Study selection and characteristics
A total of 13’166 records were identified; 4’228 items in Embase, 2’223 in Medline, 3’541 in Global Health, 2’603 items in PubMed, and 571 in CINAHL Plus and additional 71 records were identified, 68 items in IMEMR and 3 items in the MSF Field Research Website (40). After removing duplicates, titles and abstracts of 6’684 records were screened, of which only 431 were eligible for full text screening. Of the 431 articles read in full text, 132 met the inclusion criteria (Figure 2) and were included.
Table 1 describes the study characteristics. All the articles selected for inclusion in the study were in English except one study that was in French (100). More than half of included studies were conducted in only three countries: Egypt (26.5%), KSA (18.9%), and Iraq (18.2%). In all the other countries included in the review only between one and less than ten articles were included, and two countries, Djibouti and Qatar, had no included article.
A higher proportion (61.4%) of articles were from stable countries and 51 articles, described AMR in fragile and conflict-affected countries and almost half of these published in Iraq (Table 1).
Out of the 132 articles, 107 did not specify the nationality of the individuals from which the specimens were obtained. No assumption was made that a study conducted in one country would describe exclusively the population native of the same country, in virtue of the intense migration patterns from outside and within the Middle East (173, 174).
The vast majority of studies included were conducted in hospital and hospital’s laboratory settings (50.0% and 41.7% respectively). Community and primary health care settings where largely underrepresented (5.3%). Inpatients constituted the source of half of the data (50.0%) compared to outpatients. The source of microbiological data was not specified in 21.2% of studies.
Similarly, almost half (48.5%) of the included articles did not specify the age groups and/or genders included in the study population. When specified, adults were the most commonly studied population (33.3% of the included papers), followed by children (25.0%). Only one paper reported results from a study population entirely represented by men (48).
Only five articles were investigating AMR in immunosuppressed populations, represented either by HIV-infected patients (59), subjects who underwent liver transplantation (47), or patients affected by different types of malignancies (91, 115, 172).
Appraisal of risk of bias
The detailed summary of the assessment of risk of bias performed is available in Appendix 2 and 3.
All studies included in this review were screened for the completeness of information provided including; the study design, research question and objectives, and description of the findings in terms of person, place and time, along with the justification of the sample size included and the provision of a measure of random variation of the presented results.
The study design was clearly stated in 29 out of the 132 articles included (Table 1). Although the study question was rarely focused in terms of population of interest, geographic area and time span of observation (18.9%, 25/132), the study objectives were detailed in the majority of papers (77.3%, 102/132). Time period of observation and geographical area of interest were clearly specified in 96.2% (127/132) of articles, while detailed demographic information on the study population, at least in terms of age groups and gender included, were available for 32.6% (43/132).
Only one study reported sample size calculation and employed representative population although not powered for the prevalence detected nor for the comparison between the sub-groups performed (95). Measures of sampling variation in the results were reported by only four studies (95, 161, 165, 168).
Statistical methods were in general poorly detailed, and mostly summarized in terms of descriptive statistics. Only three studies described the use of multivariate analysis – usually logistic regression – to control for confounding (45, 147, 168). None of the studies took missing data into account in the analysis nor in the discussion, and similarly limitations and potential sources of bias were never mentioned in the discussion, with the exception of one surveillance study performed in Lebanon (123).
A substantial number of articles mentioned exclusively clinical suspicion from the attending physician as diagnostic criteria and rarely a clear case definition was detailed for the clinical syndrome under study. No clear sources for patient’s socio-demographic data were descripted.
In general, cross-sectional studies presented major flaws in terms of sample selection, statistical methods employed (not described in the majority of cases), and possibility of misclassification of both exposures and outcome under study.
The cohort studies included in the review (52, 131, 161) showed overall lower risk of bias compared to other study designs. Two out of the three studies (52, 131) lacked detailed description of the statistical methods employed, in particular if and how control for confounding was performed and how losses to follow up were dealt with. Remarkably, one of the cohort studies included was conducted in a non-conflict affected country, but describing findings on a conflict-affected population, as it investigated AMR in war wounded civilian Syrian patients admitted to a Jordanian hospital (161).
The two case-control studies included (57, 165) poorly described the inclusion criteria for the cases and did not clearly detail the recruitment of controls. Data extracted from these two studies were not included in the quantitative analysis.
Synthesis of results
Patterns of ABR in the Middle East and differences between conflict-affected and non-conflict-affected countries
The most commonly described pathogens were, in order of frequency, E. coli (40.9% of the included studies, 54/132), S. aureus (34.1%, 45/132), Acinetobacter spp. (29.5%, 39/132), and K. pneumoniae (25.8%, 34/132), both in conflict and non-conflict affected settings. Other bacteria were less represented in this review, and no study investigated AMR in N. gonorrhoeae.
Antibiotic susceptibility testing (AST) was performed for a total of 73 different antibiotics, with substantial inhomogeneity across studies for each pathogen. Such heterogeneity, together with that observed in terms of study settings, laboratory methodologies employed, and clinical conditions investigated, along with the high overall risk of selection bias, misclassification of the exposure categories, and lack of control for confounding, did not allow to perform a meta-analysis as quantitative synthesis of the results from the standpoint of AMR profile of the studied pathogens. However, in Table 2 median percentages and IQR of specific AMR profiles are reported for selected GLASS pathogens, focusing on the most frequently reported ones.
A third (32.3%) of E. coli isolates showed ESBL-producing profile, and 8.1% resistance to carbapenems (IQR 0.6-25.3). In both cases, the proportion of resistant isolates appeared to be higher in non-conflict affected countries compared to conflict-affected ones.
Regarding S. aureus, the overall median percentage of methicillin-resistance was 45.1% (IQR 26.0-61.6). Unfortunately, the number of articles was too small in conflict affected compared to non-conflict affected countries to make a reliable comparison, and again study populations were very heterogeneous (44, 95, 144). However, from the few studies included, no major difference was detected between the two contexts.
Proportion of carbapenem resistance among Acinetobacter spp. isolates was significant (74.2%) and appeared slightly higher in conflict-affected settings compared to stable countries (78.0% vs 72.7%, respectively).
Only a small proportion of samples of K. pneumoniae was tested for carbapenem resistance and ESBL-production (56.5% and 42.7% respectively) and 28.0% of samples appeared to be ESBL producers, with a substantial difference between conflict and non-conflict affected countries (75.5% vs 22.5%, respectively). The carbapenem resistance profile did not vary significantly between the two contexts, with an overall proportion of cases of 15.4%.
As for S. pneumoniae, half of all isolates were detected as non-susceptible to penicillin, with substantially higher proportion in stable contexts (55.0%) compared to conflict-affected settings (10.9%).
Salmonella spp. and Shigella spp. AMR were investigated in too few studies (five and three respectively) to allow a valid summary to be included.
Table 3 and 4 in the annex offer more details on the number of studies on each pathogen per country in conflict-affected vs non-conflict-affected countries, respectively. Table 5 and 6 provide details on the number of studies per site of infections in conflict-affected vs non-conflict-affected countries, respectively. From these tables we can observe that studies conducted in non-conflict affected countries appeared to be slightly more focused in terms of site of infection under investigation compared to studies conducted in conflict affected countries, where in a substantial proportion of cases the site of infection was not specified, or the authors were analysing resistance profiles of bacteria isolated from a miscellanea of specimens.