Methods for Alerting Infection Preventionists in a Health Information Exchange About Multidrug-Resistant Gram-Negative Bacterial Infections
Marc B. Rosenman, Kinga A. Szucs, S. Maria E. Finnell, Shahid Khokhar, David C. Shepherd, Jeff Friedlin, Larry Lemmon, Mark Tucker, Xiaochun Li, Abel N. Kho
Abstract
Recent outbreaks of carbapenem-resistant gram-negative bacteria (CRGNB) among hospitalized patients have elicited national attention and underscored the danger of healthcare-associated infections. Whenever patients visit more than one hospital, a multidrug-resistant organism (MDRO) may spread. Within hospitals, spread is also a scourge. Therefore, our goal is to build a regional system that parses laboratory microbiology culture data to make the data usable for decision support and then alerts hospitals when a patient with a history of MDRO is admitted. The previous methicillin-resistant Staphylococcus aureus/vancomycin-resistant enterococci (MRSA/VRE) alert system in our region relied on hospital infection preventionists (IPs) to enter data; our new system uses microbiology culture data generated in the normal course of health care at more than 25 hospitals. We cull the microbiology data from the Health Level Seven version 2 (HL7v2) messages that hospitals send to a health information exchange. The principal informatics problem is that for microbiology data, more than for simpler types of laboratory results data, most of the messages are not structured in standard HL7v2 format by the sending hospitals. We therefore built an HL7v2 correction engine that deals with incorrect message structure and/or content in order to generate new, standardized microbiology content, which we append to the existing message. The engine uses natural language processing and other methods to parse key data elements needed for infection control alerts: organism, antibiotics tested, minimum inhibitory concentrations, susceptibility interpretation, body source of the culture, and health care facility where drawn. These standardized data elements can then be integrated into enhanced email alerts to IPs. We solicited suggestions from IPs in various hospitals across the State regarding how they would like to receive information and which organisms to include. We subsequently will evaluate the perceived utility of the alerts, the rate and timeliness of the use of isolation, and the geographic patterns of gram-negative MDRO infections.
Introduction
Recent outbreaks and deaths caused by multidrug-resistant gram-negative "superbugs" among hospitalized patients in the United States have elicited national attention and have underscored the danger posed by these infections.1 In the past year, new, multidrug-resistant strains of carbapenem-resistant Enterobacteriaceae (CRE) have appeared at an accelerating rate. Of the 37 strains of CRE in the United States, 15 have been discovered since July 2012, which prompted a February 2013 Centers for Disease Control and Prevention (CDC) advisory encouraging health care providers to "act aggressively to prevent the emergence and spread" of these bacteria.2 In 2013, the Chief Medical Officer of England warned of a dystopian future: Antibiotic resistance "is a growing problem, and if we don't get it right, we will find ourselves in a health system not dissimilar from the early 19th century."3 The focus of this project is on gram-negative multidrug-resistant organisms (MDROs), but from the beginning we worked to ensure that our methods would generalize to gram-positive MDROs, such as methicillin-resistant Staphylococcus aureus (MRSA) and vancomycin-resistant enterococci (VRE), or to any other bacteria for which alerts may be desired in the future.
The increasing prevalence of gram-negative rods (GNR) that are MDROs has been documented worldwide. Carbapenem resistance was preceded by increasing recognition, over the past 10–15 years, of extended-spectrum beta-lactamase production among Enterobacteriaceae (ESBL-E). In France, the national infection control program found that from 2003 to 2013, while the incidence of MRSA declined (from 0.72 to 0.41 per 1,000 patient days), the incidence of ESBL-E climbed alarmingly (from 0.17 to 0.48 per 1,000 patient days).4 Increasing rates of ESBL-E among urinary tract5,6 and bloodstream7,8 pathogens have been documented in many nations. Based on U.S. surveillance data reports to the CDC, the prevalence of multidrug-resistant Klebsiella pneumonia and Escherichia coli increased from 7 percent in 2000 to 13 percent in 2008.9 Beyond the Enterobacteriaceae, beta-lactamase production is one of the key mechanisms underlying multidrug resistance in other gram-negative organisms, such as Pseudomonas aeruginosa10,11 and Acinetobacter baumannii.12
In a general sense, the epidemiologic understanding of GNR MDROs is beginning to parallel that of MRSA. Early in the history of MRSA, it was considered a problem centered primarily in hospitals and intensive care units (ICUs); with time, community-associated disease came to the fore.13 Similarly, it now is increasingly understood that a substantial minority of GNR MDROs arise in the outpatient setting, and that many of the affected patients have no healthcare-associated risk factors.14 From 2006 to 2011, there was a 10-fold increase in the rate of carriage of ESBL-producing E. coli among healthy adults in Paris.15 A recent (2010) analysis of diapers documented what was probably transmission from one child to another of ESBL-producing E. coli;16 almost 3 percent of healthy preschool children were colonized, as were 8 percent of children of the same age at Uppsala University Hospital in Sweden. Among international travelers studied by the Weill Cornell travel medicine clinic in New York City during 2009–2010, 28 percent acquired ESBL-E overseas and imported it into the United States.17
Because people can carry GNR MDROs in their gastrointestinal tracts asymptomatically for months or even years, they can bring these bacteria into hospitals upon admission. After discharge from a university hospital in Paris (based on data from 1997 to 2010), the median time to clearance of ESBL-E (based on rectal screening results) was 6.6 months; based on the high rate of readmission while still colonized, the authors recommended that "screening for ESBL-E and contact isolation precautions at hospital readmission are advisable for all patients identified as testing positive for ESBL-E infection during an earlier hospital stay."18 Substantial rates of inter-hospital spread of Clostridium difficile19 and MRSA20 have been documented in California. A study of five hospital systems in the Indiana Network for Patient Care (INPC) reported that in 10 percent of hospitalizations, the admitting hospital was not aware of the patient's previous history of MRSA at a different institution; as a result, in a 1-year period there were 3,600 inpatient days in which contact isolation was indicated but not ordered.21
In light of the growing danger of GNR MDRO outbreaks caused by patients who carry these organisms into the hospital, we drew upon our experience (since 2007) delivering alerts when patients with a history of MRSA or VRE are admitted to the major hospitals in Indianapolis, IN. In this paper, we describe the development of a new alert system for GNR MDRO.
Methods
Overview
With sponsorship from the Agency for Healthcare Research and Quality (AHRQ), we are developing a new system with these aims, to: parse microbiology culture data from more than 25 hospitals (in 12 hospital systems) in a regional health information exchange; check whether patients being admitted to hospitals have a history of GNR MDRO anywhere in the system; and send email alerts to notify local infection control personnel when a patient with a history of GNR MDRO is newly admitted.
Indiana Network for Patient Care
The INPC is a leading operational regional health information exchange, formed in 1994 by Regenstrief Institute and the five major hospital systems in Indianapolis.22 Its primary purpose is clinical data exchange to improve the quality and efficiency of health care; a secondary purpose is research.23 In 2002, the INPC was the site of an early randomized trial that examined the value of data shared across institutions for patients in emergency departments.24 The INPC has stored more than 1 billion data elements, and since 2009, the number of hospital systems has expanded beyond the original five to more than 25 (though not all send microbiology data). The microbiology culture parsing and email alert system described in this paper is built on the newer INPC infrastructure in Oracle; the MRSA/VRE registry and alert system that preceded it was built on the legacy Virtual Address eXtension (VAX) INPC database, which is being phased out.
Existing MRSA/VRE Registry and Alert System
Beginning in 2007, Kho and colleagues built a MRSA/VRE registry and alert system into the original five INPC hospital systems.21,25,26 Seven new variables were embedded in the INPC for infection preventionists (IPs) to enter information about patients infected or colonized with MRSA or VRE. Then, whenever a patient is admitted to a hospital in the original set of five hospital systems, the INPC's receipt of the Admission, Discharge, and Transfer (ADT) Health Level Seven (HL7) message initiates an automated process that checks the MRSA/VRE registry. If the patient's registry status shows MRSA or VRE infection or colonization that has not been "cleared" through a subsequent data entry by an IP, an automated email alert is sent to the IP at the admitting hospital. In this way, the MRSA/VRE system helps hospitals place patients into contact isolation sooner and also facilitates better monitoring of infection rates in the region and the spread of MRSA and VRE between hospitals.21
New Microbiology Culture Data Processing and Alert System
In light of the growing danger of GNR MDRO outbreaks caused by patients who carry these organisms into hospitals, we began building a new system that would, among other purposes, notify hospitals when a patient with a history of GNR MDRO is being admitted. The MRSA/VRE registry relied upon IPs to enter data. Our new system is designed to avoid this manual data entry step. The new design parses, and then stores in a usable format, microbiology culture results data generated in the normal course of health care and sent to the INPC in HL7 format. There were two additional reasons why the new microbiology culture data processing and email alert system was needed: the INPC's more than 20-year-old VAX database was in the process of being decommissioned in favor of a new INPC schema in Oracle; and the expansion of the INPC to additional hospital systems afforded an opportunity to provide microbiology alerts to new hospitals. The new message parser is designed for use within the overall HL7 message processing infrastructure developed by Regenstrief Institute, the Health Open Source Software (HOSS) pipeline. The microbiology data then are stored in an infection control database schema, the Regional Electronic Infection Control Network (REICON) database, which is modeled exactly on the main INPC database schema. We can therefore use INPC concepts ("dictionary terms") and concept mapping tables.
The Informatics Problem With Microbiology Culture Data
The principal informatics problem is that for microbiology data, more than for simpler types of laboratory results data, most of the messages are not structured in standard HL7v2 format by the hospitals that send data to the health information exchange (INPC). When the data (in the inbound HL7 Observation Result [ORU] messages) are not structured according to the HL7 standard, the existing INPC can store the data only as text "blobs." The blobs are human-readable in the electronic medical record and therefore are useful for clinicians taking care of patients one at a time. But because the blobs are not structured, they cannot serve as the basis of an automated program that would check the database to see if the patient has ever had a microbiology culture positive for GNR MDRO (or for gram-positive bacteria such as MRSA or VRE, or for any other culture result).
HL7 Correction Engine (the "REICON Transform")
We therefore built a microbiology HL7v2 correction engine, which deals with incorrect HL7 message structure and/or content in the inbound ORU messages that contain microbiology culture data. The engine generates new, standardized microbiology content that we append to the existing ORU HL7 message. This process is an enhancement to the existing methods that the INPC uses (in many cases only partially) for parsing ORU HL7 messages that contain microbiology culture data. The new engine uses natural language processing and other methods to parse the six key data types and elements needed for infection control alerts: the organism, the antibiotics tested, the minimum inhibitory concentrations of the antibiotics, the susceptibility interpretation for each antibiotic, the body source from which the patient's culture was drawn, and the health care facility where it was drawn. In addition, we look for a seventh data type: evidence that an assay for ESBL or CRE was positive. These standardized data elements can then be stored in the INPC schema in Oracle and, as structured data, can subsequently be integrated into enhanced email alerts to IPs whenever a patient with a history of GNR MDRO is admitted to a hospital. The GNR MDROs are our initial focus for alerts, but we designed the new parser to deal with any bacterial culture; therefore, it could be applied to MRSA, VRE, and other infections in the future. It will be necessary to deal with MRSA and VRE once the existing VAX-based MRSA/VRE registry is shut down.
As a first step, we obtained a dataset consisting of all inbound ORU HL7 messages from a 2-month period. These messages were culled downstream from a Regenstrief "pre-processor," which, as one of its functions, adds Logical Observation Identifiers Names and Codes (LOINC®) codes to the messages. We wrote a LOINC-code filter to separate the microbiology culture messages from all other laboratory results. For inclusion in this project, we selected the top 12 hospital systems based on microbiology message volume. As of 2012, these 12 hospital systems included 27 hospitals plus some smaller facilities. They extend from northernmost to southernmost Indiana. The five hospital systems in the original MRSA/VRE registry are included; therefore, the 12 hospital systems reflect an expansion of the infection control network both phylogenetically (to gram-negative organisms) and geographically (beyond the Indianapolis area).
Evaluating HL7 Patterns and Building a Library of Concepts
A vital step was to thoroughly study a large batch of the messages from each of the 12 hospital systems. By scrutinizing the structure and content of the HL7 messages that each institution sent to the INPC, we identified the "canonical forms"—the main patterns—that each institution was using. We analyzed how these patterns deviated from the HL7 standard. We then wrote programs to address these patterns. The programs use a combination of two natural language processing (NLP) methods—the REX (Regenstrief Extraction) tool developed by Dr. Friedlin,27 and the open-source General Architecture for Text Engineering (GATE) software28—plus additional Java steps to extract relevant content and to generate the corrected HL7 structure, which we append to the messages.
With regard to the content within the HL7 messages, we built an empirical library of all of the variants (including abbreviations and misspellings) of organism names, antibiotic names, body sources, and hospital building names/abbreviations found in the hospitals' messages. Our process maps the wide variety of nonstandardized content in the incoming messages (e.g., among bacteria: "Prt mirabilis," "Acinetobacter baum./haemol.," "Ec faecalis"; among antibiotics: "Piperacillin/T," or brand rather than generic names) into information that the decision support engine can act upon. We also use the open-source software Organism Tagger,29 but our own library for variant names of bacteria goes beyond what Organism Tagger can address. Where applicable, we map to existing concepts in the Regenstrief dictionary.
As an addition to INPC's exception browser tool, we are building in warnings for exceptions. Exceptions are situations when, despite all of the mappings and processes developed thus far, there is an incorrect or unexpected value (e.g., an unrecognized value for the organism or the antibiotic, or an incorrect value of minimum inhibitory concentration or the susceptibility interpretation sent by the hospital). Our team will review these exceptions as they are generated to give us a sense of the volume of exceptions and to help us plan for the long-term sustainability of our enhanced microbiology processing. During the development phase, we have been analyzing the exceptions as we analyze the large batches of messages in order to enhance the parsers.
A Dictionary Term for Gram-Negative Superbug
When the REICON engine appends standardized microbiology content to the existing ORU HL7 message, it also evaluates that content against criteria, developed in consultation with the IPs, for five categories of GNR MDRO (Table 1 [go to Appendix for a listing of organisms in each category]). If the criteria are met, an additional data element (GNR_MDRO) is written into the REICON database. GNR_MDRO stores which of the five "rules" was met and which version of the rules was applied. Because the criteria will evolve, our approach will enable investigators in future years to see which criteria were applied in any given period.
Table 1. Five rules for labeling, in REICON, a bacterium as a GNR MDRO
| Rule | Organism Category* | Definition |
|---|---|---|
| 1 | Enterobacteriaceae | Confirmed production of an extended-spectrum beta-lactamase(ESBL) |
| 2 | Enterobacteriaceae | Confirmed carbapenemase production |
| 3 | Pseudomonas aeruginosa | Resistant to three or more classes of the following:
|
| 4 | Acinetobacter baumannii | Resistant to three or more classes of the following:
|
| 5 | Other gram-negative bacterium not listed above | Resistant to all antibiotics tested, excluding colistin or tigecycline |
* For the list of organisms in each category, go to the appendix.
Note: GNR = gram-negative rod; MDRO = multidrug-resistant organism
ADT Hospitalization Messages and Master Patient Index Look-Up
We had to adapt to the new Oracle environment what we had in the VAX environment: a procedure that takes inbound ADT messages for hospitalizations and compares the patient identifiers with those in the INPC master patient index, to determine whether the patient being admitted has any medical record numbers (MRNs) in any other INPC institution. Then, using all of the newly admitted patient's MRNs, the REICON database is searched for any history of GNR MDRO. At the suggestion of the IPs, we do not limit the look-back period in querying for a history of GNR MDRO.
Automated Email Alerts
The emails being developed for the IPs at the admitting hospital include these key elements: organism, antibiotics tested, minimum inhibitory concentrations, susceptibility interpretation, body source of the culture, and health care facility where drawn. A disclaimer notes that the alerts were generated by an automated process and encourages the IPs to validate the results before acting upon them. At the request of the IPs, we include, in the email's subject line, an abbreviation for which rule applies: (1) ESBL-E, (2) CRE, (3) Pseudomonas, (4) Acinetobacter, or (5) Other. Not all of the hospital systems are interested in all five rules. If a hospital's policy is not to isolate patients with ESBL-positive Enterobacteriacea, its personnel can save time by not opening emails marked "ESBL-E."
Evaluation
In future work, we will evaluate the perceived utility of the alerts, the rate and timeliness with which isolation is used, and the geographic patterns of gram-negative MDRO infections. Because the "live" email alerts will require some time to accumulate prospectively, we are also using the results of our retrospective message processing to analyze how many patients with a history of GNR MDRO were subsequently admitted to any of the 12 hospital systems. In that way, we can model how many email alerts would have been generated in a retrospective timeframe. We also have begun planning a study of cost effectiveness to estimate whether REICON might, by increasing the rate and/or timeliness of isolation, reduce the spread, sequelae, and costs of GNR MDROs.
Results
We have processed a batch of 20 million ORU HL7 messages in order to extract, structure, and store the microbiology data. We also have written a 21-page deployment guide (not included here). Our message processing pipeline is summarized in Figure 1. The top panel depicts, in its long rectangle, the processing of ORU messages, with the HL7 correction engine described above (the "REICON Transform") as the center box. The bottom panel depicts, in its long rectangle, the processing of ADT messages to determine (in the "Decision Support" box) whether the patient being hospitalized is one who previously had ORU data for a GNR MDRO. A forthcoming paper will report our initial retrospective results in depth.
Discussion
Both microbiology data informatics and infection control notification are taking place amid larger-scale shifts in the hospital marketplace and in the health information exchange environment. Some hospitals are being acquired by others or are being brought into a larger hospital network. In some cases, these shifts increase the efficiency of our work; in other cases, they may create extra work. When one small hospital joined a larger network (one of the original five), we got an efficiency boost, in that some of the methods already developed for the larger network were immediately applicable to the new hospital joining it. By contrast, when another small hospital was acquired by a larger network, the hospital decided to keep its HL7 message processing and interfacing with the INPC distinct from that of its new network, at least for a few years.
Figure 1: REICON message processing
At the health information exchange level, within the past 2 years Indiana Health Information Exchange (IHIE), Inc. (which now runs the main INPC) employed a subcontractor, AT&T/Covisint, to process inbound HL7 messages from the INPC hospitals. As a result, the REICON project works with a separate stream of HL7 messages (a copy of the actual messages). This situation was part of the reason that we decided to create a separate REICON database to be housed separately but with the same schema as the main INPC.
This project underscores a key informatics principle: When dealing with many hospitals, the wide variety of electronic medical records (EMRs) and other information technology (IT) infrastructure that they use, and the wide variety of (often not fully standardized) HL7 message structure and content that they deliver, generate much complexity in message parsing. The relationship between the number of hospitals and the complexity of work is not linear, it is exponential. For this reason, it is always essential to study the message patterns at the outset, to help shape the subsequent coding as efficiently as possible. In one of our programmers' words, "If you've seen one [hospital] interface, you've seen one."
Microbiology data are particularly challenging because one culture result may contain multiple layers of results (body source, organism, susceptibility), many antibiotic assays, and additional elements such as ESBL and CRE. Even readers familiar with these complexities may be unaware of the variability in some seemingly straightforward data elements. The hospital facility where a culture was drawn, or where a patient is being admitted, is sometimes represented in HL7 messages as one of an alphabet soup of abbreviations that requires detective work to decode.
It is important, of course, to collaborate with the IPs (the end-users) in designing the system and to be flexible. Hospital systems are always re-evaluating their definitions of what constitutes an MDRO. A year ago, the criteria for MDRO Enterobacteriaceae at one of the larger hospital systems were [{resistant toCeftazidime or Ceftriaxone} and/or {confirmed production of an ESBL}] or [{resistant toImipenem or Meropenem} and/or {confirmed carbapenemase production}]. Subsequently, the hospital system restricted the rules to just ESBL or carbapenemase. It is desirable to build decision support structures that facilitate future revisions of the rules (and the addition of new organisms and rules).
We also found that it was not always necessary to create a new process for functions that the REICON project had not used before. Occasionally, by examining existing tools and infrastructure in the HOSS pipeline (the exception browser) or the INPC database (its concept dictionary), we found a way to tailor a solution that had already been written.
Conclusion
Although there are many methodological challenges in building a flexible microbiology data processing, storage, and alert system for IPs, the results may contribute to epidemiologic understanding of the patterns of gram-negative (and ultimately other) MDROs across a wide region. The email alerts may make some contribution to reducing the emergence and spread of these dangerous bacteria.
Acknowledgments
This project was funded under grant HS20014 from the Agency for Healthcare Research and Quality (AHRQ), U.S. Department of Health and Human Services. The findings and conclusions in this document are those of the authors, who are responsible for its content, and do not necessarily represent the views of AHRQ. No statement in this report should be construed as an official position of AHRQ or of the U.S. Department of Health and Human Services.
The authors gratefully acknowledge Mahesh Nalkande and his team at Persistent Systems, including Sumit Gurav, Preeti Lodha, Lakhan Pardeshi, Gunwant Walbe, and Aditya Joshi, for their creative and intensive work and many vital contributions in building the REICON system. We also gratefully acknowledge the excellent work and insightful consultation provided at the Regenstrief Institute by James Egg, Tony French, Bea Boxley, Andrew Martin, Sharmila Jothirajah, Barbara Smith, and Andy Feeney. We thank Amber McMahon at Indiana University for her important help as project coordinator, and Dr. Gunther Schadow of Pragmatic Data, LLC, for his invaluable advice at a number of key moments. We also are indebted to Suzanne Tolliver, Janet Reynolds, and Dr. Douglas H. Webb of Indiana University Health for sharing their knowledge about the criteria for multidrug-resistance and for their insights on other questions that ranged from the details of particular antibiotic susceptibility tests to larger aspects of project design.
Authors' Affiliations
Indiana University School of Medicine, Indianapolis, IN (MBR, KAS, SMEF, XL). Regenstrief Institute, Inc., Indianapolis, IN (MBR, SMEF, SK, LL, MT, ANK). Shepherd Internal Medicine, Indianapolis, IN (DCS). Indiana University School of Public Health, Indianapolis, IN (XL). Feinberg School of Medicine, Northwestern University, Chicago, IL (ANK).
Address correspondence to: Marc Rosenman, MD, Children's Health Services Research, Department of Pediatrics, Indiana University School of Medicine, 410 W. 10th Street Suite 1020, Indianapolis, IN 46202; Email: mrosenma@iu.edu.
References
1. Snitkin ES, Zelazny AM, Thomas PJ, et al. Tracking a hospital outbreak of carbapenem-resistant Klebsiella pneumoniae with whole-genome sequencing. Sci Transl Med 2012 Aug 22;4:148ra116. PMID: 22914622.
2. Centers for Disease Control and Prevention. New Carbapenem-Resistant Enterobacteriaceae Warrant Additional Action by Healthcare Providers. CDC Health Advisory. Updated 2013 Feb 14. www.bt.cdc.gov/HAN/han00341.asp. Accessed March 11, 2013.
3. McCarthy M. Resistance to antibiotics is ‘ticking time bomb' – stark warning from Chief Medical Officer Dame Sally Davies. The Independent 2013 Mar 11. www.independent.co.uk/news/science/resistance-to-antibiotics-is-ticking-time-bomb--stark-warning-from-chief-medical-officer-dame-sally-davies-8528469.html. Accessed March 11, 2013.
4. Carbonne A, Arnaud I, Maugat S, et al.; on behalf of the MDRB Surveillance National Steering Group (BMR-Raisin). National multidrug-resistant bacteria (MDRB) surveillance in France through the RAISIN network: a 9 year experience. J Antimicrob Chemother Epub 2012 Nov 29.
5. Bonkat G, Müller G, Braissant O, et al. Increasing prevalence of ciprofloxacin resistance in extended-spectrum-β-lactamase-producing Escherichia coli urinary isolates. World J Urol Epub 2013 Jan 29. PMID: 23358791.
6. Al-Assil B, Mahfoud M, Hamzeh AR. Resistance trends and risk factors of extended spectrum β-lactamases in Escherichia coli infections in Aleppo, Syria. Am J Infect Control Epub 2013 Jan 19. PMID: 23337303.
7. Kanoksil M, Jatapai A, Peacock SJ, et al. Epidemiology, microbiology and mortality associated with community-acquired bacteremia in northeast Thailand: a multicenter surveillance study. PLoS One 2013;8(1):e54714. Epub 2013 Jan 18. PMID: 23349954.
8. Karfunkel D, Carmeli Y, Chmelnitsky I, et al. The emergence and dissemination of CTX-M-producing Escherichia coli sequence type 131 causing community-onset bacteremia in Israel. Eur J Clin Microbiol Infect Dis Epub 2012 Nov 2. PMID: 23117265.
9. Kallen AJ, Srinivasan A. Current epidemiology of multidrug-resistant gram-negative bacilli in the United States. Infect Control Hosp Epidemiol 2010 Nov;31(Suppl 1):S51-4. PMID: 20929371.
10. Poole K. Pseudomonas aeruginosa: resistance to the max. Front Microbiol 2011;2:65. Epub 2011 Apr 5. PMID: 21747788.
11. Mesaros N, Nordmann P, Plésiat P, et al. Pseudomonas aeruginosa: resistance and therapeutic options at the turn of the new millennium. Clin Microbiol Infect 2007 Jun;13(6):560-78. PMID: 17266725.
12. Zavascki AP, Carvalhaes CG, Picão RC, et al. Multidrug-resistant Pseudomonas aeruginosa and Acinetobacter baumannii: resistance mechanisms and implications for therapy. Expert Rev Anti Infect Ther 2010 Jan;8(1):71-93. PMID: 20014903.
13. Klevens RM, Morrison MA, Nadle J, et al. Invasive methicillin-resistant Staphylococcus aureus infections in the United States. JAMA 2007 Oct 17;298(15):1763-71. PMID: 17940231.
14. Doi Y, Park YS, Rivera JI, et al. Community-associated extended-spectrum β-lactamase-producing Escherichia coli infection in the United States. Clin Infect Dis 2013 Mar;56(5):641-8. PMID: 23150211.
15. Nicolas-Chanoine MH, Gruson C, Bialek-Davenet S, et al. 10-fold increase (2006-11) in the rate of healthy subjects with extended-spectrum β-lactamase-producing Escherichia coli faecal carriage in a Parisian check-up centre. J Antimicrob Chemother 2013 Mar;68(3):562-8. PMID: 23143897.
16. Kaarme J, Molin Y, Olsen B, Melhus A. Prevalence of extended-spectrum beta-lactamase-producing Enterobacteriaceae in healthy Swedish preschool children. Acta Paediatr Epub 2013 Feb 19. PMID: 23419070.
17. Weisenberg SA, Mediavilla JR, Chen L, et al. Extended spectrum beta-lactamase-producing Enterobacteriaceae in international travelers and non-travelers in New York City. PLoS One 2012;7(9):e45141. PMID: 23028808.
18. Birgand G, Armand-Lefevre L, Lolom I, et al. Duration of colonization by extended-spectrum β-lactamase-producing Enterobacteriaceae after hospital discharge. Am J Infect Control Epub 2012 Sep 18. PMID: 22998785.
19. Huang SS, Taliser RA, Song Y, et al. Quantifying interhospital patient sharing as a mechanism for infectious disease spread. Infect Control Hospital Epidemiol 2010 Nov;31(11):1160-9. PMID: 20874503.
20. Lee BY, McGlone SM, Wong KF, et al. Modeling the spread of methicillin-resistant Staphylococcus aureus (MRSA) outbreaks throughout the hospitals in Orange County, California. Infect Control Hosp Epidemiol 2011 Jun;32(6):562-72. PMID: 21558768.
21. Kho AN, Lemmon L, Commiskey M, et al. Use of a regional health information exchange to detect crossover of patients with MRSA between urban hospitals. J Am Med Inform Assoc 2008 Mar-Apr;15(2): 212-6. PMID: 18096903.
22. McDonald CJ, Overhage JM, Barnes M, et al. The Indiana Network For Patient Care: A working local health information infrastructure. Health Aff 2005 Sep-Oct;24(5):1214-20. PMID: 16162565.
23. Sittig DF, Hazlehurst BL, Brown J, et al. A survey of informatics platforms that enable distributed comparative effectiveness research using multi-institutional heterogen[e]ous clinical data. Med Care 2012 Jul;50 Suppl:S49-59. PMID: 22692259.
24. Overhage JM, Dexter PR, Perkins SM, et al. A randomized, controlled trial of clinical information shared from another institution. Ann Emerg Med 2002 Jan;39(1):14-23. PMID: 11782726.
25. Kho AN, Dexter P, Lemmon L, et al. Connecting the dots: creation of an electronic regional infection control network. Stud Health Technol Inform 2007;129(Pt 1):213-7. PMID: 17911709.
26. Kho AN, Lemmon L, Dexter PR, et al. An operational citywide electronic infection control network: results from the first year. AMIA Annu Symp Proc 2008 Nov 6:1222. PMID: 18999011.
27. Were MC, Gorbachev S, Cadwallader J, et al. Natural language processing to extract follow-up provider information from hospital discharge summaries. AMIA Annu Symp Proc 2010 Nov 13;2010:872-6. PMID: 21347103.
28. GATE: General Architecture for Text Engineering; c2011. http://gate.ac.uk/. Accessed January 7, 2013.
29. Naderi N, Kappler T, Baker CJ, et al. OrganismTagger: detection, normalization and grounding of organism entities in biomedical documents. Bioinformatics 2011 Oct 1;27(19):2721-9. PMID: 21828087.
Appendix
List of Organisms from Table 1
Enterobacteriaceae
Escherichia
Escherichia coli, E coli, E coli O157:H7
Escherichia vulneris
Escherichia hermannii
(or other species)
Klebsiella
Klebsiella pneumoniae
Klebsiella oxytoca
(or other species)
Enterobacter
Enterobacter sp
Enterobacter cloacae
Enterobacter species
Enterobacter aerogenes
Enterobacter agglomerans
(or other species)
Proteus
Proteus mirabilis
Proteus vulgaris
(or other species)
Serratia
Serratia marcescens
or other species)
Citrobacter
Citrobacter freundii
Citrobacter koseri
(or other species)
Salmonella
Salmonella (any species)
Shigella
Shigella (any species)
Yersinia
Yersinia (any species)
Morganella
Morganella (any species)
Providencia
Providencia (any species)
Hafnia
Hafnia (any species)
Edwardsiella
Edwardsiella (any species)
Other gram-negative bacteria not listed:
Pseudomonas (where the species is not aeruginosa)
Acinetobacter (where the species is not baumannii)
Stenotrophomonas maltophilia
Burkholderia cepacia (or other species)
Legionella pneumophila
Campylobacter jejuni (or other species)
Moraxella catarrhalis
Branhamella catarrhalis
Haemophilus influenzae (or other species)
Vibrio cholerae (or other species)
