Ontology-driven weak supervision for clinical entity classification in electronic health records.
Ontology highlight
ABSTRACT:
SUBMITTER: Fries JA
PROVIDER: S-EPMC8016863 | biostudies-literature | 2021 Apr
REPOSITORIES: biostudies-literature
Publications
Ontology-driven weak supervision for clinical entity classification in electronic health records.
Fries Jason A JA Steinberg Ethan E Khattar Saelig S Fleming Scott L SL Posada Jose J Callahan Alison A Shah Nigam H NH
Nature communications 20210401 1
In the electronic health record, using clinical notes to identify entities such as disorders and their temporality (e.g. the order of an event relative to a time index) can inform many important analyses. However, creating training data for clinical entity tasks is time consuming and sharing labeled data is challenging due to privacy concerns. The information needs of the COVID-19 pandemic highlight the need for agile methods of training machine learning models for clinical notes. We present Tro ...[more]
Similar Datasets
Project description:ObjectiveTo develop deep learning models to recognize ophthalmic examination components from clinical notes in electronic health records (EHR) using a weak supervision approach.MethodsA corpus of 39,099 ophthalmology notes weakly labeled for 24 examination entities was assembled from the EHR of one academic center. Four pre-trained transformer-based language models (DistilBert, BioBert, BlueBert, and ClinicalBert) were fine-tuned to this named entity recognition task and compared to a baseline regular expression model. Models were evaluated on the weakly labeled test dataset, a human-labeled sample of that set, and a human-labeled independent dataset.ResultsOn the weakly labeled test set, all transformer-based models had recall > 0.93, with precision varying from 0.815 to 0.843. The baseline model had lower recall (0.769) and precision (0.682). On the human-annotated sample, the baseline model had high recall (0.962, 95 % CI 0.955-0.067) with variable precision across entities (0.081-0.999). Bert models had recall ranging from 0.771 to 0.831, and precision >=0.973. On the independent dataset, precision was 0.926 and recall 0.458 for BlueBert. The baseline model had better recall (0.708, 95 % CI 0.674-0.738) but worse precision (0.399, 95 % CI -0.352-0.451).ConclusionWe developed the first deep learning system to recognize eye examination components from clinical notes, leveraging a novel opportunity for weak supervision. Transformer-based models had high precision on human-annotated labels, whereas the baseline model had poor precision but higher recall. This system may be used to improve cohort and feature identification using free-text notes.Our weakly supervised approach may help amass large datasets of domain-specific entities from EHRs in many fields.
| S-EPMC9901505 | biostudies-literature
Project description:Pneumonia and pulmonary thromboembolism (PTE) are both respiratory diseases; their diagnosis is difficult due to their similarity in symptoms, medical subjectivity, and the large amount of information from different sources necessary for a correct diagnosis. Analysis of such clinical data using computational tools could help medical staff reduce time, increase diagnostic certainty, and improve patient care during hospitalization. In addition, no studies have been found that analyze all clinical information on the Mexican population in the Spanish language. Therefore, this work performs automatic diagnosis of pneumonia and pulmonary thromboembolism using machine-learning tools along with clinical laboratory information (structured data) and clinical text (unstructured data) obtained from electronic health records. A cohort of 173 clinical records was obtained from the Mexican Social Security Institute. The data were preprocessed, transformed, and adjusted to be analyzed using several machine-learning algorithms. For structured data, naïve Bayes, support vector machine, decision trees, AdaBoost, random forest, and multilayer perceptron were used; for unstructured data, a BiLSTM was used. K-fold cross-validation and leave-one-out were used for evaluation of structured data, and hold-out was used for unstructured data; additionally, 1-vs.-1 and 1-vs.-rest approaches were used. Structured data results show that the highest AUC-ROC was achieved by the naïve Bayes algorithm classifying PTE vs. pneumonia (87.0%), PTE vs. control (75.1%), and pneumonia vs. control (85.2%) with the 1-vs.-1 approach; for the 1-vs.-rest approach, the best performance was reported in pneumonia vs. rest (86.3%) and PTE vs. rest (79.7%) using naïve Bayes, and control vs. diseases (79.8%) using decision trees. Regarding unstructured data, the results do not present a good AUC-ROC; however, the best F1-score were scored for control vs. disease (72.7%) in the 1-vs.-rest approach and control vs. pneumonia (63.6%) in the 1-to-1 approach. Additionally, several decision trees were obtained to identify important attributes for automatic diagnosis for structured data, particularly for PTE vs. pneumonia. Based on the experiments, the structured datasets present the highest values. Results suggest using naïve Bayes and structured data to automatically diagnose PTE vs. pneumonia. Moreover, using decision trees allows the observation of some decision criteria that the medical staff could consider for diagnosis.
| S-EPMC9601338 | biostudies-literature
Project description:OBJECTIVE:To develop an ontology to identify pregnant women from computerised medical record systems with dissimilar coding systems in a primary care sentinel network. MATERIALS AND METHODS:We used a three-step approach to develop our pregnancy ontology in two different coding schemata, one hierarchical and the other polyhierarchical. We developed a coding system-independent pregnancy case identification algorithm using the Royal College of General Practitioners Research and Surveillance Centre sentinel network database which held 1.8 million patients' data drawn from 150 primary care providers. We tested the algorithm by examining individual patient records in a 10% random sample of all women aged 29 in each year from 2004 to 2016. We did an external comparison with national pregnancy data. We used ?2 test to compare results obtained for the two different coding schemata. RESULTS:243 005 women (median age 29 years at start of pregnancy) had 405 591 pregnancies from 2004 to 2016 of which 333 689 went to term. We found no significant difference between results obtained for two populations using different coding schemata. Pregnancy mean ages did not differ significantly from national data. DISCUSSION:This ontologically driven algorithm enables consistent analysis across data drawn from populations using different coding schemata. It could be applied to other hierarchical coding systems (eg, International Classification of Disease) or polyhierarchical systems (eg, SNOMED CT to which our health system is currently migrating). CONCLUSION:This ontological approach will improve our surveillance in particular of influenza vaccine exposure in pregnancy.
| S-EPMC7062332 | biostudies-literature
Project description:Reducing the burden of late-life morbidity requires an understanding of the mechanisms of ageing-related diseases (ARDs), defined as diseases that accumulate with increasing age. This has been hampered by the lack of formal criteria to identify ARDs. Here, we present a framework to identify ARDs using two complementary methods consisting of unsupervised machine learning and actuarial techniques, which we applied to electronic health records (EHRs) from 3,009,048 individuals in England using primary care data from the Clinical Practice Research Datalink (CPRD) linked to the Hospital Episode Statistics admitted patient care dataset between 1 April 2010 and 31 March 2015 (mean age 49.7 years (s.d. 18.6), 51% female, 70% white ethnicity). We grouped 278 high-burden diseases into nine main clusters according to their patterns of disease onset, using a hierarchical agglomerative clustering algorithm. Four of these clusters, encompassing 207 diseases spanning diverse organ systems and clinical specialties, had rates of disease onset that clearly increased with chronological age. However, the ages of onset for these four clusters were strikingly different, with median age of onset 82 years (IQR 82-83) for Cluster 1, 77 years (IQR 75-77) for Cluster 2, 69 years (IQR 66-71) for Cluster 3 and 57 years (IQR 54-59) for Cluster 4. Fitting to ageing-related actuarial models confirmed that the vast majority of these 207 diseases had a high probability of being ageing-related. Cardiovascular diseases and cancers were highly represented, while benign neoplastic, skin and psychiatric conditions were largely absent from the four ageing-related clusters. Our framework identifies and clusters ARDs and can form the basis for fundamental and translational research into ageing pathways.
| S-EPMC7859412 | biostudies-literature
Project description:ObjectiveTo measure the rate of non-publication and assess possible publication bias in clinical trials of electronic health records.MethodsWe searched ClinicalTrials.gov to identify registered clinical trials of electronic health records and searched the biomedical literature and contacted trial investigators to determine whether the results of the trials were published. Publications were judged as positive, negative, or neutral according to the primary outcome.ResultsSeventy-six percent of trials had publications describing trial results; of these, 74% were positive, 21% were neutral, and 4% were negative (harmful). Of unpublished studies for which the investigator responded, 43% were positive, 57% were neutral, and none were negative; the lower rate of positive results was significant (p<0.001).ConclusionThe rate of non-publication in electronic health record studies is similar to that in other biomedical studies. There appears to be a bias toward publication of positive trials in this domain.
| S-EPMC3662474 | biostudies-literature
Project description:AimsElectronic health records (EHR) linked to Digital Imaging and Communications in Medicine (DICOM), biological specimens, and deep learning (DL) algorithms could potentially improve patient care through automated case detection and surveillance. We hypothesized that by applying keyword searches to routinely stored EHR, in conjunction with AI-powered automated reading of DICOM echocardiography images and analysing biomarkers from routinely stored plasma samples, we were able to identify heart failure (HF) patients.Methods and resultsWe used EHR data between 1993 and 2021 from Tayside and Fife (~20% of the Scottish population). We implemented a keyword search strategy complemented by filtering based on International Classification of Diseases (ICD) codes and prescription data to EHR data set. We then applied DL for the automated interpretation of echocardiographic DICOM images. These methods were then integrated with the analysis of routinely stored plasma samples to identify and categorize patients into HF with reduced ejection fraction (HFrEF), HF with preserved ejection fraction (HFpEF), and controls without HF. The final diagnosis was verified through a manual review of medical records, measured natriuretic peptides in stored blood samples, and by comparing clinical outcomes among groups. In our study, we selected the patient cohort through an algorithmic workflow. This process started with 60 850 EHR data and resulted in a final cohort of 578 patients, divided into 186 controls, 236 with HFpEF, and 156 with HFrEF, after excluding individuals with mismatched data or significant valvular heart disease. The analysis of baseline characteristics revealed that compared with controls, patients with HFrEF and HFpEF were generally older, had higher BMI, and showed a greater prevalence of co-morbidities such as diabetes, COPD, and CKD. Echocardiographic analysis, enhanced by DL, provided high coverage, and detailed insights into cardiac function, showing significant differences in parameters such as left ventricular diameter, ejection fraction, and myocardial strain among the groups. Clinical outcomes highlighted a higher risk of hospitalization and mortality for HF patients compared with controls, with particularly elevated risk ratios for both HFrEF and HFpEF groups. The concordance between the algorithmic selection of patients and manual validation demonstrated high accuracy, supporting the effectiveness of our approach in identifying and classifying HF subtypes, which could significantly impact future HF diagnosis and management strategies.ConclusionsOur study highlights the feasibility of combining keyword searches in EHR, DL automated echocardiographic interpretation, and biobank resources to identify HF subtypes.
| S-EPMC11424299 | biostudies-literature
Project description:BackgroundAutism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized by restricted, repetitive behavior, and impaired social communication and interactions. However, significant challenges remain in diagnosing and subtyping ASD due in part to the lack of a validated, standardized vocabulary to characterize clinical phenotypic presentation of ASD. Although the human phenotype ontology (HPO) plays an important role in delineating nuanced phenotypes for rare genetic diseases, it is inadequate to capture characteristic of behavioral and psychiatric phenotypes for individuals with ASD. There is a clear need, therefore, for a well-established phenotype terminology set that can assist in characterization of ASD phenotypes from patients' clinical narratives.MethodsTo address this challenge, we used natural language processing (NLP) techniques to identify and curate ASD phenotypic terms from high-quality unstructured clinical notes in the electronic health record (EHR) on 8499 individuals with ASD, 8177 individuals with non-ASD psychiatric disorders, and 8482 individuals without a documented psychiatric disorder. We further performed dimensional reduction clustering analysis to subgroup individuals with ASD, using nonnegative matrix factorization method.ResultsThrough a note-processing pipeline that includes several steps of state-of-the-art NLP approaches, we identified 3336 ASD terms linking to 1943 unique medical concepts, which represents among the largest ASD terminology set to date. The extracted ASD terms were further organized in a formal ontology structure similar to the HPO. Clustering analysis showed that these terms could be used in a diagnostic pipeline to differentiate individuals with ASD from individuals with other psychiatric disorders.ConclusionOur ASD phenotype ontology can assist clinicians and researchers in characterizing individuals with ASD, facilitating automated diagnosis, and subtyping individuals with ASD to facilitate personalized therapeutic decision-making.
| S-EPMC9128253 | biostudies-literature
Project description:BackgroundElectronic health records (EHRs) contain valuable information for clinical research; however, the sensitive nature of healthcare data presents security and confidentiality challenges. De-identification is therefore essential to protect personal data in EHRs and comply with government regulations. Named entity recognition (NER) methods have been proposed to remove personal identifiers, with deep learning-based models achieving better performance. However, manual annotation of training data is time-consuming and expensive. The aim of this study was to develop an automatic de-identification pipeline for all kinds of clinical documents based on a distant supervised method to significantly reduce the cost of manual annotations and to facilitate the transfer of the de-identification pipeline to other clinical centers.MethodsWe proposed an automated annotation process for French clinical de-identification, exploiting data from the eHOP clinical data warehouse (CDW) of the CHU de Rennes and national knowledge bases, as well as other features. In addition, this paper proposes an assisted data annotation solution using the Prodigy annotation tool. This approach aims to reduce the cost required to create a reference corpus for the evaluation of state-of-the-art NER models. Finally, we evaluated and compared the effectiveness of different NER methods.ResultsA French de-identification dataset was developed in this work, based on EHRs provided by the eHOP CDW at Rennes University Hospital, France. The dataset was rich in terms of personal information, and the distribution of entities was quite similar in the training and test datasets. We evaluated a Bi-LSTM + CRF sequence labeling architecture, combined with Flair + FastText word embeddings, on a test set of manually annotated clinical reports. The model outperformed the other tested models with a significant F1 score of 96,96%, demonstrating the effectiveness of our automatic approach for deidentifying sensitive information.ConclusionsThis study provides an automatic de-identification pipeline for clinical notes, which can facilitate the reuse of EHRs for secondary purposes such as clinical research. Our study highlights the importance of using advanced NLP techniques for effective de-identification, as well as the need for innovative solutions such as distant supervision to overcome the challenge of limited annotated data in the medical domain.
| S-EPMC10870625 | biostudies-literature
Project description:The privacy of patients and the security of their information is the most imperative barrier to entry when considering the adoption of electronic health records in the healthcare industry. Considering current legal regulations, this review seeks to analyze and discuss prominent security techniques for healthcare organizations seeking to adopt a secure electronic health records system. Additionally, the researchers sought to establish a foundation for further research for security in the healthcare industry. The researchers utilized the Texas State University Library to gain access to three online databases: PubMed (MEDLINE), CINAHL, and ProQuest Nursing and Allied Health Source. These sources were used to conduct searches on literature concerning security of electronic health records containing several inclusion and exclusion criteria. Researchers collected and analyzed 25 journals and reviews discussing security of electronic health records, 20 of which mentioned specific security methods and techniques. The most frequently mentioned security measures and techniques are categorized into three themes: administrative, physical, and technical safeguards. The sensitive nature of the information contained within electronic health records has prompted the need for advanced security techniques that are able to put these worries at ease. It is imperative for security techniques to cover the vast threats that are present across the three pillars of healthcare.
| S-EPMC5522514 | biostudies-other
Project description:BackgroundPatterns of disease incidence can identify new risk factors for the disease or provide insight into the etiology. For example, allergies and infectious diseases have been shown to follow periodic temporal patterns due to seasonal changes in environmental or infectious agents. Previous work searching for seasonal or other temporal patterns in disease diagnosis rates has been limited both in the scope of the diseases examined and in the ability to distinguish unexpected seasonal patterns. Electronic Health Records (EHR) compile extensive longitudinal clinical information, constituting a unique source for discovery of trends in occurrence of disease. However, the data suffer from inherent biases that preclude an identification of temporal trends.MethodsMotivated by observation of the biases in this data source, we developed a method (Lomb-Scargle periodograms in detrended data, LSP-detrend) to find periodic patterns by adjusting the temporal information for broad trends in incidence, as well as seasonal changes in total hospitalizations. LSP-detrend can sensitively uncover periodic temporal patterns in the corrected data and identify the significance of the trend. We apply LSP-detrend to a compilation of records from 1.5 million patients encoded by ICD-9-CM (International Classification of Diseases, Ninth Revision, Clinical Modification), including 2,805 disorders with more than 500 occurrences across a 12 year period, recorded from 1.5 million patients.Results and conclusionsAlthough EHR data, and ICD-9 coded records in particular, were not created with the intention of aggregated use for research, these data can in fact be mined for periodic patterns in incidence of disease, if confounders are properly removed. Of all diagnoses, around 10% are identified as seasonal by LSP-detrend, including many known phenomena. We robustly reproduce previous findings, even for relatively rare diseases. For instance, Kawasaki disease, a rare childhood disease that has been associated with weather patterns, is detected as strongly linked with winter months. Among the novel results, we find a bi-annual increase in exacerbations of myasthenia gravis, a potentially life threatening complication of an autoimmune disease. We dissect the causes of this seasonal incidence and propose that factors predisposing patients to this event vary through the year.
| S-EPMC4158606 | biostudies-literature