Gut mucosal colonisation with extended-spectrum beta-lactamase producing Enterobacteriaceae in sub-Saharan Africa: a systematic review and meta-analysis

Background: Extended-spectrum beta-lactamase producing Enterobacteriaceae (ESBL-E) threaten human health; and, in areas of sub-Saharan Africa (sSA) where carbapenems are not available, may render ESBL-E infections untreatable. Gut mucosal colonisation probably occurs before infection, making prevention of colonisation an attractive target for intervention, but the epidemiology of ESBL-E in sSA is poorly described. Objectives: Describe ESBL-E colonisation prevalence in sSA and risk factors associated with colonisation. Methods: Studies included were prospective cross-sectional or cohort studies reporting gut mucosal ESBL-E colonisation in any population in sSA. We searched PubMed and Scopus on 18 December 2018. We summarise the range of prevalence across sites and tabulated risk factors for colonisation. The protocol was registered (Prospero ID CRD42019123559). Results: From 2975 abstracts we identified 32 studies including a total of 8619 participants from a range of countries and settings. Six studies were longitudinal; no longitudinal studies followed patients beyond hospital discharge. Prevalence varied between 5 and 84% with a median of 31%, with a relationship to setting: pooled ESBL-E colonisation in community studies was 18% (95% CI 12 to 28, 12 studies); in studies recruiting people at admission to hospital colonisation was 32% (95% CI 24 to 41% 8 studies); and for inpatients, colonisation was 55% (95% CI 49 to 60%, 7 studies). Antimicrobial use was associated with increased risk of ESBL-E colonisation, and protected water sources or water treatment by boiling may reduce risk. Conclusions: ESBL-E colonisation is common in sSA, but how people become carriers and why is not well understood. To inform the design of interventions to interrupt transmission in this setting requires longitudinal, community studies.


Introduction
Extended-spectrum beta-lactamase producing Enterobacteriaceae (ESBL-E) are a significant threat to human health, and have been identified by the World Health Organisation as pathogens of critical importance 1 . In sub-Saharan Africa (sSA), it is increasingly clear that a significant proportion of invasive Enterobacteriaceae infections are ESBL-E and the absence of second line antimicrobials can render infections with these pathogens locally untreatable 2 . Strategies to interrupt ESBL-E transmission that can be practically deployed at scale in low resource settings are urgently needed.
Gut mucosal colonisation with Enterobacteriaceae is thought to precede invasive infection 3,4 , and so preventing ESBL-E colonisation is an attractive strategy for prevention of invasive disease. Data describing the basic epidemiology of ESBL-E colonisation in sSA, will help inform the design of interventions targeted at reducing colonisation. A 2016 meta-analysis of community ESBL-E colonisation prevalence among healthy individuals found only four studies from sSA with a pooled prevalence of 15% (95% CI 4-31%), and significant betweenstudy heterogeneity 5 . No studies described risk factors from Africa. We were aware of a number of studies that had been published since 2016 including a number that described ESBL-E colonisation in any population, so undertook a systematic review and meta-analysis with two aims: firstly, to describe the prevalence of ESBL-E gut mucosal colonisation in sSA; and secondly, to describe any risk factors associated with colonisation. In terms of the PRISMA (preferred reporting items for systematic reviews and meta analyses) PICOS (participants, interventions, comparisons, outcomes and study design) approach, our questions can be framed as: what is the prevalence of ESBL-E gut mucosal colonisation (the outcome) and risk factors for colonisation (comparisons) in any population in sSA (the population) as measured in prospective cross-sectional or cohort studies (study design).

Methods
Inclusion criteria were any prospective cross-sectional or cohort study that had screened for gut mucosal colonisation of ESBL-E in any population in sSA for which it was possible to extract a numerator and denominator to calculate an ESBL-E colonisation prevalence. Exclusion criteria were studies in which the sampled population was not clearly defined in a reproducible way (i.e. laboratory-based studies), or if the laboratory techniques aimed to isolate only a particular organism or type of organism (e.g. Enteropathogenic E. coli). PubMed and Scopus were searched in all fields using the search terms given in Table 1, on 18 December 2018. Abstracts were extracted into Endnote X7.8 (Thomson Reuters, United States) and independently reviewed against the inclusion criteria by two authors (JL and RL), with disagreements settled by consensus.
Full-text review of included studies was then undertaken, with studies assessed against the same inclusion criteria, again with disagreements settled by consensus. Data were then extracted into a Microsoft Excel for Mac v16.27 spreadsheet (Microsoft, United States): study title and authors, year of publication, dates of sample collection, inclusion criteria, median age or participants, details of microbiologic testing procedures, number of participants and number of participants from whom ESBL-E were isolated, and any risk factors for ESBL-E that were assessed and/or found to be associated with ESBL-E colonisation. Two authors extracted data independently (RL and JL) and any inconsistencies corrected by re-review of the original paper. For cohort studies only the baseline prevalence was included. Prevalence was presented as forest plots with exact binomial confidence intervals. Age group (neonate, child, adult, as per study definition) and location of sampling (community, outpatient [including health centre attendees], on hospital admission, [defined as a hospital inpatient for < 24hr] hospitalised, [defined as a hospital inpatient for > 24hr]) were selected as a priori subgroups that we hypothesised may explain heterogeneity in ESBL-E prevalence, and analyses were stratified by these subgroups. Studies were additionally classified as being carried out in a special population if they were carried out in a subpopulation of a subgroup (for example, pregnant women in the community). Effect size of risk factors for ESBL-E colonisation were presented as odds ratios; if odds ratios were not provided by the original studies then they were calculated, with 0.5 added to zero cells. Pooled random effect summary estimates of prevalence, where calculated, were generated using the metaprop package in R using the inverse variance method with a logit transformation. All analysis was undertaken

Amendments from Version 1
We have expanded the limitations section of the discussion to highlight points made by the reviewers: in particular that the available data are scanty and hence may not be generalisable across sub-Saharan Africa, and that some risk factors for ESBL-E carriage (in particular the role of livestock and HIV infection) are not well assessed in the identified studies. Their role in driving colonisation, therefore, remains unclear.
Any further responses from the reviewers can be found at the end of the article REVISED using R v3.5.1 (R Foundation for Statistical Computing, Vienna, Austria).
Risk of bias of included studies was assessed with a modified Critical Appraisal Skills Programme (CASP) checklist, designed to fit our research question (full tool available as extended data). The risk of bias assessment was performed by JL and RL, and any disagreements were resolved by consensus.
The protocol of this review was published on PROSPERO (PROSPERO ID CRD42019123559) and the review was undertaken as per Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (PRISMA checklist available Reporting guidelines).

Results
Of 2975 identified unique studies, 32 were included in this review 6-37 (Figure 1), from 19 countries in sSA ( Table 2). Studies from three countries -Tanzania (n=7), Madagascar (n=4) and Cameroon (n=4) -together made up 15/32 (47%) of the available studies. In total, 8619 participants were included and for 7232/8619 (84%) it was possible to disaggregate the participants into age groups: 4313/7232 (60%) were adults, 2470/7232 (34%) children and 449/7232 (6%) neonates. 2302/8619 (27%) of included participants were community members, 1729/8619 (20%) were outpatients, 2836/8619 (33%) were sampled on admission to hospital, and 1534/8619 (18%) were inpatients. 6/32 studies were cohort studies; all of these studies followed patients up whilst hospitalised only. Many studies were carried out in special populations, including the majority of community studies: 9/12 community studies were in special populations, as well as 3/7 outpatient studies, 3/8 studies of participants on hospital admission and 2/7 inpatient studies. It was not possible to classify patients from two studies into our predefined categories: one sampled staff and children of an orphanage, and the other hospital workers and their families. These studies were excluded from the pooled analyses. Details of the microbiological testing procedures are shown in Table 3.
The results of the risk of bias assessment are shown in Figure 2. The most notable potential for biased ESBL-E prevalence estimates resulted from selection of study populations. Several studies recruited a selected group, which we defined as a special population: pregnant women, street children, children and staff of an orphanage, or food handlers in schools. These are likely to produce a biased estimate of community prevalence. Though microbiological culture methods were frequently described in a reproducible manner, few studies reported quality control procedures, resulting in an assessment of moderate risk of bias for the majority of studies across this domain.
Overall ESBL-E colonisation prevalence was extremely heterogeneous across studies ranging from 5-84% (median 31%) with no trend by year of publication ( Figure 3). Some heterogeneity was explained by location of sampling ( Figure 4): inpatients tended to have the highest colonisation prevalence with community members the least. There was no clear difference in prevalence between neonates, children or adults ( Figure 5). Pooled random-effect summary estimates were therefore calculated for differing location of sampling: community members (18% [95% CI 11-28%]), outpatients (23% [95% CI 13-39%]), inpatients on hospital admission (32% [95% CI 24-41%]) and inpatients (55% [95% CI 49-60%]), though in each stratum significant heterogeneity remained (I 2 76-97%) so these summary estimates should be treated with caution ( Figure 4). Two-thirds (21/32) of studies performed an analysis to identify factors associated with ESBL-E colonisation (Table 4). Prior hospitalisation was assessed as a risk factor in 13     studies, and a statistically significant association found in 4/13, with odds ratios of 2.1-8.5. Antimicrobial exposure was assessed in 13 studies, and a statistically significant association found in 5/13 with odds ratios of 1.6-27.0. Using water from a borehole 28 , boiling water before drinking 14 and having private inside access to drinking water 10 were found to be associated with a lower prevalence of ESBL-E colonisation in three different studies. One study found that a higher socio-economic status was associated with a lower ESBL-E prevalence 29 , and one the opposite 13 . Only two studies addressed the association between HIV status and ESBL-E colonisation status; one, in adults found no association 9 , whereas the other, in children, found a strong association 17 . Only one study assessed the association between animals in the home as ESBL-E colonisation 10 , finding no association.
Of the 6 cohort studies, all sampled participants on admission to hospital and on discharge, a median 5.6-8 days later, and all found an increase in ESBL-E colonisation prevalence between the two sampling points (Table 5). No study longitudinally sampled ESBL colonisation in the community, either in community dwellers or in those discharged from hospital.

Discussion
ESBL-E colonisation is common across sSA, though with significant unexplained heterogeneity between study locations and populations. Community ESBL-E colonisation ranges from 5% in adults in Gambia in 2015 to 59% in children in the Central African Republic in 2013, the latter comparable to the highest described colonisation prevalence in the world 5 . Our pooled estimate suggests 18% (95% CI 11-29%) of people in sSA are colonised with ESBL-E, a higher prevalence than in high income settings. In Europe, community prevalence of ESBL-E colonisation is reported to range from 3.7% in Spain in 2004 to 7.3% in the UK in 2014 38-41 , similar to the United States where a community prevalence of 3.4% was reported in healthy children 42 . In many of the estimates of studies included in this review, the reported prevalence of ESBL-E is more comparable to that reported in Asia (46% [95% CI 29-63%] 5 ).
The profound differences in community ESBL-E colonisation prevalence between sSA and high-resource settings warrants further investigation, beyond the assessment of risk factors we have identified in this review. Hospitalisation and antimicrobial use are likely drivers of colonisation in the studies, with higher   prevalence seen in hospitalised individuals and prior hospitalisation and antimicrobial exposure frequently identified as risk factors for colonisation. Obversely and consistent with a putative faecal-oral transmission route, use of borehole water, a private indoor water source and boiling water before drinking were associated with reduced ESBL-E colonisation risk, and it may be that poor water, sanitation and hygiene (WASH) infrastructure and practices in sSA are driving high ESBL-E colonisation prevalence. This speaks to a role for poverty in driving ESBL-E colonisation; however, this is likely complex, and context-dependant, as evidenced by conflicting findings of the effect of socio-economic status on colonisation from two studies in different settings.
More broadly, this review highlights areas where data that could inform interventions to interrupt ESBL-E transmission are lacking. In the community, long-term longitudinal ESBL-E colonisation studies are necessary to understand the dynamics of community ESBL-E transmission, particularly the role of within household transmission, and the role of household animals. In health facilities, the determinants of apparent ESBL-E acquisition need to be clearly identified to design pragmatic intervention studies in the context of limited resources. Surprisingly, the role of HIV in driving the high ESBL-E colonisation prevalence in sSA is unknown. HIV is known to profoundly affect gut function, but we identified only two studies which have assessed HIV status as a risk factor for ESBL-E colonisation.
There are limitations of our review. Our search strategy may have missed studies that would otherwise be included. However, using broader inclusion criteria than a recent review of worldwide ESBL-E community colonisation prevalence 5 , we have identified many more studies from sSA. Risk of bias assessment in observational studies is difficult, with no gold standard, and the tool we have used may misclassify studies with regard to bias. Significant heterogeneity remaining despite stratification warrants caution in interpreting summary estimates. The number of identified studies and participants are small compared to the population of sSA and several countries are over-represented, meaning that care should be taken in generalising these findings across the diverse settings of sSA. Some potentially important risk factors for ESBL-E colonisation (HIV infection and livestock exposure, for example) are not explored in the studies we have identified, and their role in driving colonisation remains unclear.
In conclusion, ESBL-E colonisation in sSA is common, and in places comparable to the highest prevalence in the world, though with significant unexplained heterogeneity between countries and populations. Hospitalisation, antimicrobial use, and poor WASH infrastructure and practices may be contributing to high prevalence; the roles of HIV and animal-human transmission remain unknown. Given the threat to human health of ESBL-E, data to fully characterise routes and drivers of transmission in sSA are necessary to design interventions to interrupt transmission in this setting.

Data availability
Underlying data All data underlying the results are available as part of the article and no additional source data are required.

Open Peer Review
Current Peer Review Status:

Version 1
10 January 2020 Reviewer Report https://doi.org/10.21956/wellcomeopenres.16981.r37453 © 2020 Richard V. This is an open access peer review report distributed under the terms of the Creative Commons , which permits unrestricted use, distribution, and reproduction in any medium, provided the original Attribution License work is properly cited.

Vincent Richard
International Department, Institut Pasteur, Paris, France Among the targets of the AntiMicrobial Resistance, ESBL-producing Enterobacteriaceae are a real worldwide issue not well-studied in Africa. The previous review of Storberg (2014 ) showed ESBL-producing Enterobacteriaceae are a large problem in African healthcare institutions and communities. However, this author highlighted the scarcity of African data about this topic. This new review shows the same trend with only 32 considered studies from 19 countries and among them 15 studies from 3 countries (Tanzania, Madagascar, and Cameroon).
"Inclusion criteria were any prospective cross-sectional or cohort study that had screened for gut mucosal colonization of ESBL-E in any population in sSA for which it was possible to extract a numerator and denominator to calculate an ESBL-E colonization prevalence." Is that enough to explain that only 32 studies to 2975 were included? These results ask about other countries' lab capacities and about the quality of some works: 2975 identified and 32 included. Does that mean 2943 studies were laboratory-based studies? However, the methodology seems to be strong. The risk of bias was included and figure 2 a good way to assess the studies. Could we imagine from this review some proposals for better-implementing studies about this topic, leading to facilitating comparison between countries? About prevalence and risk factors the authors seem to be surprised by differences between countries but it is a picture of the high diversity of the Africa region. Once, higher socio-economic status will be a protector because of sanitation and in another country this status will be a risk factor because the load of antimicrobial exposure will be more serious.
Because of the design of the studies included in this review, the livestock transmission is not evocated as one of the such risk factors. In discussion, the authors should have to point to the lack of studies concerning assessment of risk from livestock.
Undoubtedly, the most relevant intervention to reduce the carriage of ESBL-E will be the systematic implementation of WASH infrastructures. However, this kind of intervention will need to be assessed for convincing decision-makers to involve themselves in this strategy. 1