Using chief complaints for syndromic surveillance: A review of chief complaint based classifiers in North America

Conway, Mike; Dowling, John; Chapman, Wendy W.

doi:10.1016/j.jbi.2013.04.003

Cited by 47 publications

(40 citation statements)

References 32 publications

(42 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Modern statistical classifiers can capture the decision rules employed by many keyword-based classifiers. For example, the classifier deployed in the Electronic Surveillance System for the Early Notification of Community-Based Epidemics (ESSENCE) is keyword-based (Conway et al 2013). Under this method, the record and the syndrome are represented as an unordered set of words or tokens, with a numeric value or weight assigned to each token.…”

Section: Overview Of Chief Complaint Classifiersmentioning

confidence: 99%

Chief complaint classification with recurrent neural networks

Lee

Levin

Finley

et al. 2019

Journal of Biomedical Informatics

View full text Add to dashboard Cite

Syndromic surveillance detects and monitors individual and population health indicators through sources such as emergency department records. Automated classification of these records can improve outbreak detection speed and diagnosis accuracy. Current syndromic systems rely on hand-coded keyword-based methods to parse written fields and may benefit from the use of modern supervised-learning classifier models. In this paper we implement two recurrent neural network models based on long short-term memory (LSTM) and gated recurrent unit (GRU) cells and compare them to two traditional bag-of-words classifiers: multinomial naïve Bayes (MNB) and a support vector machine (SVM). The MNB classifier is one of only two machine learning algorithms currently being used for syndromic surveillance. All four models are trained to predict diagnostic code groups as defined by Clinical Classification Software, first to predict from discharge diagnosis, then from chief complaint fields. The classifiers are trained on 3.6 million de-identified emergency department records from a single United States jurisdiction. We compare performance of these models primarily using the F1 score. We measure absolute model performance to determine which conditions are the most amenable to surveillance based on chief complaint alone. Using discharge diagnoses The LSTM classifier performs best, though all models exhibit an F1 score above 96.00. The GRU performs best on chief complaints (F1=47.38), and MNB with bigrams performs worst (F1=39.40). Certain syndrome types are easier to detect than others. For examples, chief complaints using the GRU model predicts alcohol-related disorders well (F1=78.91) but predicts influenza poorly (F1=14.80). In all instances the RNN models outperformed the bag-of-word classifiers suggesting deep learning models could substantially improve the automatic classification of unstructured text for syndromic surveillance.

show abstract

Section: Overview Of Chief Complaint Classifiersmentioning

confidence: 99%

Chief complaint classification with recurrent neural networks

Lee

Levin

Finley

et al. 2019

Journal of Biomedical Informatics

View full text Add to dashboard Cite

show abstract

“…Twelve articles were even more exploratory in nature, outlining simply the design process for a new system [19][20][21] or presenting the characterization of a new data source [22,23]. Twenty-two articles were systematic reviews, including a review of syndromic surveillance classifiers [24], the use of IIS for research [25], the use of social networking sites in public health [13], and information needs of public health practitioners [26]. Case studies were also present in 16 articles, summarizing the design or implementation of a PHI system within a single health department or group of organizations [27,28].…”

Section: Ojphimentioning

confidence: 99%

What’s Past is Prologue: A Scoping Review of Recent Public and Global Health Informatics Literature

Dixon

Pina

Kharrazi

et al. 2015

OJPHI

View full text Add to dashboard Cite

Objective: To categorize and describe the public health informatics (PHI) and global health informatics (GHI) literature between 2012 and 2014.Methods: We conducted a semi-systematic review of articles published between January 2012 and September 2014 where information and communications technologies (ICT) was a primary subject of the study or a main component of the study methodology. Additional inclusion and exclusion criteria were used to filter PHI and GHI articles from the larger biomedical informatics domain. Articles were identified using MEDLINE as well as personal bibliographies from members of the American Medical Informatics Association PHI and GHI working groups.Results: A total of 85 PHI articles and 282 GHI articles were identified. While systems in PHI continue to support surveillance activities, we identified a shift towards support for prevention, environmental health, and public health care services. Furthermore, articles from the U.S. reveal a shift towards PHI applications at state and local levels. GHI articles focused on telemedicine, mHealth and eHealth applications. The development of adequate infrastructure to support ICT remains a challenge, although we identified a small but growing set of articles that measure the impact of ICT on clinical outcomes.Discussion: There is evidence of growth with respect to both implementation of information systems within the public health enterprise as well as a widening of scope within each informatics discipline. Yet the articles also illuminate the need for more primary research studies on what works and what does not as both searches yielded small numbers of primary, empirical articles.Conclusion: While the body of knowledge around PHI and GHI continues to mature, additional studies of higher quality are needed to generate the robust evidence base needed to support continued investment in ICT by governmental health agencies.

show abstract

“…However, almost none of the existing predictive models for bronchiolitis use information embedded in clinical text. To improve the predictive models' accuracy, we can use medical information extraction techniques to automatically extract information from clinical text [118], such as by mapping chief complaint strings to syndrome categories [262]. The extracted information is added as predictors into the predictive models for bronchiolitis.…”

Section: Using Information Embedded In Clinical Textmentioning

confidence: 99%

A systematic review of predictive modeling for bronchiolitis

Luo

Nkoy

Gesteland

et al. 2014

International Journal of Medical Informatics

View full text Add to dashboard Cite

Purpose: Bronchiolitis is the most common cause of illness leading to hospitalization in young children. At present, many bronchiolitis management decisions are made subjectively, leading to significant practice variation among hospitals and physicians caring for children with bronchiolitis. To standardize care for bronchiolitis, researchers have proposed various models to predict the disease course to help determine a proper management plan. This paper reviews the existing state of the art of predictive modeling for bronchiolitis. Predictive modeling for respiratory syncytial virus (RSV) infection is covered whenever appropriate, as RSV accounts for about 70% of bronchiolitis cases.Methods: A systematic review was conducted through a PubMed search up to April 25, 2014. The literature on predictive modeling for bronchiolitis was retrieved using a comprehensive search query, which was developed through an iterative process.Search results were limited to human subjects, the English language, and children (birth-18 years). Results:The literature search returned 2,312 references in total. After manual review, 168 of these references were determined to be relevant and are discussed in this paper. We identify several limitations and open problems in predictive modeling for bronchiolitis, and provide some preliminary thoughts on how to address them, with the hope to stimulate future research in this domain. Conclusions:Many problems remain open in predictive modeling for bronchiolitis. Future studies will need to address them to achieve optimal predictive models.

show abstract

Using chief complaints for syndromic surveillance: A review of chief complaint based classifiers in North America

Cited by 47 publications

References 32 publications

Chief complaint classification with recurrent neural networks

Chief complaint classification with recurrent neural networks

What’s Past is Prologue: A Scoping Review of Recent Public and Global Health Informatics Literature

A systematic review of predictive modeling for bronchiolitis

Contact Info

Product

Resources

About