Kathrin Dentler

Comparison of reasoners for large ontologies in the OWL 2 EL profile

This paper provides a survey to and a comparison of state-of-the-art Semantic Web reasoners that succeed in classifying large ontologies expressed in the tractable OWL 2 EL profile. Reasoners are characterized along several dimensions: The first dimension comprises underlying reasoning characteristics, such as the employed reasoning method and its correctness as well as the expressivity and worst-case computational complexity of its supported language and whether the reasoner supports incremental classification, rules, justifications for inconsistent concepts and ABox reasoning tasks. The second dimension is practical usability: whether the reasoner implements the OWL API and can be used via OWLlink, whether it is available as Protege plugin, on which platforms it runs, whether its source is open or closed and which license it comes with. The last dimension contains performance indicators that can be evaluated empirically, such as classification, concept satisfiability, subsumption checking and consistency checking performance as well as required heap space and practical correctness, which is determined by comparing the computed concept hierarchies with each other. For the very large ontology SNOMED CT, which is released both in stated and inferred form, we test whether the computed concept hierarchies are correct by comparing them to the inferred form of the official distribution. The reasoners are categorized along the defined characteristics and benchmarked against well-known biomedical ontologies. The main conclusion from this study is that reasoners vary significantly with regard to all included characteristics, and therefore a critical assessment and evaluation of requirements is needed before selecting a reasoner for a real-life application.

Large-Scale Storage and Reasoning for Semantic Data Using Swarms

Scalable, adaptive and robust approaches to store and analyze the massive amounts of data expected from Semantic Web applications are needed to bring the Web of Data to its full potential. The solution at hand is to distribute both data and requests onto multiple computers. Apart from storage, the annotation of data with machine-processable semantics is essential for realizing the vision of the Semantic Web. Reasoning on webscale data faces the same requirements as storage. Swarm-based approaches have been shown to produce near-optimal solutions for hard problems in a completely decentralized way. We propose a novel concept for reasoning within a fully distributed and self-organized storage system that is based on the collective behavior of swarm individuals and does not require any schema replication. We show the general feasibility and efficiency of our approach with a proof-of-concept experiment of storage and reasoning performance. Thereby, we positively answer the research question of whether swarm-based approaches are useful in creating a large-scale distributed storage and reasoning system.

Barriers to the Reuse of Routinely Recorded Clinical Data: A Field Report

Today, clinical data is routinely recorded in vast amounts, but its reuse can be challenging. A secondary use that should ideally be based on previously collected clinical data is the computation of clinical quality indicators. In the present study, we attempted to retrieve all data from our hospital that is required to compute a set of quality indicators in the domain of colorectal cancer surgery. We categorised the barriers that we encountered in the scope of this project according to an existing framework, and provide recommendations on how to prevent or surmount these barriers. Assuming that our case is not unique, these recommendations might be applicable for the design, evaluation and optimisation of Electronic Health Records.

Evolutionary and Swarm Computing for the Semantic Web

The Semantic Web has become a dynamic and enormous network of typed links between data sets stored on different machines. These data sets are machine readable and unambiguously interpretable, thanks to their underlying standard representation languages. The expressiveness and flexibility of the publication model of Linked Data has led to its widespread adoption and an ever increasing publication of semantically rich data on the Web. This success however has started to create serious problems as the scale and complexity of information outgrows the current methods in use, which are mostly based on database technology, expressive knowledge representation formalism and high-performance computing. We argue that methods from computational intelligence can play an important role in solving these problems. In this paper we introduce and systemically discuss the typical application problems on the Semantic Web and argue that the existing approaches to address their underlying reasoning tasks consistently fail because of the increasing size, dynamicity and complexity of the data. For each of these primitive reasoning tasks we will discuss possible problem solving methods grounded in Evolutionary and Swarm computing, with short descriptions of existing approaches. Finally, we will discuss two case studies in which we successfully applied soft computing methods to two of the main reasoning tasks; an evolutionary approach to querying, and a swarm algorithm for entailment.

Influence of data quality on computed Dutch hospital quality indicators: a case study in colorectal cancer surgery

BackgroundOur study aims to assess the influence of data quality on computed Dutch hospital quality indicators, and whether colorectal cancer surgery indicators can be computed reliably based on routinely recorded data from an electronic medical record (EMR).MethodsCross-sectional study in a department of gastrointestinal oncology in a university hospital, in which a set of 10 indicators is computed (1) based on data abstracted manually for the national quality register Dutch Surgical Colorectal Audit (DSCA) as reference standard and (2) based on routinely collected data from an EMR. All 75 patients for whom data has been submitted to the DSCA for the reporting year 2011 and all 79 patients who underwent a resection of a primary colorectal carcinoma in 2011 according to structured data in the EMR were included. Comparison of results, investigating the causes for any differences based on data quality analysis. Main outcome measures are the computability of quality indicators, absolute percentages of indicator results, data quality in terms of availability in a structured format, completeness and correctness.ResultsAll indicators were fully computable based on the DSCA dataset, but only three based on EMR data, two of which were percentages. For both percentages, the difference in proportions computed based on the two datasets was significant.All required data items were available in a structured format in the DSCA dataset. Their average completeness was 86%, while the average completeness of these items in the EMR was 50%. Their average correctness was 87%.ConclusionsOur study showed that data quality can significantly influence indicator results, and that our EMR data was not suitable to reliably compute quality indicators. EMRs should be designed in a way so that the data required for audits can be entered directly in a structured and coded format.

Knowledge-Based Patient Data Generation

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.

Towards the automated calculation of clinical quality indicators

To measure the quality of care in order to identify whether and how it can be improved is of increasing importance, and several organisations define quality indicators as tools for such measurement. The values of these quality indicators should ideally be calculated automatically based on data that is being collected during the care process. The central idea behind this paper is that quality indicators can be regarded as semantic queries that retrieve patients who fulfil certain constraints, and that indicators that are formalised as semantic queries can be calculated automatically by being run against patient data. We report our experiences in manually formalising exemplary quality indicators from natural language into SPARQL queries, and prove the concept by running the resulting queries against self-generated synthetic patient data. Both the queries and the patient data make use of SNOMED CT to represent relevant concepts. Our experimental results are promising: we ran eight queries against a dataset of 300,000 synthetically generated patients, and retrieved consistent results within acceptable time.

Intra-axiom redundancies in SNOMED CT

OBJECTIVEnIntra-axiom redundancies are elements of concept definitions that are redundant as they are entailed by other elements of the concept definition. While such redundancies are harmless from a logical point of view, they make concept definitions hard to maintain, and they might lead to content-related problems when concepts evolve. The objective of this study is to develop a fully automated method to detect intra-axiom redundancies in OWL 2 EL and apply it to SNOMED Clinical Terms (SNOMED CT).nnnMATERIALS AND METHODSnWe developed a software program in which we implemented, adapted and extended readily existing rules for redundancy elimination. With this, we analysed occurence of redundancy in 11 releases of SNOMED CT (January 2009 to January 2014). We used the ELK reasoner to classify SNOMED CT, and Pellet for explanation of equivalence. We analysed the completeness and soundness of the results by an in-depth examination of the identified redundant elements in the July 2012 release of SNOMED CT. To determine if concepts with redundant elements lead to maintenance issues, we analysed a small sample of solved redundancies.nnnRESULTSnAnalyses showed that the amount of redundantly defined concepts in SNOMED CT is consistently around 35,000. In the July 2012 version of SNOMED CT, 35,010 (12%) of the 296,433 concepts contained redundant elements in their definitions. The results of applying our method are sound and complete with respect to our evaluation. Analysis of solved redundancies suggests that redundancies in concept definitions lead to inadequate maintenance of SNOMED CT.nnnCONCLUSIONSnOur analysis revealed that redundant elements are continuously introduced and removed, and that redundant elements may be overlooked when concept definitions are corrected. Applying our redundancy detection method to remove intra-axiom redundancies from the stated form of SNOMED CT and to point knowledge modellers to newly introduced redundancies can support creating and maintaining a redundancy-free version of SNOMED CT.

Formalization and computation of quality measures based on electronic medical records.

OBJECTIVEnAmbiguous definitions of quality measures in natural language impede their automated computability and also the reproducibility, validity, timeliness, traceability, comparability, and interpretability of computed results. Therefore, quality measures should be formalized before their release. We have previously developed and successfully applied a method for clinical indicator formalization (CLIF). The objective of our present study is to test whether CLIF is generalizable--that is, applicable to a large set of heterogeneous measures of different types and from various domains.nnnMATERIALS AND METHODSnWe formalized the entire set of 159 Dutch quality measures for general practice, which contains structure, process, and outcome measures and covers seven domains. We relied on a web-based tool to facilitate the application of our method. Subsequently, we computed the measures on the basis of a large database of real patient data.nnnRESULTSnOur CLIF method enabled us to fully formalize 100% of the measures. Owing to missing functionality, the accompanying tool could support full formalization of only 86% of the quality measures into Structured Query Language (SQL) queries. The remaining 14% of the measures required manual application of our CLIF method by directly translating the respective criteria into SQL. The results obtained by computing the measures show a strong correlation with results computed independently by two other parties.nnnCONCLUSIONSnThe CLIF method covers all quality measures after having been extended by an additional step. Our web tool requires further refinement for CLIF to be applied completely automatically. We therefore conclude that CLIF is sufficiently generalizable to be able to formalize the entire set of Dutch quality measures for general practice.

Semantic integration of patient data and quality indicators based on open EHR archetypes

Electronic Health Records (EHRs) contain a wealth of information, but accessing and (re)using it is often difficult. Archetypes have been shown to facilitate the (re)use of EHR data, and may be useful with regard to clinical quality indicators. These indicators are often released centrally, but computed locally in several hospitals. They are typically expressed in natural language, which due to its inherent ambiguity does not guarantee comparable results. Thus, their information requirements should be formalised and expressed via standard terminologies such as SNOMED CT to represent concepts, and information models such as archetypes to represent their agreed-upon structure, and the relations between the concepts. The two-level methodology of the archetype paradigm allows domain experts to intuitively define indicators at the knowledge level, and the resulting queries are computable across institutions that employ the required archetypes. We tested whether openEHR archetypes can represent both elements of patient data required by indicators and EHR data for automated indicator computation. The relevant elements of the indicators and our hospitals database schema were mapped to (elements of) publicly available archetypes. The coverage of the public repository was high, and editing an archetype to fit our requirements was straightforward. Based on this mapping, a set of three indicators from the domain of gastrointestinal cancer surgery was formalised into archetyped SPARQL queries and run against archetyped patient data in OWL from our hospitals data warehouse to compute the indicators. The computed indicator results were comparable to centrally computed and publicly reported results, with differences likely to be due to differing indicator definitions and interpretations, insufficient data quality and insufficient and imprecise encoding. This paper shows that openEHR archetypes facilitate the semantic integration of quality indicators and routine patient data to automatically compute indicators.