Faecal carriage of antibiotic resistant Escherichia coli in asymptomatic children and associations with primary care antibiotic prescribing: a systematic review and meta-analysis

We searched Medline, Embase, Cochrane and ISI Web of Knowledge databases for articles published in any language between 1940 and September 2015. MeSH terms for these databases included “drug resistance”, “faeces”, “carrier state” and “children”. MeSH terms were combined with text word searches which included “antibiotics”, “resistance”, “faecal/fecal”, “colonisation”, “commensal” and “paediatric/pediatric”. Grey and unpublished literature was searched for using ISI Web of Knowledge software and included journal articles, websites, conference proceedings, government and national reports and open access material. Reference lists of selected key papers were screened and authors who appeared multiple times in our search were contacted to request details of further published and unpublished work. All full-text papers were subject to citation searches. See Additional file 1 for full search strategy. The review protocol is available on PROSPERO (http://​www.​crd.​york.​ac.​uk/​PROSPERO/​), registration number CRD42014009691.

Two independent reviewers screened all titles and abstracts for eligibility. Studies were eligible if they met the following criteria: investigated and reported carriage of resistance in faecal E. coli from asymptomatic children, that is children who were not showing symptoms of infection at the time the sample was taken; or investigated associations between previous antibiotic exposure and carriage of resistant E. coli; and study participants were children aged 0–17 years, including healthy neonates with an uncomplicated vaginal birth.

Full-text papers for all eligible studies were obtained and three reviewers extracted data independently using a purpose-built spreadsheet. The following information was extracted from each paper, where provided: author, journal, year of publication, study design, study country, economic status, participants and recruitment location, recruitment time period, age range, method of faecal sample collection and testing, method of antimicrobial sensitivity testing, bacteria cultured and reported antibiotic sensitivities, previously prescribed antibiotics and time between antibiotic exposure and faecal sample collection. Economic status was measured using the OECD status of the country where the study was conducted [12]. The OECD is an international economic organisation first established in 1948, now made up of 34 countries, which aims to work together and with emerging and developing economies to reduce poverty through economic growth and financial stability [12]. OECD member countries tend to be ‘developed’ countries, whereas non-OECD countries tend to be ‘developing’. For the purpose of this review, we use OECD status as a general measure of country-level development, and a proxy marker for OTC antibiotic use. For antimicrobial exposure, time was generally recorded as a period of days, weeks or months prior to the faecal sample being taken and resistance being measured when the child had been exposed to any, or specific named antibiotics. Where any information was unclear in the paper, authors were contacted for clarification.

We reported resistance to antibiotics commonly prescribed to children in primary care, including for urinary tract infection or other indications including respiratory and skin infections. Resistance data was extracted and reported for the following antibiotics: ampicillin, co-amoxiclav (amoxicillin-clavulanic acid), co-trimoxazole (trimethoprim-sulfamethoxazole), trimethoprim, nitrofurantoin, ciprofloxacin, ceftazidime, tetracycline and chloramphenicol. Ceftazidime was the most frequently reported of all first to third generation cephalosporins, and acts as a marker for cephalosporin resistance.

Included papers were assessed for quality using a checklist based on Cochrane collaboration’s ‘risk of bias’ tool [13], We focused our quality criteria on factors we considered important for the review, namely: a reliable measure of antibiotic exposure and resistance, clear reporting of bacterial resistance, and clear reporting of children as asymptomatic or non-infected. For papers which included information on previous antibiotic exposure, we supplemented these with assessment of reporting adjustment for confounders including age, sex and socioeconomic status.

All statistical analyses were conducted using STATA version 13 software, and all methods undertaken according to PRISMA guidelines [14].

We calculated pooled prevalence of resistance estimates by generating a Forest plot for each antibiotic, stratified by OECD status. Forest plots illustrated proportion of resistant E. coli for each country, along with 95 % confidence intervals (CI), and the pooled prevalence of resistance per antibiotic per economic country group (OECD vs. non-OECD). We calculated pooled estimates for each country and for OECD and non-OECD groups using the pooled country estimates. Pooled prevalence estimates were generated for children of all age groups (0–90 days, 0–5 years and 5–17 years) and for defined time periods (1970–1979, 1980–1989, 1990–1999, 2000–2010, 2010–2015), for comparison. An I² of 25, 50 and 75 % were used to signify low-level, moderate-level and high-level heterogeneity, in line with Cochrane recommendations [13]. All pooled estimates and 95 % confidence intervals (CI) were generated using double arcsine transformation to adjust for variance instability. This avoids implausible 95 % CI for prevalence estimates when generated under the normal approximation [15].

For studies investigating the association between previous antibiotic exposure and bacterial resistance, the outcome measure was the odds ratio (OR) of bacterial resistance in children previously exposed to any antibiotic compared to those children who were previously unexposed. The crude estimates from these studies were grouped according to the reported preceding exposure time period (0–2 weeks, 0–1 month and 0–3 months). A random-effects meta-analysis was conducted where heterogeneity was moderate-to-high and a pooled OR was generated for each exposure time period measured. These were compared to adjusted OR for each time period, where reported. Variables which were adjusted for were family member antibiotic exposure, previous hospitalisation, day care attendance, nappy use, ethnicity and socio-economic status (see Additional file 2). We assessed heterogeneity using the I² statistic, and the null hypothesis of no heterogeneity was tested using the Q statistic generated from the χ² test. Finally, funnel plots were generated to explore the possibility of small study effects, which can be caused by publication bias.

We initially identified 12,997 potentially eligible articles. Of these, 8995 non-duplicate papers were assessed and 8697 excluded on basis of title (Fig. 1). The remaining 298 papers were assessed by abstract screening of which 240 were excluded. For the remaining 58, full-text papers were assessed, with 24 papers excluded. Thirty-four papers were therefore included in the review [16‐49], of which six papers reported previous antibiotic exposure data and were included in our meta-analysis [19, 27, 33, 34, 37, 49].

Table 1 summarises the characteristics of the 34 studies. Additional study characteristics can be found in Additional file 2. Twenty studies, reporting the resistance status of 3864 E. coli isolates were conducted in OECD countries (Fig. 2), and all were observational. Fourteen studies, reporting the resistance status of 6699 isolates were conducted in non-OECD studies (Fig. 2), and again, all were observational. Twenty-three studies (14 OECD vs. 9 non-OECD) reported stool sampling as the primary collection method, with 10 reporting rectal swabs, and one study accepting both methods of collection. Antimicrobial sensitivity testing was carried out using standard disk diffusion methods for all studies, which were interpreted and reported according to either the European Committee on Antimicrobial Susceptibility Testing (EUCAST) [50], or the Clinical and Laboratory Standards Institute (CLSI) [51]. All study participants were healthy, without symptoms of infection and recruited in the community, schools and day care centres, or at a primary care facility conducting routine child health surveillance check-ups. Three papers, OECD only, included healthy neonates following uncomplicated vaginal delivery recruited from maternity units.

The quality assessment ‘traffic-light’ charts for the included studies show that, for the six studies reporting antibiotic exposure information, reporting was generally good for our all our key quality indicators (Additional file 3). For studies reporting prevalence of resistance only, overall quality was good with the exception of controlling for confounding and accurate reporting of methods of analysis.

Figure 2 details the number of studies per country and shows the global variation in resistance to ampicillin by OECD status. Resistance to ampicillin in faecal E. coli from asymptomatic children was highest in Mexico (OECD) and Bolivia (non-OECD), with a pooled prevalence of 90 and 95 %, respectively. Ampicillin resistance was lowest in Sweden (OECD), with a pooled prevalence of 12 %.

Table 2 shows the pooled prevalence of resistance to antibiotics in faecal E. coli isolates and were obtained from Forest plots generated for each antibiotic, which can be found in Additional files 4, 5, 6, 7, 8, 9, 10, 11 and 12.

In OECD countries, the highest pooled resistance prevalence was for tetracycline at 37.7 % (95 % CI: 25.9–49.7 %), with ampicillin and trimethoprim resistance also high at 37.6 and 28.6 %, respectively. Resistance to ceftazidime was lowest in OECD countries at 0.3 % (0.1–0.8 %).

Similarly to OECD countries, in non-OECD countries the highest pooled prevalence of resistance was observed in the same antibiotics, with trimethoprim highest at 81.3 % (95 % CI: 40.4–100 %), followed by tetracycline and ampicillin at 80.0 and 67.2 %, respectively.

There were too few data to report pooled resistance prevalence estimates for any given age group (0–90 days, 0–5 years, 5–17 years) for any antibiotic reported in this review.

Figure 3 shows a Forest plot of the pooled resistance prevalence to ampicillin and tetracycline (for which data were most complete), by OECD status, by decade. There were too few data for all other antibiotics to report time period estimates. For OECD countries, included studies were conducted between 1970 and 2014, compared with non-OECD countries which were conducted from 1990 to 2014. Once again, the graphs show the higher resistance rates in non-OECD compared to OECD countries, however there is no evidence of a change in resistance over time as the confidence intervals for each time period and each antibiotic overlap.

Figure 4 shows a Forest plot of six studies investigating the relationship between previous exposure to antibiotics and resistance to a range of commonly used primary care antibiotics. Within all antibiotic exposure time periods, the crude odds of resistance were generally greater for children exposed to antibiotics than those unexposed, though exposure at 0–2 weeks was not found to be significantly associated with resistance. The effect sizes are reasonably similar for all time periods, with the pooled OR of resistance rising as the cumulative antibiotic exposure period increases, though confidence intervals do overlap between different time periods. Given the overlap in exposure time periods, meta-regression analysis was not appropriate.

There was no evidence of within group heterogeneity in the 0–3 month time period, with low heterogeneity in the 0–2 week period and moderate heterogeneity in the 0–1 month periods. For those studies which reported adjusted ORs, adjusting for family member antibiotic exposure, previous hospitalisation, day care attendance, nappy use, ethnicity and socio-economic status; we compared these results with our crude estimates, though we only had sufficient data to do this for exposure at 0–3 months. The pooled adjusted (OR 1.70, 95 % CI: 1.36–2.12) and crude (OR 1.65, 1.36–2.00) did not differ substantially.

There were too few studies for any given exposure time period to assess publication bias.

In asymptomatic children, we found evidence of high rates of faecal E. coli resistance to several commonly prescribed primary care antibiotics, and we have shown that resistance rates were consistently higher in non-OECD compared to OECD countries. The routine use of primary care antibiotics could be an important contributor to carriage of resistant E. coli which we showed persists at both 1 and 3 months post-antibiotic prescription.

To our knowledge, this is the first systematic review and meta-analysis to explore and report global evidence regarding faecal carriage of resistant bacteria in healthy, community-resident children and associations with the routine use of antibiotics in primary care. Our review was rigorously conducted according to the Cochrane guidelines for Systematic Reviews [13]. We chose to stratify our results by OECD status to reflect both national development and likely OTC antibiotic availability [6, 52].

We are aware of four main limitations. First, antibiotics are used very differently within OECD and non-OECD countries [53‐56], and OTC antibiotic use is difficult to measure. A systematic review conducted in 2011 reported high non-prescription antibiotic variability across countries worldwide [52], and there is not 100 % agreement between OECD status and OTC antibiotic availability. However, we are not aware of a better country-level alternative with respect to measuring global prevalence of antibiotic resistance in relation to antibiotic use, and none of the included studies reported or measured OTC antibiotic availability. There was some variation in heterogeneity for our pooled prevalence of resistance estimates. Most heterogeneity was moderate at around 50 %, higher heterogeneity was observed most frequently in our estimates from non-OECD countries. This may be due to the lack of information provided regarding the study populations; although all children were asymptomatic of infection, they may vary in other factors from country to country, for example the comorbidities. Higher heterogeneity in non-OECD countries may also be a reflection of the availability of certain antibiotics OTC in some countries. We also acknowledge that factors other than antibiotic usage and OTC availability can account for differences in carriage of resistant bacteria between OECD and non-OECD countries, including; poverty, poor sanitation, unstable governance, and lower levels of medicine regulation [57]. Additionally, the majority of our non-OECD studies were conducted in either South America or Asia, with African countries under-represented in this group.

Second, reverse causality and other confounding associations including age, sex and previous hospitalisation, could have introduced bias to our meta-analysis findings. However, analyses adjusting for confounding factors did not demonstrate substantial differences between crude and adjusted association estimates. Third, our meta-analysis of the association between antibiotic exposure and resistance reported moderate heterogeneity within the 0–1 month time period, however the difficulty in estimating a more accurate point of antibiotic exposure may have accounted for this. In addition to this, only six studies, conducted between 1987 and 2012 reported data on antibiotic exposure, therefore these findings must be interpreted with caution, but nevertheless provide some evidence to support the possibility of an association between antibiotic exposure and carriage of resistant bacteria. Finally, there were insufficient studies to adequately assess publication bias.