Ciprofloxacin resistance in community- and hospital-acquired Escherichia coli urinary tract infections: a systematic review and meta-analysis of observational studies

The protocol for conducting this review has been registered and can be accessed on the International prospective register of systematic reviews (PROSPERO) (available at http://​www.​crd.​york.​ac.​uk/​prospero/​ with registration number: CRD42014014473). Prior to registration, the protocol was reviewed by a reviewer external to the study team. Ethics approval was not sought as this review synthesized data from published studies for which approval had already been obtained.

We conducted a systematic review of observational (cross sectional, cohort and case control) studies published in the last 11 years (2004–2014) reporting on ciprofloxacin resistance in community- and hospital-acquired E. coli UTIs. This time limit is based on changes in the microbiology and epidemiology of antimicrobial resistant pathogens which occurred in the past decade with subsequent changes in treatment regimens and patient outcomes [12]. Reporting of this review complied with the preferred reporting items for systematic reviews and meta-analyses (PRISMA) [13].

The electronic bibliographic databases MEDLINE/PubMed, EMBASE, Cochrane, CINAHL and Scopus were searched. Searches were conducted for words in the title or abstract or within the full text of the papers. These included both keywords only and keywords with medical subject headings (MeSH) using the search terms ‘resistance’, ‘urinary tract infection’ and ‘Escherichia coli’ from 1st January 2004 to 31st December 2014 (see Additional file 1). The reference lists of papers identified from the electronic databases were hand-searched for additional papers.

Papers were included if they reported prevalence or incidence rates of ciprofloxacin resistance in community- or hospital-acquired E. coli UTIs. Papers reporting on urinary E. coli ciprofloxacin susceptibility in which resistance rate could be calculated were also included. We included papers involving adults and/or children. Only peer reviewed manuscripts were considered. Grey material which includes unpublished literature, conference abstracts, letters to editors, newsletters and reports were excluded. Non-peer reviewed literature were also excluded. Papers written in languages other than English were also excluded. In addition, papers not clearly specifying the setting (hospital-acquired or community-acquired); drug (ciprofloxacin) or sample (urine) were excluded. Papers that focused on specific sub-populations (e.g. diabetics and patients with recurrent UTI) were also excluded as these did not represent the general population. This review included only papers that used the Centers for Disease Control and Prevention (CDC) definition of microbiologically confirmed UTI (≥10⁵ colony forming unit/ml) [14].

The titles and abstracts of all papers identified in the electronic databases were examined and assessed for relevance and appropriateness to the principal objective of the systematic review. Irrelevant studies were excluded. Full texts of the potentially relevant papers were printed and carefully assessed against the systematic review inclusion and exclusion criteria. Those not meeting the criteria were excluded. The remaining papers deemed to have data relevant to the systematic review and meta-analysis were assessed for quality and risk of bias.

The study selection process and other stages of the review were performed by the lead author (OF). At each stage, 10 % of papers identified were also screened against the study criteria independently by other authors (AG, GM and BM). Discrepancies in either the application of inclusion or exclusion of papers, quality assessment or on data extraction were discussed among all authors to make the final decision.

The first author, year of study, country of study, study setting, age and sex distribution, co-morbidities, sample size, study design, study aim, antimicrobial susceptibility testing method, ciprofloxacin resistance rate, risk factors for ciprofloxacin resistance (i.e. previous antibiotic use) and mortality data (if reported) were extracted. Where the ciprofloxacin resistance rate was not available, the susceptibility rate was used to determine resistance.

Quality and risk of bias of the final papers included in the review was conducted using a modified version of the Newcastle-Ottawa Scale (NOS) which is a risk of bias assessment tool for observational studies recommended by the Cochrane Collaboration [21, 22]. Content validity and inter-rater reliability of this tool have been established [22]. Studies were rated by assigning a judgment of ‘Low risk’ of bias, ‘High risk’ of bias, or ‘Unclear risk’ of bias according to published criteria [21].

Pooled ciprofloxacin resistance proportions (with 95 % confidence intervals) in patients with E. coli UTI were separately calculated and compared between hospital and community settings using a random-effects meta-analysis model based on DerSimonian and Laird method [23, 24]. This method incorporates an estimate of the between-study variation into both the study weights and the standard error of the estimate of the common effect. The precision of an estimate from each included study was represented by the inverse of the variance of the outcome pooled across all studies. If the value of the pooled prevalence was within the 95 % CI, then the effect size was statistically significant at the 5 % level (P < 0.05). The heterogeneity among studies was assessed by using the I² statistic with a P value of <0.05 considered statistically significant, and I² values below 25 % indicating low heterogeneity, 25–75 % moderate heterogeneity and over 75 % high heterogeneity [25]. Subgroup analyses were done by risk of bias, study duration, age group, UTI symptoms, world region and economy of country (categorised as developed and developing using the World Bank classification [26]). A meta-regression analysis was used to determine the effect of measured covariates on the observed heterogeneity in resistance estimates across studies [23]. Assessment of publication bias was estimated using funnel plots. Further analysis was undertaken to examine pooled ciprofloxacin resistance over time using the median study year. For studies occurring over 2 years, the first year was used; for studies occurring over 4 years, the 2nd year was used; for those over 6 years, the 3rd year was used. The non-parametric Spearman’s rho correlation coefficient was calculated to determine significance in resistance trend over time. Statistical analyses were undertaken using Stata statistical softwareversion 13 [27].

Electronic database searches identified 15,062 potential studies and 31 additional studies were identified through hand searching. After 11,397 duplicates were removed, 3696 articles remained for title and abstract screening. We assessed 1134 as potentially eligible and retrieved the full text of these articles. After applying inclusion and exclusion criteria, 53 papers (5 %) were deemed to have data relevant to the systematic review and meta-analysis. These 53 papers consisted of 54 studies comprising three hospital-acquired E. coli UTI studies and 51 community-acquired E. coli UTI studies. There was one paper that compared resistance in both hospital and community settings hence reported as two studies [15]. The PRISMA flow chart describing the papers identified from the search strategy and reasons for exclusion is shown in Fig. 1.

Geographically, 53 of the 54 studies were carried out in Asia (28 %; n = 15), Europe (24 %; n = 13), Middle East (15 %; n = 8), Africa (13 %; n = 7), North America (11 %; n = 6) and South America (7 %; n = 4). The remaining study was conducted in multiple countries [28]. There were 17 (31 %) studies conducted in developed countries and 36 (67 %) in developing countries. The majority of the studies (80 %) followed a cross sectional design. The duration of studies ranged from 2 months to 84 months (median  = 15.5; IQR = 12.0-30.0). The mean age and sex proportion of patients with an E. coli UTI were stated in 13 % (n = 7) and 44 % (n = 24) of studies respectively. Most study populations included patients of both sexes although 19 % (n = 10) included only women. Antimicrobial susceptibility testing and interpretation was performed using the disk diffusion method (74 %) and Clinical and Laboratory Standards Institute (CLSI) criteria (83 %) respectively in most studies. Table 1 provides further details on the description of the included studies.

Figures 2 and 3 show the forest plots of studies reporting on ciprofloxacin resistance in community acquired E. coli UTI by region and economy, respectively. Figure 4 shows the forest plot of studies reporting on ciprofloxacin resistance in hospital acquired E. coli UTI. Compared with the community-setting, resistance to ciprofloxacin in E coli UTIs was significantly higher in the hospital-setting (P < 0.001). Overall, the pooled rate for ciprofloxacin resistance in patients with community-acquired E. coli UTIs was 0.27 (95 % CI: 0.240-0.310), compared with 0.38 (95 % CI: 0.360-0.410) in the hospital setting. There was substantial heterogeneity among the community-setting studies (I ² = 98.8 %, P < 0.0001), but very little in the hospital ones (I ² = <0.010 %, P = 0.641). Further analysis of studies reporting on community-acquired E. coli UTI by region (Fig. 3) showed that Asia had the highest pooled resistance. Analysis by economy based on the World Bank classification (Fig. 4) showed a higher pooled resistance in developing countries.

Figure 5 shows the scatter plot of ciprofloxacin resistance in 47 studies reporting on community-acquired UTI using the median study year for each study. Four studies did not provide data on the year(s) the study was conducted and were excluded from this analysis [29‐32]. The results of the Spearman’s rho correlation test showed a statistically significant rise in resistance over time (n = 47, r _s  = 0.431, P = 0.003). Similar findings were observed for developing countries. There was no significant rise in resistance over time in developed countries.

Sub-group analysis was conducted within each major setting. For community-acquired UTI studies (Table 2), there was a significant difference in the pooled resistance within each subgroup examined (risk of bias, study duration, economy, region, age group and UTI symptoms). The subgroup analyses results for studies reporting on hospital-acquired E. coli UTI (see Additional file 2) showed no difference in the pooled resistance within the subgroups examined (region, economy and UTI symptoms). When both settings were compared (see Additional file 3), there were significant differences noted for risk of bias (high), study duration (>12 months), economy (developed), region (Americas), age group (adults and children) and UTI symptoms (P < 0.001). There were no data available on mortality for comparison between settings.

Random effects meta-regression analyses of studies reporting on community-acquired E. coli UTI showed that country’s economy (P = 0.008), Asia as a region (P = 0.002), high risk of bias (P = 0.003), year of study (P = 0.020) and studies using only children as the study population (P = 0.030) were the study factors significantly accounting for the observed heterogeneity, responsible for 61 % of the between study variance (Adjusted R²) in ciprofloxacin resistance.

When studies were assessed for risk of bias using the Newcastle-Ottawa scale, 30 % (n = 16) were assessed as having a low risk of bias; 22 % (n = 12) unclear risk of bias and 48 % (n = 26) were deemed to have a high risk of bias. Further analysis of the 16 low risk studies only was consistent with findings reported from the analysis of all studies. An increasing resistance trend over time was also observed, however this increase did not reach statistical significance because of reduced statistical power.

There are a number of notable strengths to our review. To our knowledge, this is the first systematic review to compare the overall prevalence of ciprofloxacin resistance in community- and hospital-acquired E. coli UTI. We undertook a comprehensive literature search process to identify and screen articles against eligibility criteria. Given that generic versions of ciprofloxacin were first marketed at different times in various countries, our choice of 2004 as the start date was therefore made on the basis of changes in the epidemiology of antimicrobial resistant pathogens which had resulted in changes to treatment regimens. A further strength of this systematic review is the development of a peer reviewed, registered protocol prior to undertaking the review. For studies to be included in the review, they were restricted to those that used a standard laboratory UTI criterion of ≥10⁵ cfu/mL as recommended by the CDC. Although applying the internationally recognised CDC criteria may definitely be considered a strength as it ensures the quality and uniformity of included studies, this criterion limited the number of hospital-acquired UTI studies included in our systematic review. Despite this, resistance was still found to be higher in the hospital setting compared to the community setting similar to published studies. While lower counts of uropathogens are relevant for acute episodes of uncomplicated cystitis, the use of different colony counts makes comparison of data between studies difficult. Including all urinary E.coli isolates was considered but not done because this existing surveillance criterion (≥10⁵ cfu/mL and 48 h cut off) is usually applied to defining infections not isolates. Also, including all isolates carries the risk of including duplicates. This approach poses some degree of ascertainment bias as our systematic review focuses on laboratory identified UTIs which may not only underestimate the total number of UTIs but also lead to selection of samples from complicated cases thereby overestimating resistance. Another limitation is the wide variation of resistance estimates between studies and the inclusion of studies having substantial clinical and methodological heterogeneity. Visual inspection of the funnel plot (Fig. 6) showed asymmetry suggesting evidence of publication bias, with studies reporting high resistance rates being more likely to be published posing a limitation to this review. Also, the quality and risk of bias of some of the studies included in the review were assessed as high. These limitations were addressed by undertaking a random effects meta-analysis with subsequent subgroup analyses and random effects meta-regression to explain the sources of heterogeneity. For studies in which the design was not stated, the review authors faced difficulties in categorising such studies hence some of these studies were grouped as non-classifiable. These studies did not provide clear and explicit information on the methods used for conducting the studies. This emphasizes the need for implementation and adherence to clear reporting standards prior to publication of papers. Furthermore, in some included studies, adjustments were not made for important confounding factors relevant to antimicrobial resistance such as antibiotic use and patient demographics including age and sex. For this systematic review, studies on samples obtained from emergency department (ED) patients were classified as community-acquired samples. Included papers did not provide any information on whether some of these patients may have returned from a recent hospitalisation and represented to the ED. Ideally, these should be considered as hospital-acquired infections as some of these patients may have been discharged in the previous 48 h. For the purpose of this review and to overcome inherent variations in how individual studies have defined these patients, we classified all papers reporting on ED patients as community-acquired UTI studies. It was not possible to determine the potential effect of samples obtained from nursing home or residential aged care studies on the pooled resistance because this participant group did not meet the inclusion criteria for analysis. Furthermore, classification of this setting as hospital or community remains controversial. Finally, validity issues may have arisen from the use of different antimicrobial susceptibility test and interpretation methods with differing breakpoints which tend to change over the years. To date, there is still no worldwide consensus on the most suitable antimicrobial susceptibility testing method with the fact that various countries and even laboratories within the same country use different tests and interpretative criteria. Subgroup analysis for AST method was considered but not done because almost all studies used the disk diffusion method and CLSI criteria.

Data supporting the findings of this systematic review are available upon request from the corresponding author.