Extended-spectrum beta-lactamase-producing Enterobacteriaceae (ESBL-PE) among travellers to Africa: destination-specific data pooled from three European prospective studies

Study design, volunteers and samples

For all three prospective studies, the volunteers provided both pre- and post-travel stool samples/rectal swabs. Of the 14 (3.5%) travellers with pre-travel samples positive for ESBL-PE, three had the same strain detected in post-travel samples. In six volunteers, the post-travel sample was negative; these were included in the ESBL-PE(−) group. The five that contracted a different type of ESBL-PE during travel were included in the ESBL-PE(+) group. Travellers who contracted a new ESBL-PE strain during travel constituted the ESBL-PE(+) group, while all others belonged to the ESBL-PE(−) group. The following information was available from all three studies in comparable format: travel itinerary, travel duration, travel dates, age, sex, antimicrobial usage, occurrence of TD, and possible hospitalisation abroad (overnight stay or more).

In all studies, written informed consent was obtained from all participants and the Ethics Committees in the respective organisations approved the study protocols.

Collection of stool samples and identification of ESBL-PE strains

We have described earlier in detail the approaches to collection of samples (stools or swabs) and methods used for identification of ESBL-PE and carbapenemase-producing Enterobacteriaceae (CPE) [4, 6, 10].

Definition of TD and geographical subregions

For the purpose of the present study, TD was defined as three or more loose or liquid stools per day. Geographical subregions in Africa were defined according to the United Nations [16]: Southern Africa, Western Africa, Middle Africa, Eastern Africa, and Northern Africa. Travellers visiting more than one subregion in Africa were categorised on the basis of longest stay.

Statistical analyses

Statistical analyses were carried out with SPSS software version 24 (IBM Corp, Armonk, NY) and Stata version 15.1 (StataCorp. College Station, TX). Binomial regression model was used to obtain profile likelihood confidence intervals for the proportions of travellers with given risk factors and positive for ESBL-PE. The chi-square test, Fisher’s exact test or binary logistic regression analysis were used to compare categorical variables when applicable. Binary logistic regression was used with continuous variables. Variables with a p-value < 0.2 in the univariate analysis for ESBL-PE colonization were subjected to multivariable analysis together with doxycycline as antimalarial, gender and duration of travel in days. The shape of the form for travel duration and age were assessed by cubic splines and appeared log-linear. The interaction between variables of interest and studies was assessed. The final model was built using binary logistic regression analysis with a stepwise backward selection of variables by Akaike Information Criteria (AIC). Factors with 95% confidence intervals ranging only either above or below 1 were considered significant. The three studies pooled in the present paper [4, 6, 10] and the two others [7, 11] used for comparisons were all brought together to produce a forest plot analysis. Heterogeneity between studies in forest plot was measured with I²; values above 75% were considered high, 25–75% moderate, and below 25%. For our pooled data, the interaction between studies and geographical subregions was analysed in the multivariable model.

Demographic data, background characteristics, and occurrence of TD

ESBL-PE acquisition rates by subregion in the pooled data

In the pooled data (Table 2, Fig. 1), 61 (15.4%) travellers became colonized by ESBL-PE; one Dutch traveller to Egypt became colonized by CPE. The highest ESBL-PE colonization rates were seen among travellers to Northern Africa (12/28; 42.9%), followed by Middle Africa (4/15; 26.7%) and Eastern Africa (30/185; 16.2%). Of travellers to Western and Southern Africa, 10.0% (11/110) and 6.9% (4/58), respectively, acquired ESBL-PE (Tables 1 and 2).

Results of the multivariable analysis for risk factors of ESBL-PE colonization in Africa

The initial univariate analysis (Table 1) detected the following factors with p < 0.2: age, original study, sampling technique, subregion in Africa, use of fluoroquinolones, beta-lactams or other AB/regimen not known, TD, and overnight hospitalisation abroad. When all of these factors, together with gender, duration of travel, and use of doxycycline as antimalarial were subjected to multivariable analysis, the following were found to be independently associated with increased risk: travel to Northern Africa, overnight hospitalisation abroad, age, TD and use of fluoroquinolones (Table 1).

Results of meta-analysis of five studies

Table 4 and forest plot analysis (Fig. 2) show the ESBL-PE colonization rates from the three studies pooled [4, 6, 10], together with investigations by Ruppé et al. and Arcilla et al. [7, 11] in relation to geographical subregion visited. For Southern and Northern Africa, heterogeneity between the five studies appeared low (I² = 0.0 and 0.0%, respectively), for Middle Africa moderate (I² 52.0%), and Eastern and Western Africa high (I² 88.9 and 90.4%, respectively). In the multivariable regression model of our pooled data, the interaction between subregions and the three studies was not found significant at 5% significance level.

	All	Northern Africa	Middle Africa	Eastern Africa	Western Africa	Southern Africa
Kantele et al. [6] (Finnish study) 2009–2010
ESBL-PE (+) among travellers n (% of all visitors to subregion)	25/196 (12.8)	2/3 (66.7)	1/4 (25.0)	14/86 (16.3)	5/78 (6.4)	3/25 (12.0)
Paltansing et al. [4] (Dutch study I) 2011
ESBL-PE (+) among travellers n (% of all visitors to subregion)	29/103 (28.2)	7/13 (53.8)	3/7 (42.9)	14/54 (25.9)	4/12 (33.3)	1/17 (5.9)
Reuland et al. [10] (Dutch study II) 2012–2013
ESBL-PE (+) among travellers n (% of all visitors to subregion)	7/97 (7.2)	3/12 (25.0)	0/4 (0.0)	2/45 (4.4)	2/20 (10.0)	0/16 (0.0)
Ruppé et al. [7]2012–2013 (data on travellers visiting only one country)
ESBL-PE (+) among travellers n (% of all visitors to subregion)	89/182 (48.9)	N/A	24/45 (53.3)	21/37 (56.8)	44/99 (44.4)	0/1 (0)
Arcilla et al. [11]2012–2013
ESBL-PE (+) among travellers n (% of all visitors to subregion)	118/508 (23.2)	34/81 (42.0)	N/A	57/205 (27.8)	20/106 (18.9)	7/116 (6.0)
Combined total: ESBL-PE colonization rates	268/1086 (24.7)	46/109 (42.2)	28/60 (46.7)	108/427 (25.3)	75/315 (23.8)	11/175 (6.3)

ESBL-PE colonization rates in northern Africa

Our pooled data showed the highest acquisition rates (12/28; 42.9%) among visitors to Northern Africa, which accords with the results from the study by Arcilla et al. (42.0%) [11]; Ruppé et al. [7] did not report visitors to this subregion. Similar (43–44%) rates have been reported among Swedish travellers [3, 12]. Visitors to Egypt appear to be at particularly high risk; 70.6% (12/17) of our subjects, 80.0% (23/40) of those in a study by Arcilla et al., and 50% (19/38) of those in another by Tham et al. became colonized [11, 14]. Moreover, all 12 travellers colonized by ESBL-PE in Northern Africa had visited Egypt. It is noteworthy that these proportions are as high as those among travellers to India/South Asia in various investigations [4–8, 11]. Bassyouni et al. reported carriage rates as low as 21% among healthcare workers in Egypt [17].

ESBL-PE colonization rates in middle Africa

To our knowledge, only one previous study has reported ESBL-PE acquisition rates among visitors to Middle Africa; Ruppé et al. [7] found 53.3% (24/45) of travellers to be colonized. In our pooled data, colonization rates in Middle Africa ranked second (4/15; 26.7%) among the subregions. In nonclinical samples obtained from local populations, carriage rates as high as 59% have been shown among healthy children in the Central African Republic [18], and 44–57% among inpatient carers, hospital workers and their household members in Cameroon [19].

ESBL-PE rates in eastern Africa

Colonization rates among travellers to Eastern Africa (30/185; 16.2%) were lower than those reported by Arcilla et al. (57/205; 27.8%) [11], Lubbert et al. (12/47; 25.5%) [8], and Ruppé et al. (17/29; 56.8%,) [7]. Our moderate colonization rates are supported by findings among local populations: ESBL-PE carriage rates between 11.6 and 16.5% have been reported for healthy community children in Tanzania, [20, 21] and 5.3% for locals in Uganda, [22].

ESBL-PE colonization rates in western Africa

ESBL-PE acquisition rates in Western Africa appear moderately low, but the results differ between studies: our pooled data showed proportions (11/110; 10.0%) close to those presented by Arcilla et al. (20/106; 18.9%) [11], while higher rates have been found by Ruppé et al. (44/99; 44.4%) [7] and Lubbert et al. (5/12; 38.5%) [8] among German travellers to Western and Middle Africa. Moreover, in the research by Frickmann et al., 27.1% (13/48) of European military personnel with diarrhoea in Mali became colonized by ESBL-PE [23]. As for local populations, colonization rates of 22% have been reported for healthy volunteers in Burkina Faso [24] and 33% for healthy community children in Guinea-Bissau [25].

ESBL-PE colonization rates in southern Africa

Our low rates in Southern Africa (4/58; 6.8%) accord with those found by Arcilla et al. (7/116; 6.0%) [11] and Lubbert et al. (2/18; 11%) [8]. In our pooled data, the vast majority had visited South Africa or Namibia. Consistent with the low ESBL-PE acquisition rates, one study exploring local populations in South Africa reported maternal faecal carriage rates of 4.4% in South Africa [26].

Findings from multivariable analysis

Limitations of the study

As the data for the joint risk factor analysis were derived from three separate studies, some data had been collected in differing formats rendering the results incomparable. Moreover, although pooling served to increase the validity and precision of study results, the data remained insufficient in some occasions for analysis in any great detail: In Additional file 1: Table S1, we present the factors available from two out of three studies [4, 6]: purpose of travel, diet (omnivore or vegetarian), type of accommodation, use of medications (antidiarrhoeals, proton-pump inhibitors, and antiemetics) and contact with local health care (other than hospitalisation). The five investigations appeared heterogeneous in the forest plot analysis, however, in the multivariable analysis of the pooled data, the interaction between subregions and studies was not found statistically significant.

Information concerning mild gastrointestinal symptoms in the ‘no TD’ group was only available for the Finnish volunteers (48.8% of all ‘no TD’ cases). To pool the three studies, we had to define TD as three or more stools per 24 h; milder diarrhoea cases were categorised as ‘no TD’, although even mild TD also may predispose to ESBL-PE acquisition.