國家衛生研究院 NHRI:Item 3990099045/8617

NHRI > National Institute of Infectious Diseases and Vaccinology > Yee-Chun Chen > Conference Papers/Meeting Abstract > Item 3990099045/8617

Please use this identifier to cite or link to this item: http://ir.nhri.org.tw/handle/3990099045/8617

Title:	Validation and comparison of a candidemia prediction models: A case-control study
Authors:	Tseng, YJ;Ichise, R;Huang, BC;Lin, HC;Chen, MY;Shang, RJ;Sheng, WH;Chen, YC;Lai, F;Chang, SC
Contributors:	Division of Infectious Diseases
Abstract:	Purpose: Inappropriate or delayed treatment of invasive candidiasis (IC) is associated with excess mortality and resource usage. However, patients with IC share similar clinical risk factors as those infected with multidrug resistant organisms which was common in critically ill patients. Implementing active microbial surveillance to determine heavy colonization of Candida at multiple body sites is not practical. Thus, this study aims to develop a candidemia prediction model to facilitate selection of empirical antimicrobial agents before microbiological confirmation of bloodstream infection (BSI) based on a case-control study. Methods: To develop the model, electronic medical records (EMR) of hospitalized patients with candidemia (322 cases) and patients with BSI due to top 5 bacteria at a 2200-bed teaching hospital in Taiwan in 2011 (1018 controls) were systematically surveyed. A total of 27 features existing before the first blood cultures with positive results were collected: demographics, underlying conditions, and epidemiological and healthcare–associated factors. The System incorporates 3 data mining algorithms: support vector machine, decision tree, and inductive logic programming (ILP). Generalized linear model was used as a baseline algorithm. In addition, the effect of adopting clinical knowledge into ILP was also evaluated. Results: Based on F1 score, the most parsimonious model was ILP with clinical knowledge, having performance with high accuracy (0.713) and specificity (0.792), and also with optimal F1 scores (0.437) and sensitivity (0.464). In the three methods with ILP, the F1 score in using background knowledge was significantly higher than others (P=.015). Conclusion: This study demonstrated the performance of the prediction models using data mining algorithms to predict IC in patients from whom blood cultures are collected under the suspicion of BSI. Our data showed significantly improves the performance when incorporating clinical knowledge of IC.
Date:	2015-04
Relation:	Journal of Microbiology, Immunology and Infection. 2015 Apr;48(2, Suppl. 1):S43.
Link to:	http://dx.doi.org/10.1016/j.jmii.2015.02.073
JIF/Ranking 2023:	http://gateway.webofknowledge.com/gateway/Gateway.cgi?GWVersion=2&SrcAuth=NHRI&SrcApp=NHRI_IR&KeyISSN=1684-1182&DestApp=IC2JCR
Appears in Collections:	[Yee-Chun Chen] Conference Papers/Meeting Abstract

Files in This Item:

File	Description	Size	Format
SDO1684118215001061.pdf		45Kb	Adobe PDF	729	View/Open

Loading...