International Journal of Antimicrobial Agents 44 (2014) 424–430
Contents lists available at ScienceDirect
International Journal of Antimicrobial Agents journal homepage: http://www.elsevier.com/locate/ijantimicag
The bigger picture: The history of antibiotics and antimicrobial resistance displayed by scientometric data Christian Brandt a,∗ , Oliwia Makarewicz a , Thomas Fischer b , Claudia Stein a , Yvonne Pfeifer c , Guido Werner c , Mathias W. Pletz a a
Center for Infectious Diseases and Infection Control, Jena University Hospital, 07740 Jena, Germany Department of Business Informatics, Friedrich Schiller University, Jena, Germany c Nosocomial Pathogens and Antibiotic Resistance, Robert Koch Institute, Wernigerode, Germany b
a r t i c l e
i n f o
Article history: Received 26 March 2014 Accepted 12 August 2014 Keywords: Semantic search -Lactamase MRSA VRE Bacterial infection Scientometric data
a b s t r a c t Monitoring the rapid global spread of antimicrobial resistance requires an over-regional and fast surveillance tool. Data from major surveillance studies based on aggregated results of selected sentinel laboratories or retrospective strain collections are not available for the whole scientific community and are limited by time and region. Thus, we tested an alternative approach to monitor resistance trends by automated semantic and scientometric analysis of all (>100 000) related PubMed entries. A semantic search was done using ‘Gene Ontology’ and MeSH vocabulary and additional search terms for further data refinement. Data extraction was performed using the semantic search engine ‘GoPubMed’. The timely relationship between introduction of novel -lactam antibiotic classes into the market and emergence of respective resistance was investigated using nearly 22 300 publications over the last 70 years. Further analysis was done with around 54 000 publications related to ‘infectious diseases’ and an additional 50 000 publications related to ‘antimicrobial resistance’ to estimate current trends in publication interest regarding resistance development since 1940. Scientometric results were compared with data from the major surveillance network EARS-Net. Furthermore, the relationship between micro-organism, year and antibiotic market introduction was investigated for eight key antibiotics using nearly 37 500 publications. Owing to influencing factors such as availability of alternative antibiotics, scientometric analysis correlated only partly with resistance development. However, it provides a fast, reliable and global overview of the clinical and public health importance of a specific resistance including the period of the 1940s–1980s, when resistance surveillance studies were not yet established. © 2014 Elsevier B.V. and the International Society of Chemotherapy. All rights reserved.
1. Introduction The global spread of antimicrobial resistance requires fast international and nationwide surveillance. Major surveillance studies [e.g. European Antimicrobial Resistance Surveillance Network (EARS-Net), National Antimicrobial Resistance Monitoring System for Enteric Bacteria (NARMS, USA), The Surveillance Network (TSN, USA), Antibiotic Resistance Surveillance (ARS, Germany), Central Asian and Eastern European Surveillance on Antimicrobial Resistance (CAESAR) and Surveillance of Antibiotic Use and Bacterial Resistance in German Intensive Care Units (SARI)] provide only access to processed information but not to the raw data. Therefore, extraction of individually required information is impossible
∗ Corresponding author. Tel.: +49 3641 932 4793; fax: +49 3641 932 4652. E-mail address: [email protected] (C. Brandt).
or at least labour-intensive and time-consuming for scientists who are not members of the respective consortium. Moreover, the opportunity to monitor individual resistance trends is often limited to defined time periods, regions and clinical entities (e.g. bacteraemia). Most surveillance systems heavily depend on the engagement of the contributing members as well as available funding. They tend to exhaust when they become more extensive or the funding declines, resulting in a possibly inaccurate reflection of the real resistance epidemiology. Therefore, there is a need for an independent open-source-based monitoring approach that allows fast and easy access to a broad range of information. With around 23 million publications and annually rising entries (ca. 1 million in 2012), PubMed is a comprehensive and continuously growing database that might serve as a fast and suitable source for monitoring resistance trends. The disadvantage of this large amount of data is the time-consuming extraction of relevant information. Currently, systematic literature research is an increasing challenge
http://dx.doi.org/10.1016/j.ijantimicag.2014.08.001 0924-8579/© 2014 Elsevier B.V. and the International Society of Chemotherapy. All rights reserved.
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
425
Fig. 1. Example ‘search flow’ for penicillin resistance, excluding meticillin resistance and reviews (for other search terms see Supplementary Table S1).
for scientists. Monitoring of global resistance development by analysing research papers and other sources appears feasible but needs mainly manual data extraction. For a computer algorithmbased search, the widely heterogeneous structure and the inherent semantics of full papers are a major barrier. To overcome these limitations, we tested the potential of the ontology-based search of ‘GoPubMed® ’ (http://www.gopubmed.org), which offers a faceted search interface to its semantic search engine by using the ‘Gene Ontology’ (GO) and ‘Medical Subject Headings’ (MeSH) vocabulary to categorise and explore abstracts of PubMed entries [1–3]. MeSH is an online-available, open-source, controlled vocabulary and thesaurus that arranges terms not only alphabetically but also hierarchically. Those terms are called ‘descriptors’ or ‘main headings’ since they include additional information, e.g. the hierarchical tree location(s). For example, searching for abstracts using the descriptor ‘bacterial infections’ takes into consideration alternate spelling as well as related terms that are MeSH-pre-defined in a hierarchical manner (taxonomy), e.g. bacteraemia, endocarditis (bacterial), skin disease (bacterial), brucellosis, legionellosis, hordeolum, tuberculosis, etc. Therefore, the descriptor ‘bacterial infections’ enables searching for all topics related to bacterial infection, even if the specific term ‘bacterial infections’ is not present within an abstract. This approach delivers a more embracing search result compared with a simple keyword search. The aim and use of GO is similar to MeSH but focuses on contexts strongly related to molecular biology including genes, proteins, their functions and biological processes. GoPubMed presents the retrieved publications in a contextrelated, hierarchically and ontologically arranged manner allowing further data refinements. Such refinements include, for example, the alteration of descriptors by excluding terms to render unrelated subjects and to confine received results. Moreover, statistical and visualisation tools allow a user-friendly output of the information. 2. Methods We assumed that an increase or decrease in publication activity covering a certain subject is influenced by its public health significance. To describe this publication activity, we used the ‘relative research interest’, defined as the proportion (in %) of subject-related papers per year within the number of all published papers on PubMed [1]. This approach controls for the general annual increase in PubMed entries and prevents an overestimation of the publication activity. To increase the accuracy of this process, all reviews were excluded from the analysis (Fig. 1). Data sets were generated by using MeSH terms, if available, or, if necessary, additional search terms. A detailed list of all used terms can be found in Supplementary Table S1. All publications accessible on 31 January 2014 were
included in the analysis. However, not all publication titles from 2013 could be used due to yet incomplete PubMed indexing. One should note that the US antibiotic market releases retrieved for this work [4–6] are an approximate value for this scientometric approach, since market release may differ between different countries. 3. Results and discussion 3.1. Global infection and resistance trends In total, we found 49 690 publications released for the years 1940–2013 related to ‘beta-lactamases’ (22 275), ‘vancomycin resistance’ (2875), ‘methicillin resistance’ (18 706), ‘tetracycline resistance’ (2933) and ‘fluoroquinolones resistance’ (2901) (Fig. 2A). To estimate the burden of disease by individual species, 54 381 publications released between 1940 and 2012 that referred to ‘Escherichia coli infections’ (17 447), ‘Klebsiella pneumoniae infections’ (3512), ‘Acinetobacter baumannii infections’ (1444), ‘Pseudomonas aeruginosa infections’ (10 241) and ‘Staphylococcus aureus infections’ (21 737) were extracted (Fig. 2B). The relative research interest was compared with available data from EARS-Net [7] to assess the accuracy of the search results regarding the development of resistance (Fig. 2C). When comparing the results for ‘beta-lactamases’, ‘vancomycin resistance’, ‘methicillin resistance’, ‘tetracycline resistance’ and ‘fluoroquinolones resistance’, we found that the relative research interest for all of these topics except -lactamases are plateauing (meticillin resistance) or declining (Fig. 2A). Obviously, the increasing relative research interest for -lactamases correlates with the worldwide emergence of multidrug-resistant Gram-negative bacteria and their limited treatment options [8–10] and might be an explanation of the downward trends of interest in other resistances. The trend of -lactamases was also confirmed by extracting publications related to infections and their respective pathogens (Fig. 2B). Data from EARS-Net reflect the increasing extended resistance to -lactam antibiotics in E. coli and K. pneumoniae (Fig. 2C). Interestingly, we found a plateau phase in the relative research interest of meticillin-resistant S. aureus (MRSA) over the last 5 years both for the relative research interest in ‘methicillin resistance’ and ‘S. aureus infections’. The stagnated relative research interest correlates with the current decreasing MRSA resistance rates in Europe according to EARS-Net (Fig. 2C) and may be also related to the availability of novel antibiotics with anti-MRSA activity (e.g. linezolid, tigecycline, daptomycin, ceftaroline). Recent analyses of the TSN and EARS-Net hypothesised that the decreasing trends of MRSA in the USA, Germany and the UK may be explained by the effectiveness of the implemented infection prevention measures [11–13].
426
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
Fig. 2. (A) Global resistance trends calculated by scientometric analysis, (B) global infection trends calculated by scientometric analysis and (C) resistance rate for each year calculated out of the European Antimicrobial Resistance Surveillance Network (EARS-Net) database for vancomycin-resistant enterococci (115 317 isolates, 4218 resistant), meticillin-resistant Staphylococcus aureus (MRSA) (313 211 isolates, 64 618 resistant), multidrug-resistant Klebsiella pneumoniae (69 126 isolates, 12 008 resistant) and thirdgeneration cephalosporin-resistant Escherichia coli (421 986 isolates, 23 868 resistant). The x-axes in (C) begin with meaningful ascertainment (>5 contributing countries/year) of EARS-Net.
However, there is no surveillance study covering resistance development since the 1940s. EARS-Net surveillance was implemented in 1998. Therefore, comparison between both methods can cover only this timeframe. Comparisons with other surveillance studies (i.e. NARMS, CAESAR) were limited by restricted access to the raw data.
3.2. Publications related to ˇ-lactamases There is a continuously increasing publication activity covering -lactamases. Of the 22 275 related publications, 3577 reported on extended-spectrum -lactamases (ESBLs) and 1263 on carbapenemases. Fig. 3 displays the relative research interest from 1940–2013
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
427
Fig. 3. (A) Trends in resistance to -lactam antibiotics according to the relative research interest, including proportion of extended-spectrum -lactamase (ESBL) and carbapenemases, in relation to the market release of new -lactam antibiotics. (B) Publications containing -lactamase-related clinical outbreaks extracted by MeSH vocabulary in comparison with entries of the Outbreak Database (http://www.outbreak-database.com).
in relation to the market release of certain classes of -lactam antibiotics and the publications on -lactamases related to clinical outbreaks over the last 30 years. The course of the relative research interest in this figure may be explained as follows. The relative research interest increased with increasing resistance but decreased briefly after antibiotics that overcome the respective type of resistance became available (e.g. after the introduction of cephalosporins, -lactamase inhibitors and carbapenems in the 1980s). The relative research interest rose again with the appearance of ESBLs and carbapenemases. Furthermore, the increasing amount of publications related to hospital outbreaks caused by -lactam-resistant micro-organisms is alarming (Fig. 3B). Despite the number of included entries differing between the Outbreak Database (http://www.outbreakdatabase.com) and the GoPubMed search, the increasing trend of -lactamase-related outbreaks is similar until year 2007 between both sources. The difference observed after 2007 may be explained by difficulties in long-term maintenance of such manually annotated databases that depend on the engagement of contributing members and available funding. 3.3. Correlations between antibiotic market introduction and respective resistance development Current surveillance systems were implemented within the last decades and thus do not cover an appropriate time range to image the relationship between market release and resistance development for most antibiotics in clinical use. We performed a
scientometric analysis including 37 469 publications ranging from 1940–2012 for penicillin (9762), tetracycline (2833), erythromycin (1674), meticillin (17 016), gentamicin (1173), vancomycin (1977), carbapenems (952), cephalosporins (1725), levofloxacin (143) and linezolid (214) to describe the relationship between market releases and respective resistance in the main pathogens treated with these antibiotics. The current report of the US Centers for Disease Control and Prevention (CDC) (Antibiotic Resistance Threats in the United States, 2013; http://www.cdc.gov/drugresistance/ threat-report-2013/pdf/ar-threats-2013-508.pdf) was used to identify resistant pathogens with the highest public health relevance. The relative research interest in resistance regardless of the species (e.g. ‘penicillin resistance’, ‘tetracycline resistance’, etc.) was included as a reference (black area in Fig. 4). Since penicillin exhibits the most extensive timeframe, it provides some interesting insights. Shortly after market release, there is a slight ‘publication burst’ in publications related to penicillin resistance. After a ‘lag phase’ of minor relative research interest of ca. 20 years, a sudden increase in interest in penicillin resistance occurred that is followed by a slight but steady negative slope. Similar courses of relative research interest with a ‘lag phase’ of minor interest and a marked increase 20 years after market release that is afterwards slightly decreasing can also be observed for tetracycline, erythromycin and meticillin, and with a reduced ‘lag phase’ of around 10 years for gentamicin. Noteworthy, a decreasing relative research interest is not yet observed for cephalosporin and carbapenem resistance. In contrast to the antibiotics mentioned before, cephalosporins and carbapenems are the mainstay of
428
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
Fig. 4. Relative research interest based on the resistances against key antibiotics, including the most relevant pathogens. The respective market releases are included.
empirical treatment in critically ill patients and belong to the most frequently used antibiotics in hospitals. According to spot-checking of 40 titles and available abstracts, the ‘publication burst’ for penicillin and erythromycin around the time of market release reflects the scientific interest in a novel
antibiotic, including the description of individually selected laboratory or clinically derived resistant isolates. The lag phase of minor publication interest may be explained by the successful clinical use of the respective antibiotic, whereas the high increase after the ‘lag phase’ reflects the broad spread of clinically relevant resistant
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
strains and related treatment failure (proven by spot-checking of 40 titles and available abstracts per strain and species, or all if less available). The decrease in relative research interest regarding resistance probably correlates with the availability of novel antibiotics that overcome the respective type of resistance (e.g. cephalosporins for the treatment of penicillin-resistant S. aureus) or the realisation that the clinical impact was overestimated, e.g. ‘low-level’ penicillin resistance in pneumococcal pneumonia [14], that could also be mapped by the relative research interest in penicillin resistance related to Streptococcus pneumoniae with a sudden increase around 1999 and a continuous decrease briefly afterwards. Both resistance to carbapenems and cephalosporins share the lag phase that was also observed for other resistances, but the relative research interest continues to increase, reflecting the current public health threat caused by these types of resistance [15].
3.4. Factors influencing research interest and limitations of scientometric resistance data extraction As outlined above, research interest is not only driven by resistance rates but also by the clinical relevance of the respective resistance. For example, resistance to tetracyclines is frequent in Europe but does not impose a major threat to public health – in contrast to ESBLs or carbapenemases – because tetracyclines are not used for severe bacterial infections. This minor ‘medical need’ is reflected by lower research interest in tetracycline resistance compared with -lactam resistance despite high tetracycline resistance rates (Fig. 2). Other examples for a mismatch between resistance rates and the respective clinical importance include erythromycin resistance in S. aureus and penicillin resistance in S. aureus after -lactamase inhibitors and cephalosporins have become available. Scientometric data cannot easily map resistance trends on a regional scale because there are a variety of biasing factors, such as different surveillance systems, political decisions regarding research focus and funding, or the cultural influence of each country [16,17], affecting the overall results. Therefore, we were unable to correlate the relative research interest of individual European countries for MRSA with the respective EARS-Net data (not shown). A Google-based algorithm, such as the Google search index for common flu reflecting regional flu activity [18,19], can probably not be used to mirror antibiotic resistance since the majority of Google users lack the expertise for an appropriate and analysable search for antibiotic resistance. Another limitation is that logical linguistic inter-relation is elusive and biases the accuracy of the output. The limitation of disambiguation of the semantic search is reflected by the similar curve shape of research interest in vancomycin resistance for both Enterococcus faecalis and Enterococcus faecium. Vancomycin resistance is rare in E. faecalis [20]. A spot-check of 50 abstracts revealed that E. faecalis and E. faecium are frequently mentioned together in the respective abstracts in the same sentence, limiting the discriminative capacity of the semantic algorithm used by GoPubMed.
4. Conclusions Scientists all over the world are ambitious to publish their results in PubMed-listed journals thereby opening the information to a wide audience. The initial aim was to test whether scientometric data may provide a more general and robust estimation of the overall resistance development than regionally and temporally limited surveillance studies owing to the large amount of publications covering timeframes since 1940.
429
Scientometric analysis correlates partly with resistance rates. Besides resistance development, introduction of novel antibiotics, increasing or decreasing public health significance, availability of alternative antibiotics, and available research funding have an impact on the relative research interest, limiting the overall accuracy of this approach. Regarding the point of view, this is not necessarily a disadvantage, e.g. when the aim is to estimate where the medical need for development of novel antibiotics is highest. Nevertheless, scientometric analysis of data extracted by freely available search engines (e.g. ‘GoPubMed’) provides a fast and global overview and may reflect the ‘medical need’, i.e. gap between resistance development and specific treatment options, better than the numerically more accurate analysis of surveillance studies. In conclusion, the scientometric approach using freely available search tools provides a fast and global overview of the development and the clinical and public health importance of a specific resistance. In addition, it provides an estimate on the ‘durability’ of newly introduced antibiotics, which may help to support a foreseeing policy regarding research funding for anti-infectives. The scientometric approach is, however, no substitute for wellperformed surveillance studies, which of course need sufficient funding. Funding: This work was supported by grants from the German Ministry of Education and Research (BMBF) [grant no. 01KI1204 and 01EO1002]. Competing interests: None declared. Ethical approval: Not required. Acknowledgments The authors would like to acknowledge the GoPubMed team for this platform as well as Transinsight GmbH (Dresden, Germany) for providing raw data. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ijantimicag. 2014.08.001. References [1] Doms A, Schroeder M. GoPubMed: exploring PubMed with the Gene Ontology. Nucleic Acids Res 2005;33:W783–6. [2] Rogers FB. Medical subject headings. Bull Med Lib Assoc 1963;51:114–16. [3] Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, Foulger R, et al. The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res 2004;32:D258–61. [4] Lewis K. Platforms for antibiotic discovery. Nat Rev Drug Discovery 2013;12:371–87. [5] Powers JH. Antimicrobial drug development—the past, the present, and the future. Clin Microbiol Infect 2004;10:23–31. [6] Gentamicin. Br Med J 1967;1:158–9. [7] European Centre for Disease Prevention and Control. Antimicrobial resistance surveillance in Europe 2010. In: Annual report of the European Antimicrobial Resistance Surveillance Network (EARS-Net). ECDC; 2011. [8] da Silva RM, Traebert J, Galato D. Klebsiella pneumoniae carbapenemase (KPC)producing Klebsiella pneumoniae: a review of epidemiological and clinical aspects. Expert Opin Biol Ther 2012;12:663–71. [9] Lee GC, Burgess DS. Treatment of Klebsiella pneumoniae carbapenemase (KPC) infections: a review of published case series and case reports. Ann Clin Microbiol Antimicrob 2012;11:32. [10] Kaul DR, Collins CD, Hyzy RC. New developments in antimicrobial use in sepsis. Curr Pharm Des 2008;14:1912–20. [11] Master RN, Deane J, Opiela C, Sahm DF. Recent trends in resistance to cell envelope-active antibacterial agents among key bacterial pathogens. Ann NY Acad Sci 2013;1277:1–7. [12] Humphreys H. Do guidelines for the prevention and control of methicillinresistant Staphylococcus aureus make a difference. Clin Microbiol Infect 2009;15:39–43. [13] Humphreys H. National guidelines for the control and prevention of methicillin-resistant Staphylococcus aureus—what do they tell us? Clin Microbiol Infect 2007;13:846–53.
430
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
[14] Kanungo R, Rajalakshmi B. Serotype distribution & antimicrobial resistance in Streptococcus pneumoniae causing invasive & other infections in south India. Indian J Med Res 2001;114:127–32. [15] Laxminarayan R, Duse A, Wattal C, Zaidi AK, Wertheim HF, Sumpradit N, et al. Antibiotic resistance—the need for global solutions. Lancet Infect Dis 2013;13:1057–98. [16] Borg MA, Camilleri L, Waisfisz B. Understanding the epidemiology of MRSA in Europe: do we need to think outside the box? J Hosp Infect 2012;81: 251–6. [17] Hofstede GH. Culture’s consequences: comparing values, behaviors, institutions, and organizations across nations. 2nd ed. Thousand Oaks, CA: Sage Publications; 2001.
[18] Olson DR, Konty KJ, Paladini M, Viboud C, Simonsen L. Reassessing Google Flu Trends data for detection of seasonal and pandemic influenza: a comparative epidemiological study at three geographic scales. PLoS Comput Biol 2013;9:e1003256. [19] de Lange MM, Meijer A, Friesema IH, Donker GA, Koppeschaar CE, Hooiveld M, et al. Comparison of five influenza surveillance systems during the 2009 pandemic and their association with media attention. BMC Public Health 2013;13:881. [20] Cattoir V, Leclercq R. Twenty-five years of shared life with vancomycinresistant enterococci: is it time to divorce? J Antimicrob Chemother 2013;68:731–42.
Contents lists available at ScienceDirect
International Journal of Antimicrobial Agents journal homepage: http://www.elsevier.com/locate/ijantimicag
The bigger picture: The history of antibiotics and antimicrobial resistance displayed by scientometric data Christian Brandt a,∗ , Oliwia Makarewicz a , Thomas Fischer b , Claudia Stein a , Yvonne Pfeifer c , Guido Werner c , Mathias W. Pletz a a
Center for Infectious Diseases and Infection Control, Jena University Hospital, 07740 Jena, Germany Department of Business Informatics, Friedrich Schiller University, Jena, Germany c Nosocomial Pathogens and Antibiotic Resistance, Robert Koch Institute, Wernigerode, Germany b
a r t i c l e
i n f o
Article history: Received 26 March 2014 Accepted 12 August 2014 Keywords: Semantic search -Lactamase MRSA VRE Bacterial infection Scientometric data
a b s t r a c t Monitoring the rapid global spread of antimicrobial resistance requires an over-regional and fast surveillance tool. Data from major surveillance studies based on aggregated results of selected sentinel laboratories or retrospective strain collections are not available for the whole scientific community and are limited by time and region. Thus, we tested an alternative approach to monitor resistance trends by automated semantic and scientometric analysis of all (>100 000) related PubMed entries. A semantic search was done using ‘Gene Ontology’ and MeSH vocabulary and additional search terms for further data refinement. Data extraction was performed using the semantic search engine ‘GoPubMed’. The timely relationship between introduction of novel -lactam antibiotic classes into the market and emergence of respective resistance was investigated using nearly 22 300 publications over the last 70 years. Further analysis was done with around 54 000 publications related to ‘infectious diseases’ and an additional 50 000 publications related to ‘antimicrobial resistance’ to estimate current trends in publication interest regarding resistance development since 1940. Scientometric results were compared with data from the major surveillance network EARS-Net. Furthermore, the relationship between micro-organism, year and antibiotic market introduction was investigated for eight key antibiotics using nearly 37 500 publications. Owing to influencing factors such as availability of alternative antibiotics, scientometric analysis correlated only partly with resistance development. However, it provides a fast, reliable and global overview of the clinical and public health importance of a specific resistance including the period of the 1940s–1980s, when resistance surveillance studies were not yet established. © 2014 Elsevier B.V. and the International Society of Chemotherapy. All rights reserved.
1. Introduction The global spread of antimicrobial resistance requires fast international and nationwide surveillance. Major surveillance studies [e.g. European Antimicrobial Resistance Surveillance Network (EARS-Net), National Antimicrobial Resistance Monitoring System for Enteric Bacteria (NARMS, USA), The Surveillance Network (TSN, USA), Antibiotic Resistance Surveillance (ARS, Germany), Central Asian and Eastern European Surveillance on Antimicrobial Resistance (CAESAR) and Surveillance of Antibiotic Use and Bacterial Resistance in German Intensive Care Units (SARI)] provide only access to processed information but not to the raw data. Therefore, extraction of individually required information is impossible
∗ Corresponding author. Tel.: +49 3641 932 4793; fax: +49 3641 932 4652. E-mail address: [email protected] (C. Brandt).
or at least labour-intensive and time-consuming for scientists who are not members of the respective consortium. Moreover, the opportunity to monitor individual resistance trends is often limited to defined time periods, regions and clinical entities (e.g. bacteraemia). Most surveillance systems heavily depend on the engagement of the contributing members as well as available funding. They tend to exhaust when they become more extensive or the funding declines, resulting in a possibly inaccurate reflection of the real resistance epidemiology. Therefore, there is a need for an independent open-source-based monitoring approach that allows fast and easy access to a broad range of information. With around 23 million publications and annually rising entries (ca. 1 million in 2012), PubMed is a comprehensive and continuously growing database that might serve as a fast and suitable source for monitoring resistance trends. The disadvantage of this large amount of data is the time-consuming extraction of relevant information. Currently, systematic literature research is an increasing challenge
http://dx.doi.org/10.1016/j.ijantimicag.2014.08.001 0924-8579/© 2014 Elsevier B.V. and the International Society of Chemotherapy. All rights reserved.
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
425
Fig. 1. Example ‘search flow’ for penicillin resistance, excluding meticillin resistance and reviews (for other search terms see Supplementary Table S1).
for scientists. Monitoring of global resistance development by analysing research papers and other sources appears feasible but needs mainly manual data extraction. For a computer algorithmbased search, the widely heterogeneous structure and the inherent semantics of full papers are a major barrier. To overcome these limitations, we tested the potential of the ontology-based search of ‘GoPubMed® ’ (http://www.gopubmed.org), which offers a faceted search interface to its semantic search engine by using the ‘Gene Ontology’ (GO) and ‘Medical Subject Headings’ (MeSH) vocabulary to categorise and explore abstracts of PubMed entries [1–3]. MeSH is an online-available, open-source, controlled vocabulary and thesaurus that arranges terms not only alphabetically but also hierarchically. Those terms are called ‘descriptors’ or ‘main headings’ since they include additional information, e.g. the hierarchical tree location(s). For example, searching for abstracts using the descriptor ‘bacterial infections’ takes into consideration alternate spelling as well as related terms that are MeSH-pre-defined in a hierarchical manner (taxonomy), e.g. bacteraemia, endocarditis (bacterial), skin disease (bacterial), brucellosis, legionellosis, hordeolum, tuberculosis, etc. Therefore, the descriptor ‘bacterial infections’ enables searching for all topics related to bacterial infection, even if the specific term ‘bacterial infections’ is not present within an abstract. This approach delivers a more embracing search result compared with a simple keyword search. The aim and use of GO is similar to MeSH but focuses on contexts strongly related to molecular biology including genes, proteins, their functions and biological processes. GoPubMed presents the retrieved publications in a contextrelated, hierarchically and ontologically arranged manner allowing further data refinements. Such refinements include, for example, the alteration of descriptors by excluding terms to render unrelated subjects and to confine received results. Moreover, statistical and visualisation tools allow a user-friendly output of the information. 2. Methods We assumed that an increase or decrease in publication activity covering a certain subject is influenced by its public health significance. To describe this publication activity, we used the ‘relative research interest’, defined as the proportion (in %) of subject-related papers per year within the number of all published papers on PubMed [1]. This approach controls for the general annual increase in PubMed entries and prevents an overestimation of the publication activity. To increase the accuracy of this process, all reviews were excluded from the analysis (Fig. 1). Data sets were generated by using MeSH terms, if available, or, if necessary, additional search terms. A detailed list of all used terms can be found in Supplementary Table S1. All publications accessible on 31 January 2014 were
included in the analysis. However, not all publication titles from 2013 could be used due to yet incomplete PubMed indexing. One should note that the US antibiotic market releases retrieved for this work [4–6] are an approximate value for this scientometric approach, since market release may differ between different countries. 3. Results and discussion 3.1. Global infection and resistance trends In total, we found 49 690 publications released for the years 1940–2013 related to ‘beta-lactamases’ (22 275), ‘vancomycin resistance’ (2875), ‘methicillin resistance’ (18 706), ‘tetracycline resistance’ (2933) and ‘fluoroquinolones resistance’ (2901) (Fig. 2A). To estimate the burden of disease by individual species, 54 381 publications released between 1940 and 2012 that referred to ‘Escherichia coli infections’ (17 447), ‘Klebsiella pneumoniae infections’ (3512), ‘Acinetobacter baumannii infections’ (1444), ‘Pseudomonas aeruginosa infections’ (10 241) and ‘Staphylococcus aureus infections’ (21 737) were extracted (Fig. 2B). The relative research interest was compared with available data from EARS-Net [7] to assess the accuracy of the search results regarding the development of resistance (Fig. 2C). When comparing the results for ‘beta-lactamases’, ‘vancomycin resistance’, ‘methicillin resistance’, ‘tetracycline resistance’ and ‘fluoroquinolones resistance’, we found that the relative research interest for all of these topics except -lactamases are plateauing (meticillin resistance) or declining (Fig. 2A). Obviously, the increasing relative research interest for -lactamases correlates with the worldwide emergence of multidrug-resistant Gram-negative bacteria and their limited treatment options [8–10] and might be an explanation of the downward trends of interest in other resistances. The trend of -lactamases was also confirmed by extracting publications related to infections and their respective pathogens (Fig. 2B). Data from EARS-Net reflect the increasing extended resistance to -lactam antibiotics in E. coli and K. pneumoniae (Fig. 2C). Interestingly, we found a plateau phase in the relative research interest of meticillin-resistant S. aureus (MRSA) over the last 5 years both for the relative research interest in ‘methicillin resistance’ and ‘S. aureus infections’. The stagnated relative research interest correlates with the current decreasing MRSA resistance rates in Europe according to EARS-Net (Fig. 2C) and may be also related to the availability of novel antibiotics with anti-MRSA activity (e.g. linezolid, tigecycline, daptomycin, ceftaroline). Recent analyses of the TSN and EARS-Net hypothesised that the decreasing trends of MRSA in the USA, Germany and the UK may be explained by the effectiveness of the implemented infection prevention measures [11–13].
426
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
Fig. 2. (A) Global resistance trends calculated by scientometric analysis, (B) global infection trends calculated by scientometric analysis and (C) resistance rate for each year calculated out of the European Antimicrobial Resistance Surveillance Network (EARS-Net) database for vancomycin-resistant enterococci (115 317 isolates, 4218 resistant), meticillin-resistant Staphylococcus aureus (MRSA) (313 211 isolates, 64 618 resistant), multidrug-resistant Klebsiella pneumoniae (69 126 isolates, 12 008 resistant) and thirdgeneration cephalosporin-resistant Escherichia coli (421 986 isolates, 23 868 resistant). The x-axes in (C) begin with meaningful ascertainment (>5 contributing countries/year) of EARS-Net.
However, there is no surveillance study covering resistance development since the 1940s. EARS-Net surveillance was implemented in 1998. Therefore, comparison between both methods can cover only this timeframe. Comparisons with other surveillance studies (i.e. NARMS, CAESAR) were limited by restricted access to the raw data.
3.2. Publications related to ˇ-lactamases There is a continuously increasing publication activity covering -lactamases. Of the 22 275 related publications, 3577 reported on extended-spectrum -lactamases (ESBLs) and 1263 on carbapenemases. Fig. 3 displays the relative research interest from 1940–2013
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
427
Fig. 3. (A) Trends in resistance to -lactam antibiotics according to the relative research interest, including proportion of extended-spectrum -lactamase (ESBL) and carbapenemases, in relation to the market release of new -lactam antibiotics. (B) Publications containing -lactamase-related clinical outbreaks extracted by MeSH vocabulary in comparison with entries of the Outbreak Database (http://www.outbreak-database.com).
in relation to the market release of certain classes of -lactam antibiotics and the publications on -lactamases related to clinical outbreaks over the last 30 years. The course of the relative research interest in this figure may be explained as follows. The relative research interest increased with increasing resistance but decreased briefly after antibiotics that overcome the respective type of resistance became available (e.g. after the introduction of cephalosporins, -lactamase inhibitors and carbapenems in the 1980s). The relative research interest rose again with the appearance of ESBLs and carbapenemases. Furthermore, the increasing amount of publications related to hospital outbreaks caused by -lactam-resistant micro-organisms is alarming (Fig. 3B). Despite the number of included entries differing between the Outbreak Database (http://www.outbreakdatabase.com) and the GoPubMed search, the increasing trend of -lactamase-related outbreaks is similar until year 2007 between both sources. The difference observed after 2007 may be explained by difficulties in long-term maintenance of such manually annotated databases that depend on the engagement of contributing members and available funding. 3.3. Correlations between antibiotic market introduction and respective resistance development Current surveillance systems were implemented within the last decades and thus do not cover an appropriate time range to image the relationship between market release and resistance development for most antibiotics in clinical use. We performed a
scientometric analysis including 37 469 publications ranging from 1940–2012 for penicillin (9762), tetracycline (2833), erythromycin (1674), meticillin (17 016), gentamicin (1173), vancomycin (1977), carbapenems (952), cephalosporins (1725), levofloxacin (143) and linezolid (214) to describe the relationship between market releases and respective resistance in the main pathogens treated with these antibiotics. The current report of the US Centers for Disease Control and Prevention (CDC) (Antibiotic Resistance Threats in the United States, 2013; http://www.cdc.gov/drugresistance/ threat-report-2013/pdf/ar-threats-2013-508.pdf) was used to identify resistant pathogens with the highest public health relevance. The relative research interest in resistance regardless of the species (e.g. ‘penicillin resistance’, ‘tetracycline resistance’, etc.) was included as a reference (black area in Fig. 4). Since penicillin exhibits the most extensive timeframe, it provides some interesting insights. Shortly after market release, there is a slight ‘publication burst’ in publications related to penicillin resistance. After a ‘lag phase’ of minor relative research interest of ca. 20 years, a sudden increase in interest in penicillin resistance occurred that is followed by a slight but steady negative slope. Similar courses of relative research interest with a ‘lag phase’ of minor interest and a marked increase 20 years after market release that is afterwards slightly decreasing can also be observed for tetracycline, erythromycin and meticillin, and with a reduced ‘lag phase’ of around 10 years for gentamicin. Noteworthy, a decreasing relative research interest is not yet observed for cephalosporin and carbapenem resistance. In contrast to the antibiotics mentioned before, cephalosporins and carbapenems are the mainstay of
428
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
Fig. 4. Relative research interest based on the resistances against key antibiotics, including the most relevant pathogens. The respective market releases are included.
empirical treatment in critically ill patients and belong to the most frequently used antibiotics in hospitals. According to spot-checking of 40 titles and available abstracts, the ‘publication burst’ for penicillin and erythromycin around the time of market release reflects the scientific interest in a novel
antibiotic, including the description of individually selected laboratory or clinically derived resistant isolates. The lag phase of minor publication interest may be explained by the successful clinical use of the respective antibiotic, whereas the high increase after the ‘lag phase’ reflects the broad spread of clinically relevant resistant
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
strains and related treatment failure (proven by spot-checking of 40 titles and available abstracts per strain and species, or all if less available). The decrease in relative research interest regarding resistance probably correlates with the availability of novel antibiotics that overcome the respective type of resistance (e.g. cephalosporins for the treatment of penicillin-resistant S. aureus) or the realisation that the clinical impact was overestimated, e.g. ‘low-level’ penicillin resistance in pneumococcal pneumonia [14], that could also be mapped by the relative research interest in penicillin resistance related to Streptococcus pneumoniae with a sudden increase around 1999 and a continuous decrease briefly afterwards. Both resistance to carbapenems and cephalosporins share the lag phase that was also observed for other resistances, but the relative research interest continues to increase, reflecting the current public health threat caused by these types of resistance [15].
3.4. Factors influencing research interest and limitations of scientometric resistance data extraction As outlined above, research interest is not only driven by resistance rates but also by the clinical relevance of the respective resistance. For example, resistance to tetracyclines is frequent in Europe but does not impose a major threat to public health – in contrast to ESBLs or carbapenemases – because tetracyclines are not used for severe bacterial infections. This minor ‘medical need’ is reflected by lower research interest in tetracycline resistance compared with -lactam resistance despite high tetracycline resistance rates (Fig. 2). Other examples for a mismatch between resistance rates and the respective clinical importance include erythromycin resistance in S. aureus and penicillin resistance in S. aureus after -lactamase inhibitors and cephalosporins have become available. Scientometric data cannot easily map resistance trends on a regional scale because there are a variety of biasing factors, such as different surveillance systems, political decisions regarding research focus and funding, or the cultural influence of each country [16,17], affecting the overall results. Therefore, we were unable to correlate the relative research interest of individual European countries for MRSA with the respective EARS-Net data (not shown). A Google-based algorithm, such as the Google search index for common flu reflecting regional flu activity [18,19], can probably not be used to mirror antibiotic resistance since the majority of Google users lack the expertise for an appropriate and analysable search for antibiotic resistance. Another limitation is that logical linguistic inter-relation is elusive and biases the accuracy of the output. The limitation of disambiguation of the semantic search is reflected by the similar curve shape of research interest in vancomycin resistance for both Enterococcus faecalis and Enterococcus faecium. Vancomycin resistance is rare in E. faecalis [20]. A spot-check of 50 abstracts revealed that E. faecalis and E. faecium are frequently mentioned together in the respective abstracts in the same sentence, limiting the discriminative capacity of the semantic algorithm used by GoPubMed.
4. Conclusions Scientists all over the world are ambitious to publish their results in PubMed-listed journals thereby opening the information to a wide audience. The initial aim was to test whether scientometric data may provide a more general and robust estimation of the overall resistance development than regionally and temporally limited surveillance studies owing to the large amount of publications covering timeframes since 1940.
429
Scientometric analysis correlates partly with resistance rates. Besides resistance development, introduction of novel antibiotics, increasing or decreasing public health significance, availability of alternative antibiotics, and available research funding have an impact on the relative research interest, limiting the overall accuracy of this approach. Regarding the point of view, this is not necessarily a disadvantage, e.g. when the aim is to estimate where the medical need for development of novel antibiotics is highest. Nevertheless, scientometric analysis of data extracted by freely available search engines (e.g. ‘GoPubMed’) provides a fast and global overview and may reflect the ‘medical need’, i.e. gap between resistance development and specific treatment options, better than the numerically more accurate analysis of surveillance studies. In conclusion, the scientometric approach using freely available search tools provides a fast and global overview of the development and the clinical and public health importance of a specific resistance. In addition, it provides an estimate on the ‘durability’ of newly introduced antibiotics, which may help to support a foreseeing policy regarding research funding for anti-infectives. The scientometric approach is, however, no substitute for wellperformed surveillance studies, which of course need sufficient funding. Funding: This work was supported by grants from the German Ministry of Education and Research (BMBF) [grant no. 01KI1204 and 01EO1002]. Competing interests: None declared. Ethical approval: Not required. Acknowledgments The authors would like to acknowledge the GoPubMed team for this platform as well as Transinsight GmbH (Dresden, Germany) for providing raw data. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.ijantimicag. 2014.08.001. References [1] Doms A, Schroeder M. GoPubMed: exploring PubMed with the Gene Ontology. Nucleic Acids Res 2005;33:W783–6. [2] Rogers FB. Medical subject headings. Bull Med Lib Assoc 1963;51:114–16. [3] Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, Foulger R, et al. The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res 2004;32:D258–61. [4] Lewis K. Platforms for antibiotic discovery. Nat Rev Drug Discovery 2013;12:371–87. [5] Powers JH. Antimicrobial drug development—the past, the present, and the future. Clin Microbiol Infect 2004;10:23–31. [6] Gentamicin. Br Med J 1967;1:158–9. [7] European Centre for Disease Prevention and Control. Antimicrobial resistance surveillance in Europe 2010. In: Annual report of the European Antimicrobial Resistance Surveillance Network (EARS-Net). ECDC; 2011. [8] da Silva RM, Traebert J, Galato D. Klebsiella pneumoniae carbapenemase (KPC)producing Klebsiella pneumoniae: a review of epidemiological and clinical aspects. Expert Opin Biol Ther 2012;12:663–71. [9] Lee GC, Burgess DS. Treatment of Klebsiella pneumoniae carbapenemase (KPC) infections: a review of published case series and case reports. Ann Clin Microbiol Antimicrob 2012;11:32. [10] Kaul DR, Collins CD, Hyzy RC. New developments in antimicrobial use in sepsis. Curr Pharm Des 2008;14:1912–20. [11] Master RN, Deane J, Opiela C, Sahm DF. Recent trends in resistance to cell envelope-active antibacterial agents among key bacterial pathogens. Ann NY Acad Sci 2013;1277:1–7. [12] Humphreys H. Do guidelines for the prevention and control of methicillinresistant Staphylococcus aureus make a difference. Clin Microbiol Infect 2009;15:39–43. [13] Humphreys H. National guidelines for the control and prevention of methicillin-resistant Staphylococcus aureus—what do they tell us? Clin Microbiol Infect 2007;13:846–53.
430
C. Brandt et al. / International Journal of Antimicrobial Agents 44 (2014) 424–430
[14] Kanungo R, Rajalakshmi B. Serotype distribution & antimicrobial resistance in Streptococcus pneumoniae causing invasive & other infections in south India. Indian J Med Res 2001;114:127–32. [15] Laxminarayan R, Duse A, Wattal C, Zaidi AK, Wertheim HF, Sumpradit N, et al. Antibiotic resistance—the need for global solutions. Lancet Infect Dis 2013;13:1057–98. [16] Borg MA, Camilleri L, Waisfisz B. Understanding the epidemiology of MRSA in Europe: do we need to think outside the box? J Hosp Infect 2012;81: 251–6. [17] Hofstede GH. Culture’s consequences: comparing values, behaviors, institutions, and organizations across nations. 2nd ed. Thousand Oaks, CA: Sage Publications; 2001.
[18] Olson DR, Konty KJ, Paladini M, Viboud C, Simonsen L. Reassessing Google Flu Trends data for detection of seasonal and pandemic influenza: a comparative epidemiological study at three geographic scales. PLoS Comput Biol 2013;9:e1003256. [19] de Lange MM, Meijer A, Friesema IH, Donker GA, Koppeschaar CE, Hooiveld M, et al. Comparison of five influenza surveillance systems during the 2009 pandemic and their association with media attention. BMC Public Health 2013;13:881. [20] Cattoir V, Leclercq R. Twenty-five years of shared life with vancomycinresistant enterococci: is it time to divorce? J Antimicrob Chemother 2013;68:731–42.
