An inventory of supranational antimicrobial resistance surveillance networks involving low- and middle-income countries since 2000

Abstract Low- and middle-income countries (LMICs) shoulder the bulk of the global burden of infectious diseases and drug resistance. We searched for supranational networks performing antimicrobial resistance (AMR) surveillance in LMICs and assessed their organization, methodology, impacts and challenges. Since 2000, 72 supranational networks for AMR surveillance in bacteria, fungi, HIV, TB and malaria have been created that have involved LMICs, of which 34 are ongoing. The median (range) duration of the networks was 6 years (1–70) and the number of LMICs included was 8 (1–67). Networks were categorized as WHO/governmental (n = 26), academic (n = 24) or pharma initiated (n = 22). Funding sources varied, with 30 networks receiving public or WHO funding, 25 corporate, 13 trust or foundation, and 4 funded from more than one source. The leading global programmes for drug resistance surveillance in TB, malaria and HIV gather data in LMICs through periodic active surveillance efforts or combined active and passive approaches. The biggest challenges faced by these networks has been achieving high coverage across LMICs and complying with the recommended frequency of reporting. Obtaining high quality, representative surveillance data in LMICs is challenging. Antibiotic resistance surveillance requires a level of laboratory infrastructure and training that is not widely available in LMICs. The nascent Global Antimicrobial Resistance Surveillance System (GLASS) aims to build up passive surveillance in all member states. Past experience suggests complementary active approaches may be needed in many LMICs if representative, clinically relevant, meaningful data are to be obtained. Maintaining an up-to-date registry of networks would promote a more coordinated approach to surveillance.


Introduction
The burden of drug-resistant infections is increasing year on year. It has been predicted that the largest numbers of lives that will be lost as a result of these infections will be in low-and middleincome countries (LMICs). 1 A global action plan on antimicrobial resistance (AMR) was endorsed in May 2015 by the World Health Assembly and calls upon countries to strengthen AMR surveillance. It is generally accepted that we need good AMR surveillance data to be able to assess the scale of the problem accurately and to guide interventions. Many LMICs are already participating in surveillance initiatives for AMR in malaria, TB, HIV and influenza.
Attempts to kick-start global surveillance for resistance to commonly used antibacterial drugs have been made in the past but generally without success. The Global Antimicrobial Resistance Surveillance System (GLASS) was launched in 2015 with the goal of collecting comparable AMR data at country level for key bacterial pathogens. 2 At the same time, the recent catastrophic Ebola epidemic in West Africa has brought the need for surveillance for emerging or epidemic-prone diseases into sharp focus, as experience has shown the majority of these have their origins in LMICs. The interaction between different drivers in humans, animals and the environment argues for adopting a 'One Health' approach to surveillance for both AMR and emerging diseases.
Here, we summarize the supranational surveillance networks for drug-resistant infections operating in LMICs since 2000 and discuss their impacts and challenges, and any implications for the implementation of GLASS.

Methods
For the purposes of this analysis, AMR was defined as resistance to antimicrobial agents in bacteria, protozoa, fungi and viruses. Countries were categorized into income groups using the World Bank 2015 classification. 3

Search strategy
We searched for supranational networks performing AMR surveillance in LMICs from January 2000 to August 2017 in Embase, PubMed and Global Health databases. The search was performed first in May 2016 and updated in August 2017. Search terms were broad and included multiple alternative terms for AMR (e.g. drug resistance, antibiotic resistance, antifungal resistance, antimalarial drug resistance, antiviral resistance, cross resistance, multidrug resistance), as well as alternative terms for surveillance and for LMICs, which were also searched for individually by name (the complete list of search terms is available as Supplementary data at JAC Online). The titles and abstracts or full text of 20 558 (16 629 in 2016 plus 3929 in 2017) articles were screened to identify networks.
Networks did not have to collect primary samples to be included, i.e. they could collate resistance data collected by other groups. We excluded networks that occasionally reported drug resistance but did not have AMR surveillance as the major focus of their activity, e.g. a global travelassociated infection surveillance network, several One-Health networks and the Digital Disease Detection networks (e.g. ProMed). Networks were categorized by type (WHO/governmental, academic, pharmaceutical company/contract research organization-led or other), target pathogen grouping (bacteria, TB, malaria, HIV, other) and funding source. Networks performing AMR surveillance in bacteria were further characterized by pathogen sub-group (e.g. respiratory, enteric) and population under surveillance (e.g. community versus hospital-acquired infection, children). We noted the approaches to quality management taken and impacts or challenges of the networks when recorded.

Results
We identified 72 supranational networks concerned with AMR surveillance since 2000, of which 26 were WHO/governmental (global or regional), 24 academic and 22 pharma initiated ( Figure 1). Funding sources varied, with 30 networks receiving public or WHO funding, 25 corporate, 13 trust or foundation, and 4 funded from more than one type of source. The data-sharing models of the networks were open access (n " 3), closed (n " 38) and shared or unclear (n " 31).
In terms of the pathogens under surveillance, 45 networks were for AMR in bacteria or fungi (Table 1), 18 in malaria, 2 in TB, 6 in HIV and 1 for influenza ( Table 2). The median (range) duration of the networks was 6 years (1-70). In the case of the discontinued malaria networks, inability to secure sustainable funding was an important reason for their collapse. 4 Coverage of LMICs by the networks varied greatly. The median (range) number of LMICs included in the AMR surveillance networks for which the information was available was 8 (1-67). The WHO Global Influenza Surveillance and Response System (WHO GISRS) was the longest running network, established in 1947, and included the greatest number of LMICs (67), although antiviral resistance was not under surveillance at the outset.

Networks for AMR surveillance in bacterial pathogens
Of the 44 networks focused on AMR in bacteria, 6 reported data on the GLASS priority pathogens (with the exception of Salmonella spp. in 4), 2 networks were for Staphylococcus aureus, 10 were for respiratory pathogens (2 of these included Neisseria meningitidis and 1 enteric pathogens), 4 were for enteric pathogens only, 1 was for Neisseria gonorrhoeae and the remainder included a range of Gram-negative (5) or Gram-positive (2) bacteria or a mixture of the two. Seven networks collected or reported data on invasive isolates only, five non-invasive only and the remainder both. For those networks that specified the patient populations isolates came from, seven were community-acquired, five hospital-acquired, one was in women and four in children.

Differences between network categories
The networks were a heterogeneous group with different approaches to surveillance reflecting different objectives. The greatest diversity was found in the antibacterial surveillance group. Most global networks initiated and sponsored by pharmaceutical companies had the objective of evaluating susceptibility to specific drugs (registered drugs or new compounds). A variety of bacterial or fungal pathogens were collected by the pharma networks including community-and hospital-acquired isolates from both sterile and non-sterile sites. Academic networks tended to focus AMR surveillance around a specific clinical question, e.g. one project of the Asian Network for Surveillance of Resistant Pathogens (ANSORP) evaluated susceptibility of ESBL-producing isolates collected in the region to different antimicrobials (Tables 1 and 2). Other academic networks such as the WorldWide Antimalarial Resistance Network (WWARN) part of the newly established Infectious Diseases Data Observatory (IDDO) and International Epidemiologic Databases to Evaluate AIDS (IeDEA) have led analyses of individual patient data collected by other research groups.
The approaches taken for drug resistance surveillance by the major global programmes (TB, malaria, HIV, bacteria, influenza) are summarized in Table 3. As shown, the TB, malaria and HIV networks take an active approach to AMR surveillance in LMICs while the antibacterial and influenza networks rely on case-based surveillance from sentinel sites.

Networks for AMR surveillance in animals
There is one supranational European network for surveillance of food-and waterborne diseases and zoonoses that collects data on antimicrobial susceptibility in humans, animals and food. Larger networks that monitor foodborne infections [WHO Global Foodborne Infections network (GFN) and PulseNet International], including animal and environmental isolates, do not report AMR data although GFN does support an external quality assurance (EQA) programme for participating laboratories, which includes antimicrobial susceptibility testing (AST). No other supranational networks for AMR surveillance in animals were identified.

Quality management
The networks had different approaches to quality management ( Table 4). The pharma-led networks typically did not involve LMIC laboratories in EQA programmes but sent all isolates to a central Review laboratory for confirmatory testing. The global surveillance programmes for AMR in TB, HIV, influenza and gonorrhoea all had proficiency testing programmes delivered via supranational networks of reference laboratories. Among the networks for AMR surveillance in bacteria, the Latin-American network, Red Latinoamericana de Vigilancia de la Resistencia a los Antimicrobianos (ReLAVRA) has been running an EQA scheme (LA-EQAS) since 2000 and provides proficiency testing services at no cost to participating laboratories. The Central Asian and Eastern European Surveillance of Antimicrobial Resistance (CAESAR), the non-EU European network, has used the UK National External Quality Assessment Service (UK-NEQAS) for EQA. WHO-sponsored EQA efforts for AST included the discontinued WHO EQAS AST (1998)(1999)(2000)(2001) 5 and the WHO-AFRO/ NICD-SA EQAP for countries within the WHO-AFRO region. 6 Currently GLASS recommends national reference laboratories take responsibility for quality management.

Impacts and challenges of the networks
Impacts and challenges of the networks were not recorded consistently. The main themes are summarized in Table 5 with examples. The biggest challenges faced by the global networks have been achieving high coverage across LMICs and complying with the recommended frequency of reporting. The Global Project on Anti-Tuberculosis Drug Resistance Surveillance has collected resistance data from 155/194 member states since its inception in 1994. For 72 countries without routine drug susceptibility testing of cases these data come from surveys, which are ideally performed  In a detailed account of the experience of setting up the Network for Surveillance of Pneumococcal Disease in the East Africa Region (netSPEAR), an East African network funded by the GAVI Alliance, in which routine surveillance for pneumococcal disease in public hospitals was strengthened, key challenges noted were difficulty in engaging the government of one of the participating countries in the network, poor performance of some sites despite training and problems with attracting funding. 14 The importance of national and institutional ownership of surveillance activity and of framing it as part of routine activity rather than extra work was stressed. The benefits of collaboration between policymakers, academics and service providers were highlighted, a sentiment echoed by the experience of the malaria regional networks, which re-energized surveillance and also played a role in advocacy for policy change, acting as a bridge between research groups and national control programmes. 4 Individual patient data meta-analyses coordinated by WWARN have led to policy recommendations to change antimalarial drug dosing. Another impact of the academic malaria drug efficacy surveillance networks has been the establishment of successful North-South scientific partnerships. There are a few examples where the scientific leadership now comes from the South, e.g. Plasmodium Diversity Network Africa, a molecular surveillance network. 15 Surveillance networks have a positive impact by connecting laboratories in different countries. The Antibiotic Resistance in the Mediterranean Region (ARMed) network, which ran between 2003 and 2007, reported improvement in participating laboratories' capacity to perform bacterial identification and AST, as a result of the EQA programme attached to the network. 16 The HIV, mycobacteria, influenza and gonorrhoea reference laboratory networks have been created thanks to global surveillance programmes.

Discussion
Defining the global burden of AMR and monitoring the impact of interventions to counter it requires reliable surveillance data. LMICs shoulder the bulk of the global burden of infectious diseases and drug resistance but their surveillance systems tend to be weaker than those in high-income countries (HICs), because passive surveillance cannot be integrated with routine casemanagement of patients easily in many areas. This problem has been circumvented to an extent in TB, malaria and HIV AMR surveillance by using active approaches to surveillance in LMICs and gathering data intermittently to provide a snapshot of the situation. However, achieving high coverage of all LMICs and complying with the recommended frequency of surveillance has been difficult. A review of the HIV, TB and malaria surveillance systems in 2011 suggested that one risk of integrating surveillance into routine activities was that high-quality implementation was less likely. 17 By contrast, GLASS is based on building up or strengthening traditional models of passive case-based surveillance to generate data, as in HICs. Priority pathogens, drugs and specimens for surveillance are named but, unlike the other networks, GLASS does not specify minimum sample sizes or detailed selection criteria for target populations. Responsibility for quality management is devolved to national reference centres rather than a supranational body. Member states are requested to submit their AMR data to case-based surveillance can be implemented in middle-income countries but obtaining representative data may take time. It is likely that it will be many more years before most low-income Review countries have a well-functioning system for routine bacteriological surveillance with high coverage. As a result, this approach risks generating non-representative data in the short-to mediumterm, as has happened so far, and making inter-country comparisons will be difficult. The long-established WHO/International Union Against Tuberculosis and Lung Disease (WHO/IUATLD) surveillance programme had been described as the 'pathfinder' for GLASS but is at a considerable advantage with the development of robust molecular detection methods, notably the roll-out of GeneXpert V R , a PCR-based technology that can be performed   18 Molecular surveillance for drug resistance in other bacteria remains some way off but should be made a high priority in order to simplify surveillance in LMICs.
Assessing the representativeness of AMR surveillance data presents a particular challenge. This will be affected by the geographical location and number of sentinel sites, the number and characteristics of individuals sampled, prior treatment history, the incidence of the target pathogen and the methods of detection. WHO/IUATLD has developed its surveillance methodology to the point where it uses survey data to estimate MDR-TB incidence worldwide but this is exceptional for the global programmes. The global report on early warning indicators of HIV drug resistance states that data from most countries cannot be considered as representative due to the way in which the clinics sampled were selected. 11 In malaria therapeutic efficacy studies in hightransmission settings, children less than five years of age are studied since they have the lowest levels of acquired immunity to malaria to give a 'worst-case scenario' depiction of drug efficacy. AMR surveillance for the most commonly encountered bacteria, as it has been practised to date, presents more problems than for other pathogens because of the lack of agreed case definitions and standardized sampling methods. An analysis comparing trends in Escherichia coli resistance from 1997 to 2001 reported by the global Meropenem Yearly Susceptible Test Information Collection (MYSTIC) and SENTRY pharma networks showed that, despite collecting isolates from similar geographical areas, estimates of nonsusceptibility from MYSTIC were consistently higher than those from SENTRY. However, further analysis revealed this was due to a higher proportion of isolates from patients in ICUs in MYSTIC. 19 AMR surveillance in animals is still in its infancy, with the exception of foodborne infections, but some strategies have been piloted in LMICs under the guidance of the WHO Advisory Group on Integrated Surveillance of Antimicrobial Resistance (AGISAR). The challenges are great, e.g. progress towards standardizing AST breakpoints in veterinary microbiology is far behind that made in humans.
Other networks deserving of a mention that were not included in this analysis are two Digital Disease Detection networks, ProMed and HealthMap, which publish sporadic AMR reports and have an advantage over other networks for the rapidity with which they disseminate information. There is potential for overlap between the activities of networks for AMR detection, foodborne infections and emerging disease detection.
The main limitation of our approach is that the heterogeneity of the data meant meta-analysis was not possible. There are no recognized standards for the composition and activities of AMR surveillance networks. Impacts and challenges of the networks were reported infrequently and our assessment is reliant on published information, which may be more likely to report challenges. In addition, our search was only performed in English with a supplementary search in Spanish to obtain more information about the Latin-American networks.
A successful AMR surveillance network should generate up-todate comparable, representative, high-quality data on pathogens of concern from the target population(s). It should be able to detect and track unexpected events including outbreaks in real time, have rapid, effective mechanisms for communication and reporting, and have a responsible data-sharing policy. A network needs strong leadership and coordination, and it should influence guidelines and policy and ultimately impact on human and animal health. Very few networks were instigated to specifically monitor intervention programmes, e.g. the International Nosocomial Infection Control Consortium. Linking surveillance activity to interventions to combat drug resistance has the potential to increase their impact.
Pharma networks produce high-quality data, but they may not be representative and these networks do not usually support laboratory capacity building in LMICs or influence policy and guidelines. Purely academic networks also produce high-quality data; they often target a clinical or policy question, but they too have limited influence on policy and their sustainability is reliant on external funding. Most of the networks are slow to report their findings and do not give unrestricted access to their data. The experience of the larger global programmes for AMR surveillance in TB, malaria and  • Low coverage, particularly in sub-Saharan Africa and India (GASP, GISRS) • Lack of representativeness of data, e.g. due to selective sampling (HIV, GASP, some CAESAR sites) • Difficulties of implementing routine blood culture/diagnostic microbiology in clinical practice (CAESAR) • Difficulties in implementing complex surveillance methodologies, e.g. optimal in vivo methods for surveillance for artemisinin resistance in malaria, second-line drug susceptibility testing for TB • Lack of engagement by some partners (netSPEAR) • Reporting delays • Sustainability due to underfunding with consequent understaffing; surveillance has generally not been given high priority by external donors (EANMAT, netSPEAR) Review HIV suggests that options for more active surveillance may need to be considered in order to gather comparable useful data from low-income countries before reliable case-based surveillance can be established.
Maintaining an up-to-date registry of networks would promote a more coordinated approach to surveillance, reduce duplication of efforts, optimize funding investment and improve sustainability.