Process Support and Knowledge Representation in Health Care

The main rationale of biomedical terminologies and formalized clinical information models is to provide semantic standards to improve the exchange of meaningful clinical information. Whereas terminologies should express context-independent meanings of domain terms, information models are built to represent the situational and epistemic contexts in which domain terms are used. In practice, semantic interoperability is encumbered by a plurality of different encodings of the same piece of clinical information. The same meaning can be represented by single codes in different terminologies, pre- and postcoordinated expressions in the same terminology, as well as by different combinations of (partly overlapping) terminologies and information models.

Formal ontologies can support the automatically recognition and processing of such heterogeneous but isosemantic expressions. In the SemanticHealthNet Network of Excellence a semantic framework is being built which addresses the goal of semantic interoperability by proposing a generalized methodology of transforming existing resources into “semantically enhanced” ones. The semantic enhancements consist in annotations as OWL axioms which commit to an upper-level ontology that provides categories, relations, and constraints for both domain entities and informational entities. Prospects and the challenges of this approach – particularly human and computational limitations – are discussed.

Translating clinical practice guidelines into a computer-interpretable format is a challenging and laborious task. In this project we focus on supporting the early steps of the modeling process by automatically identifying conditional activities in guideline documents in order to model them automatically in further consequence. Therefore, we developed a rule-based, heuristic method that combines domain-independent information extraction rules and semantic pattern rules. The classification also uses a weighting coefficient to verify the relevance of the sentence in the context of other information aspects, such as effects, intentions, etc. Our evaluation results show that even with a small set of training data, we achieved a recall of 75 % and a precision of 88 %. This outcome shows that this method supports the modeling task and eases the translation of CPGs into a semi-formal model.

Medical protocols and guidelines can be looked upon as concurrent programs, where the patient’s state dynamically changes over time. Methods based on verification and model-checking developed in the past have been shown to offer insight into the correctness of guidelines and protocols by adopting a logical point of view. However, there is uncertainty involved both in the management of the disease and the way the disease will develop, and, therefore, a probabilistic view on medical protocols seems more appropriate. Representations using Bayesian networks capture that uncertainty, but usually concern a single patient group and do not capture the dynamic nature of care. In this paper, we propose a new method inspired by automata learning to represent and identify patient groups for obtaining insight into the care that patients have received. We evaluate this approach using data obtained from general practitioners and identify significant differences in patients who were diagnosed with a transient ischemic attack. Finally, we discuss the implications of such a computational method for the analysis of medical protocols and guidelines.

Physicians are required to interpret, abstract and present in free-text large amounts of clinical data in their daily tasks. This is especially true for chronic-disease domains, but also in other clinical domains. In our previous work, we have suggested a general framework for performing this task, given a time-oriented clinical database, and appropriate formal abstraction and summarization knowledge. We have recently developed a prototype system, CliniText, which demonstrates our ideas. Our prototype combines knowledge-based temporal data abstraction, textual summarization, abduction, and natural-language generation techniques, to generate an intelligent textual summary of longitudinal clinical data. We demonstrate both our methodology, and the feasibility of providing a free-text summary of longitudinal electronic patient records, by generating a discharge summary of a patient from the MIMIC database, who had undergone a Coronary Artery Bypass Graft operation.

The development and investigation of medical applications require patient data from various Electronic Health Records (EHR) or Clinical Records (CR). However, in practice, patient data is and should be protected and monitored to avoid unauthorized access or publicity, because of many reasons including privacy, security, ethics, and confidentiality. Thus, many researchers and developers encounter the problem to access required patient data for their research or make patient data available for example to demonstrate the reproducibility of their results. In this paper, we propose a knowledge-based approach of synthesizing large scale patient data. Our main goal is to make the generated patient data as realistic as possible, by using domain knowledge to control the data generation process. Such domain knowledge can be collected from biomedical publications such as PubMed, from medical textbooks, or web resources (e.g. Wikipedia and medical websites). Collected knowledge is formalized in the Patient Data Definition Language (PDDL) for the patient data generation. We have implemented the proposed approach in our Advanced Patient Data Generator (APDG). We have used APDG to generate large scale data for breast cancer patients in the experiments of SemanticCT, a semantically-enabled system for clinical trials. The results show that the generated patient data are useful for various tests in the system.

We present a patient-centered self-management framework that aims to assist individuals to achieve self-efficacy in terms of self-management of their chronic condition. We have incorporated an evidence-driven behavior model—i.e. the Social Cognition Theory (SCT)—to personalize the self-management educational content based on the individual’s health and psychosocial profile. We have taken a knowledge management approach to the development of the self-management framework where we have modeled the SCT, educational content and strategies, assessment tools and the personalization logic using an OWL-DL based ontology. The execution of the knowledge encapsulated within the SCT ontology allows for the dynamic generation of a patient’s profile and the selection of the relevant self-management strategies, educational and motivational messages. We applied our self-management framework to develop a self-management program for cardiac conditions.

Providing IT-based care support for elderly at home is proposed as a highly promising appraoch to address the aging population problem. With the emergence of homecare application service providers, a homecare system can be seen as a linked set of services. Configuring and composing existing homecare application services to create new homecare composite applications can reduce the application development cost. The idea even looks more promising if the service provisioning is dynamic, i.e., if applications can update their behaviors with respect to the contextual changes without or with minimum manpower. Dynamic service provisioning can play an important role to accept homecare systems in practical settings. This motivated us to develop a Dynamic Homecare Service Provisioning (DHSP) platform to address the homecare context changes in an effective and efficient manner. As a proof of concept, we have developed a software prototype of our platform. The prototype was subsequently used in a real-world field test at a care institution in the Netherlands to validate the approach. This paper describes the design of the field test and reflects on the outcome of the validation experiments.

Management of alerts triggered by unexpected or hazardous changes in a patient’s state is a key task in continuous monitoring of patients. Using domain knowledge enables us to specify more sophisticated triggering patterns for alerts, based on temporal patterns detected in a stream of patient data, which include both the temporal element and significant domain knowledge, such as "rapidly increasing fever" instead of monitoring of only raw vital signals, such as "temperature higher than 39 C". In the current study, we introduce iALARM, a two-tier computational architecture, accompanied by a language for specification of intelligent alerts, which represents an additional computational [meta] level above the temporal-abstraction level. Alerts in the iALARM language consist of (a) the target population part (Who is to be monitored?); (b) a declarative part (What is the triggering pattern?), i.e., a set of time and value constraints, specifying the triggering pattern to be computed by the bottom tier; and (c) a procedural part (How should we raise the alarm? How should we continue the monitoring and follow-up?), i.e., an action or a whole plan to apply when the alert is triggered, and a list of meta-properties of the alert and action. One of our underlying principles is to avoid alert fatigue as much as possible; for instance, one can specify that a certain alert should be activated only the first time that the triggering pattern is detected, or only if it has not been raised over the past hour. Thus, we introduce a complete life cycle for alerts. Finally, we discuss the implied requirements for the knowledge- acquisition tool and for the alert monitoring and procedural application engines to support the iALARM language. We intend to evaluate our architecture in several clinical domains, within a large project for remote patient monitoring.

In the last years, many parties have been engaged in developing models for encoding clinical practice guidelines in a computer-interpretable form. Despite the attempts involved to specify and adopt a single, common model, to date, there is no de facto standard solution. Moreover, the effort in defining new models has not been coupled by a parallel effort in supporting a seamless integration with the clinical workflow and existing health information systems. In such a direction, this paper proposes a standards-based verifiable guideline model, named GLM-CDS (GuideLine Model for Clinical Decision Support), whose main features can be summarized in the following points: i) its control-flow model is a formal Task-Network Model devised to represent guidelines on multiple levels of abstraction by focusing only on issues pertaining the clinical decision support; ii) its information model is expressly built on the top of the simplified patient information model standardized as HL7 Virtual Medical Record for Clinical Decision Support; iii) its terminological model is essentially constructed on the top of standard medical terminologies; iv) its computer-interpretable encoding is built in terms of both a formal, semantically well-defined and verifiable ontology for describing control-flow and information models, and a logical rule formalism for specifying decision criteria.