Showing papers presented at "Knowledge Representation for Health-Care in 2016"

Applying SPARQL-based inference and ontologies for modelling and execution of clinical practice guidelines: A case study on hypertension management

Abstract: Clinical practice guidelines (CPGs) constitute a systematically developed, critical body of medical knowledge which is compiled and maintained in order to assist healthcare professionals in decision making. They are available for diverse diseases/conditions and routinely used in many countries, providing reference material for healthcare delivery in clinical settings. As CPGs are paper-based, i.e. plain documents, there have been various approaches for their computerization and expression in a formal manner so that they can be incorporated in clinical information and decision support systems. Semantic Web technologies and ontologies have been extensively used for CPG formalization. In this paper, we present a novel method for the representation and execution of CPGs using OWL ontologies and SPARQL-based inference rules. The proposed approach is capable of expressing complex CPG constructs and can be used to express formalisms, such as negations, which are hard to express using ontologies alone. The encapsulation of SPARQL rules in the CPG ontology is based on the SPARQL Inference Notation (SPIN). The proposed representation of different aspects of CPGs, such as numerical comparisons, calculations, decision branches and state transitions, and their execution is demonstrated through the respective parts of comprehensive, though complex enough, CPGs for arterial hypertension management. The paper concludes by comparing the proposed approach with other relevant works, indicating its potential and limitations, as well as a future work directions.

Generating conflict-free treatments for patients with comorbidity using ASP

Abstract: Conflicts in recommended medical interventions regularly arise when multiple treatments are simultaneously needed for patients with comorbid diseases. An approach that can automatically repair such inconsistencies and generate conflict-free combined treatments is thus a valuable aid for clinicians. In this paper we propose an answer set programming based method that detects and repairs conflicts between treatments. The answer sets of the program directly correspond to proposed treatments, accounting for multiple possible solutions if they exist. We also include the possibility to take preferences based on drug-drug interactions into account while solving inconsistencies. We show in a case study that our method results in more preferred treatments.

Knowledge-Driven Paper Retrieval to Support Updating of Clinical Guidelines

Abstract: Clinical Guidelines are important knowledge resources for medical decision making. They provide clinical recommendations based on a collection of research findings with respect to a specific disease. Since, new findings are regularly published, CGs are also expected to be regularly updated. However, selecting and analysing medical publications require a huge human efforts, even when these publications are mostly regrouped and into repositories (e.g., MEDLINE database) and accessible via a search engine (e.g. PubMed). Automatically detecting those research findings from a medical search engine such as PubMed supports the guideline updating process. A simple search method is to select the medical terms that appear in the conclusions of the guideline to generate a query to search for new evidences. However, some challenges rise in this method: how to select the important terms, besides how to consider background knowledge that may be missing or not explicitly stated in those conclusions. In this paper we apply a knowledge model that formally describes elements such as actions and their effects to investigate (i) if it favors selecting the medical terms to compose queries and (ii) if a search enhanced with background knowledge can provide better result than other methods. This work explores a knowledge-driven approach for detecting new evidences relevant for the clinical guideline update process. Based on the outcomes of two experiments, we found that this approach can improve the recall by retrieving more relevant evidences than previous methods.

Detecting New Evidences for Evidence-Based Medical Guidelines with Journal Filtering

Abstract: Evidence-based medical guidelines are systematically developed recommendations with the aim to assist practitioner and patients decisions regarding appropriate health care for specific clinical circumstances, and are based on evidence described in medical research papers. Evidence-based medical guidelines should be regularly updated, such that they can serve medical practice using based on the latest medical research evidence. A usual approach to detecting new evidences is to use a set of terms which appear in a guideline conclusion or recommendation and create queries over a bio-medical search engine such as PubMed with a ranking over a selected subset of terms to search for relevant new research papers. However, the sizes of the found relevant papers are usually very large (i.e. over a few hundreds, even thousands), which results in a low precision of the search. This makes it for medical professionals quite difficult to find which papers are really interesting and useful for updating the guideline. We propose a filtering step to decrease the number of papers. More exactly we are interested in the question if we can reduce the number of papers with no or a slightly lower recall. A plausible approach is to introduce journal filtering, such that evidence appear in those top journals are preferred.

Formalization and Computation of Diabetes Quality Indicators with Patient Data from a Chinese Hospital

Abstract: Clinical quality indicators are tools to measure the quality of healthcare and can be classified into structure-related, process-related and outcome-related indicators. The objective of this study is to investigate whether Electronic Medical Record (EMR) data from a Chinese diabetes specialty hospital can be used for the automated computation of a set of 38 diabetes quality indicators, especially process-related indicators. The clinical quality indicator formalization (CLIF) method and tool and SNOMED CT were adopted to formalize diabetes indicators into executable queries. The formalized indicators were run on the patient data to test the feasibility of their automated computation. In this study, we successfully formalized and computed 32 of 38 quality indicators based on the EMR data. The results indicate that the data from our Chinese EMR can be used for the formalization and computation of most diabetes indicators, but that it can be improved to support the computation of more indicators.