Building a best-in-class automated de-identification tool for electronic health records through ensemble learning.
Ontology highlight
ABSTRACT:
SUBMITTER: Murugadoss K
PROVIDER: S-EPMC8212138 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:BACKGROUND:Adverse events in health care entail substantial burdens to health care systems, institutions, and patients. Retrospective trigger tools are often manually applied to detect AEs, although automated approaches using electronic health records may offer real-time adverse event detection, allowing timely corrective interventions. OBJECTIVE:The aim of this systematic review was to describe current study methods and challenges regarding the use of automatic trigger tool-based adverse event detection methods in electronic health records. In addition, we aimed to appraise the applied studies' designs and to synthesize estimates of adverse event prevalence and diagnostic test accuracy of automatic detection methods using manual trigger tool as a reference standard. METHODS:PubMed, EMBASE, CINAHL, and the Cochrane Library were queried. We included observational studies, applying trigger tools in acute care settings, and excluded studies using nonhospital and outpatient settings. Eligible articles were divided into diagnostic test accuracy studies and prevalence studies. We derived the study prevalence and estimates for the positive predictive value. We assessed bias risks and applicability concerns using Quality Assessment tool for Diagnostic Accuracy Studies-2 (QUADAS-2) for diagnostic test accuracy studies and an in-house developed tool for prevalence studies. RESULTS:A total of 11 studies met all criteria: 2 concerned diagnostic test accuracy and 9 prevalence. We judged several studies to be at high bias risks for their automated detection method, definition of outcomes, and type of statistical analyses. Across all the 11 studies, adverse event prevalence ranged from 0% to 17.9%, with a median of 0.8%. The positive predictive value of all triggers to detect adverse events ranged from 0% to 100% across studies, with a median of 40%. Some triggers had wide ranging positive predictive value values: (1) in 6 studies, hypoglycemia had a positive predictive value ranging from 15.8% to 60%; (2) in 5 studies, naloxone had a positive predictive value ranging from 20% to 91%; (3) in 4 studies, flumazenil had a positive predictive value ranging from 38.9% to 83.3%; and (4) in 4 studies, protamine had a positive predictive value ranging from 0% to 60%. We were unable to determine the adverse event prevalence, positive predictive value, preventability, and severity in 40.4%, 10.5%, 71.1%, and 68.4% of the studies, respectively. These studies did not report the overall number of records analyzed, triggers, or adverse events; or the studies did not conduct the analysis. CONCLUSIONS:We observed broad interstudy variation in reported adverse event prevalence and positive predictive value. The lack of sufficiently described methods led to difficulties regarding interpretation. To improve quality, we see the need for a set of recommendations to endorse optimal use of research designs and adequate reporting of future adverse event detection studies.
| S-EPMC6000482 | biostudies-literature
Project description:OBJECTIVE:Identification of drugs, associated medication entities, and interactions among them are crucial to prevent unwanted effects of drug therapy, known as adverse drug events. This article describes our participation to the n2c2 shared-task in extracting relations between medication-related entities in electronic health records. MATERIALS AND METHODS:We proposed an ensemble approach for relation extraction and classification between drugs and medication-related entities. We incorporated state-of-the-art named-entity recognition (NER) models based on bidirectional long short-term memory (BiLSTM) networks and conditional random fields (CRF) for end-to-end extraction. We additionally developed separate models for intra- and inter-sentence relation extraction and combined them using an ensemble method. The intra-sentence models rely on bidirectional long short-term memory networks and attention mechanisms and are able to capture dependencies between multiple related pairs in the same sentence. For the inter-sentence relations, we adopted a neural architecture that utilizes the Transformer network to improve performance in longer sequences. RESULTS:Our team ranked third with a micro-averaged F1 score of 94.72% and 87.65% for relation and end-to-end relation extraction, respectively (Tracks 2 and 3). Our ensemble effectively takes advantages from our proposed models. Analysis of the reported results indicated that our proposed approach is more generalizable than the top-performing system, which employs additional training data- and corpus-driven processing techniques. CONCLUSIONS:We proposed a relation extraction system to identify relations between drugs and medication-related entities. The proposed approach is independent of external syntactic tools. Analysis showed that by using latent Drug-Drug interactions we were able to significantly improve the performance of non-Drug-Drug pairs in EHRs.
| S-EPMC6913215 | biostudies-literature
Project description:BackgroundText-based patient medical records are a vital resource in medical research. In order to preserve patient confidentiality, however, the U.S. Health Insurance Portability and Accountability Act (HIPAA) requires that protected health information (PHI) be removed from medical records before they can be disseminated. Manual de-identification of large medical record databases is prohibitively expensive, time-consuming and prone to error, necessitating automatic methods for large-scale, automated de-identification.MethodsWe describe an automated Perl-based de-identification software package that is generally usable on most free-text medical records, e.g., nursing notes, discharge summaries, X-ray reports, etc. The software uses lexical look-up tables, regular expressions, and simple heuristics to locate both HIPAA PHI, and an extended PHI set that includes doctors' names and years of dates. To develop the de-identification approach, we assembled a gold standard corpus of re-identified nursing notes with real PHI replaced by realistic surrogate information. This corpus consists of 2,434 nursing notes containing 334,000 words and a total of 1,779 instances of PHI taken from 163 randomly selected patient records. This gold standard corpus was used to refine the algorithm and measure its sensitivity. To test the algorithm on data not used in its development, we constructed a second test corpus of 1,836 nursing notes containing 296,400 words. The algorithm's false negative rate was evaluated using this test corpus.ResultsPerformance evaluation of the de-identification software on the development corpus yielded an overall recall of 0.967, precision value of 0.749, and fallout value of approximately 0.002. On the test corpus, a total of 90 instances of false negatives were found, or 27 per 100,000 word count, with an estimated recall of 0.943. Only one full date and one age over 89 were missed. No patient names were missed in either corpus.ConclusionWe have developed a pattern-matching de-identification system based on dictionary look-ups, regular expressions, and heuristics. Evaluation based on two different sets of nursing notes collected from a U.S. hospital suggests that, in terms of recall, the software out-performs a single human de-identifier (0.81) and performs at least as well as a consensus of two human de-identifiers (0.94). The system is currently tuned to de-identify PHI in nursing notes and discharge summaries but is sufficiently generalized and can be customized to handle text files of any format. Although the accuracy of the algorithm is high, it is probably insufficient to be used to publicly disseminate medical data. The open-source de-identification software and the gold standard re-identified corpus of medical records have therefore been made available to researchers via the PhysioNet website to encourage improvements in the algorithm.
| S-EPMC2526997 | biostudies-literature
Project description:The definition of severe acute respiratory infection (SARI) - a respiratory illness with fever and cough, occurring within the past 10 days and requiring hospital admission - has not been evaluated for critically ill patients. Using integrated electronic health records data, we developed an automated search algorithm to identify SARI cases in a large cohort of critical care patients and evaluate patient outcomes. We conducted a retrospective cohort study of all admissions to a medical intensive care unit from August 2009 through March 2016. Subsets were randomly selected for deriving and validating a search algorithm, which was compared with temporal trends in laboratory-confirmed influenza to ensure that SARI was correlated with influenza. The algorithm was applied to the cohort to identify clinical differences for patients with and without SARI. For identifying SARI, the algorithm (sensitivity, 86.9%; specificity, 95.6%) outperformed billing-based searching (sensitivity, 73.8%; specificity, 78.8%). Automated searching correlated with peaks in laboratory-confirmed influenza. Adjusted for severity of illness, SARI was associated with more hospital, intensive care unit and ventilator days but not with death or dismissal to home. The search algorithm accurately identified SARI for epidemiologic study and surveillance.
| S-EPMC5997502 | biostudies-literature
Project description:electronic Medical Records & Genomics (eMERGE)
| PRJNA158491 | ENA
Project description:OBJECTIVE:The traditional fee-for-service approach to healthcare can lead to the management of a patient's conditions in a siloed manner, inducing various negative consequences. It has been recognized that a bundled approach to healthcare - one that manages a collection of health conditions together - may enable greater efficacy and cost savings. However, it is not always evident which sets of conditions should be managed in a bundled manner. In this study, we investigate if a data-driven approach can automatically learn potential bundles. METHODS:We designed a framework to infer health condition collections (HCCs) based on the similarity of their clinical workflows, according to electronic medical record (EMR) utilization. We evaluated the framework with data from over 16,500 inpatient stays from Northwestern Memorial Hospital in Chicago, Illinois. The plausibility of the inferred HCCs for bundled care was assessed through an online survey of a panel of five experts, whose responses were analyzed via an analysis of variance (ANOVA) at a 95% confidence level. We further assessed the face validity of the HCCs using evidence in the published literature. RESULTS:The framework inferred four HCCs, indicative of (1) fetal abnormalities, (2) late pregnancies, (3) prostate problems, and (4) chronic diseases, with congestive heart failure featuring prominently. Each HCC was substantiated with evidence in the literature and was deemed plausible for bundled care by the experts at a statistically significant level. CONCLUSIONS:The findings suggest that an automated EMR data-driven framework conducted can provide a basis for discovering bundled care opportunities. Still, translating such findings into actual care management will require further refinement, implementation, and evaluation.
| S-EPMC5771885 | biostudies-literature
Project description:BACKGROUND:Circulating biomarkers can facilitate diagnosis and risk stratification for complex conditions such as heart failure (HF). Newer molecular platforms can accelerate biomarker discovery, but they require significant resources for data and sample acquisition. OBJECTIVES:The purpose of this study was to test a pragmatic biomarker discovery strategy integrating automated clinical biobanking with proteomics. METHODS:Using the electronic health record, the authors identified patients with and without HF, retrieved their discarded plasma samples, and screened these specimens using a DNA aptamer-based proteomic platform (1,129 proteins). Candidate biomarkers were validated in 3 different prospective cohorts. RESULTS:In an automated manner, plasma samples from 1,315 patients (31% with HF) were collected. Proteomic analysis of a 96-patient subset identified 9 candidate biomarkers (p < 4.42 × 10-5). Two proteins, angiopoietin-2 and thrombospondin-2, were associated with HF in 3 separate validation cohorts. In an emergency department-based registry of 852 dyspneic patients, the 2 biomarkers improved discrimination of acute HF compared with a clinical score (p < 0.0001) or clinical score plus B-type natriuretic peptide (p = 0.02). In a community-based cohort (n = 768), both biomarkers predicted incident HF independent of traditional risk factors and N-terminal pro-B-type natriuretic peptide (hazard ratio per SD increment: 1.35 [95% confidence interval: 1.14 to 1.61; p = 0.0007] for angiopoietin-2, and 1.37 [95% confidence interval: 1.06 to 1.79; p = 0.02] for thrombospondin-2). Among 30 advanced HF patients, concentrations of both biomarkers declined (80% to 84%) following cardiac transplant (p < 0.001 for both). CONCLUSIONS:A novel strategy integrating electronic health records, discarded clinical specimens, and proteomics identified 2 biomarkers that robustly predict HF across diverse clinical settings. This approach could accelerate biomarker discovery for many diseases.
| S-EPMC6501811 | biostudies-literature
Project description:Real-time automated continuous sampling of electronic medical record data may expeditiously identify patients at risk for death and enable prompt life-saving interventions. We hypothesized that a real-time electronic medical record-based alert could identify hospitalized patients at risk for mortality.An automated alert was developed and implemented to continuously sample electronic medical record data and trigger when at least 2 of 4 systemic inflammatory response syndrome criteria plus at least one of 14 acute organ dysfunction parameters was detected. The systemic inflammatory response syndrome and organ dysfunction alert was applied in real time to 312,214 patients in 24 hospitals and analyzed in 2 phases: training and validation datasets.In the training phase, 29,317 (18.8%) triggered the alert and 5.2% of such patients died, whereas only 0.2% without the alert died (unadjusted odds ratio 30.1; 95% confidence interval, 26.1-34.5; P < .0001). In the validation phase, the sensitivity, specificity, area under the curve, and positive and negative likelihood ratios for predicting mortality were 0.86, 0.82, 0.84, 4.9, and 0.16, respectively. Multivariate Cox-proportional hazard regression model revealed greater hospital mortality when the alert was triggered (adjusted hazards ratio 4.0; 95% confidence interval, 3.3-4.9; P < .0001). Triggering the alert was associated with additional hospitalization days (+3.0 days) and ventilator days (+1.6 days; P < .0001).An automated alert system that continuously samples electronic medical record data can be implemented, has excellent test characteristics, and can assist in the real-time identification of hospitalized patients at risk for death.
| S-EPMC4916370 | biostudies-literature
Project description:Predictive modeling with electronic health record (EHR) data is anticipated to drive personalized medicine and improve healthcare quality. Constructing predictive statistical models typically requires extraction of curated predictor variables from normalized EHR data, a labor-intensive process that discards the vast majority of information in each patient's record. We propose a representation of patients' entire raw EHR records based on the Fast Healthcare Interoperability Resources (FHIR) format. We demonstrate that deep learning methods using this representation are capable of accurately predicting multiple medical events from multiple centers without site-specific data harmonization. We validated our approach using de-identified EHR data from two US academic medical centers with 216,221 adult patients hospitalized for at least 24?h. In the sequential format we propose, this volume of EHR data unrolled into a total of 46,864,534,945 data points, including clinical notes. Deep learning models achieved high accuracy for tasks such as predicting: in-hospital mortality (area under the receiver operator curve [AUROC] across sites 0.93-0.94), 30-day unplanned readmission (AUROC 0.75-0.76), prolonged length of stay (AUROC 0.85-0.86), and all of a patient's final discharge diagnoses (frequency-weighted AUROC 0.90). These models outperformed traditional, clinically-used predictive models in all cases. We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios. In a case study of a particular prediction, we demonstrate that neural networks can be used to identify relevant information from the patient's chart.
| S-EPMC6550175 | biostudies-literature
Project description:Demand for clinical decision support systems in medicine and self-diagnostic symptom checkers has substantially increased in recent years. Existing platforms rely on knowledge bases manually compiled through a labor-intensive process or automatically derived using simple pairwise statistics. This study explored an automated process to learn high quality knowledge bases linking diseases and symptoms directly from electronic medical records. Medical concepts were extracted from 273,174 de-identified patient records and maximum likelihood estimation of three probabilistic models was used to automatically construct knowledge graphs: logistic regression, naive Bayes classifier and a Bayesian network using noisy OR gates. A graph of disease-symptom relationships was elicited from the learned parameters and the constructed knowledge graphs were evaluated and validated, with permission, against Google's manually-constructed knowledge graph and against expert physician opinions. Our study shows that direct and automated construction of high quality health knowledge graphs from medical records using rudimentary concept extraction is feasible. The noisy OR model produces a high quality knowledge graph reaching precision of 0.85 for a recall of 0.6 in the clinical evaluation. Noisy OR significantly outperforms all tested models across evaluation frameworks (p?<?0.01).
| S-EPMC5519723 | biostudies-literature