Dataset Information

Combining knowledge- and data-driven methods for de-identification of clinical narratives.

ABSTRACT: A recent promise to access unstructured clinical data from electronic health records on large-scale has revitalized the interest in automated de-identification of clinical notes, which includes the identification of mentions of Protected Health Information (PHI). We describe the methods developed and evaluated as part of the i2b2/UTHealth 2014 challenge to identify PHI defined by 25 entity types in longitudinal clinical narratives. Our approach combines knowledge-driven (dictionaries and rules) and data-driven (machine learning) methods with a large range of features to address de-identification of specific named entities. In addition, we have devised a two-pass recognition approach that creates a patient-specific run-time dictionary from the PHI entities identified in the first step with high confidence, which is then used in the second pass to identify mentions that lack specific clues. The proposed method achieved the overall micro F1-measures of 91% on strict and 95% on token-level evaluation on the test dataset (514 narratives). Whilst most PHI entities can be reliably identified, particularly challenging were mentions of Organizations and Professions. Still, the overall results suggest that automated text mining methods can be used to reliably process clinical notes to identify personal information and thus providing a crucial step in large-scale de-identification of unstructured data for further clinical and epidemiological studies.

SUBMITTER: Dehghan A

PROVIDER: S-EPMC4976126 | biostudies-literature | 2015 Dec

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Combining knowledge- and data-driven methods for de-identification of clinical narratives.

Dehghan Azad A Kovacevic Aleksandar A Karystianis George G Keane John A JA Nenadic Goran G

Journal of biomedical informatics 20150722

A recent promise to access unstructured clinical data from electronic health records on large-scale has revitalized the interest in automated de-identification of clinical notes, which includes the identification of mentions of Protected Health Information (PHI). We describe the methods developed and evaluated as part of the i2b2/UTHealth 2014 challenge to identify PHI defined by 25 entity types in longitudinal clinical narratives. Our approach combines knowledge-driven (dictionaries and rules) ...[more]

PMID: 26210359

Similar Datasets

Project description:Recently, computational approaches integrating copy number aberrations (CNAs) and gene expression (GE) have been extensively studied to identify cancer-related genes and pathways. In this work, we integrate these two data sets with protein-protein interaction (PPI) information to find cancer-related functional modules. To integrate CNA and GE data, we first built a gene-gene relationship network from a set of seed genes by enumerating all types of pairwise correlations, e.g. GE-GE, CNA-GE, and CNA-CNA, over multiple patients. Next, we propose a voting-based cancer module identification algorithm by combining topological and data-driven properties (VToD algorithm) by using the gene-gene relationship network as a source of data-driven information, and the PPI data as topological information. We applied the VToD algorithm to 266 glioblastoma multiforme (GBM) and 96 ovarian carcinoma (OVC) samples that have both expression and copy number measurements, and identified 22 GBM modules and 23 OVC modules. Among 22 GBM modules, 15, 12, and 20 modules were significantly enriched with cancer-related KEGG, BioCarta pathways, and GO terms, respectively. Among 23 OVC modules, 19, 18, and 23 modules were significantly enriched with cancer-related KEGG, BioCarta pathways, and GO terms, respectively. Similarly, we also observed that 9 and 2 GBM modules and 15 and 18 OVC modules were enriched with cancer gene census (CGC) and specific cancer driver genes, respectively. Our proposed module-detection algorithm significantly outperformed other existing methods in terms of both functional and cancer gene set enrichments. Most of the cancer-related pathways from both cancer data sets found in our algorithm contained more than two types of gene-gene relationships, showing strong positive correlations between the number of different types of relationship and CGC enrichment [Formula: see text]-values (0.64 for GBM and 0.49 for OVC). This study suggests that identified modules containing both expression changes and CNAs can explain cancer-related activities with greater insights.

Project description:Hypothesis generation is a critical step in research and a cornerstone in the rare disease field. Research is most efficient when those hypotheses are based on the entirety of knowledge known to date. Systematic review articles are commonly used in biomedicine to summarize existing knowledge and contextualize experimental data. But the information contained within review articles is typically only expressed as free-text, which is difficult to use computationally. Researchers struggle to navigate, collect and remix prior knowledge as it is scattered in several silos without seamless integration and access. This lack of a structured information framework hinders research by both experimental and computational scientists. To better organize knowledge and data, we built a structured review article that is specifically focused on NGLY1 Deficiency, an ultra-rare genetic disease first reported in 2012. We represented this structured review as a knowledge graph and then stored this knowledge graph in a Neo4j database to simplify dissemination, querying and visualization of the network. Relative to free-text, this structured review better promotes the principles of findability, accessibility, interoperability and reusability (FAIR). In collaboration with domain experts in NGLY1 Deficiency, we demonstrate how this resource can improve the efficiency and comprehensiveness of hypothesis generation. We also developed a read-write interface that allows domain experts to contribute FAIR structured knowledge to this community resource. In contrast to traditional free-text review articles, this structured review exists as a living knowledge graph that is curated by humans and accessible to computational analyses. Finally, we have generalized this workflow into modular and repurposable components that can be applied to other domain areas. This NGLY1 Deficiency-focused network is publicly available at http://ngly1graph.org/. AVAILABILITY AND IMPLEMENTATION:Database URL: http://ngly1graph.org/. Network data files are at: https://github.com/SuLab/ngly1-graph and source code at: https://github.com/SuLab/bioknowledge-reviewer. CONTACT:asu@scripps.edu.

Project description:INTRODUCTION:A common bottleneck during ontology evaluation is knowledge acquisition from domain experts for gold standard creation. This paper contributes a novel semi-automated method for evaluating the concept coverage and accuracy of biomedical ontologies by complementing expert knowledge with knowledge automatically extracted from clinical practice guidelines and electronic health records, which minimizes reliance on expensive domain expertise for gold standards generation. METHODS:We developed a bacterial clinical infectious diseases ontology (BCIDO) to assist clinical infectious disease treatment decision support. Using a semi-automated method we integrated diverse knowledge sources, including publically available infectious disease guidelines from international repositories, electronic health records, and expert-generated infectious disease case scenarios, to generate a compendium of infectious disease knowledge and use it to evaluate the accuracy and coverage of BCIDO. RESULTS:BCIDO has three classes (i.e., infectious disease, antibiotic, bacteria) containing 593 distinct concepts and 2345 distinct concept relationships. Our semi-automated method generated an ID knowledge compendium consisting of 637 concepts and 1554 concept relationships. Overall, BCIDO covered 79% (504/637) of the concepts and 89% (1378/1554) of the concept relationships in the ID compendium. BCIDO coverage of ID compendium concepts was 92% (121/131) for antibiotic, 80% (205/257) for infectious disease, and 72% (178/249) for bacteria. The low coverage of bacterial concepts in BCIDO was due to a difference in concept granularity between BCIDO and infectious disease guidelines. Guidelines and expert generated scenarios were the richest source of ID concepts and relationships while patient records provided relatively fewer concepts and relationships. CONCLUSIONS:Our semi-automated method was cost-effective for generating a useful knowledge compendium with minimal reliance on domain experts. This method can be useful for continued development and evaluation of biomedical ontologies for better accuracy and coverage.

Dataset Information

Combining knowledge- and data-driven methods for de-identification of clinical narratives.

Publications

Combining knowledge- and data-driven methods for de-identification of clinical narratives.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets