Articles
Global burden of bacterial antimicrobial resistance in 2019:
a systematic analysis
Antimicrobial Resistance Collaborators*
Summary Lancet 2022; 399: 629–55
Background Antimicrobial resistance (AMR) poses a major threat to human health around the world. Previous Published Online
publications have estimated the effect of AMR on incidence, deaths, hospital length of stay, and health-care costs for January 20, 2022
specific pathogen–drug combinations in select locations. To our knowledge, this study presents the most https://doi.org/10.1016/
comprehensive estimates of AMR burden to date. S0140-6736(21)02724-0
Methods We estimated deaths and disability-adjusted life-years (DALYs) attributable to and associated with bacterial This online publication has been
AMR for 23 pathogens and 88 pathogen–drug combinations in 204 countries and territories in 2019. We obtained corrected. The corrected version
data from systematic literature reviews, hospital systems, surveillance systems, and other sources, covering first appeared at thelancet.com
471 million individual records or isolates and 7585 study-location-years. We used predictive statistical modelling to on September 29, 2022
produce estimates of AMR burden for all locations, including for locations with no data. Our approach can be
divided into five broad components: number of deaths where infection played a role, proportion of infectious deaths See Comment page 606
attributable to a given infectious syndrome, proportion of infectious syndrome deaths attributable to a given
pathogen, the percentage of a given pathogen resistant to an antibiotic of interest, and the excess risk of death or *Collaborators are listed at the
duration of an infection associated with this resistance. Using these components, we estimated disease burden end of the paper
based on two counterfactuals: deaths attributable to AMR (based on an alternative scenario in which all drug-
resistant infections were replaced by drug-susceptible infections), and deaths associated with AMR (based on an Correspondence to:
alternative scenario in which all drug-resistant infections were replaced by no infection). We generated Dr Mohsen Naghavi, Institute for
95% uncertainty intervals (UIs) for final estimates as the 25th and 975th ordered values across 1000 posterior draws, Health Metrics and Evaluation,
and models were cross-validated for out-of-sample predictive validity. We present final estimates aggregated to the University of Washington,
global and regional level. Seattle, WA 98195, USA
[email protected]
Findings On the basis of our predictive statistical models, there were an estimated 4·95 million (3·62–6·57) deaths
associated with bacterial AMR in 2019, including 1·27 million (95% UI 0·911–1·71) deaths attributable to bacterial
AMR. At the regional level, we estimated the all-age death rate attributable to resistance to be highest in western sub-
Saharan Africa, at 27·3 deaths per 100 000 (20·9–35·3), and lowest in Australasia, at 6·5 deaths (4·3–9·4) per 100 000.
Lower respiratory infections accounted for more than 1·5 million deaths associated with resistance in 2019, making
it the most burdensome infectious syndrome. The six leading pathogens for deaths associated with resistance
(Escherichia coli, followed by Staphylococcus aureus, Klebsiella pneumoniae, Streptococcus pneumoniae, Acinetobacter
baumannii, and Pseudomonas aeruginosa) were responsible for 929 000 (660 000–1 270 000) deaths attributable to AMR
and 3·57 million (2·62–4·78) deaths associated with AMR in 2019. One pathogen–drug combination, meticillin-
resistant S aureus, caused more than 100 000 deaths attributable to AMR in 2019, while six more each caused
50 000–100 000 deaths: multidrug-resistant excluding extensively drug-resistant tuberculosis, third-generation
cephalosporin-resistant E coli, carbapenem-resistant A baumannii, fluoroquinolone-resistant E coli, carbapenem-
resistant K pneumoniae, and third-generation cephalosporin-resistant K pneumoniae.
Interpretation To our knowledge, this study provides the first comprehensive assessment of the global burden of
AMR, as well as an evaluation of the availability of data. AMR is a leading cause of death around the world, with the
highest burdens in low-resource settings. Understanding the burden of AMR and the leading pathogen–drug
combinations contributing to it is crucial to making informed and location-specific policy decisions, particularly
about infection prevention and control programmes, access to essential antibiotics, and research and development of
new vaccines and antibiotics. There are serious data gaps in many low-income settings, emphasising the need to
expand microbiology laboratory capacity and data collection systems to improve our understanding of this important
human health threat.
Funding Bill & Melinda Gates Foundation, Wellcome Trust, and Department of Health and Social Care using UK aid
funding managed by the Fleming Fund.
Copyright © 2022 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY
4.0 license.
www.thelancet.com Vol 399 February 12, 2022 629
Articles
Research in context disease incidence, prevalence, and mortality from the Global
Burden of Diseases, Injuries, and Risk Factors Study 2019, our
Evidence before this study findings on the burden of bacterial AMR can be compared with
To identify previous estimates of antimicrobial resistance other causes of death, offering crucial context on the
(AMR) burden before this study, we did a systematic review and magnitude of the burden of this important health issue.
consulted with content experts. We searched the evidence Improvements to the input data and models compared with
available in PubMed for published works that evaluate exposure previous publications make our AMR estimates the most robust
to antimicrobial resistant organisms (bacteria only) and of any to date. Finally, this study is the first to quantify the
evaluated all human-focused publications with more than ten burden of AMR using two different AMR counterfactual
cases, all genders, and all age groups. From these findings, we scenarios.
extracted study type, pathogen–drug combinations,
counterfactuals, locations, methods, outcomes, and Implications of all the available evidence
population. Extensive literature exists estimating incidence, Our estimates indicate that bacterial AMR is a health problem
deaths, hospital length of stay, and health-care costs associated whose magnitude is at least as large as major diseases such as
with AMR from a small number of drug-resistant infections in HIV and malaria, and potentially much larger. Bacterial AMR is a
select locations. There is widespread agreement that AMR poses problem in all regions; we estimated that, in 2019, the highest
a serious potential threat to human health around the world. rates of AMR burden were in sub-Saharan Africa. Six pathogens
The Review on Antimicrobial Resistance, published in 2016, accounted for 73·4% (95% uncertainty interval 66·9–78·8) of
estimated that as many as 10 million people could die annually deaths attributable to bacterial AMR. Seven pathogen–drug
from AMR by 2050. Recent estimates of the burden of drug- combinations each caused more than 50 000 deaths,
resistant infections covering several pathogens have also been highlighting the importance of developing policies that
published for the USA, Thailand, the EU and European specifically target the deadliest pathogen–drug combinations,
Economic Area, and several other locations, as well as estimates particularly through expansion of infection prevention and
for several pathogen–drug combinations for a wider range of control programmes, improving access to essential second-line
locations. To our knowledge, however, there have been no antibiotics where needed, and through vaccine and antibiotic
comprehensive estimates covering all locations and a broad development. Additionally, our comprehensive data collection
range of pathogens and pathogen–drug combinations. effort shows that high-quality data on infectious disease,
pathogens, and AMR are only sparsely available in many
Added value of this study low-income settings. Both preventing bacterial AMR and
This study is the most comprehensive analysis of the burden of increasing microbiological laboratory and data collection
AMR to date, producing estimates for 204 countries and capacity to improve scientific understanding of this health
territories, 23 bacterial pathogens, and 88 pathogen–drug threat should be a very high priority for global health policy
combinations, in 2019. This study uses major methodological makers.
innovations to provide important new insights into the AMR
burden. Additionally, since this analysis builds on estimates of
Introduction where surveillance is minimal and data are sparse.
Extensive literature exists estimating the effects of AMR
Bacterial antimicrobial resistance (AMR)—which occurs on incidence, deaths, hospital length of stay, and health-
when changes in bacteria cause the drugs used to treat care costs for select pathogen–drug combinations in
infections to become less effective—has emerged as one specific locations,1,2,6,9–12 but, to our knowledge, no
of the leading public health threats of the 21st century. comprehensive estimates covering all locations and
The Review on Antimicrobial Resistance, commissioned a broad range of pathogens and pathogen–drug
by the UK Government, argued that AMR could kill combinations have ever been published. For instance, the
10 million people per year by 2050.1,2 Although these US Centers for Disease Control and Prevention (CDC)
forecasts have been criticised by some,3,4 WHO and published a 2019 report on AMR infections and deaths in
numerous other groups and researchers agree that the the USA for 18 AMR threats using surveillance data,6
spread of AMR is an urgent issue requiring a global, while Cassini and colleagues10 estimated the burden of
coordinated action plan to address.5–8 Information about eight bacterial pathogens and 16 pathogen–drug
the current magnitude of the burden of bacterial AMR, combinations in the EU and European Economic Area
trends in different parts of the world, and the leading for 2007–15. Likewise, Lim and colleagues estimated the
pathogen–drug combinations contributing to bacterial burden of multidrug resistance in six bacterial pathogens
AMR burden is crucial. If left unchecked, the spread of in Thailand in 2010,11 and Temkin and colleagues
AMR could make many bacterial pathogens much more estimated the incidence of Escherichia coli and Klebsiella
lethal in the future than they are today. pneumoniae resistant to third-generation cephalosporins
and carbapenems in 193 countries in 2014.12
One major challenge to tackling AMR is understanding
the true burden of resistance, particularly in locations
630 www.thelancet.com Vol 399 February 12, 2022
Articles
Although these publications are important contribu For the first counterfactual scenario—where all drug-
tions to the body of work on AMR, they are insufficient to resistant infections are replaced by susceptible infections—
understand the global burden of AMR and identify and we estimated only deaths and DALYs directly attributable
target the highest priority pathogens in different to resistance. For the second counterfactual scenario—
locations. Additionally, existing studies have generally where all drug-resistant infections are replaced by no
considered only one measure of AMR burden.13 Because infection—we estimated all deaths and DALYs associated
we do not know the extent to which drug-resistant with resistant infection. Estimates of AMR burden based
infections would be replaced by susceptible infections or on each counterfactual are useful in different ways for
by no infection in a scenario in which all drug resistance informing the development of potential intervention
was eliminated, it is important to quantify the burden on strategies to control AMR.13,17,18
the basis of both these counterfactual scenarios.
Input data See Online for appendix
In this study, we present the first global estimates of
the burden of bacterial AMR covering an extensive set of We used several data collection strategies. Through our
pathogens and pathogen–drug combinations using large collaborator networks, we obtained datasets not
consistent methods for both counterfactual scenarios. previously available for AMR research, including hospital
and laboratory data, as well as datasets published previously
Methods and those outlined in research articles.19 Each component
of the estimation process had different data requirements
Overview and, as such, the input data used for each modelling
We developed an approach for estimating the burden of component differed. The diverse data sought included the
AMR that makes use of all available data and builds on following sources: pharmaceutical companies that run
death and incidence estimates for different underlying surveillance networks, diagnostic laboratories, and clinical
conditions from the Global Burden of Diseases, Injuries, trial data; high-quality data from researchers including
and Risk Factors Study (GBD) 2019, which provides large multisite research collaborations, smaller studies,
age-specific and sex-specific estimates of disease burden clinical trials, and well established research institutes
for 369 diseases and injuries in 204 countries and based in low-income and middle-income countries
territories in 1990–2019.14 Our approach can be divided (LMICs); data from public and private hospitals and public
into ten estimation steps that occur within five broad health institutes providing diagnostic testing; global
modelling components (a flowchart of the estimation surveillance networks; enhanced surveillance systems;
steps is given in the appendix p 123). First, we obtained national surveillance systems; and surveillance systems for
data from multiple data sources, including from specific organisms such as Mycobacterium tuberculosis and
published studies (eg, microbiology data, inpatient data, Neisseria gonorrhoeae (all sources are listed by data type in
data on multiple causes of death, and pharmaceutical the appendix pp 8–15).
sales data) and directly from collaborators on the Global
Research on Antimicrobial Resistance project,15 members Figure 1 shows a summary of the distinct data types
of the GBD Collaborator Network, and other data gathered and for which estimation step each data type
providers. was used. Also shown in figure 1 is the number of unique
study-location-years and individual records or isolates
We estimated the disease burdens associated with and available for each data type. Location-years of data refer
attributable to AMR for 12 major infectious syndromes to unique GBD locations and years for which we have
(lower respiratory infections and all related infections in records or isolates. In total, 471 million individual records
the thorax; bloodstream infections; peritoneal and intra- or isolates covering 7585 study-location-years were used
abdominal infections; meningitis and other bacterial as input data to the estimation process. Table 1 shows the
CNS infections; typhoid, paratyphoid, and invasive non- number of individual records or isolates used and
typhoidal Salmonella spp; urinary tract infections and number of countries covered in each of the five broad
pyelonephritis; diarrhoea; tuberculosis [not including modelling components separately by GBD region. Two of
tuberculosis associated with HIV]; bacterial infections of five components included data from every GBD region
the skin and subcutaneous systems; endocarditis and and two of five included data from 19 of 21 GBD regions.
other cardiac infections; infections of bones, joints, Our models of sepsis and infectious syndrome were the
and related organs; and gonorrhoea and chlamydia) and most geographically sparse, covering 16 countries from
one residual category, 23 bacterial pathogens, 18 drug ten regions; the input data for these models were highly
categories or combinations of drugs for which there detailed microdata that are only sparsely available.
is resistance, and 88 pathogen–drug combinations However, our framework for estimating the total
(appendix pp 45–46). We modelled all-age and age-specific envelope of infectious syndrome mortality used GBD
deaths and disability-adjusted life-years (DALYs) for cause-specific mortality estimates to minimise reliance
204 countries and territories, and we present aggregated on these sparse data.
estimates for 21 GBD regions, seven GBD super-regions,
and globally in 2019 (a full list of GBD locations by region All data inputs for the models were empirical data, not
is available in the appendix pp 100–05).16 modelled estimates, except for a custom meta-analysis of
www.thelancet.com Vol 399 February 12, 2022 631
Articles
Source type Number Sample size Sample Estimation step
of size units
study- 1: 2: 3: 4: 5: 6: 7: 8: 9:
location- sepsis infectious case- pathogen antibiotic prevalence resistance relative relative
years syndrome fatality distribution use of profiles risk of length
ratio resistance death of stay
Multiple cause 2980 120871372 Deaths
of death
Hospital 391 192533415 Discharges
discharge 1102
Microbial or 2302 3060802 Isolates
laboratory data 607
with outcome 145067113 Isolates
Microbial or 158
laboratory data 1536 701 356 Cases,
without 203 8 648 390 isolates, or
outcome pathogen–
7 1536 drug
Literature susceptibility
studies 38 tests
9324 Pathogen–
Single drug drug
resistance susceptibility
profiles tests
Study-
Pharmaceutical country-
sales years
Antibiotic use
among children 151 455 Households
younger than surveyed
5 years with
reported illness 870 Deaths
Mortality
surveillance 264010 Deaths
(minimally 471 300 319
invasive tissue
sampling from
Child Health
and Mortality
Prevention
Surveillance)
Linkage
(mortality only)
Grand total
Figure 1: Data inputs by source type
Total sample size for each source type, regardless of specific inclusion criteria for a given estimation step. Individual isolates that were tested multiple times for
resistance to different antibiotics are listed only once here whenever isolates were identified uniquely in the data. For datasets where isolates could not be uniquely
identified across pathogen–drug combinations, such as some antimicrobial resistance surveillance systems, some isolates might be double counted. Yellow boxes
indicate that the source type was used in that estimation step. A full list of data sources included in this study, organised by data type, is included in the appendix
(pp 8–15).
For the data input citations see vaccine probe data that we did to estimate the fraction appendix (pp 17–18, 31, 34–35, 44, 54). Data input citations
http://ghdx.healthdata.org/ of pneumonia caused by Streptococcus pneumoniae are available online.
record/ihme-data/global- (appendix pp 37–38). All study-level covariates for
bacterial-antimicrobial- models, such as age and sex, were extracted from Estimation steps one and two: deaths in which
resistance-burden- empirical data. All country-level covariates were modelled infection played a role by infectious syndrome
estimates-2019 estimates that were produced previously for GBD 2019,20,21 First, to define the number of deaths where infection
or those that were modelled by Browne and colleagues.22 plays a role, we used GBD 2019 cause of death estimates14
We describe data inputs for each of ten estimation steps to determine the number of deaths by age, sex, and
in greater detail in the following subsections and in the location for which either the underlying cause of death
632 www.thelancet.com Vol 399 February 12, 2022
Articles
Component Fraction of Component Fraction of Component Fraction of Component Fraction of Component Fraction of
1: sepsis and countries 2: case- countries 3: pathogen countries 4: fraction of countries 5: relative countries
infectious represented in fatality ratio represented in distribution represented in resistance† represented in risk represented in
syndrome component 1 component 2 component 3 component 5
component 4
models*
Andean Latin America 0 0/3 1784 2/3 12 010 2/3 538 644 3/3 4338 2/3
Australasia 320 909 1/2 94 818 1/2 6 294 677 2/2 4 653 832 2/2 5211 2/2
Caribbean 0/19 5/19 5/19 10/19 529 1/19
Central Asia 0 0/9 2858 2/9 6225 1/9 68 078 9/9 6065 1/9
Central Europe 0 0/13 43 852 10/13 2785 11/13 304 341 13/13 397 885 10/13
Central Latin America 0 2/9 371 112 9/9 627 844 8/9 3 148 864 9/9 20 210 5/9
Central sub-Saharan Africa 8 130 066 0/6 3 932 601 0/6 11 641 626 2/6 829 686 6/6 0/6
East Asia 0 1/3 2/3 2/3 40 243 3/3 0 2/3
Eastern Europe 1 189 309 0/7 0 4/7 770 5/7 2 501 536 7/7 185 980 4/7
Eastern sub-Saharan 0 3/15 385 443 4/15 257 522 9/15 968 565 14/15 102 904 2/15
Africa 292 118 754 64 212 474 280
High-income Asia Pacific 0/4 68 791 3436
High-income North 0 2/3 6388
America 84 520 574
North Africa and Middle 0/21 135 907 3/4 99 042 3/4 18 909 332 4/4 7577 3/4
East 0 7 184 424 3/3 7 255 147 2/3 32 205 001 3/3 14 071 025 2/3
Oceania 0/18
South Asia 0 1/5 209 479 13/21 53 833 16/21 531 120 21/21 90 079 10/21
Southeast Asia 54 0/13
Southern Latin America 0 0/3 0 0/18 20 1/18 4297 12/18 0 0/18
Southern sub-Saharan 0 1/6 77 811 4/5 51 810 4/5 1 413 840 5/5 97 131 4/5
Africa 4 696 789 195 087 9/13 91 259 8/13 3 128 014 12/13 172 947 8/13
Tropical Latin America 1/2 200 665 3/3 73 512 2/3 3/3 5000 1/3
Western Europe 17 224 511 2/24 80 717 2/6 4 699 304 2/6 740 385 6/6 1/6
Western sub-Saharan 10 599 906 2/19 910 509 1051
Africa
83 3 988 611 1/2 20 956 932 2/2 286 450 2/2 6443 1/2
94 506 554 20/24 105 183 184 21/24 18 909 732 21/24 932 016 21/24
9/19 10/19 18/19 14 880 2/19
26 985 21 896 369 482
Total sample size and fraction of countries covered for each modelling component by GBD region. The units for sample size are deaths for sepsis and infectious syndrome models; cases for case-fatality ratios; cases,
deaths, or isolates for pathogen distribution; pathogen–drug tests for fraction of resistance; and pathogen–drug tests for relative risk. Sample sizes reflect model-specific selection criteria, resulting in lower totals for
the sepsis, infectious syndrome, case-fatality ratio, and pathogen distribution models in this table than those in figure 1. Totals for fraction of resistance and relative risk are higher in this table than in figure 1 because
of the difference in units for certain source types, such as microbial data (isolates in figure 1, pathogen–drug tests here). Several data sources inform multiple components; therefore, data points should not be
summed across a row as that will lead to duplication. More information on the data types used and the components that they inform is presented in the appendix (pp 8–15). GBD=Global Burden of Diseases, Injuries,
and Risk Factors Study. *The data points listed in the sepsis and infectious syndrome models include only sources used to determine the fraction of sepsis in non-communicable diseases; maternal, neonatal, and
nutritional diseases; and injuries, as well as the distribution of infectious syndromes; final estimates of the number of deaths in each infectious syndrome were generated by multiplying the fractions of sepsis and
infection syndromes on GBD 2019 death estimates; GBD 2019 death estimates include 7417 sources with 28 106 location-years of data for under-5 mortality and 7355 sources with over 7322 location-years of data.
†For sources in the fraction of resistance modelling component, de-duplication across antibiotic resistance tests was not possible, leading to potential double counting, as seen in the high-income Asia Pacific region.
Table 1: Data included in each modelling component by region and the fraction of countries represented in each region
was infectious or—for non-communicable, maternal, logistic regression models to predict the fraction of
neonatal, nutritional, and injury deaths—for which the sepsis occurring in each communicable, maternal,
pathway to death was through sepsis. Sepsis is defined as neonatal, and nutritional underlying cause of death;
a life-threatening organ dysfunction due to a dysregulated non-communicable underlying cause of death; and
host response to infection.23 The methods used to injury underlying cause of death. This approach follows
estimate infectious underlying causes of death and sepsis the methods validated by many researchers in sepsis
deaths have been published previously14,24 and are epidemiology25–28 and used by Rudd and colleagues.24
summarised in the appendix (pp 17–18).
We then multiplied the fraction of sepsis predicted
In estimation step one, we used data for multiple from the logistic regression models onto GBD cause-
causes of death covering 121 million deaths, 5·54 million specific mortality estimates to determine the mortality
hospital discharges with discharge status of death, and envelope for our analysis. Our mortality envelope
264 000 records of multiple causes of death linked to consisted of all deaths in which infection played a
hospital records from ten countries and territories, as role, which included all sepsis deaths with non-
well as 870 deaths from Child Health and Mortality infectious underlying causes, plus all deaths with an
Prevention Surveillance (CHAMPS) sites across six infectious underlying cause in GBD 2019 (appendix
countries (appendix pp 17–18), to develop random effects pp 21–23).
www.thelancet.com Vol 399 February 12, 2022 633
Articles
In estimation step two, we used details on the Estimation steps three and four: pathogen distribution
pathways of disease provided in multiple causes of for deaths and incident cases
death and hospital discharge data in a second stage of To estimate the pathogen distribution of each infectious
random effects logistic regression models to further syndrome separately for deaths and incident cases for
subdivide deaths in which infection played a role into each age, sex, and location, we made use of multiple data
12 major infectious syndromes and one residual sources. For estimation step three, we took data that
category. These regressions predicted the proportion of linked pathogen-specific disease incidence to deaths to
sepsis-related deaths that were caused by a given develop models for pathogen-specific CFRs that varied
infectious syndrome separately for each communicable, by age, location, and syndrome. We used the Bayesian
maternal, neonatal, and nutritional underlying cause of meta-regression tool MR-BRT29 to estimate CFRs as a
death; non-communicable underlying cause of death; function of the Healthcare Access and Quality Index and
and injury underlying cause of death. We used this various bias covariates (appendix pp 31–34).21 These
fraction to subdivide sepsis deaths with non-infectious CFRs allowed us to integrate sources that reported
underlying causes into specific infectious syndromes. pathogen distribution only for deaths and those that
For underlying causes of death that are themselves reported only incidence by mapping the reported deaths
infectious, all deaths were assigned to their single by pathogen into implied cases by pathogen. After
corresponding infectious syndrome (eg, the GBD cause mapping, we had 157 million isolates and cases from
“lower respiratory infections” was assigned to the 118 countries and territories to estimate the pathogen
infectious syndrome “lower respiratory infections and distribution of each infectious syndrome (estimation
all related infections in the thorax”; appendix pp 21–23). step four), with each dataset including a unique
spectrum of pathogens and groups of pathogens. To
Due to the pathogen distributions varying substantially incorporate all these heterogeneous data, we used a new
for hospital-acquired and community-acquired infections modelling environment, termed multinomial estimation
in two infectious syndromes—lower respiratory and with partial and composite observations. This modelling
thorax infections and urinary tract infections—we environment allows for the inclusion of covariates in the
further estimated the subdivision of these syndromes network analysis29 and for Bayesian prior probability
into community-acquired and hospital-acquired distributions to be incorporated. To model the infectious
infections (appendix pp 17–30; table with community- syndrome pathogen distribution comprehensively, we
acquired and hospital-acquired subdivisions presented estimated, where applicable, the incidence and death
on pp 24–25). proportions attributable to viral, fungal, parasitic, and
bacterial pathogens; however, AMR burden was
Incidence of infectious syndromes disaggregated by calculated only for selected bacteria for which resistance
age, sex, and location is clinically relevant and sufficient data are available.
For the nine infectious syndromes in this study that were More details on this approach are provided in the
estimated as one or more causes of death and disability appendix (pp 34–44).
in GBD 2019 (lower respiratory and thorax infections;
CNS infections; typhoid, paratyphoid, and invasive non- Estimation steps five to seven: prevalence of resistance
typhoidal Salmonella spp; urinary tract infections; by pathogen
diarrhoea; tuberculosis; bacterial skin infections; cardiac We used data from 52·8 million isolates to analyse the
infections; and gonorrhoea and chlamydia), we used proportion of phenotypic AMR for each pathogen—the
GBD 2019 incidence estimates as a baseline for infectious proportion of infections that were drug resistant, hereafter
syndrome incidence (appendix p 16).14 To this baseline, referred to as prevalence of resistance—for 88 pathogen–
we added the number of incident cases of each infectious drug combinations. We chose these 88 combinations by
syndrome that co-occurred with underlying non- first creating an exhaustive list of all clinically relevant
communicable diseases (NCDs); maternal, neonatal, and combinations for which we had any data and then
nutritional diseases (MNNDs); and injuries, which we eliminating combinations that did not meet minimum
calculated by dividing the number of infectious syndrome data availability and computational feasibility requirements
deaths that occurred with underlying NCDs, MNNDs, for accurate statistical modelling (appendix pp 59–60).
and injuries (by age, sex, location, and GBD cause) by
syndrome-specific and pathogen-specific case-fatality For the pathogen–drug combinations in the 2014 WHO
ratios (CFRs; estimation described in the following AMR global report on surveillance,30 as well as
subsection). Bloodstream infections, bone and joint fluoroquinolone and multidrug resistance in Salmonella
infections, and intra-abdominal infections are not enterica serotypes Typhi and Paratyphi, we supplemented
estimated in GBD, so for these infectious syndromes, we microbial datasets from collaborators and surveillance
exclusively used the number of incident cases of each networks with aggregate microbiology data from sys
infectious syndrome that co-occurred with underlying tematic reviews and published surveillance reports. The
NCDs, MNNDs, and injuries to estimate incidence number of positive isolates identified for each pathogen–
(appendix pp 56–60). drug combination is shown in the appendix (pp 90–91).
634 www.thelancet.com Vol 399 February 12, 2022
Articles
Clinical and Laboratory Standard Institute (CLSI) binomial distributions that define the prevalence of
guidelines were used to define minimum inhibitory resistance of every combination of resistance to the
concentration breakpoints when these minimums were antibiotics analysed. Every such distribution was
provided. When only a phenotypic disk interpretation was characterised by a contingency table specifying
available, we used the interpretation as provided. We probabilities of all combinations of resistance and
used two categories of susceptibility: susceptible and susceptibility among the antibiotics analysed. The
non-susceptible. The non-susceptible group includes observed prevalence of each drug overall and Pearson
isolates reported as “non-susceptible”, “intermediate”, and correlations between drugs provided noisy partial
“resistant”. To account for bias in resistance data provided observations of combinations of these entries. We
by tertiary care facilities, we adjusted tertiary rates of optimised over the space of such contingency tables to
resistance by crosswalking them to data from non-tertiary find the nearest feasible distribution given the data,
and mixed facilities using MR-BRT as described in the producing, for each pathogen, a set of resistance
appendix (pp 45–48).31 profiles: the proportions of bacteria with each com
bination of resistance and susceptibility among all the
We used a two-stage spatiotemporal modelling antibiotics analysed (appendix pp 48–49).
framework to estimate the prevalence of resistance in
each pathogen–drug combination by location for 2018. Estimation steps eight and nine: relative risk of death
Given the many challenges to data collection and for drug-resistant infection compared with drug-
reporting caused by the COVID-19 pandemic,32,33 as well sensitive infections
as our collaborators’ process of data collation and Using data from 164 sources representing 511 870 patients
cleaning, we were unable to collect more contemporary with known outcome and resistance information, we
data; we assumed no change in prevalence of resistance estimated the relative risk of death for each pathogen–
for 2019. First, we fitted a stacked ensemble model drug combination for a resistant infection compared
between the input data and selected covariates from the with that of a drug-sensitive infection using MR-BRT.
list of plausible and health-related covariates available in Because of data sparsity, we assumed the relative risk
GBD 2019 (appendix pp 48–49, 92–93); the estimates was the same for every syndrome, location, and age
from the stacked ensemble model were then inputted group; the assumptions on location and age group risk
into a spatiotemporal Gaussian process regression are consistent with those in the estimation process
model31 to smooth the estimates in space and time. The previously used by Cassini and colleagues.10 We used a
exceptions to this modelling approach were multidrug- two-stage nested mixed effects meta-regression model to
resistant (MDR) excluding extensively drug-resistant estimate relative risk of death for each pathogen–drug
(XDR) tuberculosis and XDR tuberculosis, for which combination that was adjusted for age, admission
published GBD 2019 estimates were already available.14 diagnosis, hospital-acquired versus community-acquired
infection, and site of infection (appendix pp 54–56). For
Given the strong relationship between antibiotic the non-fatal excess risk, we estimated the relative
consumption levels and the proliferation of resistance, increase in length of stay associated with a resistant
we modelled antibiotic consumption at the national infection compared with that of a drug-sensitive
level to use as a covariate in the stacked ensemble infection, adjusted for length of stay prior to culture
model of prevalence of resistance. We analysed data being drawn. Data on length of stay were available from
from 65 Demographic and Health Surveys and 59 sources representing 455 906 admissions. We used
138 Multiple Indicator Cluster Surveys using model- the same modelling framework for excess length of stay
based geostatistics to quantify antibiotic usage in as we used for relative risk of death. Due to data sparsity
LMICs. These LMIC-specific estimates of antibiotic on the excess risk of death associated with drug-resistant
usage were combined with pharmaceutical sales data N gonorrhoeae, we did not produce a fatal estimate for
from IQVIA, WHO, and the European Centre for this pathogen.
Disease Prevention and Control (ECDC) by use of an
ensemble spatiotemporal Gaussian process regression To produce burden estimates of multiple pathogen–
model to produce a location-year covariate on antibiotic drug combinations that were mutually exclusive within
consumption for all 204 countries and territories a given pathogen (and thus could be added), we
included in this study.22 Additional details on our produced a population-attributable fraction (PAF) for
estimation method for prevalence of resistance are each resistance profile with resistance to at least one
available in the appendix (pp 44–53). drug (appendix pp 56–60). The PAF represents the
proportional reduction in deaths or years lived with
To account for multidrug resistance, we used line- disability (YLDs) that would occur if all infections
level microbiology data that tested multiple antibiotics with the resistance profile of interest were instead
for the same isolate to produce Pearson correlation susceptible to all antibiotics included in the analysis.
coefficients of the co-occurrence of resistance to When two or more antibiotics were resistant in a single
different antibiotics. With these Pearson correlations profile, we used the relative risk for the antibiotic class
and our prevalence of resistance estimates, we used an
optimisation-based approach to solve for multivariate
www.thelancet.com Vol 399 February 12, 2022 635
Articles
that was the largest as the relative risk for calculating infection and the burden associated with bacterial AMR
the PAF: based on the counterfactual of no infection (appendix
pp 56–60). Briefly, to estimate the burden attributable to
PAF= RKd(RRKD–1) AMR, we first calculated the deaths attributable to
resistance by taking the product of deaths for each
1+Σdn=1 RKd (RRKD–1) underlying cause, the proportion of these deaths in
which infection played a role, the proportion of infectious
Where R is prevalence of resistance, RR is relative risk, K deaths attributable to each infectious syndrome, the
is a pathogen with d=1, …, n resistance profiles with proportion of infectious syndrome deaths attributable to
resistance to at least one antibiotic class, and D is the each pathogen, and the mortality PAF for each resistance
antibiotic class in profile d with the highest relative risk profile. We used previously described GBD methods14 to
(appendix pp 56–60). convert age-specific deaths into years of life lost (YLLs)
using the standard counterfactual life expectancy at each
Estimation step ten: computing burden attributable to age.34 To calculate attributable YLDs, we took the product
drug resistance and burden associated with drug- of the infectious syndrome incidence, the proportion of
resistant infections infectious syndrome incident cases attributable to each
We computed two counterfactuals to estimate the drug- pathogen, YLDs per incident case, and the non-fatal PAF.
resistant burden: the burden attributable to bacterial For resistance profiles that had resistance to more than
AMR based on the counterfactual of drug-sensitive one antibiotic class, we redistributed burden to the
Associated with resistance Attributable to resistance
Deaths YLLs DALYs YLDs Deaths YLLs DALYs YLDs
Counts, thousands 4950 189 000 192 000 2290 1270 47 600 47 900 275
Global (3620–6570) (145 000–245 000) (146 000–248 000) (1520–3450) (911–1710) (35 000–63 400) (35 300–63 700) (161–439)
Central Europe, eastern 283 7530 7630 102 73·7 1980 1990 9·95
Europe, and central Asia (190–403) (5240–10 500) (5320–10 600) (69–140) (48·7–105) (1350–2790) (1360–2800) (4·79–16·8)
High income 20·2
604 10 100 10 300 123 141 2390 2410 (12·7–31·2)
Latin America and Caribbean (434–824) (6960–14 200) (7040–14 400) (79·7–183) (98·6–197) (1620–3400) (1640–3420) 16
(9·79–24·9)
North Africa and Middle East 338 9550 9640 97·2 84·3 2370 2380 20·7
(243–453) (6770–12 900) (6830–13 100) (63·2–146) (60·3–117) (1660–3310) (1680–3330) (12–33·5)
South Asia 111
256 9970 10 100 116 68·3 2590 2610 (58·5–188)
Southeast Asia, east Asia, (174–362) (6880–13 900) (6970–14 000) (73·4–176) (45·6–99) (1770–3700) (1790–3720) 45·6
and Oceania 1390 1000 389 16 000 16 100 (25–80·1)
Sub-Saharan Africa (1030–1830) 58 900 59 900 (638–1550) (273–538) (11 500–21 600) (11 600–21 700)
1020 (44 800–76 300) (45 700–77 500) 51·1
Rates, per 100 000 (678–1460) 437 254 6830 6870 (30·2–81·8)
Global 1070 27 500 27 900 (256–776) (167–369) (4620–9840) (4670–9890)
(847–1340) (18 700–38 600) (19 100–39 100) 416 15 400 15 500 3·6
Central Europe, eastern (270–599) 255 (11 700–19 900) (11 800–20 000) (2·1–5·7)
Europe, and central Asia 64·0 65 800 66 200 (196–331)
High income (46·8–84·9) (51 400–83 600) (51 800–84 000) 618·7 2·4
(455·7–823·2) (1·1–4·0)
Latin America and Caribbean 67·7 2448·1 2477·7 29·6 16·4 615·1
(45·4–96·6) (1868·9–3170·3) (1889·9–3199·1) (19·7–44·5) (11·8–22·0) (452·4–819·1) 476·7 1·9
North Africa and Middle East (325·2–671·0) (1·2–2·9)
55·7 1802·5 1826·9 24·4 17·6 474·3
South Asia (40·1–76·0) (1253·9–2515·1) (1274·5–2545·4) (16·5–33·6) (11·7–25·3) (323·0–667·3) 222·3 2·7
(151·5–315·9) (1·7–4·3)
Southeast Asia, east Asia, and 57·9 935·3 946·7 11·3 13·0 220·4
Oceania (41·6–77·6) (641·9–1310·1) (649·8–1327·2) (7·3–16·9) (9·1–18·2) (149·9–314·0) 408·1 3·4
Sub-Saharan Africa (286·9–570·0) (2·0–5·5)
42·0 1633·8 1650·5 16·6 14·4 405·3
(28·7–59·5) (1158·7–2215·9) (1169·0–2236·6) (10·8–25·0) (10·3–20·0) (284·8–566·6) 429·0 6·2
(293·7–611·5) (3·2–10·4)
76·8 1637·5 1656·6 19·1 11·2 425·6
(57·2–101·2) (1130·4–2283·2) (1145·2–2300·9) (12·1–28·9) (7·5–16·3) (291·2–608·4) 892·0 2·1
(643·1–1200·2) (1·2–3·7)
47·1 3262·6 3318·1 55·4 21·5 885·8
(31·4–67·7) (2482·4–4228·2) (2532·9–4291·7) (35·4–86·0) (15·1–29·8) (636·3–1194·6) 318·2 4·7
(216·1–458·0) (2·8–7·6)
98·9 1272·6 1292·8 20·2 11·7 316·1
(78·6–124·2) (866·8–1789·0) (884·7–1811·4) (11·8–35·9) (7·8–17·1) (213·9–455·7) 1436·7
(1090·0–1853·5)
6105·3 6143·9 38·6 23·7 1432·0
(4770·2–7749·1) (4802·8–7792·2) (25·1–55·6) (18·2–30·7) (1084·6–1848·1)
DALYs=disability-adjusted life-years. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study. YLDs=years lived with disability. YLLs=years of life lost.
Table 2: Deaths, YLLs, YLDs, and DALYs (in counts and all-age rates) associated with and attributable to bacterial antimicrobial resistance, globally and by GBD super-region, 2019
636 www.thelancet.com Vol 399 February 12, 2022
Articles
individual antibiotic classes proportionally on the basis globally in 2019 (including those directly attributable to Resistance
of excess risk, providing a mutually exclusive burden for AMR). Table 2 provides estimates of deaths, YLLs, and Associated with resistance
each pathogen–drug combination (appendix pp 56–60). DALYs from AMR for each counterfactual. Attributable to resistance
To calculate DALYs, we took the sum of YLLs and YLDs.
To estimate the overall AMR burden of the drug-sensitive Deaths (rate per 100000 population)We estimated that among the 21 GBD regions,
counterfactual, we added the burden estimates of all the Australasia had the lowest AMR burden in 2019, with
pathogen–drug combinations. 6·5 deaths per 100 000 (95% UI 4·3–9·4) attributable to
AMR and 28·0 deaths per 100 000 (18·8–39·9) associated
The approach for calculating the fatal burden with AMR in 2019 (figure 2). Western sub-Saharan Africa
associated with AMR was identical to that for fatal had the highest burden, with 27·3 deaths per
burden attributable to AMR, except we replaced the 100 000 (20·9–35·3) attributable to AMR and 114·8 deaths
mortality PAF for each resistance profile with the per 100 000 (90·4–145·3) associated with AMR. Five
prevalence of resistance in deaths. For the number of regions had all-age death rates associated with bacterial
incident infections associated with resistance, we took AMR higher than 75 per 100 000: all four regions of
the product of infectious syndrome incidence, the sub-Saharan Africa and south Asia. Although
proportion of infectious incident cases attributable to sub-Saharan Africa had the highest all-age death rate
each pathogen, and the prevalence of resistance in attributable to and associated with AMR, the percentage
incident cases. On the basis of these death and incidence of all infectious deaths attributable to AMR was lowest in
estimates, we then computed YLLs, YLDs, and DALYs this super-region (appendix p 97).
associated with drug-resistant infections. We calculated
YLLs using the same methods used to calculate YLLs Three infectious syndromes dominated the global
attributable to AMR. We converted incidence into YLDs burdens attributable to and associated with AMR in 2019:
using a YLDs per incident case ratio for each infectious lower respiratory and thorax infections, bloodstream
syndrome based on a proxy GBD cause (a simplified infections, and intra-abdominal infections (figure 3).
YLD calculation compared with the standard sequelae- Combined, these three syndromes accounted for 78·8%
based method; appendix pp 56–60). Finally, we calculated (95% UI 70·8–85·2) of deaths attributable to AMR
DALYs by summing YLLs and YLDs. To estimate the in 2019; lower respiratory infections alone accounted for
overall AMR burden of this counterfactual, we repeated more than 400 000 attributable deaths and 1·5 million
the described calculations with the prevalence of associated deaths (figure 3).
resistance to one or more antibiotics estimated and
summed across all pathogens. GBD super-region
Central Europe, eastern Europe, and central Asia
Uncertainty analysis and out-of-sample validation
Following previously described GBD methods,14 we 150 High income
propagated uncertainty from each step of the analysis into Latin America and Caribbean
the final estimates of deaths and infections attributable to North Africa and Middle East
and associated with drug resistance by taking the 25th and South Asia
975th of 1000 draws from the posterior distribution of each Southeast Asia, east Asia, and Oceania
quantity of interest. Out-of-sample validity estimates are Sub-Saharan Africa
provided in the appendix for our models of sepsis
(pp 25–30), infectious syndrome distribution (pp 25–30), 100
pathogen distribution (pp 43–44), prevalence of resistance
(pp 51–53), and relative risk (pp 55–56). 50
Role of the funding source 0
The funders of the study had no role in study design,
data collection, data analysis, data interpretation, or the GBD region
writing of the report. Figure 2: All-age rate of deaths attributable to and associated with bacterial antimicrobial resistance by GBD
region, 2019
Results Estimates were aggregated across drugs, accounting for the co-occurrence of resistance to multiple drugs. Error
bars show 95% uncertainty intervals. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study.
We estimated that, in 2019, 1·27 million deaths SNHiooEWCgruaetethssH-hnhttiSiteeegorTnAArrrrCfhucannnlnre-toioidssssnphcniteuuuueamcrcraebbbbaaao----llnnEWSCnSSSSNaeLLLLmeodaaaassoaaaaeunrtthhhhttttCttiiiiteeMaaaaASAerirrrrrrhCnnhnnsouanidaaaaEaelnnstuarAAAAAaadOrtnnnnitEEElssrcaPbmmmmmeltthuuuaAAAAealrrrbeeeefefffaEcrrrrrrrrrAAAAaeoooiiiiiiiiiiassnsssapppiiiciiccccccccifiscataaaaenaaeaaeaaaaaa
(95% uncertainty interval [UI] 0·911–1·71) were directly
attributable to resistance (ie, based on the counterfactual
scenario that drug-resistant infections were instead drug
susceptible) in the 88 pathogen–drug combinations
evaluated in this study. On the basis of a counterfactual
scenario of no infection, we estimated that 4·95 million
deaths (3·62–6·57) were associated with bacterial AMR
www.thelancet.com Vol 399 February 12, 2022 637
Articles
Resistance
Associated with resistance
Attributable to resistance
2 000 000
1 500 000
Deaths (count) 1 000 000
500 000
0 BSI Intra- UTI Tuberculosis Skin CNS TF–PF–iNTS Diarrhoea Cardiac Bone+
LRI+ abdominal
Infectious syndrome
Figure 3: Global deaths (counts) attributable to and associated with bacterial antimicrobial resistance by infectious syndrome, 2019
Estimates were aggregated across drugs, accounting for the co-occurrence of resistance to multiple drugs. Error bars show 95% uncertainty intervals. Does not
include gonorrhoea and chlamydia because we did not estimate the fatal burden of this infectious syndrome. Bone+=infections of bones, joints, and related organs.
BSI=bloodstream infections. Cardiac=endocarditis and other cardiac infections. CNS=meningitis and other bacterial CNS infections. Intra-abdominal=peritoneal and
intra-abdominal infections. LRI+=lower respiratory infections and all related infections in the thorax. Skin=bacterial infections of the skin and subcutaneous systems.
TF–PF–iNTS= typhoid fever, paratyphoid fever, and invasive non-typhoidal Salmonella spp. UTI=urinary tract infections and pyelonephritis.
In 2019, six pathogens were each responsible for more leading pathogens were distinct from those of the high-
than 250 000 deaths associated with AMR (figure 4): income super-region, and each represented a smaller
E coli, Staphylococcus aureus, K pneumoniae, S pneumoniae, share of the AMR burden; S pneumoniae contributed to
Acinetobacter baumannii, and Pseudomonas aeruginosa, by 15·9% (11·4–21·0) of the deaths attributable to AMR and
order of number of deaths. Together, these six pathogens 19·0% (17·1–21·1) of the deaths associated with AMR,
were responsible for 929 000 (95% UI 660 000–1 270 000) whereas K pneumoniae contributed to 19·9% (15·1–25·4)
of 1·27 million deaths (0·911–1·71) attributable to AMR of the deaths attributable to AMR and 17·5% (16·3–18·7)
and 3·57 million (2·62–4·78) of 4·95 million deaths of the deaths associated with AMR.
(3·62–6·57) associated with AMR globally in 2019.
Six more pathogens were each responsible for between In 2019, meticillin-resistant S aureus was the one
100 000 and 250 000 deaths associated with AMR: pathogen–drug combination in our analysis with more
M tuberculosis, Enterococcus faecium, Enterobacter spp, than 100 000 deaths and 3·5 million DALYs attributable
Streptococcus agalactiae (group B Streptococcus), S Typhi, to resistance (figure 6; appendix pp 121–22, 129). Six
and Enterococcus faecalis. For deaths attributable to AMR, more pathogen–drug combinations each caused
E coli was responsible for the most deaths in 2019, between 50 000 and 100 000 resistance-attributable
followed by K pneumoniae, S aureus, A baumannii, deaths in 2019: MDR excluding XDR tuberculosis, third-
S pneumoniae, and M tuberculosis. generation cephalosporin-resistant E coli, carbapenem-
resistant A baumannii, fluoroquinolone-resistant E coli,
The share of AMR burden caused by each of the carbapenem-resistant K pneumoniae, and third-genera
six leading pathogens differed substantially across GBD tion cephalosporin-resistant K pneumoniae (figure 6).
super-regions. In the high-income super-region, In the next tier of pathogen–drug combinations,
approximately half of the fatal AMR burden was linked to ten combinations each caused between 25 000 and
two pathogens: S aureus (constituting 26·1% [95% UI 50 000 deaths attributable to AMR. Four of these ten
17·4–34·1] of deaths attributable to AMR and 25·4% combinations included fluoroquinolone resistance,
[24·1–27·0] of deaths associated with AMR) and E coli three included carbapenem resistance, and two had
(constituting 23·4% [19·5–28·2] of deaths attributable to trimethoprim-sulfamethoxazole resistance.
AMR and 24·3% [22·9–25·8] of deaths associated with
AMR; figure 5). By contrast, in sub-Saharan Africa, the In the appendix, we present the equivalent AMR
findings for DALYs instead of deaths (pp 124–29), as well
638 www.thelancet.com Vol 399 February 12, 2022
Articles
Resistance
Associated with resistance
Attributable to resistance
Deaths (count) 900 000
600 000
300 000
0
SalmSoalnemlloaneelMnlSNtytaPorceAsreeconiHien-pGcbGSaKtnutrtrtaElaeeeEdyooceaornstcinotpmtbupueotcoeOesorhhprpiebmctraiEpryooecaoolhlisutBAnohlccuCEonieetMcdiytlassrocSSrametoeaoputtccrrlsorroppcterrsScSecPhootgceennSaiurbenchuebbyarieeuPepptarboansurliraasttpgnatuueesfuflcccarrrufeeeooettamlltuchccaammillagouimeeeioeTteuoooorsaarauccaalnnyyirncccnnossssssonecceciizoaupplclsppppppslonaaauiuuihchliissisaseimseiaeipppppp
Pathogen
Figure 4: Global deaths (counts) attributable to and associated with bacterial antimicrobial resistance by pathogen, 2019
Estimates were aggregated across drugs, accounting for the co-occurrence of resistance to multiple drugs. Error bars show 95% uncertainty intervals.
as the burden attributable to and associated with specific deaths were directly attributable to drug resistance. In
pathogen–drug combinations by age group (neonatal, other words, if all drug-resistant infections were replaced
post-neonatal, age 1–4 years, and age 5 years or older) and by no infection, 4·95 million deaths could have been
super-region (pp 106–18). prevented in 2019, whereas if all drug-resistant infections
were replaced by drug-susceptible infections, 1·27 million
Among the seven leading pathogen–drug combinations deaths could have been prevented. Compared with all
for deaths attributable to resistance, the proportion of underlying causes of death in GBD 2019, AMR would
isolates estimated to be resistant varied substantially by have been the third leading GBD Level 3 cause of death
country and territory (figure 7A–G). For meticillin- in 2019, on the basis of the counterfactual of no infection;
resistant S aureus, resistance was generally highest only ischaemic heart disease and stroke accounted for
(60% to less than 80%) in countries in north Africa and more deaths that year.14 Using the counterfactual of
the Middle East (eg, Iraq and Kuwait) and lowest (less susceptible infection, AMR would have been the 12th
than 5%) in several countries in Europe and sub-Saharan leading GBD Level 3 cause of death globally, ahead of
Africa (figure 7A). For isoniazid and rifampicin co- both HIV and malaria (more information on GBD causes
resistant (MDR excluding XDR) M tuberculosis, isolate by level presented in the appendix pp 18, 67–75).14 By any
resistance was highest (primarily 10% to less than 30%) metric, bacterial AMR is a leading global health issue.12
in eastern Europe and under 5% in many countries Additionally, our analysis showed that AMR all-age death
around the world (figure 7B). To show where data are rates were highest in some LMICs, making AMR not
available and how the modelled estimates differ from the only a major health problem globally but a particularly
input data, figure 7 also shows the raw, unadjusted serious problem for some of the poorest countries in the
prevalence of resistance for each of the seven leading world.
pathogen–drug combinations.
All six of the leading pathogens contributing to the
Discussion burden of AMR in 2019 (E coli, S aureus, K pneumoniae,
S pneumoniae, A baumannii, and P aeruginosa) have
The global burden associated with drug-resistant been identified as priority pathogens by WHO34 and
infections assessed across 88 pathogen–drug combina AMR has been highlighted in the political arena through
tions in 2019 was an estimated 4·95 million (95% UI the Global Action Plan on AMR,8 the UN Interagency
3·62–6·57) deaths, of which 1·27 million (0·911–1·71)
www.thelancet.com Vol 399 February 12, 2022 639
Articles
A
Pathogen
Acinetobacter baumannii
Pathogen-attributable fraction of AMR deaths attributable to resistance Escherichia coli
Klebsiella pneumoniae
Pseudomonas aeruginosa
0·30 Staphylococcus aureus
Streptococcus pneumoniae
0·20
0·10
0
B
Pathogen-attributable fraction of AMR deaths associated with resistance 0·20
0·10
0 High income Latin America North Africa South Asia Southeast Asia, Sub-Saharan Africa
Central Europe, and Caribbean and Middle East east Asia, and Oceania
eastern Europe, and
central Asia
Figure 5: Pathogen-attributable fraction of deaths attributable to (A) and associated with (B) bacterial AMR for the six leading pathogens by GBD super-
region, 2019
Error bars show 95% uncertainty intervals. AMR=antimicrobial resistance. GBD=Global Burden of Diseases, Injuries, and Risk Factors Study.
Coordination Group,35 the One Health Global Leaders to date, clearly show that drug resistance in each of these
Group,36 and several others. However, only one of these leading pathogens is a major global health threat that
pathogens has been the focus of a major global health warrants more attention, funding, capacity building,
intervention programme—S pneumoniae, primarily research and development, and pathogen-specific priority
through pneumococcal vaccin ation.37 Furthermore, the setting from the broader global health community.
first Sustainable Development Goal38 indicator for
antimicrobial resistance was only proposed in 2019, and Resistance to fluoroquinolones and β-lactam antibiotics
this indicator (3.d.2) is very limited in scope.39,40 Our (ie, carbapenems, cephalosporins, and penicillins)—
findings, which—to our knowledge—are the most antibiotics often considered first line for empirical
comprehensive estimates of the burden of bacterial AMR therapy of severe infections41—accounted for more than
70% of deaths attributable to AMR across pathogens.
640 www.thelancet.com Vol 399 February 12, 2022
Articles
Count (thousands)
≥100 75 to <100 50 to <75 25 to <50 10 to <25 5 to <10 <5 NA
Acinetobacter baumannii 132 000 6860 3280 10 400 13 300 811 57 700 40 000
Citrobacter spp 10 600 1840 1340 411 2170 2300 2510
Enterobacter spp 46 100 5320 3070 9950 15 300 7800 4650
Enterococcus faecalis 30 200 26 800 3420
Enterococcus faecium 51 500 37 200 14 300
Other enterococci 14 500
Escherichia coli 219 000 59 900 12 200 2220
11 700 10 500 21 300 29 500 56 000 30 200
Group A Streptococcus 3630 3630
Group B Streptococcus 25 800
Haemophilus influenzae 6760 2470 11 500 13 500 799
4290
Klebsiella pneumoniae 193 000 50 100 26 300 7930 55 700 29 000 23 500
Morganella spp 749 168 154 427
Mycobacterium tuberculosis 84 800
Proteus spp 11 500 4730 64 600 11 600 3350 5210
887 1,330 2970 1620
Pseudomonas aeruginosa 84 600 10 400 4370 3010 10 300 38 100 18 300
S Paratyphi 4110
S Typhi 23 700 4040 64
Non-typhoidal Salmonella 5620 17 200 6460
Serratia spp 10 700 1100 2610 953 5620
Shigella spp 5990 2450 1080
Staphylococcus aureus 178 000 5990
Streptococcus pneumonia 122 000 3330 2480 15 900 19 600 121 000 18 700 3120
2040 41 900 11 200 12 500 12 400 38 700
All pathogens 1 270 000 141 000 17 100 56 800 16 100 38 200 32 100 243 000 306 000 49 300 64 600 6530 121 000 11 600 3350 13 200 117 000 23 100 5210
Resistance to 1+
3GC
AntAiA−mipminsenouogldpyeocnoimsciio4ldlniGeaCnls
MDRMeDxRcliundiSnTgypXhFDilRuaionnrCdtoarSuqbbuPiaMeaarpnrccBeoarlLutnlooy−leoinpBsdemiLhesIsis
Meticillin
Mono INH
Mono RIF
Penicillin
XDR in tVuaTbnecMrocP-umlSyocsMiinXs
Figure 6: Global deaths (counts) attributable to bacterial antimicrobial resistance by pathogen–drug combination, 2019
For this figure, only deaths attributable to resistance, not deaths associated with resistance, are shown due to the very high levels of correlation for resistance patterns between some drugs. 3GC=third-
generation cephalosporins. 4GC=fourth-generation cephalosporins. Anti-pseudomonal=anti-pseudomonal penicillin or beta-lactamase inhibitors. BL-BLI=β-lactam or β-lactamase inhibitors.
MDR=multidrug resistance. Mono INH=isoniazid mono-resistance. Mono RIF=rifampicin mono-resistance. NA=not applicable. Resistance to 1+=resistance to one or more drug. S Paratyphi=Salmonella
enterica serotype Paratyphi. S Typhi=S enterica serotype Typhi. TMP-SMX=trimethoprim-sulfamethoxazole. XDR=extensive drug resistance.
In 2017, WHO published a priority list for developing pathogen–drug combinations can inform future work on
new and effective antibiotic treatments. The list was WHO priority pathogen–drug combinations.
intended to inform research and development priorities
related to new antibiotics and put the most emphasis on Intervention strategies for addressing the challenge of
pathogens with multidrug resistance that cause severe bacterial AMR fall into five main categories. First, the
and often deadly infections in health-care and nursing principles of infection prevention and control remain a
home settings. Although the intention of this list was to foundation for preventing infections broadly and a
set new antibiotic research and development priorities cornerstone in combating the spread of AMR.44 These
rather than identify the most burdensome pathogen– include both hospital-based infection prevention and
drug combinations, its utility in dictating priorities has control programmes focused on preventing health-care-
still been limited by the absence of a global assessment acquired infections, and community-based programmes
of the burden of bacterial AMR. Only five of the seven focused on water, sanitation, and hygiene. Community-
pathogen–drug combinations that we estimated to have based programmes are particularly important in LMICs
caused the most deaths attributable to bacterial AMR where the AMR burden is highest and clean water and
in 2019 are currently on the list; MDR tuberculosis and sanitation infrastructure is weak; sustained support for
fluoroquinolone-resistant E coli are not included.34 these programmes is an essential element of combating
Additionally, meticillin-resistant S aureus—the leading AMR.
pathogen–drug combination in our analysis for
attributable deaths in 2019—is listed as “high” but not Second, preventing infections through vaccinations is
“critical” priority.34 WHO has explained that the absence paramount for reducing the need for antibiotics. Vaccines
of MDR tuberculosis from its priority list is because it are available for only one of the six leading pathogens
has already been established globally as a top priority for (S pneumoniae), although new vaccine programmes are
innovative treatments, but this exclusion remains a underway for S aureus, E coli, and others.45 Vaccination
source of considerable debate.42,43 Although many factors programmes are an important strategy for preventing
were considered in producing the WHO priority list, S pneumoniae,46 and vaccine development is crucial for
these new estimates of the global burden of specific pathogens that currently have no vaccine. Other vaccines,
such as the influenza or rotavirus vaccines, also play a
role in preventing febrile illness, which can lead to a
www.thelancet.com Vol 399 February 12, 2022 641
Articles
A Meticillin-resistant Staphylococcus aureus
Raw data
Percentage of isolates with resistance
<5% 40 to <50%
5 to <10% 50 to <60%
10 to <20% 60 to <70%
20 to <30% 70 to <80%
30 to <40% ≥80%
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
Modelled estimates
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
642 Mediterranean
Northern Europe
(Figure 7 continues on next page)
www.thelancet.com Vol 399 February 12, 2022
Articles
B Isoniazid and rifampicin co-resistant (excluding XDR) Mycobacterium tuberculosis
Raw data
Percentage of isolates with resistance
<5% 40 to <50%
5 to <10% 50 to <60%
10 to <20% 60 to <70%
20 to <30% 70 to <80%
30 to <40% ≥80%
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
Modelled estimates
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
www.thelancet.com Vol 399 February 12, 2022 (Figure 7 continues on next page)
643
Articles
C Third-generation cephalosporin-resistant Escherichia coli
Raw data
Percentage of isolates with resistance
<5% 40 to <50%
5 to <10% 50 to <60%
10 to <20% 60 to <70%
20 to <30% 70 to <80%
30 to <40% ≥80%
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
Modelled estimates
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
644 Mediterranean
Northern Europe
(Figure 7 continues on next page)
www.thelancet.com Vol 399 February 12, 2022
Articles
D Carbapenem-resistant Acinetobacter baumannii
Raw data
Percentage of isolates with resistance
<5% 40 to <50%
5 to <10% 50 to <60%
10 to <20% 60 to <70%
20 to <30% 70 to <80%
30 to <40% ≥80%
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
Modelled estimates
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
www.thelancet.com Vol 399 February 12, 2022 (Figure 7 continues on next page)
645
Articles
E Fluoroquinolone-resistant Escherichia coli
Raw data
Percentage of isolates with resistance
<5% 40 to <50%
5 to <10% 50 to <60%
10 to <20% 60 to <70%
20 to <30% 70 to <80%
30 to <40% ≥80%
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
Modelled estimates
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
646 Mediterranean
Northern Europe
(Figure 7 continues on next page)
www.thelancet.com Vol 399 February 12, 2022
Articles
F Carbapenem-resistant Klebsiella pneumoniae
Raw data
Percentage of isolates with resistance
<5% 40 to <50%
5 to <10% 50 to <60%
10 to <20% 60 to <70%
20 to <30% 70 to <80%
30 to <40% ≥80%
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
Modelled estimates
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
www.thelancet.com Vol 399 February 12, 2022 (Figure 7 continues on next page)
647
Articles
G Third-generation cephalosporin-resistant Klebsiella pneumoniae
Raw data
Percentage of isolates with resistance
<5% 40 to <50%
5 to <10% 50 to <60%
10 to <20% 60 to <70%
20 to <30% 70 to <80%
30 to <40% ≥80%
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
Modelled estimates
Caribbean and central America Persian Gulf Balkan Peninsula Southeast Asia West Africa Eastern
Mediterranean
Northern Europe
648 www.thelancet.com Vol 399 February 12, 2022
Articles
reduction in antibiotic prescribing and can reduce AMR The higher burden in low-resource health systems
emergence even for pathogens without vaccines.45 highlights the importance—both for the management of
individual patients and for the surveillance of AMR—of
Third, reducing exposure to antibiotics unrelated to well developed national action plans and laboratory
treating human disease is an important potential way to infrastructure in all regions and countries. The pattern of
reduce risk. Increased use of antibiotics in farming has AMR varies geographically, with different pathogens and
been identified as a potential contributor to AMR in pathogen–drug combinations dominating in different
humans,2,47–49 although the direct causal link remains locations. Our regional estimates could prove useful for
controversial.50,51 tailoring local responses as a one size fits all approach
might be inappropriate. Although antibiotic stewardship
Fourth, minimising the use of antibiotics when they is a foundational aspect for preventing the spread of
are not necessary to improve human health—such as AMR, limiting access to antibiotics is not a suitable
treating viral infections—should be prioritised. To this response to AMR in all settings. In fact, it could be
end, building infrastructure that allows clinicians to argued that an increase in access to antibiotics would
diagnose infection accurately and rapidly is crucial so decrease the AMR burden in some locations where
that antimicrobial use can be narrowed or stopped when second-line antibiotics are unavailable and would be
appropriate.52 The notion of antibiotic stewardship lifesaving; this might well be the case in western
remains a core strategy in most national and international sub-Saharan Africa. By contrast, limiting access to
AMR management plans, although barriers to antibiotics in south Asia through stewardship
implementing stewardship programmes in LMICs programmes might be the appropriate response for that
should be addressed.53,54 region because antibiotic overuse or misuse is believed to
be a major driver of AMR there.58 AMR is a global
Fifth, maintaining investment in the development problem and one that requires both global action and
pipeline for new antibiotics—and access to second-line nationally tailored responses.
antibiotics in locations without widespread access—is
essential. In the past few decades, investments have This study evaluated both the burden of bacterial
been small compared with those in other public health infections associated with drug resistance and the burden
issues with similar or less impact.55 Given the global directly attributable to drug resistance.13 At the global
importance of bacterial AMR, more assessment of level, the difference is nearly four-times that attributable
which policies have worked, and where, is urgently to AMR. We estimated both measures of burden because
needed. there is insufficient evidence to determine the extent to
which drug-resistant infections would be replaced by no
Many might expect that with higher antibiotic infection or susceptible infection if drug resistance was
consumption in high-resource settings, the burden of eliminated. Some evidence from the spread of meticillin-
bacterial AMR would be correspondingly higher in those resistant S aureus and meticillin-susceptible S aureus
settings. We found, however, that the highest rates of suggests that drug-resistant infections do not simply
death were in sub-Saharan Africa and south Asia. High replace drug-susceptible infections,63,64 but this finding
bacterial AMR burdens are a function of both the might not generalise to all other pathogens and other
prevalence of resistance and the underlying frequency of mechanisms of resistance.
critical infections such as lower respiratory infections,
bloodstream infections, and intra-abdominal infections, Both measures are informative in different ways. For
which are higher in these regions.14 Other drivers of the instance, when considering the specific burden of each
observed higher burden in LMICs include the scarcity of pathogen–drug combination, we believe that the burden
laboratory infrastructure making microbiological testing attributable to resistance is more appropriate because
unavailable to inform treatment to stop or narrow very high levels of co-resistance among some drugs lead
antibiotics,56 the inappropriate use of antibiotics driven to many deaths being duplicated across drugs when
by insufficient regulations and ease of acquisition,57 considering burden associated with resistance. When
inadequate access to second-line and third-line thinking about the role of vaccination to combat AMR,
antibiotics, counterfeit or substandard antibiotics that the no-infection counterfactual is more appropriate
can drive resistance,52,58,59 and poor sanitation and because infections would be eliminated, whereas
hygiene.60–62 interventions based on antimicrobial stewardship might
be better informed by the susceptible infection
Figure 7: Raw data and modelled estimates for the percentage of pathogen counterfactual because some resistant bacteria might be
isolates that are resistant by country and territory, 2019 replaced by susceptible bacteria.22 In either case, the
Meticillin-resistant Staphylococcus aureus (A), isoniazid and rifampicin co-resistant magnitude of the global bacterial AMR problem is very
(excluding XDR) Mycobacterium tuberculosis (B), third-generation cephalosporin- large and likely bounded by the two measures.
resistant Escherichia coli (C), carbapenem-resistant Acinetobacter baumannii (D),
fluoroquinolone-resistant E coli (E), carbapenem-resistant Klebsiella pneumoniae Our ability to compare our estimates with previous
(F), and third-generation cephalosporin-resistant K pneumoniae (G). Locations estimates is somewhat limited. The only global burden
with no data or modelled estimates are presented in white. XDR=extensively drug estimates for AMR are from the Review on Antimicrobial
resistant.
www.thelancet.com Vol 399 February 12, 2022 649
Articles
For the pathogen–drug Resistance,1 which did not provide death estimates by ECDC, we found the burden of third-generation
combinations estimates see pathogen–drug combination, making direct comparison cephalosporin-resistant E coli to exceed that of
challenging. The Review on Antimicrobial Resistance carbapenem-resistant A baumannii in high-income
http://ghdx.healthdata.org/ estimated 700 000 deaths in 2014 attributable to resistance settings, whereas the inverse pattern was found in south
record/ihme-data/global- to six pathogens: HIV, tuberculosis, malaria, S aureus, Asia, where a higher relative burden of carbapenem-
bacterial-antimicrobial- E coli, and K pneumoniae. We produced estimates for four resistant A baumannii than that in high-income regions
resistance-burden- of those pathogens—tuberculosis, S aureus, E coli, and has been documented.11 Our global burden was strongly
estimates-2019 K pneumoniae—and estimated 670 000 deaths attributable influenced by this higher relative burden of carbapenem-
to resistance to those pathogens in 2019. resistant A baumannii in south Asia and other LMICs.
Cassini and colleagues10 produced an estimate for the Our estimate for the burden of resistance is confined to
EU of 16 pathogen–antibiotic combinations in 2015. We the 88 pathogen–drug combinations we analysed.
produced estimates for 11 of these 16 combinations; we Expanding our resistance analysis to more pathogen–
did not estimate colistin resistance in E coli, P aeruginosa, drug combinations—particularly adding viruses,
or A baumannii because of the paucity of data on colistin parasites, and fungi—would increase our estimate of the
resistance in LMICs, or multidrug resistance in burden and could alter some of the results reported,
P aeruginosa or A baumannii because of our approach to depending on the correlation structure of resistance
MDR infections. For the 11 pathogen–drug combinations between the newly added and original 88 pathogen–drug
that overlap, Cassini and colleagues estimated approx combinations. It would provide a more thorough account
imately 30 000 deaths and 796 000 DALYs caused by of the threat of AMR and improve the accuracy of our
resistance in the EU in 2015. For these same 11 pathogen– estimates for the combinations reported here that share a
drug combinations, we estimated 23 100 deaths (95% UI high degree of co-resistance with combinations not yet
14 600–34 600) and 393 000 DALYs (246 000–595 000) analysed.
attributable to bacterial AMR for western and central
Europe combined. Cassini and colleagues used a mix of This study has several limitations, the most important
both counterfactuals to inform their estimates, so it is being the sparsity of data from many LMICs on the
expected that their EU estimate is somewhat higher than distribution of pathogens by infectious syndrome, the
ours for the susceptible counterfactual. This comparison prevalence of resistance for key pathogen–drug
is not perfect because there is not complete overlap in the combinations, and the number of deaths involving
locations included in western and central Europe and EU infection; and the severe scarcity of data linking
member countries (ie, Switzerland is included in our laboratory results to outcomes such as death. 19 of
estimate and not in the EU designation, whereas Estonia 204 countries and territories had no data available for any
is in the EU but is part of our eastern Europe region; of our modelling components. Limited availability of
appendix pp 100–05), but it offers some idea of how our data in some parts of the world was particularly
estimates compare with those of previous publications. consequential for the prevalence of resistance and
relative risk modelling components; we assumed that the
Some of our estimates might be unexpected and relative risk for each pathogen–drug combination, as
deserve special attention, particularly the high burden in well as the correlation structure of resistance between
sub-Saharan Africa and the burden of carbapenem- drugs, was the same in every location, age, and infectious
resistant A baumannii. Although we estimated syndrome. This might underestimate the AMR burden
sub-Saharan Africa to be the super-region with the lowest for LMICs, since the relative risk might be higher in
percentage of infectious deaths attributable to AMR locations where fewer second-line and third-line
(appendix p 97), the rate of deaths in which infection antibiotics are available. Assuming a single relative risk
plays a role was so much greater in sub-Saharan Africa for all infectious syndromes is a potentially strong
than in other super-regions that it overcame a relatively assumption; it is not immediately clear what direction
low prevalence of resistance and was the super-region this biases results, but it might lead to overestimation.
with the highest estimated AMR burden in 2019. Another substantial assumption we made due to
insufficient linked data was that the relative risk of death
Regarding carbapenem-resistant A baumannii, we or length of stay for infection from an MDR organism
estimated that it was the fourth leading pathogen–drug was assumed to be equal to the highest individual relative
combination globally for 2019, responsible for slightly risk among the drugs assessed. This mostly likely
fewer deaths than third-generation cephalosporin- underestimates the relative risk of MDR infections
resistant E coli. At first glance, this finding seems to because fewer effective antibiotic options remain as
contrast with other estimates such as those from Cassini resistance accumulates. In light of data sparsity, we made
and colleagues or the CDC, who have estimated the several additional methodological assumptions
burden of carbapenem-resistant A baumannii to be (appendix pp 17–60). Despite scarcity, our estimates are
substantially lower than that of third-generation informed by data from all regions (figure 7, table 1).
cephalosporin-resistant E coli.6,10 When assessed by super- These figures, and the appendix (pp 26–30, 43–44, 52–53,
region, however, our results are much more consistent and 56, which provides out-of-sample model validation),
with the published literature: similar to the CDC and
650 www.thelancet.com Vol 399 February 12, 2022
Articles
suggest that our modelled estimates fit the data, where or unclassifiable facilities, so it is possible that we did not
available. fully adjust for all potential tertiary bias. Further
limitations specific to each modelling component can be
Our analysis echoes that of another paper in found in the appendix (pp 119–20).
highlighting critical AMR data gaps in several regions.65
There are many well described barriers to good-quality Despite these limitations, this study is the most
clinical bacteriology in LMICs, and proper quality comprehensive analysis of bacterial AMR burden to date,
assurance and quality-control measures are crucial for reflecting the best and widest range of available data and
quality care and accurate laboratory-based surveillance.66 the use of models that have been tested and iterated over
Many lab-based surveillance systems are not linked to years of GBD analysis to incorporate disparate data
patient diagnoses or outcomes, limiting the inferences sources. Individually, these sources do not fully address
that are possible to obtain from such data. Selection bias the burden of AMR but, when used collectively, they
in how samples get incorporated into surveillance provide a more complete estimate with robust
systems; scarcity of laboratory facilities to test for AMR geographical coverage. To our knowledge, our study is
and other challenges in identifying AMR;17 insufficient the first to report burden both attributable to and
data linking prevalence of resistance to infectious associated with AMR for an extensive list of pathogens
syndrome, underlying cause, and outcome; barriers to and pathogen–drug combinations, with global and
sharing data that have been collected; and other data- regional findings based on estimates for 204 countries
linking and data optimisation issues continue to and territories. In the future, these estimates could be
complicate the assessment and interpretation of the used to better inform treatment guidelines. The
results in many cases. dominant bacterial pathogens for a given infectious
syndrome and the antibiotics that would offer effective
A second limitation of our study was the several treatment could be identified using the data for this
potential sources of bias we noted when combining and study, which, along with estimates of pathogen–drug
standardising data from a wide variety of providers. Our burden, could be used to inform empirical syndromic
estimates of the proportion of infections that were treatment guidelines tailored to a specific location.
community acquired versus hospital acquired for lower
respiratory and thorax infections and urinary tract Our analysis clearly shows that bacterial AMR is a
infections were based on the coding of data from multiple major global health problem. It poses the largest threat to
causes of death and hospital discharge data. This human health in sub-Saharan Africa and south Asia, but
approach could lead to misclassification, since the it is important in all regions. A diverse set of pathogens
criteria used in this coding are not strictly related to are involved, and resistance is high for multiple classes
community versus hospital acquisition. In future of essential agents, including beta-lactams and fluoro
iterations of the project, we hope to improve on the quinolones. Efforts to build laboratory infrastructure are
identification of community-acquired and hospital- paramount to addressing the large and universal burden
acquired infections. of AMR, by improving the management of individual
patients and the quality of data in local and global AMR
Additionally, no universal laboratory standard exists to surveillance and bolstering national AMR plans of action.
demarcate resistance versus susceptibility, and we often Enhanced infrastructure would also expand AMR
had to defer to laboratory interpretation to classify the research in the future to evaluate the indirect effects of
isolates in our data, resulting in heterogeneous AMR, such as the effect of AMR on perioperative
classification. Whenever possible, we classified resistance prophylaxis or prophylaxis of infections in transplant
using the most recent CLSI guidelines based on the recipients, the effects of AMR on transmission, the
minimum inhibitory concentrations provided in the impact and prevalence of specific variants evaluated
data; however, CLSI breakpoints have changed over time, through genotypic epidemiology, and more. Identifying
and many datasets did not provide sufficient detail to strategies that can work to reduce the burden of bacterial
allow for retrospective reanalysis of the data.67 AMR—either across a wide range of settings or those
that are specifically tailored to the resources available and
Finally, there is a possibility of selection bias in passive leading pathogen–drug combinations in a particular
microbial surveillance data, particularly if cultures are setting—is an urgent priority.
not routinely drawn. It might be that, in certain locations,
cultures are drawn only if a patient does not respond to Antimicrobial Resistance Collaborators
initial antibiotic therapy, which might lead to an
overestimate of the prevalence of resistance. Christopher J L Murray, Kevin Shunji Ikuta, Fablina Sharara,
Furthermore, in LMICs, hospital microbial data might Lucien Swetschinski, Gisela Robles Aguilar, Authia Gray, Chieh Han,
skew towards more urban populations or more severe Catherine Bisignano, Puja Rao, Eve Wool, Sarah C Johnson,
disease, which might not be representative of the broader Annie J Browne, Michael Give Chipeta, Frederick Fell, Sean Hackett,
population. We also received various data from tertiary Georgina Haines-Woodhouse, Bahar H Kashef Hamadani,
care facilities; although we adjusted for bias in the Emmanuelle A P Kumaran, Barney McManigal, Sureeruk Achalapong,
prevalence of resistance data collected from these Ramesh Agarwal, Samuel Akech, Samuel Albertson, John Amuasi,
sources, much of our data came from mixed-classification Jason Andrews, Aleskandr Aravkin, Elizabeth Ashley,
François-Xavier Babin, Freddie Bailey, Stephen Baker, Buddha Basnyat,
Adrie Bekker, Rose Bender, James A Berkley, Adhisivam Bethou,
www.thelancet.com Vol 399 February 12, 2022 651
Articles
Julia Bielicki, Suppawat Boonkasidecha, James Bukosia, Institute of Antimicrobial Research (Prof T Walsh DSc), University of
Cristina Carvalheiro, Carlos Castañeda-Orjuela, Vilada Chansamouth, Oxford, Oxford, UK; Infection Control Unit (S Achalapong PhD),
Suman Chaurasia, Sara Chiurchiù, Fazle Chowdhury, Department of Medicine (S Khusuwan MD), Department of Pediatrics
Rafai Clotaire Donatien, Aislinn J Cook, Ben Cooper, Tim R Cressey, (P Ounchanum MD), Department of Internal Medicine
Elia Criollo-Mora, Matthew Cunningham, Saffiatou Darboe, (S Seekaew MD), Chiang Rai Prachanukroh Hospital, Chiang Rai,
Nicholas P J Day, Maia De Luca, Klara Dokova, Angela Dramowski, Thailand; Department of Pediatrics (R Agarwal DM), All India Institute
Susanna J Dunachie, Thuy Duong Bich, Tim Eckmanns, Daniel Eibach, of Medical Sciences, New Delhi, India; Health Services Research Unit
Amir Emami, Nicholas Feasey, Natasha Fisher-Pearson, Karen Forrest, (S Akech PhD), Nairobi Programme (J Bukosia MSc), Kenya Medical
Coralith Garcia, Denise Garrett, Petra Gastmeier, Ababi Zergaw Giref, Research Institute (KEMRI)—Wellcome Trust Research Programme,
Rachel Claire Greer, Vikas Gupta, Sebastian Haller, Andrea Haselbeck, Nairobi, Kenya; Department of Global Health (J Amuasi PhD), Kwame
Simon I Hay, Marianne Holm, Susan Hopkins, Yingfen Hsia, Nkrumah University of Science and Technology, Kumasi, Ghana; Global
Kenneth C Iregbu, Jan Jacobs, Daniel Jarovsky, Fatemeh Javanmardi, Health and Infectious Diseases (J Amuasi), Kumasi Centre for
Adam W J Jenney, Meera Khorana, Niranjan Kissoon, Elsa Kobeissi, Collaborative Research in Tropical Medicine, Kumasi, Ghana;
Tomislav Kostyanev, Koukeo Phommasone, Suwimon Khusuwan, Department of Medicine (J Andrews MD), Stanford University, Stanford,
Fiorella Krapp, Ralf Krumkamp, Ajay Kumar, Hmwe H Kyu, Cherry CA, USA; Lao-Oxford-Mahosot Hospital-Wellcome Trust Research Unit
Lim, Kruy Lim, Direk Limmathurotsakul, Michael James Loftus, (E Ashley FRCPath, V Chansamouth MSc, K Phommasone PhD,
Miles Lunn, Jianing Ma, Anand Manoharan, Florian Marks, Jürgen May, P Newton FRCP, S Rattanavong MD, T Roberts PhD,
Mayfong Mayxay, Neema Mturi, Tatiana Munera-Huertas, A J Simpson FRCPath, M Vongsouvath MD) Mahosot Hospital,
Patrick Musicha, Lillian A Musila, Marisa Marcia Mussi-Pinhata, Vientiane, Laos; Mérieux Foundation, Lyon, France (F Babin PharmD);
Ravi Narayan Naidu, Tomoka Nakamura, Ruchi Nanavati, Department of Medicine (S Baker PhD), Cambridge Institute of
Sushma Nangia, Paul Newton, Chanpheaktra Ngoun, Amanda Novotney, Therapeutic Immunology and Infectious Disease (CITIID)
Davis Nwakanma, Christina W Obiero, Theresa J Ochoa, (F Marks PhD), University of Cambridge, Cambridge, UK; Oxford
Antonio Olivas-Martinez, Piero Olliaro, Ednah Ooko, University Clinical Research Unit-Nepal (B Basnyat FRCPE), Oxford
Edgar Ortiz-Brizuela, Pradthana Ounchanum, Gideok D Pak, University, Kathmandu, Nepal; Department of Paediatrics and Child
Jose Luis Paredes, Anton Yariv Peleg, Carlo Perrone, Thong Phe, Health (A Bekker PhD, A Dramowski PhD), Stellenbosch University,
Nishad Plakkal, Alfredo Ponce-de-Leon, Mathieu Raad, Cape Town, South Africa; Clinical Research Department
Tanusha Ramdin, Sayaphet Rattanavong, Amy Riddell, Tamalee Roberts, (J A Berkley, N Mturi MRCPCH, C Obiero MPH), Department of
Julie Victoria Robotham, Anna Roca, Victor Daniel Rosenthal, Epidemiology & Demography (Prof J Scott FMedSci), Department of
Kristina E Rudd, Neal Russell, Helio S Sader, Weerawut Saengchan, Biomedical Sciences (C Tigoi MSc), KEMRI - Wellcome Trust Research
Jesse Schnall, John Anthony Gerard Scott, Samroeng Seekaew, Programme, Kilifi, Kenya (E Ooko PhD); Department of Neonatology
Mike Sharland, Madhusudhan Shivamallappa, Jose Sifuentes-Osornio, (A Bethou PhD, N Plakkal MD), Jawaharlal Institute of Postgraduate
Andrew J Simpson Nicolas Steenkeste, Andrew James Stewardson, Medical Education & Research, Puducherry, India; Paediatric Infectious
Temenuga Stoeva, Nidanuch Tasak, Areerat Thaiprakong, Guy Thwaites, Disease Department (J Bielicki PhD), University of Basel Children’s
Caroline Tigoi, Claudia Turner, Paul Turner, H Rogier van Doorn, Hospital, Basel, Switzerland; Paediatric Infectious Diseases Research
Sithembiso Velaphi, Avina Vongpradith, Manivanh Vongsouvath, Group (J Bielicki, A J Cook MSc, A Riddell PhD, N Russell MBBS),
Huong Vu, Timothy Walsh, Judd L Walson, Seymour Waner, Institute for Infection and Immunity (T Munera-Huertas PhD)
Tri Wangrangsimakul, Prapass Wannapinij, Teresa Wozniak, M Sharland MD, St George’s University of London, London, UK;
Tracey E M W Young-Sharma, Kalvin C Yu, Peng Zheng, Benn Sartorius, Department of Pediatrics (S Boonkasidecha MD, M Khorana MD),
Alan D Lopez, Andy Stergachis, Catrin Moore*, Christiane Dolecek*, Queen Sirikit National Institute of Child Health, Bangkok, Thailand;
Mohsen Naghavi. Department of Pediatrics (C Carvalheiro PhD), University of Sao Paulo,
*Contributed equally. Ribeirao Preto, Brazil; Colombian National Health Observatory
(C Castañeda-Orjuela MD), Instituto Nacional de Salud,
Affiliations Bogota, Colombia; Epidemiology and Public Health Evaluation Group
(C Castañeda-Orjuela), Universidad Nacional de Colombia,
Institute for Health Metrics and Evaluation (Prof C J L Murray DPhil, Bogota, Colombia; Department of Neonatology (S Chaurasia PhD),
K S Ikuta MD, F Sharara MS, L Swetschinski MSc, A Gray BS, All India Institute of Medical Sciences, Rishikesh, India; Immunology
C Han BA, C Bisignano MPH, P Rao MPH, E Wool MPH, and Infectious Disease Unit Academic Department of Pediatrics
S C Johnson MSc, S Albertson BS, A Aravkin PhD, R Bender BS, (S Chiurchiù MD), Academic Hospital Pediatric Department
M Cunningham MSc, Prof S I Hay FMedSci, H H Kyu PhD, J Ma MS, (M De Luca MD), Bambino Gesù Children’s Hospital, Rome, Italy;
A Novotney MPH, A Vongpradith BA, P Zheng PhD, A Stergachis PhD) Department of Internal Medicine (F Chowdhury PhD), Bangabandhu
Prof M Naghavi PhD, Department of Health Metrics Sciences, School of Sheikh Mujib Medical University, Dhaka, Bangladesh; Chiang Rai
Medicine (Prof C J L Murray, A Aravkin, Prof S I Hay, B Sartorius, Clinical Research Unit (R Greer MRCGP, T Wangrangsimakul
Prof M Naghavi PhD), Department of Applied Mathematics (A Aravkin), FRCPath), Department of Microbiology (S Dunachie PhD, C Lim MSc,
Department of Global Health (J Walson MD, A Stergachis), Department Prof D Limmathurotsakul PhD, N Tasak BNS, A Thaiprakong BS),
of Pediatrics (J Walson), Department of Pharmacy, School of Pharmacy Faculty of Tropical Medicine (N Day DM, C Perrone MD),
(A Stergachis), University of Washington, Seattle, WA, USA; Mahidol-Oxford Tropical Medicine Research Unit, Bangkok, Thailand
Department of Infectious Diseases (K S Ikuta), Veterans Affairs Greater (F Chowdhury PhD, Prof C Dolecek FRCP, P Wannapinij MIS);
Los Angeles Healthcare System, Los Angeles, CA, USA; Department of Laboratoire National de Biologie Clinique et de Santé Publique (National
Infectious Diseases (K S Ikuta), University of California, Los Angeles, Laboratory of Clinical Biology and Public Health) (R Clotaire Donatien
Los Angeles, CA, USA; Nuffield Department of Medicine, Big Data MD), Health Ministry of Central African Republic, Bangui, Central
Institute (F Bailey MBChB, A Browne MPH, M Chipeta PhD, African Republic; University of Bangui (R Clotaire Donatien MD),
F Fell MSc, N Fisher-Pearson BA, S Hackett PhD, G Haines-Woodhouse Faculté des Sciences de la Santé (Faculty of Health Science),
MRes, E Kobeissi MPH, E Kumaran MSc, M Lunn BSc, Bangui, Central African Republic; Department of Molecular & Clinical
Prof M Mayxay PhD, B McManigal PhD, C E Moore DPhil, Pharmacology (T R Cressey), University of Liverpool, Liverpool, UK;
P Olliaro PhD, G Robles Aguilar DPhil), Nuffield Department of Department of Pharmacy (E Criollo-Mora BSc), Department of Medicine
Medicine, Centre for Tropical Medicine and Global Health (A Olivas-Martinez MD, E Ortiz-Brizuela MSc), Department of
(E Ashley FRCPath, J A Berkley PhD, V Chansamouth MSc, Infectious Diseases (A Ponce-de-Leon MD), Instituto Nacional de
B Cooper PhD, Prof C Dolecek FRCP, S Dunachie PhD, Ciencias Medicas y Nutricion Salvador Zubiran, Mexico City, Mexico;
B Kashef Hamadani MPH, C Lim MSc, Prof D Limmathurotsakul PhD, Disease Control and Elimination Department (S Darboe MSc,
P Newton FRCP, C Perrone MD, G Thwaites FMedSci, A Roca PhD), Clinical Services Department (K Forrest FRCP),
P Turner FRCPath, B Sartorius PhD, N Day DM, R Greer MRCGP, Laboratory Services Department (D Nwakanma PhD), Medical Research
H van Doorn PhD, T Wangrangsimakul FRCPath), Nuffield Department
of Clinical Medicine (A J Simpson FRCPath, C Tigoi MSc), Ineos Oxford
652 www.thelancet.com Vol 399 February 12, 2022
Articles
Council Unit The Gambia at the London School of Hygiene & Tropical Department of Immunization, Vaccines, and Biologicals
Medicine, Banjul, The Gambia; Department of Social Medicine and (T Nakamura MSPH), World Health Organization, Geneva, Switzerland;
Health Care Organization (K Dokova PhD), Department of Microbiology Department of Infectious Disease Epidemiology (T Nakamura), London
and Virology (T Stoeva PhD), Medical University of Varna, School of Hygiene and Tropical Medicine, London, UK; Department of
Varna, Bulgaria; Adult Intensive Care Unit, Hospital for Tropical Neonatology (R Nanavati MD), Seth GSMC & KEM Hospital,
Diseases, Ho Chi Minh City, Vietnam (T Duong Bich PhD); Infectious Mumbai, India; Medical Department (C Ngoun MD), Executive Office
Disease Epidemiology (T Eckmanns PhD, S Haller MPH), (C Turner FRCPCH), Cambodia Oxford Medical Research Unit
Robert Koch Institute, Berlin, Germany; Department of Infectious (P Turner), Angkor Hospital for Children, Siem Reap, Cambodia;
Disease Epidemiology (D Eibach MD, R Krumkamp DrPH), J May MD, Department of Global Health (C W Obiero), University of Amsterdam,
Bernhard Nocht Institute for Tropical Medicine, Hamburg, Germany; Amsterdam, Netherlands; Infectious Disease Department (A Y Peleg,
Microbiology Department (A Emami PhD), Shiraz University of Medical A J Stewardson), The Alfred Hospital, Melbourne, VIC, Australia;
Sciences, Shiraz, Iran; Clinical Sciences (N Feasey PhD), Liverpool Department of Medicine (A Ponce-de-Leon), Universidad Panamericana,
School of Tropical Medicine, Liverpool, UK; Malawi Liverpool Wellcome Mexico City, Mexico; International Operations Department
Trust Clinical Research Programme, Blantyre, Malawi (N Feasey); (M Raad MD), International Operations Direction (N Steenkeste PhD),
Instituto de Medicina Tropical Alexander von Humboldt Fondation Mérieux, Lyon, France; Department of Paediatric and Child
(Alexander von Humboldt Institute of Tropical Medicine) (C Garcia PhD, Health (T Ramdin MBBCh), University of Witwatersrand, Parktown,
F Krapp MSc, T Ochoa PhD, J Paredes MD), Universidad Peruana South Africa; Department of Public Health Sciences (V Rosenthal MD),
Cayetano Heredia (Cayetano Heredia Peruvian University), Lima, Peru; University of Miami, Miami, FL, USA; International Nosocomial
Departamento de Enfermedades Infecciosas, Tropicales y Infection Control Consortium (V Rosenthal MD), International
Dermatológicas (Department of Infectious, Tropical, and Dermatological Nosocomial Infections Control Consortium (INICC) Foundation,
Diseases) (C Garcia PhD), Hospital Cayetano Heredia (Cayetano Heredia Miami, FL, USA; Department of Critical Care Medicine (K E Rudd MD),
Hospital), Lima, Peru; Applied Epidemiology Programs (D Garrett MD), University of Pittsburgh, Pittsburgh, PA, USA; Department of
Sabin Vaccine Institute, Washington, DC, USA; Institute of Hygiene Microbiology and Surveillance (H S Sader MD), JMI Laboratories,
(Prof P Gastmeier MD), Charité University Medicine Berlin, North Liberty, IA, USA; Laboratory (W Saengchan BS), Chiang Rai
Berlin, Germany; Department of Health Policy and Management Prachnukroh Hospital, Chiang Rai, Thailand; Doctors in Training
(A Z Giref PhD), Addis Ababa University, Addis Ababa, Ethiopia; (J Schnall MBBS), Austin Health, Heidelberg, VIC, Australia;
National Data Management Center (A Z Giref), Ethiopian Public Health Department of Infectious Disease Epidemiology
Institute, Addis Ababa, Ethiopia; MMS Medical Affairs (Prof J A G Scott FMedSci), London School of Hygiene & Tropical
(V Gupta PharmD), Becton, Dickinson and Company, Franklin Lakes, Medicine, London, UK; Department of Neonatology
NJ, USA; Epidemiology & Public Health Research Department (M Shivamallappa DM), King Edward Memorial Hospital Mumbai,
(A Haselbeck Dr rer medic, M Holm PhD), Epidemiology, Public Health, Mumbai, India; Department of Medicine (J Sifuentes-Osornio MD),
Impact (EPIC) (F Marks PhD), Department of Biostatistics & Data Instituto Nactional de Ciencias Medicas, Mexico City, Mexico;
Management (G D Pak MSc), International Vaccine Institute, Microbiology Laboratory (T Stoeva), Varna University Hospital, Varna,
Seoul, South Korea; National Infection Service (S Hopkins FRCP), Bulgaria; Oxford University Clinical Research Unit Viet Nam
Antimicrobrial Resistance Division (J V Robotham PhD), Public Health (G Thwaites, H Vu PhD), University of Oxford, Ho Chi Minh City,
England, London, UK; School of Pharmacy (Y Hsia PhD), Queen’s Vietnam; Cambodia Oxford Medical Research Unit, Siem Reap,
University Belfast, Belfast, Ireland; Department of Medical Microbiology Cambodia (C Turner); Oxford University Clinical Research Unit,
(K C Iregbu MD), National Hospital, Abuja, Nigeria; Department of Hanoi, Vietnam (H R van Doorn); School of Clinical Medicine, Faculty
Medical Microbiology (K C Iregbu), University of Abuja, Abuja, Nigeria; of Health Sciences (S Velaphi PhD), University of the Witwatersrand,
Department of Clinical Sciences (Prof J Jacobs PhD), Institute of Johannesburg, South Africa; Department of Paediatrics (S Velaphi),
Tropical Medicine, Antwerp, Belgium; Department of Microbiology, Chris Hani Baragwanath Academic Hospital,
Immunology, and Transplantation (Prof J Jacobs), KU Leuven, Johannesburg, South Africa; Department of Microbiology
Leuven, Belgium; Pediatric Infectious Disease Department (S Waner MMed), Lancet Laboratories, Johannesburg, South Africa;
(D Jarovsky MD), Santa Casa de São Paulo, São Paulo, Brazil; Department of Global Tropical Health (T Wozniak PhD), Menzies
Microbiology Department (F Javanmardi PhDc), Shiraz University of School of Health Research, Brisbane, QLD, Australia. Medical and
Medical sciences, Shiraz, Iran; Infectious Disease Department Scientific Affairs (K C Yu MD), Becton Dickinson, Venice, CA, USA”
(A Jenney PhD, A Y Peleg, A J Stewardson), The Alfred Hospital, following Australia.
Melbourne, VIC, Australia; Department of Pediatrics (N Kissoon MBBS),
University of British Columbia, Vancouver, BC, Canada; Laboratory of Contributors
Medical Microbiology (T Kostyanev MD), University of Antwerp, Detailed information about individual author contributions to the
Antwerp, Belgium; Department of Neonatology (A Kumar MD, research are available in the appendix (pp 65–66). Members of the core
S Nangia MD), Lady Hardinge Medical College & Kalawati Saran’s research team for this topic area had full access to the underlying data
Children’s Hospital, New Delhi, India; Sihanouk Hospital Center of used to generate estimates presented in this paper. All other authors had
Hope, Phnom Penh, Cambodia (K Lim MD, T Phe MSc); Department of access to, and reviewed, estimates as part of the research evaluation
Infectious Diseases (M J Loftus MBBS, A Y Peleg PhD, process, which includes additional stages of formal review.
A J Stewardson PhD), Monash University, Melbourne, VIC, Australia;
KEMRI—Wellcome Trust Research Programme, Kilifi, Kenya Declaration of interests
(E Ooko PhD); Research Department (A Manoharan PhD), The CHILDS E Ashley reports that Lao-Oxford-Mahosot Hospital—Wellcome Trust
Trust Medical Research Foundation, Chennai, India; Site Research Unit received financial support from the Global Research on
Hamburg-Borstel-Lübeck-Riems (J May MD), German Center for Antimicrobial Resistance Project (GRAM) to extract and prepare data for
Infection Research (DZIF), Hamburg, Germany; Department of the present manuscript. J Bielicki reports grants from the European and
Research and Educational Development (Prof M Mayxay PhD), Developing Countries Clinical Trials Partnership, Horizon 2020, and
University of Health Sciences, Vientiane, Laos; Parasites and Microbes Swiss National Science Foundation, and a contract from the National
Programme (P Musicha PhD), Wellcome Sanger Institute, Institute for Health Research (NIHR), outside of the submitted work;
Cambridge, UK; Department of Emerging Infectious Diseases and consulting fees from Shionogi and Sandoz and speaking fees from
(L A Musila PhD), United States Army Medical Research Directorate - Pfizer and Sandoz, outside the submitted work. C Carvalheiro reports
Africa, Nairobi, Kenya; Center for Clinical Research (L A Musila PhD), financial support for the present manuscript from the Global Antibiotic
Kenya Medical Research Institute, Nairobi, Kenya; Deparment of Research and Development Partnership, who provided payments to
Pediatrics (M M Mussi-Pinhata MD), University of São Paulo, Fundação de Apoio ao Ensino, Pesquisa e Assistência of the Clinical
Ribeirão Preto, Brazil; Colonial War Memorial Hospital, Ministry Of Hospital of the Faculty of Medicine of Ribeirão Preto, University of
Health, Suva, Fiji (R Naidu MMed, T E Young-Sharma MMed); São Paulo, Brazil. S Dunachie reports financial support for the present
manuscript from UL Flemming Fund at the Department of Health and
www.thelancet.com Vol 399 February 12, 2022 653
Articles
Social Care, the Bill & Melinda Gates Foundation, and the Wellcome Programme grant. S Dunachie acknowledges funding from NIHR Global
Trust; a paid membership role for the Wellcome Trust Vaccines Advisory Research Professorship (NIHR300791). F Krapp was supported by
Selection Panel Vaccines and AMR in November, 2019; and an unpaid role Framework Agreement Belgian Directorate of Development Cooperation-
as an expert adviser to WHO’s Global Antimicrobrial Resistance Institute of Tropical Medicine in Antwerp. M Khorana and S
Surveillance System, from November, 2018 onwards, outside the Boonkasidecha would like to acknowledge GARDP. A Peleg
submitted work. A Haselbeck reports support for the present manuscript acknowledges the support from an Australian National Health and
from the Bill & Melinda Gates Foundation (OPP1205877). Y Hsia reports Medical Research Council Practitioner Fellowship. A Stewardson is
support for the present manuscript from National Institutes of Health for supported by an Australian National Health and Medical Research
the Global Pediatric Point Prevalence Survey (GARPEC-PPS) study and Council Early Career Fellowship (GNT1141398). P Turner acknowledges
Blood Stream Infections (GARPEC-BSI) study, paid directly to St George’s that the Cambodia Oxford Medical Research Unit is part of the Mahidol-
University of London and grants or contract from the World Health Oxford Tropical Medicine Research Unit Tropical Health Network and is
Organisation to support the Global Pediatric Formulation Project. C Lim core funded by Wellcome (220211/Z/20/Z). T Wangrangsimakul
was supported by the Wellcome Trust Training Fellowship between acknowledges funding from the Wellcome Trust, as part of the MORU
September, 2017 and March 2020 (206736/Z/17/Z), outside the submitted Tropical Health Network institutional funding support. The Medical
work. M Mussi-Pinhata reports support for the present manuscript from Research Council Unit—The Gambia at the LSHTM acknowledges all
research from grant funding from Fondazione PENTA—Onlus and the the staff in the microbiology clinical laboratory for their support.
Clinical Trial Manager Global Antibiotic R&D Partnership (GARDP). P We acknowledge The Australian Group for Antimicrobial Resistance, and
Newton reports support for the present manuscript from research grant The Australian Commission on Safety and Quality in Healthcare, Sydney,
funding from the Wellcome Trust. T J Ochoa reports support for the Australia. We acknowledge John Murray, Becton, Dickinson and
present manuscript from Pfizer research grant for invasive pneumococcal Company. We give thanks to the late Rattanaphone Phetsouvanh, the
surveillance, paid directly to their institution. F Marks reports unpaid Lao Ministry of Health, and the Directorate of Mahosot Hospital who
participation on the London School of Hygiene and Tropical Medicine enabled the collection and sharing of Lao data, Vientiane, Lao People’s
data safety monitoring board for the Ebola vaccine study, outside the Democratic Republic. We thank Sabrina Bacci, Liselotte Diaz Högberg,
submitted work. J May reports support for the present manuscript from Marlena Kaczmarek, Maria Keramarou, Favelle Lamb,
the Bill & Melinda Gates Foundation, German Federal Ministry of Dominique L Monnet, Gianfranco Spiteri, Carl Suetens,
Research & Education, German Federal Ministry of Health, and German Therese Westrell and Klaus Weist at the ECDC, Solna, Sweden, for
Research Association and support for attending meetings and/or travel providing information on databases and discussions on data
from the German Federal Ministry of Health, outside the submitted work. interpretation. We acknowledge Jennifer R Verani and team, CDC,
J Robotham is a member of the UK Government Advisory Committee on Nairobi, Kenya; Allan Audi and team, Centre for Global Health Research,
Antimicrobial Prescribing Resistance and Healthcare Associated KEMRI, Kisumu, Kenya. We acknowledge Jephté Kaleb and
Infections, outside the submitted work. J Scott reports that the London Giscard Wilfried Koyaweda, National Laboratory of Clinical Biology and
School of Hygiene & Tropical Medicine (LSHTM) received financial Public Health, Bangui, Central African Republic. We acknowledge
support from Emory University to support CHAMPS projects in Ethiopia Tien Viet Dung Vu and Nguyen Minh Trang Nghiem, Oxford University
for the present manuscript; reports a paid fellowship from the Wellcome Clinical Research Unit, Wellcome Africa Asia Programme, National
Trust, research grants from Gavi, the Vaccine Alliance, and NIHR paid to Hospital for Tropical Diseases, Hanoi, Vietnam; and the VINARES
LSHTM, and an African research leader fellowship paid to LSHTM by the Consortium. We acknowledge the Department of Pathology and
Medical Research Council, outside the submitted work; and reports being Laboratory Medicine, and Department of Paediatrics and Child Health,
a member of the data safety and monitoring board for PATH Vaccines Aga Khan University, Karachi, Pakistan. We acknowledge Samuel Akech,
Solutions for SII PCV10 in The Gambia. J Sifuentes-Osornio reports Ednah Ooko, James Bukosia, Neema Mturi, J Anthony G Scott,
financial support from Oxford University for the present manuscript; Philip Bejon, Lynette Isabella Oyier, Salim Mwarumba,
research grants from Oxford, CONACYT, Sanofi, and Novartis, outside of Esther Muthumbi, Christina Obiero, Robert Musyimi,
the study; consulting fees from Senosiain and speaker fees from Merck, Shebe Mohammed, Caroline Ogwang, Christopher Maronga,
outside of the study; and membership of the Sanofi advisory board of Ambrose Agweyu, KEMRI Wellcome Trust Research Programme, Kilifi,
COVID-19 Vaccine Development, which is currently in progress, outside Kenya. We acknowledge scientific contributions to this work from the
of the study. A J Stewardson reports grants or contracts from Merck, Pan American Health Organization. We would like to acknowledge the
Sharp, & Dohme paid to Monash University, Melbourne, outside of the scientific contributions made from the GRAM advisory committee,
study. P Turner reports grants, consulting fees, and support for attending specifically Neil Ferguson and Sharon Peacock. We would like to
meetings or travel from Wellcome Trust, outside the study. H van Doorn acknowledge Tomislav Mestrovic for his significant contributions to this
reports grants or contracts from the University of Oxford and is the manuscript and the overall research enterprise. We thank the Kenya
principal investigator for the Fleming Fund pilot grant; and he is a board Medical Research Institute, United States Army Medical Research
member of Wellcome Trust’s Surveillance and Epidemiology of Drug Directorate-Africa, Kenya, Nairobi, Kenya; we acknowledge the
Resistant Infections. T Walsh reports financial support from the Bill & International Nosocomial Infection Control Consortium,
Melinda Gates Foundation for the BARNARDS (neonatal sepsis and Buenos Aires, Argentina; we would like to thank the CHILDS Trust
mortality) study for the present manuscript. J L Walson reports grants or Medical Research Foundation, Chennai, India; we acknowledge the
contracts from the Bill & Melinda Gates Foundation and National Childhood Acute Illness & Nutrition Network investigators; we thank
Institutes of Health, paid directly to their institution, outside of the Institut Pasteur and Laboratoire National de Biologie Clinique et de Santé
submitted work. All other authors declare no competing interests. Publique in Bangui, Central African Republic; we would like to
acknowledge the Global Tuberculosis Programme of WHO, Geneva,
Data sharing Switzerland; we thank the SENTRY Antimicrobial Surveillance Program,
Citations for the data used in the study can be accessed from the Global JMI Laboratories, North Liberty, Iowa, USA; we acknowledge the
Health Data Exchange AMR website. Access to the data are also provided Sihanouk Hospital Center of Hope, Phnom Penh, Cambodia. Data
as data use agreements permit. provided for this manuscript was generated by work funded by the
Armed Forces Health Surveillance Division, Global Emerging Infections
Acknowledgments Surveillance (GEIS) Branch PROMIS ID P0153 KY 2015-2019.
Funding was provided by the Bill & Melinda Gates Foundation
(OPP1176062), the Wellcome Trust (A126042), and the UK Department of Editorial note: the Lancet Group takes a neutral position with respect to
Health and Social Care using UK aid funding managed by the Fleming territorial claims in published maps and institutional affiliations.
Fund (R52354 CN001). E Ashley acknowledges that Lao-Oxford-Mahosot
Hospital–Wellcome Trust Research Unit receives core funding from References
Wellcome (20211/Z/20/Z). N Feasey acknowledges that the Malawi 1 O’Neill J. Tackling drug-resistant infections globally: final report
Liverpool Wellcome Trust Clinical Research Programme diagnostic
microbiology service is funded by a Wellcome Asia and Africa and recommendations. London: Review on Antimicrobial
Resistance, 2016.
654 www.thelancet.com Vol 399 February 12, 2022
Articles
2 O’Neill J. Antimicrobial resistance: tackling a crisis for the health and 25 WHO. Global report on the epidemiology and burden of sepsis:
wealth of nations. London: Review on Antimicrobial Resistance, current evidence, identifying gaps and future directions. Geneva:
2014. World Health Organization, 2020.
3 de Kraker MEA, Stewardson AJ, Harbarth S. Will 10 million people 26 Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J,
die a year due to antimicrobial resistance by 2050? PLoS Med 2016; Pinsky MR. Epidemiology of severe sepsis in the United States:
13: e1002184. analysis of incidence, outcome, and associated costs of care.
Crit Care Med 2001; 29: 1303–10.
4 National Office for Animal Health. NOAH response to final
O’Neill AMR review report July 2016. Middlesex: National Office for 27 Martin GS, Mannino DM, Eaton S, Moss M. The epidemiology of
Animal Health, 2016. sepsis in the United States from 1979 through 2000. N Engl J Med
2003; 348: 1546–54.
5 WHO. Antimicrobial resistance. 2021. https://www.who.int/news-
room/fact-sheets/detail/antimicrobial-resistance (accessed 28 Rhee C, Dantes R, Epstein L, et al. Incidence and trends of sepsis
Feb 23, 2021). in US hospitals using clinical vs claims data, 2009–2014. JAMA
2017; 318: 1241–49.
6 US Centers for Disease Control and Prevention. Antibiotic resistance
threats in the United States, 2019. Atlanta, GA: US Department of 29 Zheng P, Barber R, Sorensen RJD, Murray CJL, Aravkin AY.
Health and Human Services, 2019. Trimmed constrained mixed effects models: formulations and
algorithms. J Comput Graph Stat 2021; 0: 1–13.
7 Prestinaci F, Pezzotti P, Pantosti A. Antimicrobial resistance: a global
multifaceted phenomenon. Pathog Glob Health 2015; 109: 309–18. 30 WHO. Antimicrobial resistance global report on surveillance 2014.
Geneva: World Health Organization, 2014.
8 WHO. Global action plan on antimicrobial resistance. Geneva: World
Health Organization, 2015. 31 Murray CJL, Aravkin AY, Zheng P, et al. Global burden of 87 risk
factors in 204 countries and territories, 1990-2019: a systematic
9 Naylor NR, Atun R, Zhu N, et al. Estimating the burden of analysis for the Global Burden of Disease Study 2019. Lancet 2020;
antimicrobial resistance: a systematic literature review. 396: 1223–49.
Antimicrob Resist Infect Control 2018; 7: 58.
32 New York Department of Health. Hospital associated infection
10 Cassini A, Högberg LD, Plachouras D, et al. Attributable deaths and reporting persuant to Public Health Law section 2819. 2020.
disability-adjusted life-years caused by infections with antibiotic- https://www.gnyha.org/wp-content/uploads/2020/06/FINAL-DAL-
resistant bacteria in the EU and the European Economic Area in 20-10-PHL-2819-HAI-Reporting-6_11_20.pdf (accessed
2015: a population-level modelling analysis. Lancet Infect Dis 2019; May 19, 2021).
19: 56–66.
33 Centers for Medicare and Medicaid Services. COVID-19
11 Lim C, Takahashi E, Hongsuwan M, et al. Epidemiology and burden emergency declaration blanket waivers for health care providers.
of multidrug-resistant bacterial infection in a developing country. Baltimore, MD: Centers for Medicare and Medicaid Services, 2021.
eLife 2016; 5: e18082.
34 WHO. Global priority list of antibiotic-resistant bacteria to guide
12 Temkin E, Fallach N, Almagor J, Gladstone BP, Tacconelli E, Carmeli research, discovery, and development of new antibiotics. 2017.
Y. Estimating the number of infections caused by antibiotic-resistant https://www.who.int/medicines/publications/WHO-PPL-Short_
Escherichia coli and Klebsiella pneumoniae in 2014: a modelling study. Summary_25Feb-ET_NM_WHO.pdf?ua=1 (accessed
Lancet Glob Health 2018; 6: e969–79. April 23, 2021).
13 de Kraker MEA, Lipsitch M. Burden of antimicrobial resistance: 35 UN. Interagency Coordination Group on Antimicrobial
compared to what? Epidemiol Rev 2021; published online March 12. Resistance. United Nations. 2017. https://www.un.org/sg/en/
https://doi.org/10.1093/epirev/mxab001. content/sg/personnel-appointments/2017-03-17/interagency-
coordination-group-antimicrobial-resistance (accessed
14 Vos T, Lim SS, Abbafati C, et al. Global burden of 369 diseases and April 29, 2021).
injuries in 204 countries and territories, 1990-2019: a systematic
analysis for the Global Burden of Disease Study 2019. Lancet 2020; 36 WHO. Global Leaders Group on antimicrobial resistance. 2021.
396: 1204–22. https://www.who.int/groups/one-health-global-leaders-group-on-
antimicrobial-resistance (accessed April 29, 2021).
15 Hay SI, Rao PC, Dolecek C, et al. Measuring and mapping the global
burden of antimicrobial resistance. BMC Med 2018; 16: 78. 37 WHO. Pneumococcal conjugate vaccines in infants and children
under 5 years of age: WHO position paper—February 2019.
16 Murray CJ, Ezzati M, Flaxman AD, et al. GBD 2010: design, Geneva: World Health Organization, 2019.
definitions, and metrics. Lancet 2012; 380: 2063–66.
38 UN. SDG Indicators—global indicator framework for the
17 Dunachie SJ, Day NP, Dolecek C. The challenges of estimating the Sustainable Development Goals and targets of the 2030 Agenda
human global burden of disease of antimicrobial resistant bacteria. for Sustainable Development. https://unstats.un.org/sdgs/
Curr Opin Microbiol 2020; 57: 95–101. indicators/indicators-list/ (accessed Jan 27, 2020).
18 Limmathurotsakul D, Dunachie S, Fukuda K, et al. Improving the 39 WHO. Indicator 3.d.2: percentage of bloodstream infections due
estimation of the global burden of antimicrobial resistant infections. to selected antimicrobial-resistant organisms. Geneva: World
Lancet Infect Dis 2019; 19: e392–98. Health Organization, 2020.
19 Ashley EA, Recht J, Chua A, et al. An inventory of supranational 40 UN. Report of the Inter-Agency and Expert Group on Sustainable
antimicrobial resistance surveillance networks involving low- and Development Goal indicators. New York, NY: United Nations
middle-income countries since 2000. J Antimicrob Chemother 2018; Economic and Social Council, 2020.
73: 1737–49.
41 WHO. The selection and use of essential medicines: report of the
20 Institute for Health Metrics and Evaluation. Global Burden of WHO Expert Committee, 2017 (including the 20th WHO Model
Disease Study 2019 (GBD 2019) data input sources tool. 2021. http:// List of Essential Medicines and the 6th WHO Model List of
ghdx.healthdata.org/gbd-2019/data-input-sources (accessed Essential Medicines for Children). Geneva: World Health
June 4, 2021). Organization, 2017.
21 Fullman N, Yearwood J, Abay SM, et al. Measuring performance on 42 Burki TK. Tuberculosis missing from WHO bacteria list.
the Healthcare Access and Quality Index for 195 countries and Lancet Respir Med 2017; 5: 252.
territories and selected subnational locations: a systematic analysis
from the Global Burden of Disease Study 2016. Lancet 2018; 43 Castro J. The WHO made a big mistake on tuberculosis. It must
391: 2236–71. fix it. STAT. 2017. https://www.statnews.com/2017/03/13/
tuberculosis-who-antibiotic-resistance/ (accessed April 23, 2021).
22 Browne A, Chipeta M, Haines-Woodhouse G, et al. Global antibiotic
consumption in humans, 2000 to 2018: a spatial modelling study. 44 Organisation for Economic Co-operation and Development,
Lancet Planet Health 2021; published online Nov 11. https://doi.org/ European Centre for Disease Prevention and Control.
10.1016/S2542-5196(21)00280-1. Antimicrobial resistance: tackling the burden in the European
Union. Briefing note for EU/EEA countries. Paris: OECD
23 Singer M, Deutschman CS, Seymour CW, et al. The third Publications, 2019.
international consensus definitions for sepsis and septic shock
(sepsis-3). JAMA 2016; 315: 801–10. 45 Jansen KU, Knirsch C, Anderson AS. The role of vaccines in
preventing bacterial antimicrobial resistance. Nat Med 2018;
24 Rudd KE, Johnson SC, Agesa KM, et al. Global, regional, and 24: 10–19.
national sepsis incidence and mortality, 1990-2017: analysis for the
Global Burden of Disease Study. Lancet 2020; 395: 200–11.
www.thelancet.com Vol 399 February 12, 2022 655
Articles
Estimating the number of infections caused by
antibiotic-resistant Escherichia coli and Klebsiella pneumoniae
in 2014: a modelling study
Elizabeth Temkin, Noga Fallach, Jonatan Almagor, Beryl Primrose Gladstone, Evelina Tacconelli, Yehuda Carmeli, on behalf of the DRIVE-AB
Consortium
Summary Lancet Glob Health 2018;
6: e969–79
Background The number of infections caused by resistant organisms is largely unknown. We estimated the number
of infections worldwide that are caused by the WHO priority pathogens third-generation cephalosporin-resistant and See Comment page e934
carbapenem-resistant Escherichia coli and Klebsiella pneumoniae.
Department of Epidemiology
Methods We calculated a uniform weighted mean incidence of serious infections caused by antibiotic-susceptible and Preventive Medicine,
E coli and K pneumoniae using data from 17 countries. Using this uniform incidence, as well as population sizes and Tel Aviv Sourasky Medical
country-specific resistance levels, we estimated the number of infections caused by third-generation cephalosporin- Center, Tel Aviv, Israel
resistant and carbapenem-resistant E coli and K pneumoniae in 193 countries in 2014. We also calculated interval (E Temkin DrPH, N Fallach MA,
estimates derived from changing the fixed incidence of susceptible infections to 1 SD below and above the weighted J Almagor PhD,
mean. We compared an additive model with combination models in which resistant infections were replaced by Prof Y Carmeli MD); Division of
susceptible infections. We distinguished between higher-certainty regions (those with good-quality data sources for Infectious Diseases,
resistance levels and resistance ≤30%), moderate-certainty regions (those with good-quality data sources for Department of Internal
resistance levels and including some countries with resistance >30%), and low-certainty regions (those in which Medicine 1, German Center for
good-quality data sources for resistance levels were unavailable for countries comprising at least 20% of the region’s Infection Research, University
population, regardless of resistance level). Hospital Tuebingen,
Tuebingen, Germany
Findings Using the additive model, we estimated that third-generation cephalosporin-resistant E coli and K pneumoniae (B P Gladstone PhD,
caused 6·4 million (interval estimate 3·5–9·2) bloodstream infections and 50·1 million (27·5–72·8) serious infections Prof E Tacconelli MD); and
in 2014; estimates were 5·5 million (3·0–7·9) bloodstream infections and 43·1 million (23·6–62·2) serious infections Sackler School of Medicine,
in the 25% replacement model, 4·6 million (2·5–6·6) bloodstream infections and 36·0 million (19·7–52·2) serious Tel Aviv University, Tel Aviv,
infections in the 50% replacement model, and 3·7 million (2·0–5·3) bloodstream infections and 28·9 million Israel (Prof Y Carmeli)
(15·8–41·9) serious infections in the 75% replacement model. Carbapenem-resistant strains caused 0·5 million
(0·3–0·7) bloodstream infections and 3·1 million (1·8–4·5) serious infections based on the additive model, Correspondence to:
0·5 million (0·3–0·7) bloodstream infections and 3·0 million (1·7–4·3) serious infections based on the 25% Dr Elizabeth Temkin, Department
replacement model, 0·4 million (0·2–0·6) bloodstream infections and 2·8 million (1·6–4·1) serious infections based of Epidemiology and Preventive
on the 50% replacement model, and 0·4 million (0·2–0·6) bloodstream infections and 2·7 million (1·5–3·8) serious Medicine, Tel Aviv Sourasky
infections based on the 75% replacement model. Medical Center, Tel Aviv 64239,
Israel
[email protected]
Interpretation To our knowledge, this study is the first to report estimates of the global number of infections caused
by antibiotic-resistant priority pathogens. Uncertainty stems from scant data on resistance levels from low-income
and middle-income countries and insufficient knowledge regarding resistance dynamics when resistance is high.
Funding Innovative Medicines Initiative.
Copyright © 2018 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND
4.0 license.
Introduction Antim icrobial Resistance Surveillance Network (EARS-
Net)3 and the Central Asian and Eastern European
Antimicrobial resistance has been recognised as a Surveillance of Antimicrobial Resistance (CAESAR),4
global public health crisis by organisations such as the track the proportion of isolates within a given species
UN and WHO.1,2 WHO’s Global Action Plan on that are resistant to an antibiotic, and not the number of
Antimicrobial Resistance calls for research to fill the infections caused by antimicrobial-resistant organisms,
knowledge gaps regarding the incidence of infections which are much harder data to collect. Likewise, in
caused by antimicrobial-resistant pathogens.2 Data on its 2014 global report on surveillance of antimicrobial
the number and incidence of infections caused by these resistance, WHO presented country-level data only on
organisms are scarce. Multicountry antimicrobial the proportion of resistant isolates.5 This proportion is
resistance surveillance systems, such as the European
www.thelancet.com/lancetgh Vol 6 September 2018 e969
Articles
Research in context caused by antimicrobial-resistant organisms in 1 year (2014).
We estimated the number of infections caused by
Evidence before this study third-generation cephalosporin-resistant and
We searched PubMed between Jan 1, 2000, and May 31, 2017, to carbapenem-resistant E coli and K pneumoniae in
identify studies that estimated the number of infections caused 193 countries. We presented methods to convert the
by antimicrobial-resistant organisms in a country, a region, proportion of resistance (a value that is often known through
or worldwide. A search using the terms (“antibiotic-resistant laboratory-based surveillance) into the number of resistant
bacteria” OR “multidrug-resistant bacteria” OR infections (data that are difficult to obtain and collected
“multidrug-resistant organisms”) AND (“estimates” OR infrequently). When possible, our calculations were based on
“estimating”) yielded three relevant studies. These studies country-specific data on the level of resistance, and we graded
estimated the number of bloodstream infections caused by the quality of these data. Because of the lack of empirical
third-generation cephalosporin-resistant Escherichia coli and evidence describing resistance dynamics when the level of
meticillin-resistant Staphylococcus aureus in 31 European resistance is high, we compared the results of four models:
countries in 2007, the number of inpatient infections caused by an additive model in which resistant infections supplement
seven antimicrobial-resistant bacteria in France in 2012, or the antibiotic-susceptible infections, and 25%, 50%, and
number of health-care-associated resistant infections in Finland 75% replacement models in which resistant infections
in 2010. We also knew of three relevant reports that are not supplant susceptible infections once resistance rises above
indexed in PubMed. The first, by the European Centre for Disease 30%. We validated our results for eight countries or states
Prevention and Control, used data from the European that collect incidence data based on mandatory reporting of
Antimicrobial Resistance Surveillance Network to estimate the resistant infections.
number of infections caused by six resistant species in
29 European countries in 2007. A 2013 report by the US Centers Implications of all the available evidence
for Disease Control and Prevention presented estimates of the According to our estimates, the number of infections caused
annual number of hospital-acquired infections in the USA caused by antibiotic-resistant E coli and K pneumoniae in 2014 was
by antimicrobial-resistant organisms, which included high: 50·1 million serious third-generation cephalosporin-
26 000 infections caused by third-generation resistant infections and 3·1 million serious carbapenem-
cephalosporin-resistant, and 9300 infections caused by resistant infections by the additive model, decreasing to 36·0
carbapenem-resistant, E coli and Klebsiella pneumoniae. The 2014 million and 2·8 million, respectively, in the 50% replacement
O’Neill Report commissioned by the UK Government estimated model. Accurate global estimates depend on strengthening
that 700 000 deaths per year worldwide are attributable to surveillance of antimicrobial resistance in low-income and
infections caused by six antimicrobial-resistant species, including middle-income countries. Laboratory-based antimicrobial
E coli and K pneumoniae; estimating the number of infections was resistance surveillance systems must incorporate
a step in the analysis, but those results were not published. epidemiological data to improve estimates of the incidence of
resistant infections.
Added value of this study
To our knowledge, this study is the first to report worldwide,
pathogen-specific estimates of the number of infections
important to clinicians choosing empirical therapy but Methods
does not provide the necessary information for policy Study overview
makers and antibiotic developers to act on the number We followed the Guidelines for Accurate and Transparent
and incidence of infections (ie, market size and, when Health Estimates Reporting.9 A list of the guidelines’
combined with associated morbidity and mortality, the elements and where they can be found in this manuscript
See Online for appendix 1 burden of disease).6 is in appendix 1. A detailed description of our methods is
Third-generation cephalosporin-resistant and carba- also presented in appendix 1.
penem-resistant Enterobacteriaceae appear in the highest We hypothesised that the incidence per 1000 population
category on WHO’s list of priority pathogens for research of serious infections caused by third-generation cephalo-
and development of new antibiotics.7 The US Centers for sporin-susceptible or carbapenem-susceptible E coli and
Disease Control and Prevention (CDC) classify third- K pneumoniae is similar across countries. This hypothesis
generation cephalosporin-resistant Entero-bacteriaceae is biologically plausible because most infections caused
as a serious threat and carbapenem-resistant Entero by antibiotic-susceptible Enterobacteriaceae result from
bacteriaceae as an urgent threat.8 We aimed to estimate translocation of indigenous gut flora to extraintestinal
the annual number and incidence per million population sites;10 we would not expect such events to vary greatly
of infections caused by third-generation cephalosporin- by country. We tested our hypothesis using country-
resistant and carbapenem-resistant Escherichia coli and level data. We then used this uniform incidence of
Klebsiella pneumoniae worldwide. susceptible infections, as well as population sizes and
e970 www.thelancet.com/lancetgh Vol 6 September 2018
Articles
country-specific resistance levels (ie, proportion of We estimated the number and incidence of
isolates resistant to a given antibiotic class), to calculate bloodstream infections and serious infections (which
the number and incidence of infections caused by third- include bloodstream infections) caused by third-
generation cephalosporin-resistant and carbapenem- generation cephalosporin-resistant and carbapenem-
resistant E coli and K pneumoniae. resistant E coli and K pneumoniae. We defined serious
infections as those that ideally should be treated in a
Whether antimicrobial resistance follows an additive hospital, while recognising that not all patients with
model, a replacement model, or a combination of both is infections live in an area with access to inpatient
uncertain. In an additive model, resistant infections care. For third-generation cephalosporin-resistant E coli
occur in addition to susceptible infections, whereas in and K pneumoniae, we also estimated the number
a replacement model, resistant infections supplant of outpatient infections; we assumed that all infections
susceptible infections. Previous studies11–13 have provided caused by carbapenem-resistant E coli and K pneum
evidence in support of the additive model but were oniae, for which oral therapy is rarely available, are
limited to countries with resistance levels lower than serious and would require inpatient treatment.
30%; thus, it is possible that replacement occurs in Our estimates encompassed all patients, regardless of
settings with higher levels of resistance. We compared age and sex.
four models: a fully additive model and models of 25%,
50%, and 75% replacement, beginning once resistance We produced country-level estimates for the 193 member
surpasses 30%. Figure 1 presents a conceptual overview states in the UN,14 which we grouped into 18 regions
of the analysis using the example of third-generation according to the UN’s classification system.15 The period
cephalosporin-resistant E coli. for our estimates was 2014: whenever available, we used
Inputs Outputs
For the 17 countries that met inclusion criteria: Using data from 17 countries that met inclusion criteria, estimate the mean
Number of hospital beds covered by surveillance system and number of incidence of serious infections caused by 3GC-susceptible E coli (a constant rate
hospital beds in country or reported proportion of country’s hospital beds to be applied to all countries); this rate remains fixed in the additive model and
covered by surveillance starts to fall in the replacement models once resistance is >30%
Number of Escherichia coli blood isolates submitted to surveillance system
Proportion of E coli blood isolates that are susceptible to 3GC
Proportion of all serious E coli infections that are bloodstream infections
(constant value for all countries)
Country population size
For all 193 countries: For 193 countries, estimate the number of serious infections caused by
Mean incidence of serious infections caused by 3GC-susceptible E coli 3GC-susceptible E coli using the additive model
Country population size
Number of serious infections caused by 3GC-susceptible E coli For each country, estimate the number of serious infections caused by
Proportion of E coli isolates that are resistant to 3GC 3GC-resistant E coli using the additive model
Mean incidence of serious 3GC-susceptible E coli infections Calculate the incidence of serious total (susceptible plus resistant) infections
when resistance is 31%; in a 100% replacement model this incidence remains
fixed once resistance is >30%
Calculate the incidence of serious total infections in models of 25%, 50%, and
75% replacement by weighting the average of the additive and 100%
replacement models
Incidence of serious total (susceptible plus resistant) E coli infections For countries with resistance >30%, estimate the number of serious
Proportion of E coli isolates that are resistant to 3GC in the country 3GC-resistant E coli infections using the 25%, 50%, and 75% replacement models
Number of serious 3GC-resistant E coli infections (by additive and replacement For each country, estimate the number of 3GC-resistant E coli bloodstream
models) infections
Proportion of all serious E coli infections that are bloodstream infections For each country, estimate the number of outpatient 3GC-resistant E coli
(constant value for all countries) infections
Number of serious 3GC-resistant E coli infections (by additive and replacement
models)
Ratio of outpatient to serious 3GC-resistant E coli infections (constant value for
all countries)
Figure 1: Conceptual overview of the data analysis method
3GC=third-generation cephalosporin.
www.thelancet.com/lancetgh Vol 6 September 2018 e971
Articles
Panel: Classification scheme to grade data sources for surveillance and the proportion that were susceptible,
countries’ levels of resistance the proportion of the country covered by surveillance, the
population size, and a constant value for the proportion
Grade 1 of serious infections that are bloodstream infections
Multicountry surveillance systems (eg, European (derived with data from our hospital).
Antimicrobial Resistance Surveillance Network, Central Asian
and Eastern European Surveillance of Antimicrobial The second major input was resistance levels in each
Resistance, Pan-American Health Organization antimicrobial country, which were gathered from various sources,
resistance report) or national surveillance systems including EARS-Net and CAESAR, national antimicrobial
resistance surveillance systems, WHO’s global report on
Grade 2 antimicrobial resistance surveillance,5 and scientific
National data obtained for the 2014 WHO antimicrobial articles. For countries with no such data, we used the
resistance report or privately sponsored multicountry median for the region. The process for identifying
surveillance systems (eg, Tigecycline Evaluation and sources is shown in appendix 1. We graded the quality of
Surveillance Trial) with more than three sites in the country, the sources using the scheme shown in the panel and
or data from the Center for Disease Dynamics, Economics and then classified regions as higher certainty, moderate
Policy if more than three sites certainty, and low certainty. Higher-certainty regions had
grade 1–2 data on level of resistance for countries
Grade 3 comprising at least 80% of the region’s population, and
Published articles with data collected from more than three no country within the region had resistance higher than
sites in the country 30%; for these regions, the additive and replacement
models were identical. We classified moderate-certainty
Grade 4 regions as those in which grade 1–2 data on resistance
Published articles with data collected from three sites or fewer levels were available for countries comprising at least
in the country or privately sponsored multicountry 80% of the region’s population and in which some
surveillance systems with three sites or fewer in the country countries had resistance higher than 30%. Low-certainty
regions were those in which grade 1–2 data were
Grade 5 unavailable for countries comprising at least 20% of the
Any data source with sample size of less than 100 isolates or region’s population, regardless of resistance level.
sample size not reported, or all data collected before 2011
Data analysis
Grade 6 We estimated the mean incidence of serious infections
No data available, value imputed caused by third-generation cephalosporin-susceptible and
carbapenem-susceptible E coli and K pneumoniae. We first
2014 data on levels of resistance and World Bank estimated the annual number of antibiotic-susceptible
population estimates for 2014.16 bloodstream infections in each of the 17 countries by
dividing the number of susceptible isolates reported to
Data inputs surveillance systems by the proportion of the country that
Descriptions of the process for identifying and accessing was represented in surveillance. Next, we divided the
all data inputs, of inclusion and exclusion criteria, and of number of susceptible bloodstream infections by the
the methods used to imput missing values are in constant value for the proportion of serious infections
appendix 1. The first data inputs were those required to that are bloodstream infections (0·12 for E coli and
estimate, for each of the four species–resistance pairs, 0·17 for K pneumoniae) to estimate the annual number of
the mean incidence of serious infections caused by serious infections caused by susceptible strains. We
susceptible organisms, to be applied to all countries. We expressed this value as incidence per 1000 population. We
identified countries with surveillance data, mainly from then calculated the weighted mean of the incidence of
EARS-Net, that allowed estimation of the proportion of serious susceptible infections in the 17 countries to
the country’s hospital beds covered by surveillance. We generate the value to be applied to all countries. Weighting
excluded countries in which surveillance covered less was done according to the proportion of the country that
than 30% of the country’s hospital beds. EARS-Net is was covered by surveillance. We displayed each country’s
based on blood isolates, which was advantageous for the incidence and weighted mean incidence in funnel plots
purpose of this study because such isolates represent to see if values clustered around the mean, which would
serious infections rather than colonisation or outpatient support our primary hypothesis that the incidence of
infections. For each of the 17 countries that met inclusion serious susceptible infections is similar in all countries.
criteria, we estimated the incidence of serious infections We displayed upper and lower control limits that were
caused by third-generation cephalosporin-susceptible one or two SEs from the mean. For all subsequent
and carbapenem-susceptible E coli and K pneumoniae calculations, we used one SD below and above the
using five inputs: the number of isolates submitted to weighted mean incidence as our interval estimate (ie, we
e972 www.thelancet.com/lancetgh Vol 6 September 2018
Articles
AB
10
Incidence per 1000 population per year 9
8 Malta Denmark
Finland
7 Malta Sweden Denmark Sweden
Austria Israel
6 Switzerland Slovenia Finland Switzerland Slovenia Ireland
5 Portugal Czech Republic Portugal
4 Czech Republic Austria Ireland Slovakia
Cyprus
Estonia Israel Estonia
Latvia
3 Slovakia Serbia
2 Cyprus Hungary Hungary
1 Latvia
0 Serbia
C D
Incidence per 1000 population per year 1·8 Czech Republic Malta Israel
1·6
1·4 Malta Portugal
1·2
1·0 Denmark Slovakia Denmark
0·8 Cyprus
0·6 Israel Estonia Slovenia
0·4 Czech Republic Portugal Sweden Austria Switzerland Sweden Austria
0·2 Latvia Finland
Cyprus Switzerland Finland
0 Ireland Ireland
0·3 Estonia Slovenia
Slovakia Latvia Hungary
Hungary Serbia
Serbia 1·0 0·3 0·4 0·5 0·6 0·7 0·8 0·9 1·0
Proportion of country represented in surveillance
0·4 0·5 0·6 0·7 0·8 0·9
Proportion of country represented in surveillance
Figure 2: Annual incidence of serious infections caused by antibiotic-susceptible Escherichia coli and Klebsiella pneumoniae in 17 countries
Black solid lines are the weighted means; green lines are 1 SE and blue lines are 2 SEs from the weighted means. (A) Third-generation cephalosporin-susceptible E coli.
(B) Carbapenem-susceptible E coli. (C) Third-generation cephalosporin-susceptible K pneumoniae. (D) Carbapenem-susceptible K pneumoniae.
set a middle, low, and high value for the incidence of outpatient infections, we used a constant ratio of
serious susceptible infections). outpatient to serious infections derived from Swiss
surveillance data (4·0 for third-generation cephalosporin-
The major steps to calculate the number of infections resistant and 2·1 for carbapenem-resistant infections).17
caused by resistant strains were as follows. First, for the
additive model, we calculated the number of serious For all indicators, interval estimates were based on
resistant infections in each country using the fixed changing only the incidence of serious susceptible
incidence of serious susceptible infections, the popu infections. We did not vary three other important
lation size, and the country-specific resistance level. parameters: the proportion of isolates that were resistant
Second, we modelled 100% replacement, in which every in each country, the proportion of serious infections that
resistant infection replaces a susceptible infection once were bloodstream infections, or the ratio of outpatient to
resistance is above 30%. At that point, the incidence of all serious infections.
serious infections (resistant plus susceptible) remains
fixed at a value set slightly above that reached by the Role of the funding source
additive model at 30% resistance. To calculate the
incidence of all serious infections for the 25%, 50%, and The funder had no role in study design, data collection,
75% replacement models, we weighted the average of the data analysis, data interpretation, or writing of the
additive and 100% replacement models accordingly. manuscript. The corresponding author had full access
Using the incidence of all serious infections, the to all of the data in the study and the final responsibility
population size, and the country-specific resistance level, for the decision to submit for publication.
we calculated the number of serious resistant infections
for each country. To calculate the number of bloodstream Results See Online for appendix 2
infections caused by resistant strains, we multiplied the
number of serious resistant infections by the constant The annual incidence of serious infections caused by third-
value for the proportion of serious infections that are generation cephalosporin-susceptible and carbapenem-
bloodstream infections. To calculate the number of susceptible E coli and K pneumoniae in the 17 countries that
met inclusion criteria are shown in figure 2 and appendix 2.
For all four species–resistance pairs, most countries fell
www.thelancet.com/lancetgh Vol 6 September 2018 e973
Articles
Additive model 100% replacement model The effect of the additive, replacement, and
combination models on the estimated incidence of
Incidence of Incidence of Incidence of Incidence of Incidence of Incidence of resistant infections are presented in table 1 and figure 3,
susceptible resistant all susceptible resistant all using third-generation cephalosporin-resistant E coli as
infections infections infections infections infections infections an example. As resistance levels increase, differences
between the models in estimates of the number of
10% 4·26 0·47 4·73 4·26 0·47 4·73 resistant infections become more pronounced. When
resistance is 40%, estimates from the additive model are
20% 4·26 1·07 5·33 4·26 1·07 5·33 12% higher than those from the 75% replacement model;
when resistance reaches 77%, estimates from the additive
30% 4·26 1·83 6·09 4·26 1·83 6·09 model are twice those from the 75% replacement model.
Thus, uncertainty in estimating the number of resistant
40% 4·26 2·84 7·10 3·67 2·45 6·12 infections is much greater in countries with very high
resistance.
50% 4·26 4·26 8·52 3·06 3·06 6·12
Table 2 presents estimates of the number of bloodstream
60% 4·26 6·39 10·65 2·45 3·67 6·12 and serious infections from the additive and combination
models, by region for carbapenem-resistant K pneumoniae
Data are incidence per 1000 population. In the additive model, the incidence of serious susceptible infections remains and by certainty level for third-generation cephalosporin-
fixed and resistant infections occur in addition to susceptible infections. In the 100% replacement model, once the resistant E coli and K pneumoniae and carbapenem-
level of resistance is greater than 30%, the incidence of total serious infections remains fixed as resistant infections resistant E coli. Estimates by region for third-generation
replace susceptible infections. The incidence of resistant infections is higher in the additive model than in the 100% cephalosporin-resistant E coli and K pneumoniae and
replacement model after 30% resistance, and the gap widens as resistance rises. carbapenem-resistant E coli are presented in appendix 1.
Estimates by country for cephalosporin-resistant E coli
Table 1: Comparison of the additive and 100% replacement models by percentage of Escherichia coli and K pneumoniae and carbapenem-resistant E coli and
isolates that are resistant to third-generation cephalosporins K pneumoniae are presented in appendix 2.
Resistant infections per 1000 population100 Additive For carbapenem-resistant K pneumoniae (for which
25% replacement few countries had resistance >30%), the worldwide point
estimate of serious infections was 2·1 million (interval
90 50% replacement estimate 1·1–3·0) with the additive model versus 1·6
75% replacement million (0·9–2·3) with the 75% replacement model
(table 2); the difference was driven by high levels of
80 100% replacement resistance in India. For third-generation cephalosporin-
resistant E coli (for which high resistance was more
70 common), the difference between models was wider
than for carbapenem-resistant K pneumoniae: from 43·1
60 million (23·5–62·7) in the additive model to 24·9 million
(13·5–36·2) in the 75% replacement model. Country-
50 level estimates of outpatient infections are shown in
appendix 3.
40
We identified six countries and two US states with
30 mandatory reporting systems for third-generation
cephalosporin-resistant or carbapenem-resistant Entero-
20 bacteriaceae (table 3). Those systems differ from
laboratory-based systems such as EARS-Net in that they
10 count all cases (defined by different systems as
bloodstream infections only, inpatient infections only, or
0 all infections) throughout the country (or US state), and
0 10 20 30 40 50 60 70 80 90 100 not only the proportion of resistant isolates in a sample.
Resistance level (%) We compared our estimates with the observed number
of infections in 2014 (table 3). In 21 of 32 instances,
Figure 3: Estimated incidence of serious infections caused by third-generation cephalosporin-resistant the observed value definitely or probably fell within
Escherichia coli with the various models our estimated range and, in six other instances, we
In the combination models, replacement begins once the level of resistance is greater than 30%. underestimated or overestimated by fewer than 30 cases
per year. Only one datapoint involved a resistance level of
within 1 SE of the weighted mean and all fell within 2 SEs greater than 30% (third-generation cephalosporin-
of the weighted mean, except for Malta and Israel where resistant K pneumoniae in Israel, with 54% resistance).
the annual incidence of infections caused by carbapenem- The observed number of bloodstream infections caused
susceptible K pneumoniae was more than 2 SEs higher
than the weighted mean. The plots confirmed our
hypothesis of a similar incidence of serious susceptible
infections in these countries, allowing use of the weighted
mean incidence in subsequent calculations.
Our grading of sources reporting levels of resistance
in individual countries is presented, by region, in
appendix 1. Grade 1–2 data were available for about half
of the world’s population, and were rare or absent for
countries in the Caribbean, all regions of Asia except for
east Asia, and all regions of Africa except for southern
Africa.
e974 www.thelancet.com/lancetgh Vol 6 September 2018
Articles
Bloodstream infections (in thousands) Serious infections (in thousands)
Additive 25% replacement 50% replacement 75% replacement Additive 25% replacement 50% replacement 75% replacement
Carbapenem-resistant Klebsiella pneumoniae*
World total 351 (195–506) 324 (180–468) 297 (165–430) 270 (150–392) 2062 (1146–2978) 1905 (1058–2753) 1747 (971–2529) 1590 (884–2305)
24 (13–34) 24 (13–34) 138 (77–200) 138 (77–200) 138 (77–200) 138 (77–200)
Higher certainty 24 (13–34) 24 (13–34) 0·3 (0·2–0·4) 0·3 (0·2–0·4) 2 (1–2) 2 (1–2) 2 (1–2)
0·1 (0·1–0·2) 0·1 (0·1–0·2) 0·8 (0·5–1·2) 0·8 (0·5–1·2) 2 (1–2) 0·8 (0·5–1·2)
Western Europe 0·3 (0·2–0·4) 0·3 (0·2–0·4) 0·8 (0·5–1·2)
4 (2–6) 4 (2–6)
Northern 0·1 (0·1–0·2) 0·1 (0·1–0·2)
Europe 2 (1–3) 2 (1–3)
0·3 (0·2–0·5) 0·3 (0·2–0·5)
Northern 4 (2–6) 4 (2–6) 16 (9–23) 16 (9–23) 25 (14–36) 25 (14–36) 25 (14–36) 25 (14–36)
America 36 (20–52) 34 (19–50)
Central America 2 (1–3) 2 (1–3) 8 (5–12) 8 (4–11) 14 (8–20) 14 (8–20) 14 (8–20) 14 (8–20)
2 (1–3) 2 (1–3) 2 (1–3) 2 (1–3)
Southern Africa 0·3 (0·2–0·5) 0·3 (0·2–0·5) 7 (4–10) 6 (4–9)
21 (11–30) 20 (11–29) 95 (53–138) 95 (53–138) 95 (53–138) 95 (53–138)
Eastern Asia 16 (9–23) 16 (9–23) 238 (132–344) 212 (118–308) 226 (125–326) 218 (121–315) 210 (117–304) 202 (112–293)
6 (3–8) 6 (3–8)
Moderate 38 (21–55) 37 (21–54)
certainty 216 (120–312) 190 (106–276)
Southern 9 (5–13) 9 (5–13) 3 (2–4) 3 (2–4) 54 (30–78) 51 (28–74) 48 (27–70) 45 (25–66)
Europe 3 (2–4) 3 (2–4)
0·4 (0·2–0·5) 0·4 (0·2–0·5)
Eastern Europe 8 (4–11) 7 (4–11) 8 (4–11) 8 (4–11) 45 (25–66) 43 (24–62) 41 (23–59) 38 (21–55)
2 (1–4) 2 (1–4) 127 (70–183) 124 (69–179) 121 (67–175) 118 (66–172)
South America 22 (12–31) 21 (12–30) 0·05 0·05 1698 (944–2452) 1549 (861–2239) 1399 (778–2025) 1250 (695–1812)
(0·03–0·8) (0·03–0·8) 33 (19–48) 33 (19–48) 33 (19–48)
Low certainty 289 (160–417) 263 (146–381) 0·05 0·05 33 (19–48)
(0·03–0·7) (0·03–0·7)
Southeastern 6 (3–8) 6 (3–8)
Asia 3718 2986
(2024–5410) (1626–4345)
South-central 266 (148–385) 241 (134–348) 1567 (871–2263) 1418 (788–2050) 1269 (705–1837) 1120 (623–1624)
Asia 56 (31–82) 56 (31–82)
1025 923 (503–1343)
Western Asia 3 (2–4) 3 (2–4) (558–1491) 17 (10–25) 17 (10–25) 17 (10–25) 17 (10–25)
2637 2007 17 (10–25) 17 (10–25) 17 (10–25) 17 (10–25)
Eastern Africa 3 (2–4) 3 (2–4) (1436–3837) (1092–2920) 2 (1–3) 2 (1–3) 2 (1–3) 2 (1–3)
45 (25–66) 45 (25–66) 45 (25–66) 44 (25–64)
Middle Africa 0·4 (0·2–0·5) 0·4 (0·2–0·5) 851 677 15 (8–21) 15 (8–21) 15 (8–21) 15 (8–21)
(486–1214) (387–966) 0·3 0·3 0·3 0·3
Northern Africa 8 (4–11) 8 (4–11) (0·2–0·5) (0·2–0·5) (0·2–0·5) (0·2–0·5)
8 (5–12) 8 (5–12) 0·3 0·3 0·3 0·3
Western Africa 2 (1–4) 2 (1–4) 227 (130–325) 196 (112–279) (0·2–0·4) (0·2–0·4) (0·2–0·4) (0·2–0·4)
Caribbean 0·05 0·05 615 (351–878) 474 (271–675)
(0·03–0·8) (0·03–0·8)
130 (76–184) 130 (76–184)
Oceania 0·05 0·05 13 (7–18) 13 (7–18)
(0·03–0·7) (0·03–0·7) 117 (69–166) 117 (69–166)
Third-generation cephalosporin-resistant Escherichia coli
World total 5173 4447 43 107 37 060 30 985 24 887
(2817–7529) (2423–6476) (23 476–62 737) (20 194–53 966) (16 871–45 087) (13 548–36 207)
Higher certainty 56 (31–82) 56 (31–82) 469 (255–682) 469 (255–682) 469 (255–682) 469 (255–682)
10 167 9364 8542 7696
Moderate 1220 1124 (5537–14 797) (5112–13 661) (4650–12 428) (4189–11 195)
certainty (664–1776) (613–1639) 31 784 27 227 21 975
(17 309–46 258) (14 826–39 623) (11 965–31 976) 16 723
Low certainty 3896 3267 (9104–24 329)
(2122–5671) (1779–4755)
Third-generation cephalosporin-resistant K pneumoniae
World total 1197 1024 7042 6023 5003 3985
(684–1710) (585–1462) (4024–10 061) (3442–8601) (2859–7141) (2277–5680)
Higher certainty 8 (5–12) 8 (5–12) 48 (28–69) 48 (28–69) 48 (28–69) 48 (28–69)
1709 (977–2442) 1527 (872–2180) 1338 (765–1910) 1151 (658–1640)
Moderate 291 (166–415) 260 (148–371)
certainty
Low certainty 898 (513–1283) 756 (432–1080) 5285 (3020–7550) 4448 (2542–6353) 3617 (2067–5162) 2785 (1591–3971)
Carbapenem-resistant E coli
World total 130 (76–184) 130 (76–184) 1084 (634–1533) 1084 (634–1533) 1084 (634–1533) 1084 (634–1533)
106 (62–150) 106 (62–150) 106 (62–150) 106 (62–150)
Higher certainty 13 (7–18) 13 (7–18) 978 (572–1383) 978 (572–1383) 978 (572–1383) 978 (572–1383)
Low certainty 117 (69–166) 117 (69–166)
Data in parentheses are interval estimates derived from changing the fixed incidence of susceptible infections to 1 SD below and above the weighted mean. Estimates were the same for all models when
resistance was ≤30%. *Data are also shown by UN region.
Table 2: Estimates of resistant bloodstream infections and serious infections in 2014, by model, WHO priority pathogen, and certainty level
www.thelancet.com/lancetgh Vol 6 September 2018 e975
Articles
Observed values, 2014 Estimated values, 2014 (additive model) Summary
Data type Infections Data type Estimate*
Ireland3,18 Invasive isolates (blood or CSF) 357 Bloodstream infections 313 (170–455) In range
3GC-resistant Escherichia coli Invasive isolates (blood or CSF) 46 Bloodstream infections 74 (42–106) In range
3GC-resistant Klebsiella pneumoniae Invasive isolates (blood or CSF) Bloodstream infections 4 (2–6) In range
Carbapenem-resistant K pneumoniae 4
Finland19 Bloodstream infections Bloodstream infections 156 (85–227) Underestimate
3GC-resistant E coli Bloodstream infections 232 Bloodstream infections 19 (11–28) In range
3GC-resistant K pneumoniae All infection types; inpatients and outpatients; includes 20 All infection types; 6511 (3546–9475) Probably in range
3GC-resistant E coli unknown percentage of screening cultures† 4190 inpatients and outpatients
All infection types; inpatients and outpatients; includes All infection types; 353 (201–504) Probably in range
3GC-resistant K pneumoniae unknown percentage of screening cultures† 312 inpatients and outpatients
Denmark3,20 Bloodstream infections (first infection per patient per year) 314 Bloodstream infections 244 (133–355) In range
3GC-resistant E coli Bloodstream infections (first infection per patient per year) 75 Bloodstream infections 59 (34–84) In range
3GC-resistant K pneumoniae Inpatient blood and urine (first infection per patient per year) All infection types; inpatients only Probable
3GC-resistant E coli 2953 2034 (1108–2960) underestimate
All infection types; inpatients only Underestimate
3GC-resistant K pneumoniae Inpatient blood and urine (first infection per patient per year) 520 Bloodstream infections 347 (198–496) Underestimate
Carbapenem-resistant E coli Bloodstream infections 3 Bloodstream infections 1 (0–1) In range
Carbapenem-resistant K pneumoniae Bloodstream infections 4 3 (2–5)
Sweden3,21
3GC-resistant E coli All infection types; inpatients and outpatients; includes 8161 All infection types; inpatients and 13 417 Overestimate
about 28% screening cultures† outpatients (7307–19 526) Overestimate
3GC-resistant K pneumoniae All infection types; inpatients and outpatients; includes about 668 All infection types; inpatients and 1166 (666–1666)
28% screening cultures† outpatients
ESBL-producing Enterobacteriaceae Bloodstream infections 520 Bloodstream infections; E coli and 396 (212–560) Probably in range
K pneumoniae only
Carbapenem-resistant All infection types 19 All infection types; E coli and 32 (19–46) Probably in range
Enterobacteriaceae K pneumoniae only
95
Norway3,22 26
4
3GC-resistant E coli Bloodstream infections (6 months only) 6 Bloodstream infections (12 months) 174 (95–253) Probably in range
Bloodstream infections (12 months) 43 (24–61) Probably in range
3GC-resistant K pneumoniae Bloodstream infections (9 months only) 1526 All infection types In range
1213 All infection types 7 (4–10) In range
Carbapenem-resistant E coli All infection types 5 (3–7)
11
Carbapenem-resistant K pneumoniae All infection types 113
Israel‡ 249
2470
3GC-resistant E coli Bloodstream infections 372 Bloodstream infections 1757 (957–2558) In range
Bloodstream infections 1033 (590–1476) In range
3GC-resistant K pneumoniae Bloodstream infections 69 Bloodstream infections In range
242 Bloodstream infections 14 (8–20) Underestimate
Carbapenem-resistant E coli Bloodstream infections 61 (34–88)
2
Carbapenem-resistant K pneumoniae Bloodstream infections 14
New York, USA23,24
Carbapenem-resistant E coli All infection types All infection types 386 (226–546) In range
All infection types 2941 (1634–4248) In range
Carbapenem-resistant K pneumoniae All infection types Bloodstream infections 546 (305–788) In range
Carbapenem-resistant E coli and Bloodstream infections
K pneumoniae
Maryland, USA25§
Carbapenem-resistant E coli All infection types All infection types 0 Underestimate
All infection types 166 (92–240) Underestimate
Carbapenem-resistant K pneumoniae All infection types Bloodstream infections Underestimate
Bloodstream infections 0 Overestimate
Carbapenem-resistant E coli Bloodstream infections (first infection per patient per year) (first infection per patient per year) 28 (16–41)
Carbapenem-resistant K pneumoniae Bloodstream infections (first infection per patient per year)
3GC=third-generation cephalosporin. ESBL=extended-spectrum β-lactamase. *Data in parentheses are interval estimates derived from changing the fixed incidence of susceptible infections to 1 SD below and
above the weighted mean.†Screening cultures detect colonisation, not infection. ‡Data for Israel were obtained from the National Center for Infection Control, Israel, and are unpublished. §The observed data
came from personal communication with David Blythe and Elisabeth Vaeth at Maryland Department of Health and Mental Hygiene.
Table 3: Comparison of estimated number of infections and number observed by surveillance systems based on mandatory reporting
e976 www.thelancet.com/lancetgh Vol 6 September 2018
Articles
by third-generation cephalosporin-resistant K pneumoniae inclusion of all species of carbapenem-resistant Entero
in Israel (1213) fell within the ranges estimated by the bacteriaceae. Colomb-Cotinat and colleagues31 estimated
additive (1033 [interval estimate 590–1476]) and 25% the number of inpatient infections caused by multidrug-
(946 [541–1351]) and 50% (860 [491–1226]) replacement resistant bacteria in France in 2012, and reported estimates
models; the estimate from the 75% replacement model for infections caused by third-generation cephalosporin-
was lower than the observed value (773 [442–1101]). resistant K pneumoniae (16 314 infections) and carbapenem-
resistant K pneumoniae (602 infections) that overlap
Discussion with ours (estimates for France from this study are in
appendix 2). By contrast, our estimate for infections caused
In 2014, the O’Neill report estimated that 700 000 deaths per by third-generation cephalosporin-resistant E coli was
year are attributable to antimicrobial-resistant infections slightly lower than Colomb-Cotinat and colleagues’
(including malaria, HIV, and tuberculosis).26 Although estimate (50 916 infections).
estimation of the number of resistant infections was a step
in this calculation, those methods and results were not We recognise the many potential sources of bias or
published.26,27 We aimed to estimate the global number of uncertainty in our estimations. First, use of a single
infections caused by third-generation cephalosporin- incidence of serious infections caused by susceptible
resistant and carbapenem-resistant E coli and K pneumoniae, E coli and K pneumoniae for all countries might have
and to provide an explicit assessment of the assumptions been an oversimplification. Although the funnel plots
and the quality of data on which the estimates were based. supported our hypothesis of a uniform incidence, data
were available only from high-income and upper-middle-
The first report to present a multicountry estimate of income countries. We confirmed that estimates based on
the incidence of antibiotic-resistant infections was The this fixed incidence were generally accurate for sites with
Bacterial Challenge: Time to React, issued by the incidence data from mandatory reporting, but no such
European Centre for Disease Prevention and Control comparisons were available for low-income or lower-
(ECDC) in 2009.28 Using data from EARS-Net, the report middle-income countries. The incidence of susceptible
estimated the number of inpatient infections caused infections in these countries might be higher (eg, because
by resistant organisms in 29 European countries in of poorer sanitation) or lower (eg, because of differences
2007, including 32 500 by third-generation cephalosporin- in population age structure) than our fixed incidence.
resistant E coli and 18 900 by third-generation cephalo
sporin-resistant K pneumoniae. The methods differed Second, we based the calculation of the incidence of
from ours in that the ECDC calculated the number of serious susceptible infections in each country on the
resistant bloodstream infections in each country using number of isolates submitted for surveillance by
the number of resistant isolates submitted to EARS-Net 17 countries. This number is subject to ascertainment
and the proportion of the country that was covered by bias stemming from differences in countries’ practices
surveillance. By contrast, we defined a constant mean regarding the frequency of taking blood cultures. Indeed,
incidence of susceptible infections that enabled when we used EARS-Net data32 to estimate the number of
generation of estimates for all countries, based solely on blood-culture sets processed in participating laboratories
the level of resistance. per 1000 population covered by surveillance, the ratio
was two times lower for Latvia and Hungary (which were
On a national scale, in 2013, the US CDC issued consistently near the bottom of our funnel plots) than for
estimates of the number of hospital-acquired infections other countries. Notably, most countries with a high
caused by antibiotic-resistant bacteria based on data from a surveillance coverage are wealthy countries where we
point-prevalence study done in acute-care hospitals in ten would expect high ascertainment of infections. The
US states.29 The US CDC’s estimates were lower than the incidence of serious susceptible infections in these
estimates in this study—for example, the US CDC countries was often higher than the mean, suggesting
estimated that 9300 hospital-acquired infections were that the mean incidence of susceptible infections in our
caused by carbapenem-resistant E coli and K pneumoniae, study (and thus the calculated number of resistant
whereas we estimated that 33 994 (interval estimate 19 163– infections) might be an underestimate.
48 824) inpatient infections were caused by these
microorganisms. In New York, USA, half of all infections Third, there were potential sources of bias regarding
caused by carbapenem-resistant E coli and K pneumoniae our estimates for each country. Grade 1–2 quality data on
in 2014 were hospital acquired.23 If the same is true levels of resistance were available for only 31% (for
throughout the USA, then our point estimate of no carbapenem-resistant E coli) to 43% (for third-generation
socomial carbapenem-resistant infections (16 997) is cephalosporin-resistant E coli) of countries. We imputed
almost twice as high as the US CDC’s. Another study30 resistance levels for 31–41% of countries. Additionally,
based on data from electronic medical records from 192 US samples might not have been representative of the
hospitals reported an estimate for carbapenem-resistant whole country because resistance can vary by region,
E coli and K pneumoniae that was slightly higher than our and tertiary care hospitals, which are the source
point estimate but within our range: 42 852 infections in of most academic reports, might have higher levels of
the USA in 2014. The difference might be due to their resistance than would community hospitals (although a
www.thelancet.com/lancetgh Vol 6 September 2018 e977
Articles
US study33 found no such difference). The possibility of the incidence of infections caused by the susceptible
sampling bias is particularly relevant to India, whose form of these pathogens differs substantially between
resistance estimates (82% in the case of third-generation locations.35,36 To generate estimates of the incidence and
cephalosporin-resistant E coli) were obtained from small number of infections caused by these organisms, and to
samples of isolates. Because India comprises 18% of the improve our estimates for E coli and K pneumoniae,
world’s population, overestimation of resistance in India surveillance systems must gather both numerator data
would substantially inflate the world totals; for this on the number of resistant infections per time period
reason, we distinguished between higher-certainty, and population-level denominator data. Two new WHO
moderate-certainty, and low-certainty regions when initiatives to strengthen and standardise antimicrobial
presenting global estimates. Moreover, some data on resistance surveillance worldwide, the Global Antim i
resistance levels were collected before 2014, which might crobial Resistance Surveillance System37 and ESBL Ec
have led to underestimation of the number of resistant Tricycle,38 might help to provide the necessary additional
infections in countries where levels of resistance are data. Improving the accuracy of global estimates of
rapidly increasing. Furthermore, anatomical sources of the number of antibiotic-resistant infections will enhance
isolates varied, although the bias that this variation efforts to prioritise infection prevention activities, limit
introduces might be minor given that antimicrobial the spread of antibiotic resistance, and develop new
resistance monitors in Sweden reported that resistance antibiotics.
among E coli isolates from outpatient urine and inpatient
blood cultures was quite similar.34 Contributors
YC and ETe conceived the study and developed the models. ETe and
Finally, we note that our 25%, 50%, and 75% replacement NF prepared the first draft. All other authors collected data and reviewed
models are, indeed, models. No empirical evidence exists results. All authors reviewed the final manuscript.
to validate which, if any, of these models approximates
resistance dynamics in countries where resistance levels Declaration of interests
are high. High resistance in the most populated countries, YC reports grants or personal fees from MSD, AstraZeneca, DaVoltera,
particularly China (for third-generation cephalosporin- Intercell AG, Allecra Therapeutics, BioMerieux SA, Rempex
resistant E coli) and India (for all but carbapenem-resistant Pharmaceuticals, Nariva, Achoagen, Roche, Pfizer, and Shionogi.
E coli), drives the wide variation in our global estimates. All other authors declare no competing interests.
Although we acknowledge these limitations, we believe
that we have generated the best possible estimates with Acknowledgments
the available data. The research leading to these results has received support from the
Innovative Medicines Initiative Joint Undertaking under grant agreement
In their article, de Kraker and colleagues27 were critical number 115618 (Driving re-investment in research and development and
of the methods used to estimate antimicrobial-resistant responsible antibiotic use [DRIVE-AB]), resources of which are composed
infections in the O’Neill and ECDC reports.26,28 Some of of financial contribution from the European Union’s Seventh Framework
these weaknesses, such as the use of data that might not Programme (FP7/2007-2013) and European Federation of Pharmaceutical
be representative or possible error in estimating Industries and Associations companies’ in kind contribution. This work
infections in all anatomical sites from surveillance based does not necessarily represent the view of all DRIVE-AB partners.
on blood cultures, might also apply to our study.
However, we have overcome other limitations cited by References
de Kraker and colleagues. First, O’Neill26 estimated the 1 UN. Political declaration of the high-level meeting of the General
number of resistant infections in countries not in EARS-
Net by applying the EARS-Net mean incidence, whereas Assembly on antimicrobial resistance. 2016. http://www.un.org/en/
we searched other sources to identify country-specific ga/search/view_doc.asp?symbol=A/RES/71/3 (accessed Feb 9, 2018).
estimates of resistance. For highly populated countries
such as China, Russia, and Brazil, we cited grade 1 data 2 WHO. Global action plan on antimicrobial resistance. 2015.
sources that reported resistance levels higher than those http://www.who.int/antimicrobial-resistance/global-action-plan/en
in most EARS-Net countries, suggesting that O’Neill’s (accessed Feb 9, 2018).
methods produced underestimates. Second, de Kraker
and colleagues observed that the reports by the ECDC 3 European Centre for Disease Prevention and Control. Surveillance
and O’Neill did not clearly state the uncertainties Atlas of Infectious Diseases. http://atlas.ecdc.europa.eu/public/
inherent in each step of their calculations. We have index.aspx (accessed Feb 9, 2018).
stated uncertainties and potential sources of bias, and
have supplied our calculations for scrutiny and revision. 4 WHO. Central Asian and Eastern European Surveillance of
Antimicrobial Resistance (CAESAR). http://www.euro.who.int/en/
Our approach of using a uniform incidence of serious health-topics/disease-prevention/antimicrobial-resistance/
susceptible infections to estimate the number of resistant publications/2015/central-asian-and-eastern-european-surveillance-of-
infections is probably not suitable for other resistant antimicrobial-resistance.-annual-report-2014 (accessed July 29, 2018).
pathogens such as carbapenem-resistant Acinetobacter
and Pseudomonas and vancomycin-resistant enterococci; 5 WHO. Antimicrobial resistance: global report on surveillance, 2014.
Geneva, 2014 http://apps.who.int/iris/bitstream/10665/112642/1/
9789241564748_eng.pdf?ua=1 (accessed Feb 9, 2018).
6 Schwaber MJ, De-Medina T, Carmeli Y. Epidemiological
interpretation of antibiotic resistance studies—what are we
missing? Nat Rev Microbiol 2004; 2: 979–83.
7 WHO. Global priority list of antibiotic-resistant bacteria to guide
research, discovery, and development of new antibiotics. 2017
http://www.who.int/medicines/publications/global-priority-list-
antibiotic-resistant-bacteria/en (accessed July 30, 2018).
8 Centers for Disease Control and Prevention. Antibiotic/
antimicrobial resistance: biggest threats. 2017. https://www.cdc.gov/
drugresistance/biggest_threats.html (accessed Feb 9, 2018).
9 Stevens GA, Alkema L, Black RE, et al. Guidelines for Accurate and
Transparent Health Estimates Reporting: the GATHER statement.
Lancet 2016; 388: e19–23.
e978 www.thelancet.com/lancetgh Vol 6 September 2018
Articles
10 Russo TA, Johnson JR. Medical and economic impact of 25 Maryland Department of Health and Mental Hygiene. Status report:
extraintestinal infections due to Escherichia coli: focus on an antibiotic resistance in Maryland. Addressing the urgent threats.
increasingly important endemic problem. Microbes Infect 2003; http://phpa.dhmh.maryland.gov/IDEHASharedDocuments/Status
5: 449–56. Report - Antibiotic Resistance in Maryland.pdf (accessed
Feb 9, 2018).
11 Ammerlaan HSM, Harbarth S, Buiting AGM, et al. Secular trends in
nosocomial bloodstream infections: antibiotic-resistant bacteria 26 O’Neill J. Review on Antimicrobial Resistance. Antimicrobial
increase the total burden of infection. Clin Infect Dis 2013; 56: 798–805. resistance: tackling a crisis for the health and wealth of nations.
2014. https://amr-review.org/sites/default/files/AMR Review
12 de Kraker MEA, Jarlier V, Monen JCM, Heuer OE, van de Sande N, Paper—Tackling a crisis for the health and wealth of nations_1.pdf
Grundmann H. The changing epidemiology of bacteraemias in (accessed Feb 9, 2018).
Europe: trends from the European Antimicrobial Resistance
Surveillance System. Clin Microbiol Infect 2013; 19: 860–68. 27 de Kraker MEA, Stewardson AJ, Harbarth S. Will 10 million people
die a year due to antimicrobial resistance by 2050? PLoS Med 2016;
13 Ho PL, Chow KH, Lai EL, Lau EH, Cheng VC. Extended-spectrum- 13: e1002184.
β-lactamase-positive Escherichia coli mainly adds to, rather than
replaces, extended-spectrum-β-lactamase-negative E coli in causing 28 ECDC/EMEA Joint Working Group. The bacterial challenge: time to
bacteraemia in Hong Kong, 2000–10. J Antimicrob Chemother 2012; react. 2009. http://ecdc.europa.eu/en/publications/Publications/0909_
67: 778–80. TER_The_Bacterial_Challenge_Time_to_React.pdf (accessed
Feb 9, 2018).
14 UN. Member states. http://www.un.org/en/member-states/index.
html (accessed Feb 9, 2018). 29 Centers for Disease Control and Prevention. Antibiotic resistance
threats in the United States, 2013. https://www.cdc.gov/
15 UN. Classification of countries by major area and region of the drugresistance/threat-report-2013 (accessed Feb 9, 2018).
world. http://www.un.org/en/development/desa/population/
publications/pdf/mortality/WMR2011/WMR2011_ 30 Schneider G, Patel HN, Zilberberg M. Epidemiological trends of
DOCUMENTATION/WMR2011_DEFINITION_OF_MAJOR_ carbapenem-resistant Enterobacteriaceae (CRE) in the US from
AREAS_AND_REGIONS.PDF (accessed Feb 9, 2018). 2009–2015. Society for Healthcare Epidemiology in America; St
Louis, MO; March 31, 2017. https://static1.squarespace.com/
16 World Bank. Health, nutrition and population statistics (updated static/51199d96e4b084d1d0b105c3/t/58de64c2893fc0598a
4/17/2017). http://databank.worldbank.org/data/reports. 6f4620/1490969796999/AKAO_SHEA_CRE_Presentation_
aspx?source=health-nutrition-and-population-statistics# (accessed Schneider2017.pdf (accessed Feb 9, 2018).
Feb 9, 2018).
31 Colomb-Cotinat M, Lacoste J, Brun-Buisson C, Jarlier V,
17 Kronenberg A, Hilty M, Endimiani A, Muhlemann K. Temporal Coignard B, Vaux S. Estimating the morbidity and mortality
trends of extended-spectrum cephalosporin-resistant Escherichia coli associated with infections due to multidrug-resistant bacteria
and Klebsiella pneumoniae isolates in in- and outpatients in (MDRB), France, 2012. Antimicrob Resist Infect Control 2016; 5: 56.
Switzerland, 2004 to 2011. Euro Surveill 2013; 18: 20484.
32 European Centre for Disease Prevention and Control. Antimicrobial
18 Health Protection Surveillance Centre (Ireland). Annual resistance surveillance in Europe 2014. Annual report of the European
epidemiological report, 2014. http://www.hpsc.ie/abouthpsc/ Antimicrobial Resistance Surveillance Network (EARS-Net).
annualreports/HPSC%20Annual%20Report%202014%20%20.pdf https://ecdc.europa.eu/en/publications-data/antimicrobial-resistance-
(accessed July 29, 2018). surveillance-europe-2014 (accessed Feb 9, 2018).
19 National Institute for Health and Welfare (Finland). Infectious 33 Gandra S, Trett A, Klein EY, Laxminarayan R. Is antimicrobial
diseases in Finland, 2014. https://www.julkari.fi/bitstream/ resistance a bigger problem in tertiary care hospitals than in small
handle/10024/128993/URN_ISBN_978-952-3024960. community hospitals in the United States? Clin Infect Dis 2017;
pdf?sequence=1 (accessed Feb 9, 2018). 65: 860–63.
20 Statens Serum Institut, National Veterinary Institute, National Food 34 Swedres-Svarm. Consumption of antibiotics and occurrence of
Institute. DANMAP 2014—use of antimicrobial agents and antibiotic resistance in Sweden. 2015. http://www.sva.se/
occurrence of antimicrobial resistance in bacteria from food globalassets/redesign2011/pdf/om_sva/publikationer/swedres_
animals, food and humans in Denmark. https://www.danmap. svarm2015.pdf (accessed Feb 9, 2018).
org/~/media/projekt%20sites/danmap/danmap%20reports/
danmap%202014/danmap_2014.ashx (accessed July 29, 2018). 35 Arias CA, Murray BE. The rise of the Enterococcus: beyond
vancomycin resistance. Nat Rev Microbiol 2012; 10: 266.
21 Swedres-Svarm. Consumption of antibiotics and occurrence of
antibiotic resistance in Sweden, 2014. http://www.sva.se/ 36 Munoz-Price LS Weinstein RA (2008). Acinetobacter infection.
globalassets/redesign2011/pdf/om_sva/publikationer/swedres_ N Engl J Med 2008; 358: 1271–81.
svarm2014.pdf (accessed Feb 9, 2018).
37 WHO. Global Antimicrobial Resistance Surveillance System
22 NORM, NORM-VET. Usage of antimicrobial agents and occurrence (GLASS). http://www.who.int/glass/en (accessed Feb 9, 2018).
of antimicrobial resistance in Norway, 2014. https://wwweng.
vetinst.no/surveillance-programmes/norm-norm-vet-report 38 AMR Control. The ESBL Tricycle AMR surveillance project:
(accessed July 29, 2018). a simple, one health approach to global surveillance.
http://resistancecontrol.info/2017/the-esbl-tricycle-amr-surveillance-
23 New York State Department of Health. Hospital-acquired infections. project-a-simple-one-health-approach-to-global-surveillance
New York State, 2014. https://www.health.ny.gov/statistics/facilities/ (accessed Feb 9, 2018).
hospital/hospital_acquired_infections/2014/docs/hospital_
acquired_infection.pdf (accessed Feb 9, 2018).
24 Centers for Disease Control and Prevention. Antibiotic Resistance
Patient Safety Atlas. https://gis.cdc.gov/grasp/PSA (accessed
Feb 9, 2018).
www.thelancet.com/lancetgh Vol 6 September 2018 e979
RESEARCH ARTICLE Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
crossm
Genomic Analysis of Hospital Plumbing Reveals Diverse
Reservoir of Bacterial Plasmids Conferring Carbapenem
Resistance
Rebecca A. Weingarten,a Ryan C. Johnson,b Sean Conlan,b Amanda M. Ramsburg,a John P. Dekker,a Anna F. Lau,a
Pavel Khil,a Robin T. Odom,a Clay Deming,b Morgan Park,c Pamela J. Thomas,c NISC Comparative Sequencing Program,c
David K. Henderson,a Tara N. Palmore,a Julia A. Segre,b Karen M. Franka
aNational Institutes of Health Clinical Center, Bethesda, Maryland, USA Received 2 November 2017 Accepted 8
bNational Human Genome Research Institute, Bethesda, Maryland, USA January 2018 Published 6 February 2018
cNIH Intramural Sequencing Center, Rockville, Maryland, USA Citation Weingarten RA, Johnson RC, Conlan S,
Ramsburg AM, Dekker JP, Lau AF, Khil P, Odom
ABSTRACT The hospital environment is a potential reservoir of bacteria with plasmids RT, Deming C, Park M, Thomas PJ, NISC
conferring carbapenem resistance. Our Hospital Epidemiology Service routinely performs Comparative Sequencing Program, Henderson
extensive sampling of high-touch surfaces, sinks, and other locations in the hospital. DK, Palmore TN, Segre JA, Frank KM. 2018.
Over a 2-year period, additional sampling was conducted at a broader range of loca- Genomic analysis of hospital plumbing reveals
tions, including housekeeping closets, wastewater from hospital internal pipes, and ex- diverse reservoir of bacterial plasmids conferring
ternal manholes. We compared these data with previously collected information from 5 carbapenem resistance. mBio 9:e02011-17.
years of patient clinical and surveillance isolates. Whole-genome sequencing and analy- https://doi.org/10.1128/mBio.02011-17.
sis of 108 isolates provided comprehensive characterization of blaKPC/blaNDM-positive iso- Editor Robert A. Bonomo, Louis Stokes
lates, enabling an in-depth genetic comparison. Strikingly, despite a very low prevalence Veterans Affairs Medical Center
of patient infections with blaKPC-positive organisms, all samples from the intensive care This is a work of the U.S. Government and is
unit pipe wastewater and external manholes contained carbapenemase-producing or- not subject to copyright protection in the
ganisms (CPOs), suggesting a vast, resilient reservoir. We observed a diverse set of spe- United States. Foreign copyrights may apply.
cies and plasmids, and we noted species and susceptibility profile differences between Address correspondence to Karen M. Frank,
environmental and patient populations of CPOs. However, there were plasmid back- [email protected].
bones common to both populations, highlighting a potential environmental reservoir of R.A.W., R.C.J., and S.C. contributed equally to
mobile elements that may contribute to the spread of resistance genes. Clear associa- this work.
tions between patient and environmental isolates were uncommon based on sequence T.N.P., J.A.S., and K.M.F. are co-senior authors.
analysis and epidemiology, suggesting reasonable infection control compliance at our
institution. Nonetheless, a probable nosocomial transmission of Leclercia sp. from the ® mbio.asm.org 1
housekeeping environment to a patient was detected by this extensive surveillance.
These data and analyses further our understanding of CPOs in the hospital environment
and are broadly relevant to the design of infection control strategies in many infrastruc-
ture settings.
IMPORTANCE Carbapenemase-producing organisms (CPOs) are a global concern
because of the morbidity and mortality associated with these resistant Gram-
negative bacteria. Horizontal plasmid transfer spreads the resistance mechanism to
new bacteria, and understanding the plasmid ecology of the hospital environment
can assist in the design of control strategies to prevent nosocomial infections. A
5-year genomic and epidemiological survey was undertaken to study the CPOs in
the patient-accessible environment, as well as in the plumbing system removed
from the patient. This comprehensive survey revealed a vast, unappreciated reservoir
of CPOs in wastewater, which was in contrast to the low positivity rate in both the
patient population and the patient-accessible environment. While there were few
patient-environmental isolate associations, there were plasmid backbones common
to both populations. These results are relevant to all hospitals for which CPO coloni-
zation may not yet be defined through extensive surveillance.
January/February 2018 Volume 9 Issue 1 e02011-17
Weingarten et al. ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
mbio.asm.org 2
KEYWORDS antimicrobial resistance, carbapenem resistant, carbapenemase-
producing organisms, environment, infection control, plasmid-mediated resistance,
plumbing, wastewater
The Centers for Disease Control and Prevention determined that nearly 4% of the
patients in acute-care hospitals in 2011 had contracted a hospital-acquired infec-
tion (HAI), with a projected estimate of 648,000 patients having had at least one HAI
in United States acute-care hospitals in 2011 (1). HAIs caused by carbapenemase-
producing organisms (CPOs) are of particular concern because of the substantial rates
of morbidity and mortality associated with these infections (2). Carbapenemases are
beta-lactamase enzymes (bla) that hydrolyze carbapenems, a family of antimicrobials
considered to be a last line of defense against infections caused by multidrug-resistant
organisms (MDROs). The most commonly detected carbapenemases in the United
States include blaKPC, blaNDM, and blaOXA (3). In 2011, the NIH Clinical Center (NIHCC)
experienced an outbreak of blaKPC-positive (blaKPCϩ) Klebsiella pneumoniae that ulti-
mately affected 19 patients. Whole-genome sequencing (WGS) was used to track the
outbreak and elucidate the chain of transmission from the index patient (4). During the
2011-2012 investigation, blaKPCϩ K. pneumoniae was also cultured from a handrail, a
ventilator, sink drains, and surfaces within a patient’s room (4, 5). Since then, extensive
routine perirectal surveillance of all high-risk patients has been performed, along with
environmental sampling throughout the hospital. All blaKPC/blaNDM PCR-positive envi-
ronmental and newly identified clinical and/or patient surveillance isolates have been
sequenced for both epidemiological and research purposes. The NIHCC has a very low
prevalence of blaKPC, blaNDM, and blaOXA CPOs (0.35% of patients for perirectal surveil-
lance and 0.14% of patients for clinical cultures between 2012 and 2016), with the
majority of CPO-colonized patients detected by culture-based screening at the time of
admission. Notably, we have identified very few blaOXAϩ isolates; therefore, our focus
for environmental screening was on CPOs that were blaKPCϩ or blaNDMϩ by PCR.
Environmental surveillance can be a useful tool to identify the source of an outbreak
(6), to better understand the microbial communities within the hospital (7), and to
evaluate the efficacy of environmental disinfection or other infection control measures
(8). Sources of outbreaks are occasionally linked to aqueous locations such as sinks and
drains (9–12), where the presence of biofilms makes remediation challenging; sink
engineering modifications have been proposed to tackle these problems (13, 14).
Beyond the sink, hospital sewage and wastewater are known reservoirs of CPOs around
the world (15–20). One hypothesis suggests that this reservoir is, in part, due to the use
of large quantities of antimicrobial agents in hospitals, which leads to the selection of
MDROs and the high likelihood of horizontal gene transfer within the hospital effluent
(21). A number of previous surveillance and outbreak studies have focused either
on hospital effluent and wastewater treatment plants (WWTPs) (19, 22–24) or on the
internal hospital environment only (25). Those studies used PCR, pulsed-field gel
electrophoresis (PFGE), and culture methods to track organisms and antimicrobial
resistance genes. WGS was often performed only on a small number of isolates and
plasmids in these studies (18, 26, 27), limiting the genomic resolution of the analysis.
Additionally, there have been limited studies comparing wastewater MDROs with
patient isolates (18, 28–31).
To conduct a more in-depth analysis, we investigated the NIHCC environment to (i)
determine if additional potential reservoirs of CPOs might provide information to
improve our surveillance strategies, (ii) characterize similarities between chromosomes
and plasmids of environmental and patient CPOs, and (iii) improve our understanding
of the microbial and genetic diversity associated with carbapenemases in the health
care environment. To the best of our knowledge, this is the first study to provide a
combined genomic analysis of blaKPC/blaNDM-positive isolates from patients, from the
accessible environment within the hospital, and from the external hospital effluent.
January/February 2018 Volume 9 Issue 1 e02011-17
Reservoir of Carbapenemase-Producing Organisms ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
TABLE 1 Summary of environmental locations surveyed and blaKPC/blaNDM-positive organisms identified between January 2012 and
December 2016
Location (no. of Total no. No. (%) with organisms Species identified
isolates) of samples carrying carbapenemase genes
Wastewater from manhole (7), pipe (8), 19 15 (78.9) Acinetobacter spp., Aeromonas spp., C. freundii
complex, Citrobacter sp., Enterobacteriaceaea
sludge (4) 128 16 (12.5) family, E. cloacae complex, E. coli,
K. pneumoniae, K. oxytoca, Pseudomonasb sp.,
Housekeeping closet floor drains (79), 340 11 (3.2) Serratiac spp.
equipment (38), surface (11) 217 3 (1.4)
704 45 (6.4) Acinetobacter spp. (blaNDM), Aeromonas spp.,
Hospital sink drains (285), trap water (5), C. freundii complex, E. cloacae complex,
aerators (32), faucets (12), water (6) Leclercia spp., Escherichiad sp., Pantoea spp.,
K. pneumoniae
High-touch surfaces
All K. pneumoniae, K. oxytoca, E. cloacae complex,
C. freundii complex,
Pantoea spp., K. pneumoniae
aNot able to assign a genus.
bNot P. aeruginosa.
cNot S. marcescens.
dNot E. coli.
Together, these data broaden our understanding of antimicrobial resistance genes in mbio.asm.org 3
multidrug-resistant (MDR) bacteria in the environment and hospital settings.
RESULTS
CPOs detected in the hospital environment. Extensive surveillance of the envi-
ronment (Table 1), perirectal surveillance of high-risk patients upon hospital admission,
and monthly whole-house surveillance of all in-house patients are performed at the
NIHCC to monitor proactively and to enable an immediate response to the presence of
CPOs. The Hospital Epidemiology Service focuses on surfaces accessible to patients and
health care providers, particularly sink-related locations and high-touch surfaces such
as countertops, handrails, furniture, patient equipment, doorknobs, carts, computers,
keyboards, phones, handles, nursing stations, break rooms, ice machines, wheelchairs,
elevators, and waiting areas. CPOs were recovered from only 3 (1.4%) of the 217
samples taken from high-touch surfaces over a 5-year period. One isolate was K. pneu-
moniae (KPNIH26) from a handrail that was sampled during the outbreak investigation,
and the remaining two isolates were blaKPCϩ Pantoea spp. (PSNIH1 and PSNIH2)
cultured from a shelf in an inpatient ward medication storage room and from the
handrail of a public staircase, respectively (5). A large number of samples were also
collected from sink components, including drains, traps, aerators, and faucets, as well
as tap water. Aerators were removed, resuspended in broth prior to culture, and
replaced with new parts. No potable water samples or faucet swabs grew CPOs, but 11
(3.2%) of the 340 samples from 10 drains and one aerator culture contained CPOs.
Surfaces to which patients do not have direct access also contained CPOs, including
12% of the samples collected from locked housekeeping storage closets (equipment
and floor drains) (Table 1).
CPOs detected in wastewater. All seven wastewater samples (100%) collected
from the intensive care unit (ICU) piping system contained at least one CPO, a
remarkable finding given the low prevalence of CPOs in our patient population. No
CPOs were detected from two wastewater collections obtained from a different non-
ICU floor of the hospital. Additionally, seven wastewater samples were collected from
two external manholes associated with the NIHCC, and CPOs were recovered from
every sample collected. The wastewater pipe system appears to be a reservoir for CPOs,
even though our data indicate infrequent input of new CPOs on the basis of surveil-
lance testing of all hospitalized patients and a recent point prevalence survey that did
not find carriage of MDROs in the intestinal flora of NIHCC health care personnel (32).
Distinct characteristics observed in CPOs from environmental and patient
samples. The 72 CPOs from the environment were compared with 36 CPOs from 30
January/February 2018 Volume 9 Issue 1 e02011-17
Weingarten et al. ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
FIG 1 Connections identified on the basis of the genome sequence similarity of CPOs isolated from the mbio.asm.org 4
environment and patients. Each rectangle in the outer ring represents a sequenced blaKPC-positive isolate and is
color coded on the basis of the source. Black, patient; red, wastewater manhole; light gray, housekeeping
equipment; green, housekeeping closet drain; purple, high-touch surface; dark blue, sink; yellow, wastewater pipe.
Four patients have multiple isolates, and these are clustered and denoted by additional black bars. Orange arcs
indicate Ͼ99.90% ANI between patient and environmental isolates. Blue arcs indicate Ͼ99.90% ANI between
environmental isolates. Arc color saturation does not have meaning and is used solely to aid the visual distinction
of links. Redundant arcs between environmental isolates are excluded for simplicity. The isolate name is provided
only if an environmental connection was found. Icon credits: Alonzo Design, Kathy Konkle, cihanterlan, pialhovik,
Panpty, istrejman/Getty Images under license.
patients (Fig. 1), as some patients were colonized with more than 1 CPO. A number of
environmental CPO species identified in the wastewater areas were not observed in our
patient population, including Serratia spp. (not Serratia marcescens) in the wastewater
and an Escherichia sp. (not Escherichia coli) found in the housekeeping closet floor
drains (see Fig. S2 in the supplemental material). Diverse blaKPCϩ Aeromonas spp. were
abundant in our environmental sampling, but only one blaKPCϩ Aeromonas sp.
(AHNIH1), which was genetically unrelated to the environmental Aeromonas spp., has
been detected in a single NIHCC patient to date (33). Surprisingly, no species desig-
nation could be assigned to three environmental Enterobacteriaceae isolates by using
three clinically relevant platforms (matrix-assisted laser desorption ionization–time of
flight mass spectrometry [MALDI-TOF MS], 16S rRNA sequencing, and WGS), highlight-
ing carbapenemase genes in uniquely environmental species. The minute volume of
wastewater collected for culture compared to the amount within the hospital effluent
provided only a narrow representation of the vast wastewater reservoir; thus, the
observed diversity of CPOs in the hospital effluent may be much greater than that
shown by our data. In contrast to the environmental samples discussed above, com-
monality was observed between CPO species from sink drains and patients, including
the Enterobacter cloacae complex, the Citrobacter freundii complex, and Klebsiella
oxytoca.
January/February 2018 Volume 9 Issue 1 e02011-17
Reservoir of Carbapenemase-Producing Organisms ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
FIG 2 Antimicrobial susceptibility data on blaKPC/blaNDM PCR-positive Enterobacteriaceae from patients mbio.asm.org 5
and the environment between January 2012 and December 2016. Enterobacteriaceae isolates were
interpreted as intermediate or resistant to the antimicrobial agents on the basis of CLSI M100 guidelines
(62). Black bars are isolates from the environment, and gray bars are isolates from patients. The values
below the bars are the total numbers of isolates tested. Pip/Tazo, piperacillin-tazobactam; Trimeth/Sulfa,
trimethoprim-sulfamethoxazole. (*, two-tailed P Ͻ 0.05; **, P Ͻ 0.001; Fisher’s exact test.)
Susceptibility profiles of environmental and patient carbapenemase-producing
members of the family Enterobacteriaceae were compared (Fig. 2; Table S1). Overall,
Enterobacteriaceae environmental isolates were more susceptible to meropenem, cip-
rofloxacin, amikacin, and tobramycin than were patient isolates (Fig. 2). Additionally, all
of the environmental Aeromonas spp. were susceptible to meropenem and imipenem
(Table S1). Four environmental Enterobacteriaceae isolates were excluded from this
specific comparison because we limited the study set to avoid collection bias that
might inaccurately show more sensitivity of environmental isolates. Isolates that did not
demonstrate growth on HardyCHROM CRE (CRE medium; Hardy Diagnostics) after
original recovery from less selective HardyCHROM ESBL (ESBL medium; Hardy Diagnos-
tics) and R2A medium were excluded because CRE medium is used for patient surveil-
lance. Patient isolates recovered on a number of clinical culture media were included.
Potential associations between patient and environmental isolates were iden-
tified. Comparison of the genomes of 36 patient isolates and 72 environmental
isolates revealed eight instances in which patient isolates showed Ͼ99.90% average
nucleotide identity (ANI) with environmental isolates (Fig. 1). One E. cloacae complex
strain (ECNIH7) from patient P, who had been hospitalized earlier in the ICU, had
Ͼ99.96% ANI with two E. cloacae complex isolates from ICU wastewater. No other
patient isolates matched any other CPOs from ICU pipe or manhole wastewater. The
remaining seven patient isolates showed Ͼ99.90% ANI with isolates from sinks, house-
keeping closets, and high-touch surfaces. One patient isolate (KPNIH24) showed high
identity with two environmental isolates; however, this may be a widely distributed
strain with little divergence since genomes with Ͻ100 single nucleotide polymor-
phisms (SNPs) have been identified at several institutions, based upon sequence data
available in public databases. Moreover, the epidemiological link is weak, with a 3-year
gap between this patient’s hospitalization and detection in housekeeping areas (Fig. 1).
Pantoea spp. are known organisms in hospital environments and can be associated
with outbreaks (34, 35). In 2015, two patients were found to be colonized with blaKPCϩ
Pantoea spp., but epidemiology and genomic data indicated that they were not clonal;
PSNIH3 and PSNIH6 showed Ͻ90% ANI. The positive surveillance data led to an
investigation that identified two environmental Pantoea spp. (PSNIH4 and PSNIH5)
associated with housekeeping equipment. These two environmental isolates, along
with Pantoea PSNIH2 isolated from handrails in 2013 (5), showed Ͼ99.95% ANI with
January/February 2018 Volume 9 Issue 1 e02011-17
Weingarten et al. ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
mbio.asm.org 6
patient isolate PSNIH3 (Fig. 1). SNP analysis identified Ͼ119 different SNPs between the
patient strain and three environmental strains, which suggested closely related, but not
clonal, strains. These findings emphasize the ability of Pantoea spp. to survive on
different hospital environmental surfaces, as previously reported (5, 34).
Sink drains are a known reservoir of CPOs (9–11). We detected three examples of
patient isolates showing Ͼ99.90% ANI with sink drain isolates (Fig. 1). E. cloacae
complex isolate ECNIH6, detected in a patient in 2014, showed Ͼ99.95% ANI with three
sink isolates (ECNIH4, ECNIH9, and ECNIH10) recovered from two different rooms in
2012; however, the patient was never housed in either room. Another common species
identified in the sink cultures, K. oxytoca, has been implicated in biofilm-associated sink
contamination in other hospitals (10, 11). K. oxytoca sink drain isolates (KONIH5 and
KONIH6) were cultured from two sink drains in a room that housed patient S, who was
identified as carrying KONIH3 four months prior. A third sink drain isolate (KONIH9) was
detected a year later, despite pipe removal and thorough cleaning with wire brushes
and bleach. (Fig. 1; Table S2). All four isolates (one from a patient, three from two sinks)
show Ͼ99.99% ANI, suggesting that they are derivative isolates (Fig. 1).
Finally, a genetic connection was established between a patient isolate and a sink
drain as part of a continuation of a previously published investigation (5). Sink drain
isolate ECNIH2 (January 2012) carried three different blaKPC-containing plasmids. One of
the blaKPC-containing plasmids (pKEC-39c) was traced to a plasmid in K. pneumoniae
KPNIH27 from patient A, who stayed in that room from November 2011 to January
2012. At the time of our original study, we had no explanation for the other two
blaKPC-containing plasmids in ECNIH2, pKPC-272 and pKPC-f91. On further analysis, we
identified a blaKPCϩ E. cloacae isolate from patient Y (ECNIH8), who occupied the same
room (November 2010 to February 2011). This isolate is the likely recipient of the
pKEC-39c plasmid. The ECNIH8 chromosome (from February 2011) differs from ECNIH2
(January 2012) by only three SNPs (99% coverage) and carries the pKPC-272 (0 SNPs;
99% coverage) and pKPC-f91 (0 SNPs; 92% coverage) plasmids. Figure 3 shows a
proposed model of the persistent colonization of the sink drain and subsequent
plasmid transfer. Together, these data demonstrate long-term persistence of CPOs in
sink drains and the transfer of CPOs from patients to the sink drain environment.
Although it is rare to capture the details of plasmid transfer as exemplified here, transfer
of plasmids in sink biofilms may be a frequent occurrence and warrants further study.
Genomic analyses of blaKPC-containing plasmids from environmental and pa-
tient samples illustrate a broad diversity of plasmid configurations in many
species. Because transfer of carbapenemases carried on plasmids between organisms
is a public health concern (36), we moved our analysis beyond a chromosomal
comparison to a plasmid analysis with greater resolution. From this data set, each
blaKPC-containing plasmid was represented with at least one fully assembled PacBio
genome. These high-quality PacBio genomes served as references and enabled us to
resolve the plasmid sequences of related genomes, which may have only received
short-read sequencing. Similarly, genomes with limited nucleotide identity to any
previously sequenced isolate were also subjected to PacBio sequencing. In total, our
analysis included 2 genomes from Roche 454 sequencing, 55 from Illumina MiSeq
sequencing, and 51 from PacBio sequencing. The plasmid sequences of patient and
environmental isolates from MiSeq and PacBio genomes were compared by using their
k-mer composition. Sixteen-base-pair k-mers were calculated for all assemblies by using
meryl (http://kmer.sourceforge.net/), and a specific blaKPC-containing plasmid was de-
fined as present if Ն95% plasmid k-mers were found in the assembly (Fig. 4; Table S2).
Some blaKPC-containing plasmids were grouped together into families of plasmids
for analysis purposes. For instance, newly detected pKPC-8bc0 and pKPC-79f0 were
typed as incompatibility group N (IncN) (PubMLST) and exhibited a high level of genetic
similarity to other IncN ST6 plasmids previously detected at NIHCC (5) (Fig. S3). We have
identified this pervasive IncN family of plasmids in 7 patient and 23 environmental
isolates, including high-touch surfaces, sink drains, housekeeping closets, and ICU pipe
wastewater (Fig. 4). Similar to the IncN family, five newly identified blaKPC-containing
January/February 2018 Volume 9 Issue 1 e02011-17
Reservoir of Carbapenemase-Producing Organisms ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
FIG 3 CPOs are able to persist and recombine in sink drains. (A) In February 2011, E. cloacae carrying two mbio.asm.org 7
blaKPCϩ plasmids was cultured from patient Y. This organism may have also colonized a sink in the
patient’s room. (B) Eleven months later (January 2012), an isolate from patient A, who was colonized with
blaKPCϩ K. pneumoniae upon admission, was likely introduced into the same sink (KPNIH27). (C) KPNIH27’s
pKPC-39c plasmid in a sink drain isolate is hypothesized to have horizontally transferred to the sink drain
isolate from patient Y, generating strain ECNIH2. Plasmids carrying the blaKPC gene are colored, and
blaKPC genes are marked by blue (KPC-2) or pink (KPC-3) circles. An insertion in the pKPC-39c plasmid is
black. Non-KPC plasmids are gray. Selected plasmids are labeled with their size in kilobases. E. cloacae
and K. pneumoniae isolates are orange and blue, respectively.
plasmids, which we have named the pENT-e56 family, could be grouped on the basis
of sequence similarity. These plasmids were detected in 1 patient and 12 environmental
isolates (Fig. 4). Interestingly, the backbone of this family of blaKPC-containing plasmids
was also detected in patient and sink isolates that lacked blaKPC and the flanking
transposon (pKPN-068, pENT-e56, and pENT-d0d) (Fig. 5), suggesting possible evolution
of these plasmids through addition of the transposon.
Our study identified 27 new plasmids with carbapenemase genes based on PacBio
sequencing (Table S2). To date, only a small number of these plasmids were recovered
from our patient population (Fig. 4). Despite limited sampling, we detected six different
blaKPC-containing plasmids in Aeromonas spp. isolated from manhole wastewater,
highlighting the high plasmid diversity within this genus. Plasmid sizes ranged from 20
to Ͼ300 kb, and large plasmids (Ͼ150 kb) were detected in both wastewater and our
patient population. Interestingly, pKpQIL, a dominant blaKPC-containing plasmid
January/February 2018 Volume 9 Issue 1 e02011-17
Weingarten et al. ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
FIG 4 Venn diagram of blaKPC-containing plasmids detected in the environment and patients. The mbio.asm.org 8
presence of plasmids in isolates from patients and/or the environment was based on a k-mer inclusion
approach for which a plasmid was considered present within a genome assembly if Ն95% of the plasmid
k-mers were contained within the MiSeq genome assembly k-mers. Each value is the number of isolates
detected in this study that carry the specified blaKPC-containing plasmid. In the overlapping region, the
ratio indicates the number of environmental isolates/number of patient isolates. Plasmids are drawn to
scale. The inner circle represents the blaKPC allele, and the outer circle represents the transposon. The
blaKPC alleles are blue for blaKPC2 and pink for blaKPC3. Color was used to differentiate the flanking
sequence as follows: orange, Tn4401a; purple, Tn4401b; red, Tn4401b-IS630; gray, Tn4401d; green,
IS26-TnpA; light blue, Ahyd_WCHAH01; black, not determined.
among K. pneumoniae strains in our patient population, was not found in any of our
environmental samples after the outbreak investigation in 2012.
Whereas most of the CPOs we detected carried the blaKPC gene, between 2014 and
2016, blaNDM-1ϩ Acinetobacter spp. were detected in two patients and two housekeep-
ing closet drains. Chromosomally, no link could be demonstrated among these isolates,
as they belonged to different strains or species. However, the blaNDM-1 gene is carried
on a plasmid that has been described previously in other Acinetobacter spp., including
Acinetobacter lwoffii, Acinetobacter baumannii, Acinetobacter bereziniae, and Acinetobac-
ter schindleri (37–41). One of the first descriptions of this plasmid backbone was in
pNDM-BJ01 (JQ001791.1) (38), and variants missing a 6-kb region encoding GroE
proteins have also been described (e.g., pNDM-JN02) (42). All four isolates in this study
carried a blaNDM-1-containing plasmid similar to pNDM-JN02 but with a second small
deletion encompassing the trpF-nagA genes and two SNPs. Genomic and epidemio-
logic data could not link plasmid transmission between the housekeeping closet drains
and patient isolates, which suggests that this plasmid is widespread among Acineto-
bacter spp. beyond our institution and demonstrates a potential to persist in the
environment.
Probable nosocomial transmission from environment to patient. In June 2016,
a blaKPCϩ Leclercia sp. was isolated from the interior of a closet floor drain. Leclercia is
a genus within the Enterobacteriaceae family. Leclercia adecarboxylata, the only species
of this genus whose complete genome sequences is available, has been described as
an opportunistic pathogen and occasional carrier of blaNDM-1 (43). A few cases of
L. adecarboxylata infection have been documented (44–46), and at least one report
describes a case of hospital-acquired pneumonia likely due to L. adecarboxylata (47). In
August 2016, blaKPCϩ Leclercia sp. bacteria grew from swabs of additional housekeep-
ing closet floor drains. In September 2016, routine perirectal surveillance revealed a
new-onset blaKPCϩ Leclercia colonization in a hospitalized stem cell transplant recipient,
patient V. In response, the Hospital Epidemiology Service collected 39 additional
cultures from sink drains, housekeeping closet floor drains, nursing station surfaces,
isolation carts, and patient care equipment. Given our knowledge of housekeeping
January/February 2018 Volume 9 Issue 1 e02011-17
Reservoir of Carbapenemase-Producing Organisms ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
FIG 5 Genetic diversity of the pENT-e56-family of plasmids. Alignment of genetically similar blaKPC-negative (top three) mbio.asm.org 9
and blaKPC-positive (bottom five) plasmids. Gray ribbons indicate regions of homology (Ͼ99.90% sequence similarity).
Antimicrobial resistance genes are yellow, tra genes are dark green, transposase/resolvase genes are brown, and
hypothetical genes are gray. Differences between the annotations of aligned regions are largely due to changes in the
annotation pipeline and databases; genomes were annotated at the time of sequencing by using PGAP (versions 2.1 to 4.2).
closet floor drain isolates found in the preceding months, we reviewed housekeeping
procedures and collected additional cultures from housekeeping equipment as poten-
tial transmission vectors. Review of housekeeping practices identified mop buckets as
a likely point source because of their use in both patient care areas and housekeeping
closets with colonized floor drains; though individual buckets are not assigned to
specific floors. Seven mop buckets were sequestered and swabbed for CPO culture. Of
the 39 additional samples collected, one culture from a single mop bucket grew
blaKPCϩ Leclercia sp.
Overall, our investigation identified eight Leclercia isolates: two from patient V
(perirectal and bronchoalveolar lavage samples), five from housekeeping closet floor
January/February 2018 Volume 9 Issue 1 e02011-17
Weingarten et al. ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
FIG 6 Phylogenetic tree of Leclercia spp. based on core genome alignment (ParSNP v 1.2). The tree mbio.asm.org 10
includes eight genomes from this study (LSNIH1 to -8), two publicly available environmental isolates
(NCBI accession no. JWJV01 and MUFS01), and two L. adecarboxylata reference strains (ATCC 23216,
USDA-ARS-USMARC-60222). Isolation sources are indicated in parentheses. Core genome alignments
supporting this tree covered 36% of the reference sequence.
drains, and one from a mop bucket. We compared the genome sequences by using
Mash and found that the patient isolates were very similar to each other (99.99%) and
to the isolate from the mop bucket (99.91%) (clade A). The remaining five isolates
fell into two clusters, B and C, that differed from the patient cluster by Ͼ100,000
SNPs, corresponding to 97 and 93% identity, respectively. All three clusters carried
a blaKPC-containing plasmid belonging to the IncN family, a Ͼ300-kb plasmid related to
plasmids (pPSP-75c and pPSP-a3c) previously identified in Pantoea spp., and one or two
additional plasmids. We constructed a core genome phylogeny (36% of the genome) by
using ParSNP (48) to compare our isolates to existing L. adecarboxylata genomes and
found that cluster C, made up of two drain isolates, is closely related to three of four
publicly available L. adecarboxylata genomes, including the type strain L. adecarboxy-
lata ATCC 23216 (Fig. 6), suggesting identification to the species level. The A and B
clades, on the other hand, may belong to a different species or subspecies. Core
genome-based phylogeny was chosen over rRNA-based phylogeny because 16S rRNA
sequences are unreliable for the differentiation of species of the genera Enterobacter,
Leclercia, and Citrobacter (49). The similarity between the patient isolates and the mop
bucket isolate suggests a possible common source in our hospital; however, the 166
SNPs identified (excluding a variable phage region) and their even distribution through-
out the genome suggest indirect transmission. Genomic data were not sufficient to
resolve the direction of transmission; however, because this organism had not been
recovered from previous extensive surveillance of patients, and given our finding of six
isolates belonging to at least three different clades in the hospital environment, we
concluded that it was likely transmitted from an environmental source to patient V.
DISCUSSION
The presence of CPOs in the hospital environment is concerning because of the
potential for spread to an immunocompromised or otherwise vulnerable patient
population and because of the transmissibility of carbapenemase genes by mobile
January/February 2018 Volume 9 Issue 1 e02011-17
Reservoir of Carbapenemase-Producing Organisms ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
mbio.asm.org 11
genetic elements to other bacteria within the hospital and into the community (50).
Whereas our study was not designed to investigate the water treatment systems in the
community, other studies have tested the efficacy of WWTPs receiving waste from
pharmaceutical drug manufacturers (51) and hospitals (24, 30, 52) or environmental
water such as rivers upstream or downstream of WWTPs (53). These studies have
detected MDROs throughout different steps of the treatment process, noting a reduc-
tion in numbers, but not elimination, after processing.
There have been many publications of hospital environmental investigations during
outbreaks and some during nonoutbreak periods (20, 25, 27, 54, 55). Findings from the
2011 blaKPCϩ K. pneumoniae outbreak investigation at NIHCC led to changes in our
infection control policies, such as implementation of environmental surveillance cul-
tures in response to newly identified CPO carriers and surveillance testing of previously
positive sites after remediation. Additional testing of wastewater from internal pipes
and external manholes was included in this study to identify possible new environ-
mental CPO reservoirs that may further characterize the plasmid ecosystem. The
genomic analyses of 108 identified CPOs derived from the screening of Ͼ9,000 patients
and Ͼ700 environmental samples over the last 5 years have improved our understand-
ing of CPO and plasmid relationships within our hospital and surrounding environment.
Overall, the NIHCC had a very low rate of CPO positivity on high-touch surfaces
sampled within the hospital (1.4%), a higher positivity rate in sinks (3.2%) and house-
keeping closets (12.5%), and an even higher positivity rate in wastewater (78.9%). For
comparison, a 2.5-year multicenter prospective survey of high-touch surfaces in patient
rooms found that ~40% of the rooms examined were contaminated with MDROs,
including Gram-positive and Gram-negative organisms such as vancomycin-resistant
enterococci, methicillin-resistant Staphylococcus aureus, MDR K. pneumoniae, and MDR
A. baumannii. We did not investigate Gram-positive organisms in this study; further-
more, WGS was not used to investigate genetic relatedness in that study, so the
possible comparisons of studies are limited (56).
In our study, environmental CPO isolates appeared to be more susceptible to
antimicrobial agents than CPOs isolated from patients, possibly related to selective
pressure of antimicrobial agents administered to patients. The diversity of blaKPCϩ
species in our hospital environment, specifically in wastewater, supports the hypothesis
that environmental species can act as carriers of antimicrobial resistance genes and
associated mobile elements, which could lead to transmission to clinically significant
species (57). Further, we detected examples of highly similar plasmid backbones with
and without the blaKPC gene. This phenomenon has been described with pKPC_UVA01-
like plasmids at another hospital (58). These findings point to the utility of sequencing
all identified plasmids, not just plasmids with carbapenemase genes, as these may
provide the backbones that evolve into new resistance gene-containing plasmids.
One goal of this study was to characterize the plasmid population in our hospital,
particularly areas proximal to the patient that are more likely to be a potential source
of transmission, as well as areas more distal and less likely to be a direct source of
transmission. In our hospital, the most clinically concerning CPO has been K. pneu-
moniae carrying the pKpQIL plasmid, the organism associated with the 2011 outbreak.
This clonal strain was recovered repeatedly for several years from two patients who
acquired the organism during the outbreak and remained persistently colonized (4, 5).
However, surprisingly, pKpQIL was not detected in any environmental isolates after July
2012. This suggests that either the pKpQIL plasmid or the K. pneumoniae host was not
well adapted to environmental persistence. Likewise, blaNDM-containing plasmids have
not been disseminated broadly in our hospital environment; only one common
blaNDM-1 plasmid backbone was detected in Acinetobacter spp. from two environmental
samples. In contrast, there were common blaKPC-containing plasmid backbones found
in both environmental and patient isolates, including the IncN family of plasmids, which
was detected in 7 patient and 23 environmental isolates, strongly supporting the broad
host range of this plasmid family. A newly detected plasmid backbone, pENT-e56
family, was also found in 1 patient and 12 environmental isolates. However, we
January/February 2018 Volume 9 Issue 1 e02011-17
Weingarten et al. ® Downloaded from mbio.asm.org on February 10, 2018 - Published by mbio.asm.org
mbio.asm.org 12
detected 23 blaKPC-containing plasmids exclusively in the environment. Others have
described even more extensive plasmid diversity in freshwater (59) and WWTPs (53).
Environmental sources have been implicated as the cause of many hospital out-
breaks (6, 60). Because wastewater is a known source of CPOs (15, 16, 19), it is striking
that so few bacterial strains or plasmids were found to be shared by wastewater and
patients in this study. Our study results align with findings from a few other investi-
gations; one was a study of sewage from four different hospitals that could not
correlate organisms or carbapenemase genes between patient and sewage isolates
(16). Our data support the conclusion that wastewater cannot be used as a marker of
patient colonization status. Another group sampled one hospital wastewater source on
16 separate occasions over a period of 2 months and analyzed 17 carbapenemase-
producing Enterobacteriaceae isolates by PFGE. WGS of 3 of the 17 isolates was
performed (18). No link could be established between wastewater and patient isolates.
Important limitations of our study that could affect the interpretation of results include
a very small cumulative sample volume, lack of replicate testing from the same source,
and variation in culture workup.
Although few clonal connections between wastewater and patient isolates were
established, we did identify several sink drain isolates that were highly similar, if not
clonal, to patient isolates. A recent study by Kotay et al. experimentally showed the ease
with which E. coli can disperse from a sink after seeding (61). We also detected other
CPOs in sinks and housekeeping closets that were highly related to patient isolates,
but an epidemiological link could not be established. This lack of an epidemiolog-
ical connection suggests the possibility that dominant but unrelated CPOs may be
in both patient and environment populations, as exemplified in a study where
ESBL-producing members of the family Enterobacteriaceae were found in hand-
washing sinks in the absence of colonized patients (34). However, the lack of an
identified epidemiological link could also be due to sampling error or low sensitivity
of patient microbial surveillance. It is a constant challenge for infection control
teams to distinguish nosocomial transmission from coincidentally recovered iso-
lates. Detailed chromosomal and plasmid data, such as the analyses presented here,
can greatly assist in the decision algorithm.
On the basis of our findings, we suggest that hospital environmental surveillance
should also include housekeeping areas in addition to high-touch surfaces and sink
drains. blaKPCϩ Leclercia spp. were not detected in NIHCC until 2016, when they were
first identified in five different housekeeping closet floor drains. These five drain
isolates, the isolate from a mop bucket, and two patient isolates formed three different
clusters. These findings suggest that multiple strains of Leclercia were present in the
hospital environment for some time before patient colonization occurred. Leclercia is
not a common pathogen, and this was our first identification of blaKPCϩ Leclercia spp.
Given that our surveillance program is more extensive than what many hospitals can
support, the finding leads one to consider what might be found if more hospitals were
investigated to this extent. Our data highlight the importance of consistent and
proactive patient and environmental CPO surveillance. The investigation of newly
identified CPO carriers, along with an immediate examination of hygiene practices,
followed by remediation of the environment, are all critical actions to control environ-
mental contamination and stop the potential spread of organisms within the patient
population.
The extensive surveillance performed at the NIHCC has revealed CPOs in the
environment; however, it is likely that most hospitals have some CPO colonization in
wastewater and drains that remains undetected, perhaps because these sites are not
sampled routinely or as part of investigative processes. The higher rates of CPO
detection in water-associated environments, such as ICU pipes, sinks, housekeeping
closet drains, and manhole wastewater, suggest that the hospital drainage system plays
a role in providing a stable and underappreciated reservoir of MDROs within hospitals.
The physical separation of wastewater from patients, combined with proper cleaning
practices and active infection control responses, was largely sufficient to prevent
January/February 2018 Volume 9 Issue 1 e02011-17
The words you are searching are inside this book. To get more targeted content, please make full-text search by clicking here.
Sewage system as a source of Germs and bacterial outbreak-new studies
Discover the best professional documents and content resources in AnyFlip Document Base.
Search