Dataset Information

An ensemble of neural models for nested adverse drug events and medication extraction with subwords.

ABSTRACT: OBJECTIVE:This article describes an ensembling system to automatically extract adverse drug events and drug related entities from clinical narratives, which was developed for the 2018 n2c2 Shared Task Track 2. MATERIALS AND METHODS:We designed a neural model to tackle both nested (entities embedded in other entities) and polysemous entities (entities annotated with multiple semantic types) based on MIMIC III discharge summaries. To better represent rare and unknown words in entities, we further tokenized the MIMIC III data set by splitting the words into finer-grained subwords. We finally combined all the models to boost the performance. Additionally, we implemented a featured-based conditional random field model and created an ensemble to combine its predictions with those of the neural model. RESULTS:Our method achieved 92.78% lenient micro F1-score, with 95.99% lenient precision, and 89.79% lenient recall, respectively. Experimental results showed that combining the predictions of either multiple models, or of a single model with different settings can improve performance. DISCUSSION:Analysis of the development set showed that our neural models can detect more informative text regions than feature-based conditional random field models. Furthermore, most entity types significantly benefit from subword representation, which also allows us to extract sparse entities, especially nested entities. CONCLUSION:The overall results have demonstrated that the ensemble method can accurately recognize entities, including nested and polysemous entities. Additionally, our method can recognize sparse entities by reconsidering the clinical narratives at a finer-grained subword level, rather than at the word level.

SUBMITTER: Ju M

PROVIDER: S-EPMC6913208 | biostudies-literature | 2020 Jan

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

An ensemble of neural models for nested adverse drug events and medication extraction with subwords.

Ju Meizhi M Nguyen Nhung T H NTH Miwa Makoto M Ananiadou Sophia S

Journal of the American Medical Informatics Association : JAMIA 20200101 1

<h4>Objective</h4>This article describes an ensembling system to automatically extract adverse drug events and drug related entities from clinical narratives, which was developed for the 2018 n2c2 Shared Task Track 2.<h4>Materials and methods</h4>We designed a neural model to tackle both nested (entities embedded in other entities) and polysemous entities (entities annotated with multiple semantic types) based on MIMIC III discharge summaries. To better represent rare and unknown words in entiti ...[more]

PMID: 31197355

Similar Datasets

Project description:BackgroundMedication errors and adverse drug events are universal problems contributing to patient harm but the magnitude of these problems in Africa remains unclear.ObjectiveThe objective of this study was to systematically investigate the literature on the extent of medication errors and adverse drug events, and the factors contributing to medication errors in African hospitals.MethodsWe searched PubMed, MEDLINE, EMBASE, Web of Science and Global Health databases from inception to 31 August, 2017 and hand searched the reference lists of included studies. Original research studies of any design published in English that investigated adverse drug events and/or medication errors in any patient population in the hospital setting in Africa were included. Descriptive statistics including median and interquartile range were presented.ResultsFifty-one studies were included; of these, 33 focused on medication errors, 15 on adverse drug events, and three studies focused on medication errors and adverse drug events. These studies were conducted in nine (of the 54) African countries. In any patient population, the median (interquartile range) percentage of patients reported to have experienced any suspected adverse drug event at hospital admission was 8.4% (4.5-20.1%), while adverse drug events causing admission were reported in 2.8% (0.7-6.4%) of patients but it was reported that a median of 43.5% (20.0-47.0%) of the adverse drug events were deemed preventable. Similarly, the median mortality rate attributed to adverse drug events was reported to be 0.1% (interquartile range 0.0-0.3%). The most commonly reported types of medication errors were prescribing errors, occurring in a median of 57.4% (interquartile range 22.8-72.8%) of all prescriptions and a median of 15.5% (interquartile range 7.5-50.6%) of the prescriptions evaluated had dosing problems. Major contributing factors for medication errors reported in these studies were individual practitioner factors (e.g. fatigue and inadequate knowledge/training) and environmental factors, such as workplace distraction and high workload.ConclusionMedication errors in the African healthcare setting are relatively common, and the impact of adverse drug events is substantial but many are preventable. This review supports the design and implementation of preventative strategies targeting the most likely contributing factors.

Project description:Importance:Adverse drug events (ADEs) account for up to 16% of emergency department (ED) visits and 7% of hospital admissions. Medication reconciliation is required for hospital accreditation because it can reduce medication discrepancies, but there is no evidence that reducing discrepancies reduces ADEs or other adverse outcomes. Objective:To evaluate whether electronic medication reconciliation reduces ADEs, medication discrepancies, and other adverse outcomes compared with usual care. Design, Setting, and Participants:This cluster randomized trial involved 3491 patients who were discharged from 2 medical units and 2 surgical units at the McGill University Health Centre, Montreal, Quebec, Canada, between October 2014 and November 2016. Data analysis took place from July 2017 to July 2019. Intervention:The RightRx intervention electronically retrieved community drugs from the provincial insurer and aligned them with in-hospital drugs to facilitate reconciliation and communication at care transitions. Main Outcomes and Measures:The primary outcome was ADEs in 30 days after discharge. Secondary outcomes included medication discrepancies, ED visits, hospital readmissions, and a composite outcome of ED visits, readmissions, and death up to 90 days after discharge. Results:Of 4656 eligible patients, 3567 (76.6%) consented to participate (2060 [57.8%] men; mean [SD] age, 69.8 [14.9] years). Overall, 76 patients died during the hospital stay, so 3491 patients were included in the analysis. There was no significant difference in the risk of ADEs between intervention and control groups (76 [4.6%] vs 73 [4.0%]; OR, 0.97; 95% CI, 0.33-1.48), ED visits (433 [26.2%] vs 488 [26.6%]; OR, 0.83; 95% CI, 0.36-1.42), hospital readmission (170 [10.3%] vs 261 [14.2%]; OR, 0.22; 95% CI, 0.06-1.14), or the composite outcome (447 [27.0%] vs 506 [27.6%]; OR, 0.75; 95% CI, 0.34-1.27) at 30 days. Medication discrepancies were significantly reduced in the intervention group compared with the control group (437 [26.4%] vs 1029 [56.0%]; OR, 0.24; 95% CI, 0.12-0.57). Changes made to community medications (OR, 1.05; 95% CI, 1.01-1.10) and new medications (OR, 1.09; 95% CI, 1.01-1.18) were significant risk factors for ADEs. Conclusions and Relevance:Electronic medication reconciliation reduced medication discrepancies but did not reduce ADEs or other adverse outcomes. Hospital accreditation should focus on interventions that reduce the risk of adverse events for patients with multiple changes to community medications. Trial Registration:ClinicalTrials.gov identifier: NCT01179867.

Project description:BackgroundInitiation and up-titration of guideline-directed medical therapies (GDMTs) for heart failure with reduced ejection fraction (HFrEF) remains suboptimal, in part because of concerns regarding tolerability and adverse events (AEs).ObjectivesThe authors sought to compare rates of AE in patients randomized to GDMT medication vs placebo in a meta-analysis of landmark cardiovascular outcomes trials.MethodsThe authors assessed rates of reported AE in 17 landmark HFrEF clinical trials across each class of GDMT in the placebo and intervention arms. The overall rates of AE for each drug class, the absolute difference in frequency in AEs between the placebo and intervention arms, and the odds of each AE according based on randomization strata were calculated.ResultsAE were reported commonly in trials across each class of GDMT, with 75% to 85% of participants reporting at least 1 AE. There was no significant difference in the frequency of AE between the intervention and placebo arms, except for angiotensin-converting enzyme inhibitors (87.0% [95% CI: 85.0%-88.8%] vs 82.0% [95% CI: 79.8%-84.0%], absolute difference: +5% with intervention; P < 0.001). There was no significant difference in drug discontinuation because of AE between placebo and intervention arms in angiotensin-converting enzyme inhibitors, mineralocorticoid receptor antagonists, sodium glucose cotransporter 2 inhibitors, or angiotensin receptor neprilysin inhibitor/angiotensin II receptor blocker trials. Patients randomized to beta-blocker were significantly less likely to stop study drug because of AE than placebo (11.3% [95% CI: 10.3%-12.3%] vs 13.7% [95% CI: 12.5%-14.9%], absolute difference: -1.1%; P = 0.015). When individual types of AE were assessed, the initiation of an intervention vs placebo resulted in small differences in absolute frequency of AE that were largely not statistically significant.ConclusionsIn clinical trials of GDMT for HFrEF, AEs are observed frequently. However, rates of AE are similar between active medication and control, suggesting these may reflect the high risk nature of the heart failure disease state rather than be attributive to a specific therapy.

Dataset Information

An ensemble of neural models for nested adverse drug events and medication extraction with subwords.

Publications

An ensemble of neural models for nested adverse drug events and medication extraction with subwords.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets