The use of a Poisson hidden Markov model for automated detection of hospital outbreaks with vancomycin-resistant enterococci in routine surveillance data

IF 3.9 3区医学 Q1 INFECTIOUS DISEASES

Journal of Hospital Infection Pub Date : 2025-05-06 DOI:10.1016/j.jhin.2025.04.025

S.A.M. van Kessel , C.C.H. Wielders , J. van de Kassteele , A. Verbon , A.F. Schoffelen , ISIS-AR study group , SO-ZI/AMR study group

{"title":"The use of a Poisson hidden Markov model for automated detection of hospital outbreaks with vancomycin-resistant enterococci in routine surveillance data","authors":"S.A.M. van Kessel , C.C.H. Wielders , J. van de Kassteele , A. Verbon , A.F. Schoffelen , ISIS-AR study group , SO-ZI/AMR study group","doi":"10.1016/j.jhin.2025.04.025","DOIUrl":null,"url":null,"abstract":"<div><h3>Background</h3><div>Despite the low prevalence of infections due to vancomycin-resistant enterococci (VRE) in the Netherlands, VRE is a frequent source of hospital outbreaks. We investigated whether a Poisson hidden Markov model (PHMM) can detect in-hospital VRE outbreaks in routine data from the Dutch Infectious Diseases Surveillance Information System for Antimicrobial Resistance (ISIS-AR).</div></div><div><h3>Methods</h3><div>We performed a retrospective data linkage study from 2013 up to 2023, including data from 89 hospitals on VRE isolates from ISIS-AR. A PHMM was used to detect potential outbreaks based on weekly VRE counts at hospital level. Per week <em>t</em>, the model provides the probability <em>P</em> that the observed count arose from an outbreak. Thresholds of <em>P</em>(<em>t</em>) >0.5, <em>P</em>(<em>t</em>) >0.7, and <em>P</em>(<em>t</em>) >0.9 for at least two consecutive weeks were used. The PHMM's results were compared with outbreaks voluntarily reported to the ‘Early warning and response meeting on highly resistant microorganism outbreaks in healthcare institutes’. Detection percentages were calculated and VRE counts of reported but undetected outbreaks, and detected but unreported outbreaks were described.</div></div><div><h3>Findings</h3><div>Of the 85 reported outbreaks, the model detected 87%, 86%, and 81% for thresholds <em>P</em>(<em>t</em>) >0.5, <em>P</em>(<em>t</em>) >0.7, and <em>P</em>(<em>t</em>) >0.9, respectively. Undetected outbreaks were mainly small outbreaks. The PHMM detected 66, 55, and 44 unreported potential outbreaks, respectively, with 44%, 35%, and 30% involving only one to two VRE-positive patients.</div></div><div><h3>Conclusion</h3><div>Overall, the PHMM shows potential for detecting in-hospital VRE outbreaks in routine surveillance data, with high detection rates. A prospective study is needed for further optimization for clinical practice.</div></div>","PeriodicalId":54806,"journal":{"name":"Journal of Hospital Infection","volume":"162 ","pages":"Pages 62-67"},"PeriodicalIF":3.9000,"publicationDate":"2025-05-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Hospital Infection","FirstCategoryId":"3","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0195670125001288","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"INFECTIOUS DISEASES","Score":null,"Total":0}

引用次数: 0

Abstract

Background

Despite the low prevalence of infections due to vancomycin-resistant enterococci (VRE) in the Netherlands, VRE is a frequent source of hospital outbreaks. We investigated whether a Poisson hidden Markov model (PHMM) can detect in-hospital VRE outbreaks in routine data from the Dutch Infectious Diseases Surveillance Information System for Antimicrobial Resistance (ISIS-AR).

Methods

We performed a retrospective data linkage study from 2013 up to 2023, including data from 89 hospitals on VRE isolates from ISIS-AR. A PHMM was used to detect potential outbreaks based on weekly VRE counts at hospital level. Per week t, the model provides the probability P that the observed count arose from an outbreak. Thresholds of P(t) >0.5, P(t) >0.7, and P(t) >0.9 for at least two consecutive weeks were used. The PHMM's results were compared with outbreaks voluntarily reported to the ‘Early warning and response meeting on highly resistant microorganism outbreaks in healthcare institutes’. Detection percentages were calculated and VRE counts of reported but undetected outbreaks, and detected but unreported outbreaks were described.

Findings

Of the 85 reported outbreaks, the model detected 87%, 86%, and 81% for thresholds P(t) >0.5, P(t) >0.7, and P(t) >0.9, respectively. Undetected outbreaks were mainly small outbreaks. The PHMM detected 66, 55, and 44 unreported potential outbreaks, respectively, with 44%, 35%, and 30% involving only one to two VRE-positive patients.

Conclusion

Overall, the PHMM shows potential for detecting in-hospital VRE outbreaks in routine surveillance data, with high detection rates. A prospective study is needed for further optimization for clinical practice.

查看原文本刊更多论文

使用泊松隐马尔可夫模型在常规监测数据中自动检测医院暴发的万古霉素耐药肠球菌。

背景：尽管荷兰万古霉素耐药肠球菌（VRE）感染的流行率较低，但VRE是医院暴发的常见来源。我们研究了泊松隐马尔可夫模型（PHMM）能否在荷兰传染病耐药性监测信息系统（ISIS-AR）的常规数据中检测出院内VRE暴发。方法：我们进行了2013年至2023年的回顾性数据链接研究，包括来自89家医院的ISIS-AR VRE分离株的数据。PHMM用于根据每周医院一级VRE计数检测潜在疫情。每周t，该模型提供观察到的计数由爆发引起的概率p。使用至少连续两周的阈值p(t) > 0.5, p(t) > 0.7和p(t) > 0.9。将PHMM的结果与自愿报告给“卫生保健机构高耐药微生物疫情预警和反应会议”的疫情进行了比较。计算检出率，并描述已报告但未发现的疫情和已发现但未报告的疫情的VRE计数。结果：在85例报告的爆发中，该模型分别检测到87%、86%和81%的阈值p(t) >.5、p(t) >.7和p(t) >.9。未被发现的爆发主要是小规模的爆发。PHMM分别检测到66例、55例和44例未报告的潜在暴发，其中44%、35%和30%仅涉及1-2例vre阳性患者。结论：总体而言，PHMM在常规监测数据中显示出发现院内VRE暴发的潜力，检出率高。需要前瞻性研究进一步优化临床实践。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Journal of Hospital Infection 医学-传染病学

CiteScore

12.70

自引率

5.80%

发文量

271

审稿时长

19 days

期刊介绍： The Journal of Hospital Infection is the editorially independent scientific publication of the Healthcare Infection Society. The aim of the Journal is to publish high quality research and information relating to infection prevention and control that is relevant to an international audience. The Journal welcomes submissions that relate to all aspects of infection prevention and control in healthcare settings. This includes submissions that: provide new insight into the epidemiology, surveillance, or prevention and control of healthcare-associated infections and antimicrobial resistance in healthcare settings; provide new insight into cleaning, disinfection and decontamination; provide new insight into the design of healthcare premises; describe novel aspects of outbreaks of infection; throw light on techniques for effective antimicrobial stewardship; describe novel techniques (laboratory-based or point of care) for the detection of infection or antimicrobial resistance in the healthcare setting, particularly if these can be used to facilitate infection prevention and control; improve understanding of the motivations of safe healthcare behaviour, or describe techniques for achieving behavioural and cultural change; improve understanding of the use of IT systems in infection surveillance and prevention and control.