Unified Medical Language System Metathesaurus Research Articles

АВТОМАТИЗИРОВАННАЯ СИСТЕМА ИЗВЛЕЧЕНИЯ КЛИНИЧЕСКИ РЕЛЕВАНТНЫХ ТЕРМИНОВ UMLS ИЗ ТЕКСТОВ АНГЛОЯЗЫЧНЫХ СТАТЕЙ НА ПРИМЕРЕ АКСИАЛЬНОГО СПОНДИЛОАРТРИТА

Significance. Early detection of axial spondyloarthritis (axSpA) is a complex clinical task. Quality improvement of axSpA diagnostics in primary care settings is possible with the help of decision-making information systems based on the ontological approach application. The key stage of the decision-making system development consists of the elaboration of a set of clinical terms. This set should fully describe clinical area or sub-area under study. One of the essential requirements is compliance of the clinical terms used with the existing clinical nomenclatures. Currently, the largest set of clinical terms is the Unified Medical Language System (UMLS) metathesaurus. The majority of UMLS terms is presented in English only. The development of tools for the analysis of unstructured texts and recognition of clinically relevant UMLS entities make it possible to elaborate a set of terms describing axSpA diagnostic aspects. This will also help to compile a list of UMLS terminology nomenclatures for their priority adaptation and expert translation into Russian. The purpose of this study is to develop an automated system for recognizing clinically relevant UMLS terms in texts of the English-language articles. Material and methods. The research material has included English terms (11.2 million) aggregated from 76 nome6nclatures of the current UMLS (2022AB) version. In addition, the study has used texts of PubMed clinical abstracts in English. Queries to the UMLS graph model, semantic algorithms for unstructured texts and machine-learning methods have been applied for data collection and analysis. Results. The study has elaborated a set of high-accuracy regular expressions (F1-score=98%) for metadata elimination from the text corpus. Then the authors have identified patterns for searching clinically relevant terms in the aggregated set of UMLS concepts. Using a logistic regression algorithm, the authors have trained a binary classification model. Input data for the created classificatory are information about an UMLS term. Output data are a label indicating the presence or absence of clinical relevance. Conclusion. The binary classification model has been validated individually and double-tested on different data samples. Values of the accuracy, sensitivity and specificity of metrics equal to 91%, 90% and 91%, respectively, for the validation sample (a number of axSpA terms). In addition, this model has been tested on the sets of terms aggregated for any two diseases. Values of the accuracy metric equal to 91% and 90%, respectively. With the help of the developed machine-learning model, the study has estimated that UMLS contains 1.5 million unique terms applicable to describing a clinical picture. In addition, lists of priority UMLS data sources and thematic groups have been compiled. These clinically relevant UMLS terms should be adapted and translated into Russian as soon as possible.

Performance assessment of ontology matching systems for FAIR data

BackgroundOntology matching should contribute to the interoperability aspect of FAIR data (Findable, Accessible, Interoperable, and Reusable). Multiple data sources can use different ontologies for annotating their data and, thus, creating the need for dynamic ontology matching services. In this experimental study, we assessed the performance of ontology matching systems in the context of a real-life application from the rare disease domain. Additionally, we present a method for analyzing top-level classes to improve precision.ResultsWe included three ontologies (NCIt, SNOMED CT, ORDO) and three matching systems (AgreementMakerLight 2.0, FCA-Map, LogMap 2.0). We evaluated the performance of the matching systems against reference alignments from BioPortal and the Unified Medical Language System Metathesaurus (UMLS). Then, we analyzed the top-level ancestors of matched classes, to detect incorrect mappings without consulting a reference alignment. To detect such incorrect mappings, we manually matched semantically equivalent top-level classes of ontology pairs. AgreementMakerLight 2.0, FCA-Map, and LogMap 2.0 had F1-scores of 0.55, 0.46, 0.55 for BioPortal and 0.66, 0.53, 0.58 for the UMLS respectively. Using vote-based consensus alignments increased performance across the board. Evaluation with manually created top-level hierarchy mappings revealed that on average 90% of the mappings’ classes belonged to top-level classes that matched.ConclusionsOur findings show that the included ontology matching systems automatically produced mappings that were modestly accurate according to our evaluation. The hierarchical analysis of mappings seems promising when no reference alignments are available. All in all, the systems show potential to be implemented as part of an ontology matching service for querying FAIR data. Future research should focus on developing methods for the evaluation of mappings used in such mapping services, leading to their implementation in a FAIR data ecosystem.

Context-Enriched Learning Models for Aligning Biomedical Vocabularies at Scale in the UMLS Metathesaurus.

The Unified Medical Language System (UMLS) Metathesaurus construction process mainly relies on lexical algorithms and manual expert curation for integrating over 200 biomedical vocabularies. A lexical-based learning model (LexLM) was developed to predict synonymy among Metathesaurus terms and largely outperforms a rule-based approach (RBA) that approximates the current construction process. However, the LexLM has the potential for being improved further because it only uses lexical information from the source vocabularies, while the RBA also takes advantage of contextual information. We investigate the role of multiple types of contextual information available to the UMLS editors, namely source synonymy (SS), source semantic group (SG), and source hierarchical relations (HR), for the UMLS vocabulary alignment (UVA) problem. In this paper, we develop multiple variants of context-enriched learning models (ConLMs) by adding to the LexLM the types of contextual information listed above. We represent these context types in context-enriched knowledge graphs (ConKGs) with four variants ConSS, ConSG, ConHR, and ConAll. We train these ConKG embeddings using seven KG embedding techniques. We create the ConLMs by concatenating the ConKG embedding vectors with the word embedding vectors from the LexLM. We evaluate the performance of the ConLMs using the UVA generalization test datasets with hundreds of millions of pairs. Our extensive experiments show a significant performance improvement from the ConLMs over the LexLM, namely +5.0% in precision (93.75%), +0.69% in recall (93.23%), +2.88% in F1 (93.49%) for the best ConLM. Our experiments also show that the ConAll variant including the three context types takes more time, but does not always perform better than other variants with a single context type. Finally, our experiments show that the pairs of terms with high lexical similarity benefit most from adding contextual information, namely +6.56% in precision (94.97%), +2.13% in recall (93.23%), +4.35% in F1 (94.09%) for the best ConLM. The pairs with lower degrees of lexical similarity also show performance improvement with +0.85% in F1 (96%) for low similarity and +1.31% in F1 (96.34%) for no similarity. These results demonstrate the importance of using contextual information in the UVA problem.

High Throughput Neurological Phenotyping with MetaMap

The phenotyping of neurological patients involves the conversion of signs and symptoms into machine readable codes selected from an appropriate ontology. The phenotyping of neurological patients is manual and laborious. MetaMap is used for high throughput mapping of the medical literature to concepts in the Unified Medical Language System Metathesaurus (UMLS). MetaMap was evaluated as a tool for the high throughput phenotyping of neurological patients. Based on 15 patient histories from electronic health records, 30 patient histories from neurology textbooks, and 20 clinical summaries from the Online Mendelian Inheritance in Man repository, MetaMap showed a recall of 61-89%, a precision of 84-93%, and an accuracy of 56-84% for the identification of phenotype concepts. The most common cause of false negatives (failure to recognize a phenotype concept) was an inability of MetaMap to find concepts that were represented as a description or a definition of the concept. The most common cause of false positives (incorrect identification of a concept in the text) was a failure to recognize that a concept was negated. MetaMap shows potential for high throughput phenotyping of neurological patients if the problems of false negatives and false positives can be solved.

Evaluating Biomedical Word Embeddings for Vocabulary Alignment at Scale in the UMLS Metathesaurus Using Siamese Networks.

Recent work uses a Siamese Network, initialized with BioWordVec embeddings (distributed word embeddings), for predicting synonymy among biomedical terms to automate a part of the UMLS (Unified Medical Language System) Metathesaurus construction process. We evaluate the use of contextualized word embeddings extracted from nine different biomedical BERT-based models for synonymy prediction in the UMLS by replacing BioWordVec embeddings with embeddings extracted from each biomedical BERT model using different feature extraction methods. Surprisingly, we find that Siamese Networks initialized with BioWordVec embeddings still outperform the Siamese Networks initialized with embedding extracted from biomedical BERT model.

Customizable Natural Language Processing Biomarker Extraction Tool.

Natural language processing (NLP) in pathology reports to extract biomarker information is an ongoing area of research. MetaMap is a natural language processing tool developed and funded by the National Library of Medicine to map biomedical text to the Unified Medical Language System Metathesaurus by applying specific tags to clinically relevant terms. Although results are useful without additional postprocessing, these tags lack important contextual information. Our novel method takes terminology-driven semantic tags and incorporates those into a semantic frame that is task-specific to add necessary context to MetaMap. We use important contextual information to capture biomarker results to support Community Health System's use of Precision Medicine treatments for patients with cancer. For each biomarker, the name, type, numeric quantifiers, non-numeric qualifiers, and the time frame are extracted. These fields then associate biomarkers with their context in the pathology report such as test type, probe intensity, copy-number changes, and even failed results. A selection of 6,713 relevant reports contained the following standard-of-care biomarkers for metastatic breast cancer: breast cancer gene 1 and 2, estrogen receptor, progesterone receptor, human epidermal growth factor receptor 2, and programmed death-ligand 1. The method was tested on pathology reports from the internal pathology laboratory at Henry Ford Health System. A certified tumor registrar reviewed 400 tests, which showed > 95% accuracy for all extracted biomarker types. Using this new method, it is possible to extract high-quality, contextual biomarker information, and this represents a significant advance in biomarker extraction.

Using the Unified Medical Language System to Expand the Operative Stress Score – First Use Case

The Unified Medical Language System (UMLS) maps relationships between and within >100 biomedical vocabularies, including Current Procedural Terminology (CPT) codes, creating a powerful knowledge resource which can accelerate clinical research. We used synonymy and concepts relating hierarchical structure of CPT codes within the UMLS, (1) guiding surgical experts in expanding the Operative Stress Score (OSS) from 565 originally rated CPT codes to additional, 1,853 related procedures; (2) establishing validity of the association between the added OSS ratings and 30-day outcomes in VASQIP (2015-2018). The UMLS Metathesaurus and Semantic Network was converted into an interactive graph database (https://github.com/dbmi-pitt/UMLS-Graph) delineating ontology relatedness. From this UMLS-graph, the CPT hierarchy was queried obtaining all paths from each code to the hierarchical apex. Of 1,853 added ratings, 43% and 76% were siblings and cousins of original OSS CPT codes. Of 857,577 VASQIP cases (mean age, 64±11years; 91% male; 75% white), 786,122 (92%) and 71,455 (8%) were rated in the original and added OSS. Compared to original, added OSS cases included more females (14% versus 9%) and frail patients (25% versus 19%) undergoing high stress procedures (11% versus 8%; all P <.001). Postoperative mortality consistently increased with OSS. Very low stress procedures had <0.5% (original, 0.4% [95%CI, 0.4%-0.5%] versus added, 0.9% [95%CI, 0.6%-1.2%]) and very high 3.8% (original, 3.5% [95%CI, 3.0%-4.0%] versus added, 5.8% [95%CI, 4.6-7.3%]) mortality rates. The synonymy and concepts relating biomedical data within the UMLS can be abstracted and efficiently used to expand the utility of existing clinical research tools.

PhenClust, a standalone tool for identifying trends within sets of biological phenotypes using semantic similarity and the Unified Medical Language System metathesaurus.

ObjectivesWe sought to cluster biological phenotypes using semantic similarity and create an easy-to-install, stable, and reproducible tool.Materials and MethodsWe generated Phenotype Clustering (PhenClust)—a novel application of semantic similarity for interpreting biological phenotype associations—using the Unified Medical Language System (UMLS) metathesaurus, demonstrated the tool’s application, and developed Docker containers with stable installations of two UMLS versions.ResultsPhenClust identified disease clusters for drug network-associated phenotypes and a meta-analysis of drug target candidates. The Dockerized containers eliminated the requirement that the user install the UMLS metathesaurus.DiscussionClustering phenotypes summarized all phenotypes associated with a drug network and two drug candidates. Docker containers can support dissemination and reproducibility of tools that are otherwise limited due to insufficient software support.ConclusionPhenClust can improve interpretation of high-throughput biological analyses where many phenotypes are associated with a query and the Dockerized PhenClust achieved our objective of decreasing installation complexity.

Improving early diagnosis of rare diseases using Natural Language Processing in unstructured medical records: an illustration from Dravet syndrome

BackgroundThe growing use of Electronic Health Records (EHRs) is promoting the application of data mining in health-care. A promising use of big data in this field is to develop models to support early diagnosis and to establish natural history. Dravet Syndrome (DS) is a rare developmental and epileptic encephalopathy that commonly initiates in the first year of life with febrile seizures (FS). Age at diagnosis is often delayed after 2 years, as it is difficult to differentiate DS at onset from FS. We aimed to explore if some clinical terms (concepts) are significantly more used in the electronic narrative medical reports of individuals with DS before the age of 2 years compared to those of individuals with FS. These concepts would allow an earlier detection of patients with DS resulting in an earlier orientation toward expert centers that can provide early diagnosis and care.MethodsData were collected from the Necker Enfants Malades Hospital using a document-based data warehouse, Dr Warehouse, which employs Natural Language Processing, a computer technology consisting in processing written information. Using Unified Medical Language System Meta-thesaurus, phenotype concepts can be recognized in medical reports. We selected individuals with DS (DS Cohort) and individuals with FS (FS Cohort) with confirmed diagnosis after the age of 4 years. A phenome-wide analysis was performed evaluating the statistical associations between the phenotypes of DS and FS, based on concepts found in the reports produced before 2 years and using a series of logistic regressions.ResultsWe found significative higher representation of concepts related to seizures’ phenotypes distinguishing DS from FS in the first phases, namely the major recurrence of complex febrile convulsions (long-lasting and/or with focal signs) and other seizure-types. Some typical early onset non-seizure concepts also emerged, in relation to neurodevelopment and gait disorders.ConclusionsNarrative medical reports of individuals younger than 2 years with FS contain specific concepts linked to DS diagnosis, which can be automatically detected by software exploiting NLP. This approach could represent an innovative and sustainable methodology to decrease time of diagnosis of DS and could be transposed to other rare diseases.

Clinical Term Normalization Using Learned Edit Patterns and Subconcept Matching: System Development and Evaluation.

BackgroundClinical terms mentioned in clinical text are often not in their standardized forms as listed in clinical terminologies because of linguistic and stylistic variations. However, many automated downstream applications require clinical terms mapped to their corresponding concepts in clinical terminologies, thus necessitating the task of clinical term normalization.ObjectiveIn this paper, a system for clinical term normalization is presented that utilizes edit patterns to convert clinical terms into their normalized forms.MethodsThe edit patterns are automatically learned from the Unified Medical Language System (UMLS) Metathesaurus as well as from the given training data. The edit patterns are generalized sequences of edits that are derived from edit distance computations. The edit patterns are both character based as well as word based and are learned separately for different semantic types. In addition to these edit patterns, the system also normalizes clinical terms through the subconcepts mentioned within them.ResultsThe system was evaluated as part of the 2019 n2c2 Track 3 shared task of clinical term normalization. It obtained 80.79% accuracy on the standard test data. This paper includes ablation studies to evaluate the contributions of different components of the system. A challenging part of the task was disambiguation when a clinical term could be normalized to multiple concepts.ConclusionsThe learned edit patterns led the system to perform well on the normalization task. Given that the system is based on patterns, it is human interpretable and is also capable of giving insights about common variations of clinical terms mentioned in clinical text that are different from their standardized forms.

The impact of learning Unified Medical Language System knowledge embeddings in relation extraction from biomedical texts

We explored how knowledge embeddings (KEs) learned from the Unified Medical Language System (UMLS) Metathesaurus impact the quality of relation extraction on 2 diverse sets of biomedical texts. Two forms of KEs were learned for concepts and relation types from the UMLS Metathesaurus, namely lexicalized knowledge embeddings (LKEs) and unlexicalized KEs. A knowledge embedding encoder (KEE) enabled learning either LKEs or unlexicalized KEs as well as neural models capable of producing LKEs for mentions of biomedical concepts in texts and relation types that are not encoded in the UMLS Metathesaurus. This allowed us to design the relation extraction with knowledge embeddings (REKE) system, which incorporates either LKEs or unlexicalized KEs produced for relation types of interest and their arguments. The incorporation of either LKEs or unlexicalized KE in REKE advances the state of the art in relation extraction on 2 relation extraction datasets: the 2010 i2b2/VA dataset and the 2013 Drug-Drug Interaction Extraction Challenge corpus. Moreover, the impact of LKEs is superior, achieving F1 scores of 78.2 and 82.0, respectively. REKE not only highlights the importance of incorporating knowledge encoded in the UMLS Metathesaurus in a novel way, through 2 possible forms of KEs, but it also showcases the subtleties of incorporating KEs in relation extraction systems. Incorporating LKEs informed by the UMLS Metathesaurus in a relation extraction system operating on biomedical texts shows significant promise. We present the REKE system, which establishes new state-of-the-art results for relation extraction on 2 datasets when using LKEs.

Semantic Deep Learning: Prior Knowledge and a Type of Four-Term Embedding Analogy to Acquire Treatments for Well-Known Diseases.

BackgroundHow to treat a disease remains to be the most common type of clinical question. Obtaining evidence-based answers from biomedical literature is difficult. Analogical reasoning with embeddings from deep learning (embedding analogies) may extract such biomedical facts, although the state-of-the-art focuses on pair-based proportional (pairwise) analogies such as man:woman::king:queen (“queen = −man +king +woman”).ObjectiveThis study aimed to systematically extract disease treatment statements with a Semantic Deep Learning (SemDeep) approach underpinned by prior knowledge and another type of 4-term analogy (other than pairwise).MethodsAs preliminaries, we investigated Continuous Bag-of-Words (CBOW) embedding analogies in a common-English corpus with five lines of text and observed a type of 4-term analogy (not pairwise) applying the 3CosAdd formula and relating the semantic fields person and death: “dagger = −Romeo +die +died” (search query: −Romeo +die +died). Our SemDeep approach worked with pre-existing items of knowledge (what is known) to make inferences sanctioned by a 4-term analogy (search query −x +z1 +z2) from CBOW and Skip-gram embeddings created with a PubMed systematic reviews subset (PMSB dataset). Stage1: Knowledge acquisition. Obtaining a set of terms, candidate y, from embeddings using vector arithmetic. Some n-gram pairs from the cosine and validated with evidence (prior knowledge) are the input for the 3cosAdd, seeking a type of 4-term analogy relating the semantic fields disease and treatment. Stage 2: Knowledge organization. Identification of candidates sanctioned by the analogy belonging to the semantic field treatment and mapping these candidates to unified medical language system Metathesaurus concepts with MetaMap. A concept pair is a brief disease treatment statement (biomedical fact). Stage 3: Knowledge validation. An evidence-based evaluation followed by human validation of biomedical facts potentially useful for clinicians.ResultsWe obtained 5352 n-gram pairs from 446 search queries by applying the 3CosAdd. The microaveraging performance of MetaMap for candidate y belonging to the semantic field treatment was F-measure=80.00% (precision=77.00%, recall=83.25%). We developed an empirical heuristic with some predictive power for clinical winners, that is, search queries bringing candidate y with evidence of a therapeutic intent for target disease x. The search queries -asthma +inhaled_corticosteroids +inhaled_corticosteroid and -epilepsy +valproate +antiepileptic_drug were clinical winners, finding eight evidence-based beneficial treatments.ConclusionsExtracting treatments with therapeutic intent by analogical reasoning from embeddings (423K n-grams from the PMSB dataset) is an ambitious goal. Our SemDeep approach is knowledge-based, underpinned by embedding analogies that exploit prior knowledge. Biomedical facts from embedding analogies (4-term type, not pairwise) are potentially useful for clinicians. The heuristic offers a practical way to discover beneficial treatments for well-known diseases. Learning from deep learning models does not require a massive amount of data. Embedding analogies are not limited to pairwise analogies; hence, analogical reasoning with embeddings is underexploited.

Ontological and Non-Ontological Resources for Associating Medical Dictionary for Regulatory Activities Terms to SNOMED Clinical Terms With Semantic Properties.

Background: Formal definitions allow selecting terms (e.g., identifying all terms related to “Infectious disease” using the query “has causative agent organism”) and terminological reasoning (e.g., “hepatitis B” is a “hepatitis” and is an “infectious disease”). However, the standard international terminology Medical Dictionary for Regulatory Activities (MedDRA) used for coding adverse drug reactions in pharmacovigilance databases does not beneficiate from such formal definitions. Our objective was to evaluate the potential of reuse of ontological and non-ontological resources for generating such definitions for MedDRA. Methods: We developed several methods that collectively allow a semiautomatic semantic enrichment of MedDRA: 1) using MedDRA-to-SNOMED Clinical Terms (SNOMED CT) mappings (available in the Unified Medical Language System metathesaurus or other mapping resources, e.g., the MedDRA preferred term “hepatitis B” is associated to the SNOMED CT concept “type B viral hepatitis”) to extract term definitions (e.g., “hepatitis B” is associated with the following properties: has finding site liver structure, has associated morphology inflammation morphology, and has causative agent hepatitis B virus); 2) using MedDRA labels and lexical/syntactic methods for automatic decomposition of complex MedDRA terms (e.g., the MedDRA systems organ class “blood and lymphatic system disorders” is decomposed in blood system disorders and lymphatic system disorders) or automatic suggestions of properties (e.g., the string “cyclic” in preferred term “cyclic neutropenia” leads to the property has clinical course cyclic). Results: The Unified Medical Language System metathesaurus was the main ontological resource reusable for generating formal definitions for MedDRA terms. The non-ontological resources (another mapping resource provided by Nadkarni and Darer in 2010 and MedDRA labels) allowed defining few additional preferred terms. While the Ci4SeR tool helped the curator to define 1,935 terms by suggesting potential supplemental relations based on the parents’ and siblings’ semantic definition, defining manually all MedDRA terms remains expensive in time. Discussion: Several ontological and non-ontological resources are available for associating MedDRA terms to SNOMED CT concepts with semantic properties, but providing manual definitions is still necessary. The ontology of adverse events is a possible alternative but does not cover all MedDRA terms either. Perspectives are to implement more efficient techniques to find more logical relations between SNOMED CT and MedDRA in an automated way.

Construction of Disease Similarity Networks Using Concept Embedding and Ontology.

Discovering disease similarities are beneficial for the diagnosis and treatment of mental diseases. In this research, we proposed a data driven method, that is, integrating a variety of publicly available data resources including Unified Medical Language System (UMLS) Metathesaurus, Systematized Nomenclature of Medicine -- Clinical Terms (SNOMED CT) and cui2vec concept embedding to construct a mental disease similarity network. The resulting mental disease similarity network offered a new view for navigating and investigating disease relations; it also revealed popular mental disease in the literature in terms of the number of connections and similarities with other diseases. It shows that depressive disorder is directly connected with nine other popular diseases and connects 52 other diseases in the network. The top three popular mental diseases are depressive disorder, dysthymia (now known as persistent depressive disorder), and neurosis. Future research will focus on studying the clusters generated from the similarity network.

UMLS mapping and Word embeddings for ICD code assignment using the MIMIC-III intensive care database.

Diagnosis codes are used as a billing mechanism in the Electronic Health Record and have the capability to benefit decision support systems, which aim to assist coders by suggesting a relevant subset of potential codes to choose from. Due to the large set of possible labels and length of patient records, automatic ICD code assignment is considered to be a challenging task within the field of multi-label classification. This paper introduces a baseline for automatic ICD code assignment using Support Vector Machines (SVM) and FastText with Unified Medical Language System (UMLS) metathesaurus mappings into word embedding models. Training data is obtained from the Medical Information Mart for Intensive Care (MIMIC-III) database and extended with 'is-a' relationships from ICD-9 hierarchy. FastText is evaluated with different label count estimations, of which an approach based on label cardinality yields a F1-Score of 62.2%. FastText achieves high recall results and mentionable performance improvements over previous models. Reported values are obtained through 10-fold cross-validation.

Ontological representation-oriented term normalization and standardization of the Research Domain Criteria.

The Research Domain Criteria, launched by the National Institute of Mental Health, is a new dimensional and interdisciplinary research framework for mental disorders. The Research Domain Criteria matrix is its core part. Since an ontology has the strengths of supporting semantic inferencing and automatic data processing, we would like to transform the Research Domain Criteria matrix into an ontological structure. In terms of data normalization, which is the essential part of an ontology representation, the Research Domain Criteria elements (mainly in the Units of Analysis) have some limitations. In this article, we propose a series of solutions to improve data normalization of the Research Domain Criteria elements in the Units of Analysis, including leveraging standard terminologies (i.e. the Unified Medical Language System Metathesaurus), context-combining queries, and domain expertise. The evaluation results show the positive (Yes) percentage is more than 80 percent, indicating our work is favorably received by the mental health professionals, and we have formed a good data foundation for the Research Domain Criteria ontological representation in the future work.

Feature extraction for phenotyping from semantic and knowledge resources.

Phenotyping algorithms can efficiently and accurately identify patients with a specific disease phenotype and construct electronic health records (EHR)-based cohorts for subsequent clinical or genomic studies. Previous studies have introduced unsupervised EHR-based feature selection methods that yielded algorithms with high accuracy. However, those selection methods still require expert intervention to tweak the parameter settings according to the EHR data distribution for each phenotype. To further accelerate the development of phenotyping algorithms, we propose a fully automated and robust unsupervised feature selection method that leverages only publicly available medical knowledge sources, instead of EHR data. SEmantics-Driven Feature Extraction (SEDFE) collects medical concepts from online knowledge sources as candidate features and gives them vector-form distributional semantic representations derived with neural word embedding and the Unified Medical Language System Metathesaurus. A number of features that are semantically closest and that sufficiently characterize the target phenotype are determined by a linear decomposition criterion and are selected for the final classification algorithm. SEDFE was compared with the EHR-based SAFE algorithm and domain experts on feature selection for the classification of five phenotypes including coronary artery disease, rheumatoid arthritis, Crohn's disease, ulcerative colitis, and pediatric pulmonary arterial hypertension using both supervised and unsupervised approaches. Algorithms yielded by SEDFE achieved comparable accuracy to those yielded by SAFE and expert-curated features. SEDFE is also robust to the input semantic vectors. SEDFE attains satisfying performance in unsupervised feature selection for EHR phenotyping. Both fully automated and EHR-independent, this method promises efficiency and accuracy in developing algorithms for high-throughput phenotyping.

Applying natural language processing techniques to develop a task-specific EMR interface for timely stroke thrombolysis: A feasibility study

ObjectiveTo reduce errors in determining eligibility for intravenous thrombolytic therapy (IVT) in stroke patients through use of an enhanced task-specific electronic medical record (EMR) interface powered by natural language processing (NLP) techniques. Materials and methodsThe information processing algorithm utilized MetaMap to extract medical concepts from IVT eligibility criteria and expanded the concepts using the Unified Medical Language System Metathesaurus. Concepts identified from clinical notes by MetaMap were compared to those from IVT eligibility criteria. The task-specific EMR interface displays IVT-relevant information by highlighting phrases that contain matched concepts. Clinical usability was assessed with clinicians staffing the acute stroke team by comparing user performance while using the task-specific and the current EMR interfaces. ResultsThe algorithm identified IVT-relevant concepts with micro-averaged precisions, recalls, and F1 measures of 0.998, 0.812, and 0.895 at the phrase level and of 1, 0.972, and 0.986 at the document level. Users using the task-specific interface achieved a higher accuracy score than those using the current interface (91% versus 80%, p = 0.016) in assessing the IVT eligibility criteria. The completion time between the interfaces was statistically similar (2.46 min versus 1.70 min, p = 0.754). DiscussionAlthough the information processing algorithm had room for improvement, the task-specific EMR interface significantly reduced errors in assessing IVT eligibility criteria. ConclusionThe study findings provide evidence to support an NLP enhanced EMR system to facilitate IVT decision-making by presenting meaningful and timely information to clinicians, thereby offering a new avenue for improvements in acute stroke care.

Text Simplification Using Consumer Health Vocabulary to Generate Patient-Centered Radiology Reporting: Translation and Evaluation.

BackgroundRadiology reporting is a clinically oriented form of documentation that reflects critical information for patients about their health care processes. Realizing its importance, many medical institutions have started providing radiology reports in patient portals. The gain, however, can be limited because of medical language barriers, which require a way for customizing these reports for patients. The open-access, collaborative consumer health vocabulary (CHV) is a terminology system created for such purposes and can be the basis of lexical simplification processes for clinical notes.ObjectiveThe aim of this study was to examine the comprehensibility and suitability of CHV in simplifying radiology reports for consumers. This was done by characterizing the content coverage and the lexical similarity between the terms in the reports and the CHV-preferred terms.MethodsThe overall procedure was divided into the following two main stages: (1) translation and (2) evaluation. The translation process involved using MetaMap to link terms in the reports to CHV concepts. This is followed by replacing the terms with CHV-preferred terms using the concept names and sources table (MRCONSO) in the Unified Medical Language System (UMLS) Metathesaurus. In the second stage, medical terms in the reports and general terms that are used to describe medical phenomena were selected and evaluated by comparing the words in the original reports with the translated ones. The evaluation includes measuring the content coverage, investigating lexical similarity, and finding trends in missing concepts.ResultsOf the 792 terms selected from the radiology reports, 695 of them could be mapped directly to CHV concepts, indicating a content coverage of 88.5%. A total of 51 of the concepts (53%, 51/97) that could not be mapped are names of human anatomical structures and regions, followed by 28 anatomical descriptions and pathological variations (29%, 28/97). In addition, 12 radiology techniques and projections represented 12% of the unmapped concepts, whereas the remaining six concepts (6%, 12/97) were physiological descriptions. The rate of lexical similarity between the CHV-preferred terms and the terms in the radiology reports was approximately 72.6%.ConclusionsThe CHV covered a high percentage of concepts found in the radiology reports, but unmapped concepts are associated with areas that are commonly found in radiology reporting. CHV terms also showed a high percentage of lexical similarity with terms in the reports, which contain a myriad of medical jargon. This suggests that many CHV terms might not be suitable for lay consumers who would not be facile with radiology-specific vocabulary. Therefore, further patient-centered content changes are needed of the CHV to increase its usefulness and facilitate its integration into consumer-oriented applications.

Medical subdomain classification of clinical notes using a machine learning-based natural language processing approach

BackgroundThe medical subdomain of a clinical note, such as cardiology or neurology, is useful content-derived metadata for developing machine learning downstream applications. To classify the medical subdomain of a note accurately, we have constructed a machine learning-based natural language processing (NLP) pipeline and developed medical subdomain classifiers based on the content of the note.MethodsWe constructed the pipeline using the clinical NLP system, clinical Text Analysis and Knowledge Extraction System (cTAKES), the Unified Medical Language System (UMLS) Metathesaurus, Semantic Network, and learning algorithms to extract features from two datasets — clinical notes from Integrating Data for Analysis, Anonymization, and Sharing (iDASH) data repository (n = 431) and Massachusetts General Hospital (MGH) (n = 91,237), and built medical subdomain classifiers with different combinations of data representation methods and supervised learning algorithms. We evaluated the performance of classifiers and their portability across the two datasets.ResultsThe convolutional recurrent neural network with neural word embeddings trained-medical subdomain classifier yielded the best performance measurement on iDASH and MGH datasets with area under receiver operating characteristic curve (AUC) of 0.975 and 0.991, and F1 scores of 0.845 and 0.870, respectively. Considering better clinical interpretability, linear support vector machine-trained medical subdomain classifier using hybrid bag-of-words and clinically relevant UMLS concepts as the feature representation, with term frequency-inverse document frequency (tf-idf)-weighting, outperformed other shallow learning classifiers on iDASH and MGH datasets with AUC of 0.957 and 0.964, and F1 scores of 0.932 and 0.934 respectively. We trained classifiers on one dataset, applied to the other dataset and yielded the threshold of F1 score of 0.7 in classifiers for half of the medical subdomains we studied.ConclusionOur study shows that a supervised learning-based NLP approach is useful to develop medical subdomain classifiers. The deep learning algorithm with distributed word representation yields better performance yet shallow learning algorithms with the word and concept representation achieves comparable performance with better clinical interpretability. Portable classifiers may also be used across datasets from different institutions.