Dataset Information

Generating contextual embeddings for emergency department chief complaints.

ABSTRACT: Objective:We learn contextual embeddings for emergency department (ED) chief complaints using Bidirectional Encoder Representations from Transformers (BERT), a state-of-the-art language model, to derive a compact and computationally useful representation for free-text chief complaints. Materials and methods:Retrospective data on 2.1 million adult and pediatric ED visits was obtained from a large healthcare system covering the period of March 2013 to July 2019. A total of 355 497 (16.4%) visits from 65 737 (8.9%) patients were removed for absence of either a structured or unstructured chief complaint. To ensure adequate training set size, chief complaint labels that comprised less than 0.01%, or 1 in 10 000, of all visits were excluded. The cutoff threshold was incremented on a log scale to create seven datasets of decreasing sparsity. The classification task was to predict the provider-assigned label from the free-text chief complaint using BERT, with Long Short-Term Memory (LSTM) and Embeddings from Language Models (ELMo) as baselines. Performance was measured as the Top-k accuracy from k?=?1:5 on a hold-out test set comprising 5% of the samples. The embedding for each free-text chief complaint was extracted as the final 768-dimensional layer of the BERT model and visualized using t-distributed stochastic neighbor embedding (t-SNE). Results:The models achieved increasing performance with datasets of decreasing sparsity, with BERT outperforming both LSTM and ELMo. The BERT model yielded Top-1 accuracies of 0.65 and 0.69, Top-3 accuracies of 0.87 and 0.90, and Top-5 accuracies of 0.92 and 0.94 on datasets comprised of 434 and 188 labels, respectively. Visualization using t-SNE mapped the learned embeddings in a clinically meaningful way, with related concepts embedded close to each other and broader types of chief complaints clustered together. Discussion:Despite the inherent noise in the chief complaint label space, the model was able to learn a rich representation of chief complaints and generate reasonable predictions of their labels. The learned embeddings accurately predict provider-assigned chief complaint labels and map semantically similar chief complaints to nearby points in vector space. Conclusion:Such a model may be used to automatically map free-text chief complaints to structured fields and to assist the development of a standardized, data-driven ontology of chief complaints for healthcare institutions.

SUBMITTER: Chang D

PROVIDER: S-EPMC7382638 | biostudies-literature | 2020 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Generating contextual embeddings for emergency department chief complaints.

Chang David D Hong Woo Suk WS Taylor Richard Andrew RA

JAMIA open 20200715 2

<h4>Objective</h4>We learn contextual embeddings for emergency department (ED) chief complaints using Bidirectional Encoder Representations from Transformers (BERT), a state-of-the-art language model, to derive a compact and computationally useful representation for free-text chief complaints.<h4>Materials and methods</h4>Retrospective data on 2.1 million adult and pediatric ED visits was obtained from a large healthcare system covering the period of March 2013 to July 2019. A total of 355 497 ( ...[more]

PMID: 32734154

Similar Datasets

Project description:ObjectivesVeteran persons living with dementia (PLWDs) have high acute care utilization. We aim to understand why PLWDs seek care in the emergency department (ED) and how their utilization differs from older Veterans with no dementia diagnosis. We demonstrate the use of a novel national chief complaint data set in the Veteran Affairs Health Care System.MethodsThis was a retrospective observational study of ED users 65 years or older as of FY2017. The primary outcome is presence of one or more ED visits in FYs 2017-2018 using a logistic regression model controlling for dementia and other variables. Secondary outcomes include counts of ED visits by disposition, Emergency Severity Index, chief complaints defined by a natural language processing program, and ED encounter diagnoses defined by primary International Statistical Classification of Diseases, Tenth Revision (ICD-10-CM) code.ResultsOur cohort of Veterans comprised 3,115,263 patients. Of those, 255,372 (8.2%) had a diagnosis of dementia. Logistic regression modeling demonstrated that dementia is a significant predictor of ED use (p < 0.0001), with PLWDs more likely to have an ED visit (odds ratio 1.96, 95% confidence interval 1.94-1.98). PLWDs were admitted at higher rates when accounting for age and acuity. Chief complaints that were more common among PLWDs included falls (6.7% dementia vs. 3.3% without dementia), weakness (3.6% vs. 2.2%), and abnormal mental state (2.2% vs. 0.4%). ICD-10-CM codes were largely similar between the two groups.ConclusionsOur results reinforce that the ED is a common access point for Veterans with dementia. These patients require special consideration as they are more likely to visit the ED and be admitted. Our use of a novel national chief complaint data set suggests that they more commonly present with certain geriatric syndromes and nonspecific complaints. Further work is needed to determine whether these would warrant targeted interventions to improve quality of acute care.

Project description:IntroductionKnowledge about the relationship between symptoms, diagnoses, and mortality in emergency department (ED) patients is essential for the emergency physician to optimize treatment, monitoring, and flow. In this study, we investigated the association between symptoms and discharge diagnoses; symptoms and mortality; and we then analyzed whether the association between symptoms and mortality was influenced by other risk factors.MethodsThis was a population-based, multicenter cohort study of all non-trauma ED patients ≥18 years who presented at a hospital in the Region of Southern Denmark between January 1, 2016-March 20, 2018. We used multivariable logistic regression to examine the association between symptoms and mortality adjusted for other risk factors.ResultsWe included 223,612 ED visits with a median patient age of 63 and even distribution of females and males. The frequency of the chief complaints at presentation were as follows: non-specific symptoms (19%); abdominal pain (16%); dyspnea (12%); fever (8%); chest pain (8%); and neurologic complaints (7%). Discharge diagnoses were symptom-based (24%), observational (hospital visit for observation or examination, 17%), circulatory (12%), or respiratory (12%). The overall 30-day mortality was 3.5%, with 1.7% dead within 0-7 days and 1.8% within 8-30 days. The presenting symptom was associated with mortality at 0-7 days but not with mortality at 8-30 days. Patients whose charts were missing documentation of symptoms (adjusted odds ratio [aOR] 3.5) and dyspneic patients (aOR 2.4) had the highest mortality at 0-7 days across patients with different primary symptoms. Patients ≥80 years and patients with a higher degree of comorbidity had increased mortality from 0-7 days to 8-30 days (aOR from 24.0 to 42.7 and 1.9 to 2.8, respectively).ConclusionShort-term mortality was more strongly associated with patient-related factors than with the primary presenting symptom at arrival to the hospital.

Project description:BACKGROUND:Nonspecific complaint (NSC) is a common presenting complaint in the emergency setting, especially in the elderly population. Individual studies have shown that it is associated with significant morbidity and mortality. This prognostic systematic review draws a synthesis of reported outcomes for patients presenting with NSC and compares them with outcomes for patients presenting with a specific complaint. METHODS:We conducted a literature search for publications, abstracts and conference presentations from Ovid, Scopus and Web of Science for the past 20?years. Studies were included which treated adult patients presenting to the Emergency Medical Services or Emergency Department with NSC. 2599 studies were screened for eligibility and quality was assessed using the SIGN assessment for bias tool. We excluded any low-quality studies, resulting in nine studies for quantitative analysis. We analysed the included studies for in-hospital mortality, triage category, emergency department length of stay, admission rate, hospital length of stay, intensive care admissions and re-visitation rate and compared outcomes to patients presenting with specific complaints (SC), where data were available. We grouped discharge diagnoses by ICD-10 category. RESULTS:We found that patients presenting with NSC were mostly older adults. Mortality for patients with NSC was significantly increased compared to patients presenting with SC [OR 2.50 (95% CI 1.40-4.47)]. They were triaged as urgent less often than SC patients [OR 2.12 (95% CI 1.08-4.16)]. Emergency department length of stay was increased in two out of three studies. Hospital length of stay was increased by 1-3?days. Admission rates were high in most studies, 55 to 84%, and increased in comparison to patients with SC [OR 3.86 (95% CI 1.76-8.47)]. These patients seemed to require more resources than patients with SC. The number for intensive care admissions did not seem to be increased. Data were insufficient to make conclusions regarding re-visitation rates. Discharge diagnoses were spread throughout the ICD-10 main chapters, infections being the most prevalent. CONCLUSIONS:Patients with NSC have a high risk of mortality and their care in the Emergency Department requires more time and resources than for patients with SC. We suggest that NSC should be considered a major emergency presentation.

Dataset Information

Generating contextual embeddings for emergency department chief complaints.

Publications

Generating contextual embeddings for emergency department chief complaints.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets