (clinical pathways) is not straightforward due to different information contents and semantical constructs. This article proposes a model-driven approach in conjunction with a developed knowledge acquisition tool to improve the development of guideline-compliant pathways.

The purpose of this study was to develop the tools and the methodology for a systematic analysis of usefulness of adding sonic representation of data, supplementary to visualization. This paper is mainly dedicated various temporal lenses, including the newly developed lenses with variable magnification, proposed as a tool for a better perception of short events combined with a compression of irrelevant intervals. Sonification procedures are also briefly presented. The programs were tested using various cardiac signals: ECG and heart rate HR both in humans and in rats (experimental data). The results, represented by the sound files, were uploaded in an accessible library, which contains both sonic and visual representation of the signals.

We implemented a prototype of a decision support system called SIR which has a form of a web-based classification service for diagnostic decision support. The system has the ability to select the most relevant variables and to learn a classification rule, which is guaranteed to be suitable also for high-dimensional measurements. The classification system can be useful for clinicians in primary care to support their decision-making tasks with relevant information extracted from any available clinical study. The implemented prototype was tested on a sample of patients in a cardiological study and performs an information extraction from a high-dimensional set containing both clinical and gene expression data.

Specific Language Impairment (SLI), as many other cognitive deficits, is difficult to diagnose given its heterogeneous profile and its overlap with other impairments. Existing techniques are based on different criteria using behavioral variables on different tasks. In this paper we propose a methodology for the diagnosis of SLI that uses computational cognitive modeling in order to capture the internal mechanisms of the normal and impaired brain. We show that machine learning techniques that use the information of these models perform better than those that only use behavioral variables.

Gastroesophageal reflux disease is a serious clinical problem, which can significantly impair health-related quality of life, thus having global implications for patients. The first step for a doctor is the clinical classification of the patients, divided into classes after being subjected to endoscopic examinations to control if there are lesions of the esophageal mucosa, and if present, the severity of these lesions. 269 patients were taken into consideration (4 healthy patients, 219 with non erosive reflux disease, 21 with erosive reflux disease, 15 with complicated erosive reflux disease, 10 with Barrett's disease). A set of values taken from gastroscopy, ph-metry and manometry tests were considered and a decision tree was made to classify every patient. Entropy and information gain were calculated for each node to create the most possible simple tree. The resulting tree presents some paths including a significant number of persons; the values that build these paths can be considered characteristic of each class of patient. This method can be a basis to develop a diagnostic decision support for a training doctor starting from a set of characteristics, specific to a class of patient.

The paper describes the Eurogene portal for sharing and reusing multilingual multimedia educational resources in human genetics. The content is annotated using concepts of two ontologies and a topic hierarchy. The ontology annotation is used to guide search and for calculating semantically similar content. Educational resources can be aggregated into learning packages. The system is in routine use since 2009.

A lack of acceptance has hindered the widespread adoption and implementation of clinical prediction rules (CPRs). The use of clinical decision support systems (CDSSs) has been advocated as one way of facilitating a broader dissemination and validation of CPRs. This requires computable models of clinical evidence based on open standards rather than closed proprietary content. The on-going TRANSFoRm project has developed ontological models of CPRs suitable for providing CPR based decision support. This paper presents a description of the design and implementation of the ontology model for CPRs that has been proposed. The conceptual validity of the ontology is discussed using the example of a specific CPR in the form of the Alvarado Score for acute appendicitis. We demonstrate how the model is used to query the structure of this particular rule, providing a computable representation suitable for CPRs in general.

By providing patient-specific advice, clinical decision support systems (CDSSs) are expected to promote the implementation of clinical practice guidelines (CPGs) to improve the quality of care. However, produced as texts, often incomplete and ambiguous, CPGs are difficult to translate into the formal knowledge bases (KBs) of CDSSs. The French National Authority for Health (HAS) decided to update CPGs on the management of type 2 diabetes. This work illustrates the simultaneous development of the text and its formal counterpart in a CDSS named RecosDiab. CPGs were elaborated by a working group according to the guideline development methodology. Textual recommendations were graded, either as evidence-based when evidence existed or as consensus-based when acknowledge by the working group. Knowledge modeling was performed following the steps of de-abstraction, disambiguation, and verification of completeness. This last step generated clinical situations not explicitly mentioned in the text and were graded as expert-based. The resulting KB provides therapeutic advice for 805 clinical situations, among which 2 are graded as evidence-based, 37 are consensus-based, and 766 are expert-based. However, because of the amount of expert-based propositions, the HAS did not endorse the system.

Making reliable and justified operational and strategic decisions is a really challenging task in the health care domain. So far, the decisions have been made based on the experience of managers and staff, or they are evaluated with traditional methods, using inadequate data. As a result of this kind of decision-making process, attempts to improve operations usually have failed or led to only local improvements. Health care organizations have a lot of operational data, in addition to clinical data, which is the key element for making reliable and justified decisions. However, it is progressively problematic to access it and make usage of it. In this paper we discuss about the possibilities how to exploit operational data in the most efficient way in the decision-making process. We'll share our future visions and propose a conceptual framework for automating the decision-making process.

Computerized prescription is a central component in modern clinical information systems. It allows scheduling drugs delivery, exams and other types of care. It is thought to be a useful tool for the reduction of medication errors and for the improvement of medication logistics. Whereas the success of the computerized prescription depends on the unambiguous selection of the manipulated concepts, there is a strong variability between the preferred terms of clinicians of different backgrounds. Moreover, users sometimes want to use synonyms or don't know the exact spelling of the term. This makes the search for desired procedure name through large size vocabularies time-consuming for users. In order to facilitate the prescriptions process, we have built a tool that proposes the most likely terms based on the first letters inputted by the user. The tool helps selecting the most appropriate term by ranking the possible results in a clever manner. Experimental evaluation shows promising results and indicates the tool ease the terminology manipulations.