Worldwide antibiotic resistance dynamics: how different is it from one drug-bug pair to another?
================================================================================================

* Eve Rahbe
* Laurence Watier
* Didier Guillemot
* Philippe Glaser
* Lulla Opatowski

## Summary

**Background** Antibiotic resistance (ABR) is a major concern for global health. However, factors driving its emergence and dissemination are not fully understood. Identification of such factors is crucial to explain heterogeneity in ABR rates observed across space, time and species and antibiotics.

**Methods** We analyzed count data of resistant isolates from 51 countries over 2006-2019 for nine drug-bug pairs from the surveillance database ATLAS. We characterized ABR spatial and temporal patterns and used a mixed-effect negative binomial model, accounting for country-year dependences with random effects (RE), to investigate associations with potential drivers including antibiotic sales, economic and health indicators, meteorological data, population density and tourism.

**Findings** ABR patterns were strongly country and drug-bug pair dependent. In 2019, median ABR rates ranged from 6·3% for carbapenem-resistant (CR) *Klebsiella pneumoniae* to 72·3% for CR-*Acinetobacter baumannii*, with heterogeneity across countries (up to 50·2% of interquartile range for CR-*A. baumannii*). Over 2006-2019, carbapenem resistance was on the rise in >75% of investigated countries but no global trend was observed for other resistances. Multivariable analyses identified significant associations of ABR with antibiotic sales, but only in fluoroquinolone-resistant *Escherichia coli*, CR-*A. baumannii* and penicillin-resistant *Streptococcus pneumoniae;* with temperature in investigated Enterobacterales but not in other drug-bug pairs; and with the health system quality for most drug-bug pairs, except in *Enterococci* and *S. pneumoniae*. Despite wide consideration of possible explanatory variables, drug-bug pairs still showed high spatial RE variance.

**Interpretation** These results reflect the diversity found among human bacterial pathogens and stress the difficulty to generalize global antibiotic resistance findings.

**Funding** Independent research Pfizer Global Medical Grant; ANR Labex IBEID (ANR-10-LABX-62)

## Introduction

Antibiotic resistance (ABR) in clinically relevant bacteria is a major public health threat.1 The latest scenarios show that antimicrobial resistant infections are associated with 4·95 millions deaths in 2019,2 and could cost up to 3·5 billions dollars per year by 2050 only in countries from the Organization for Economic Cooperation and Development (OECD).3 Although ABR is a global problem observed in all countries worldwide, resistance rates show important geographical differences and greatly vary depending on the bacterial species and antibiotic resistance considered.3 Factors explaining these variations are not fully understood.

Recent studies investigating factors associated with global ABR levels suggested that in addition to antibiotic consumption, sanitation, access to clean water, public health-care expenditure, governance, or corruption are correlated to ABR rates.4,5 Containing ABR spread, rather than only limiting antibiotic consumption, thus seems to be crucial to reduce its global burden but the limited knowledge about its determinants impedes the setup of efficient control measures. Other socioeconomic and behavioral factors could play a role in ABR transmission within national borders, such as human mobility or medical tourism.6 In addition, meteorological factors such as temperature have been suggested to be involved in ABR dissemination and maintenance in the United States or Europe. 7–9

Global ABR geographical patterns have been previously analyzed using data from one or two selected pathogens or aggregated resistance indices.4,5,10 However, as pathogens exhibit different resistance mechanisms, thrive in different ecological niches, and can be associated with different epidemiological sources (hospital-acquired or community-acquired infections), understanding ABR dynamics requires studying it at the bacterial species level.

In this context, data collected by global surveillance programs are needed to identify determinants of ABR dynamics globally in different pathogens. Specifically, the Antimicrobial Testing Leadership and Surveillance (ATLAS) system is a ABR hospital surveillance program,11 which gathers clinical isolates from different bacterial species of which many are listed in the WHO list of high-priority pathogens for the development of new antibiotics.12 ATLAS collects longitudinal data over all continents and includes lower middle- to high-income countries. It uses a standardized and centralized procedure of determining minimal inhibitory concentrations (MICs) across multiple species and antibiotic drugs combinations (“drug-bug pairs”).

In this study, we aimed to analyze spatial-temporal dynamics of major critical drug-bug pairs from worldwide hospital ABR surveillance data collected through ATLAS and identify key factors - antibiotic sales, meteorological and climatic variables, wealth and health metrics, population density, tourism - associated with such dynamics.

## Research in context

#### Evidence before this study

We performed a PubMed search for articles in English published between 2000 and 2021. Search terms were: (“global” OR “worldwide” OR “country” OR “country-level”) AND (“antibiotic resistance” OR “antimicrobial resistance” OR “bacterial resistance” OR “drug resistance”). We excluded studies analyzing viruses, parasites, fungi, cancers, cholera, and tuberculosis. After filtering for articles referring to ecological or prevalence studies focusing on only one country, one bacterial species or one infection source, we found 6 relevant studies. Two studies focused on ABR rates prediction at the country-level either spatially or temporally. The four other studies analyzed determinants of ABR spread: two articles found a link between climate warming and ABR in European countries, and two articles studied multiple individual socio-economic and national associated factors, both in Europe and in a worldwide dataset. From molecular studies, international travelers were found to be colonized with resistant bacteria suggesting that tourism could also play a role in ABR spread. No other publications had compared multiple bacterial species and antibiotic resistances using longitudinal data to assess differences in ABR spread factors at a worldwide level.

#### Added value of this study

ATLAS represents a unique dataset, useful to compare ABR levels worldwide and question their dynamics. This study represents longitudinal hospital surveillance data for 51 countries, not restricted to European Union countries, and over 14 years (2006-2019). To our knowledge, this is the first study to analyze ABR hospital surveillance data from nine different clinically relevant drug-bug pairs and to characterize the observed heterogeneity in ABR rates across different countries and years. Our statistical analysis approach includes both spatial and temporal dimensions of the ABR phenomenon, with country-year dependent outcome variable and co-variables, which were not considered in previous studies. Moreover, the choice of covariables is not restricted to global indices, but rather detailed metrics such as different meteorological factors (rainfall, relative humidity, temperature, and extreme climatic events such as flooding) or antibiotic sales described at the antibiotic class level.

#### Implications of all the available evidence

Analysis of the ATLAS data showed that ABR rates are highly dependent on the country and most importantly on the drug-bug pair considered. Results from our statistical analysis suggested that factors associated with ABR rates were different across drug-bug pairs but more similar within a bacterial species, reflecting different underlying ecological behaviors. Variance between countries was only explained by the proposed factors in *Escherichia coli* where antibiotic sales, temperature, extreme climatic events, and the health system quality could explain most ABR rates differences. For other drug-bug pairs, strong spatial variance remained unexplained. Overall, results from this study suggest that ABR should be considered as a plural problem whose control should be tailored regarding the country or the drug-bug pair under consideration. It also highlights the importance of robust surveillance systems for the understanding of global ABR.

## Methods

### Antibiotic resistance data

Antibiotic resistance data were obtained from the ATLAS database11. Count data of non-resistant and resistant isolates from 51 countries (appendix p 3) over 14 years (2006-2019) were extracted. Nine clinically relevant antibiotic resistance-bacterium pairs (called drug-bug pairs) were analyzed: fluoroquinolone-resistant *Escherichia coli* (FR-*Ec*), aminopenicillin-resistant *E. coli* (APR-*Ec*), third generation cephalosporin-resistant *E. coli* (3GCR-*Ec*), 3GC-resistant *Klebsiella pneumoniae* (3GCR-*Kp*), carbapenem-resistant *K. pneumoniae* (CR-*Kp*), carbapenem-resistant *Pseudomonas aeruginosa* (CR-*Pa*), carbapenem-resistant *Acinetobacter baumannii* (CR-*Ab*), vancomycin-resistant *Enterococcus faecalis* and *Enterococcus faecium* (VR-*E*) and penicillin-resistant *Streptococcus pneumoniae* (PR-*Sp*). Numbers of isolates tested for each drug-bug pair are provided in the appendix p 4. Isolates came from a variety of infection sites with blood, endotracheal aspirate, sputum, urine, and wound being the five most frequent. MIC values were reported in the ATLAS database and isolates were categorized into Susceptible (S), Intermediate (I) or Resistant (R) based on the 2019 MIC breakpoints from the *European Committee on Antimicrobial Susceptibility Testing* standards (appendix pp 5-6).13 An isolate was considered resistant if it was in the R category, except for PR-*Sp* where an isolate was considered resistant if it was in the I or R categories. An isolate was considered resistant to an antibiotic class if it was resistant to at least one of the antibiotics of the class (appendix p 7).

Data filtering was performed to ensure ABR data reliability. For each drug-bug pair, only observations with more than 10 isolates for one country and countries with more than 5 years of ABR data available over the full period were included. Data imputation of missing ABR data was computed for each country by using a moving average estimate with a time window of k = 1 year. Imputation was only performed when the number of consecutive missing years was less than 3.

### Co-variables data

The co-variables data described below were tested as explanatory variables in the statistical analyses of ABR drivers (appendix pp 8-9). Antibiotic sales data from hospitals and retail sectors were obtained from the IQVIA MIDAS database.14 Data from Denmark, Latvia, Lithuania, and the Netherlands were not available through this database, and were obtained from the ESAC-Net database by converting Defined Daily Doses into grams as presented in appendix p 10. We selected the five most frequently sold classes of antibiotics in IQVIA over all countries and years (broad-spectrum penicillins, cephalosporins, macrolides, quinolones, trimethoprim and combinations of sulfonamides); and carbapenems and glycopeptides also corresponding to resistances from the considered drug-bug pairs (appendix p 11). Global sales represented the sum of these seven classes. Average temperature, minimum temperature, average rainfall and average relative humidity per country-year were obtained through the MERRA-2 dataset, using the capital city value for each country.15 Extreme climatic events (droughts, flooding or extreme temperatures) faced by the population in 2009 was obtained from the World Bank Data.16 Population size, population density, gross domestic product (GDP) per capita, and international tourism (legal arrivals and departures) data were obtained through the WBD at the country-year level.16 As a proxy for national health system qualities, the 2019 Global Health Security index (GHS, [0-100]) was used.17

### Statistical analysis

#### Descriptive statistics

ABR rate was defined as the proportion of resistant isolates over the total number of isolates tested. Median ABR rate and interquartile ranges (IQR) across all countries were estimated for the year 2019, the most recent year of the dataset. Temporal trends of ABR rates from 2006 to 2019 for each country and each pair were estimated using a weighted linear regression, weights being the total number of isolates tested.

#### Mixed-effect negative binomial model

A mixed-effect model was used to analyze ABR rates and to identify associated factors by accounting for the dependence of the data between countries and across years.18 The count of resistant isolates per country-year was considered as the response variable. A negative binomial (NB) response distribution was chosen to account for data overdispersion. The model included spatial and temporal random effects (RE) and fixed effects (FE) with the co-variables data. The model equation used for each drug-bug pair was the following: ![Formula][1]</img>  where c is the country, t the year, *Y**c,t* is the number of resistant isolates per country and year; *μ**c,t* the expected number of resistant isolates; *θ* the dispersion parameter of the NB distribution; *n**c,t* the number of isolates tested (offset); *π**c,t* the proportion of resistant isolates (rate); *β* the fixed effects parameter vector of size *p* (*p* the number of fixed effects); ***X*** the fixed effects model matrix of size C\*T\**p* (with C total number of countries and T total number of years); *b**c* the country-dependent random intercept; ϕ*t* the year-dependent random intercept and *ε**c,t* the residuals of the model. Random effects and residuals are assumed to be normally distributed.

Intercept-only (called M_null), univariate (M_uni) and multivariable models (M_multi) were evaluated independently for all drug-bug pairs. Co-variables data were standardized (centered with their mean and scaled with their standard deviation). Correlation analysis between each co-variable was performed using the Pearson coefficient (r), and co-variables associated with *r* > |0.7| were excluded from analyses (appendix p 12). For antibiotic sales variables, only global sales and sales of the antibiotic selecting for the specific resistance (called “antibiotic sales of interest”) were tested in the univariate analysis. Co-variables associated with a p-value < 20% in the univariate analysis were included in the multivariable analysis. Backward selection was performed using hypothesis testing method (z-test). Every co-variable left in the M_multi final multivariable models were statistically significant with a p-value < 5%. Akaike Information Criterion (AIC) values were reported to compare M\_null and M_multi models for all drug-bug pairs. The model parameters were estimated using maximum likelihood by Laplace approximation using the *lme4* R package.19 For CP-*Kp* and VR-*E* where resistance levels were frequently zeros, only countries where at least two data points were different from zero over the full period were included. Observations where covariables were not available were also excluded.

#### Sensitivity analyses

To assess the potential impact of the filtering threshold related to the minimum number of isolates per country-year, we repeated analyses with a filter on 20 isolates instead of 10. To assess whether using ESAC-Net data for Denmark, Latvia, Lithuania, and the Netherlands had any major impacts on the results, we repeated analyses by excluding these countries. To assess whether the infection site might affect ABR rates and associations with covariables, we carried out analyses on a subset of the data where isolates only came from blood source.

All analyses were carried out using R, version 4.0.3.20

### Role of the funding source

*The funder had no role in data analysis, data interpretation or writing of the paper. The authors had full access to all the data in the study and had final responsibility for the decision to submit for publication*.

## Results

Distributions of antibiotic resistance (ABR) rates across countries in 2019 highlighted strong heterogeneity across the nine drug-bug pairs analyzed. The median ABR ranged from 6·3% for CR-*Kp* to 72·3% for CR-*Ab* (figure 1A). Interestingly, although CR-*Ab* and APR-*Ec* median rates were similarly high, 72·3% and 68·0% respectively, higher dispersion was found for CR-*Ab* compared to APR-*Ec* (interquartile range (IQR) of 50·2% and 18·2%, respectively), revealing different ABR levels distributions across countries (figure 1B and 1C).

![Figure 1.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2022/02/09/2022.02.09.22270726/F1.medium.gif)

[Figure 1.](http://medrxiv.org/content/early/2022/02/09/2022.02.09.22270726/F1)

Figure 1. Worldwide antibiotic resistance rates distribution, 2019, ATLAS.
ABR rate is reported in percentage of resistant isolates over total number of tested isolates per country for each drug-bug pair. (A) ABR rates for all drug-bug pairs are represented with boxplots, with medians (line) and lower and upper quartiles (box limits). (B-C) Maps of worldwide resistance rates for APR-*Ec* and CR-*Ab* respectively, two pairs exhibiting high median rates in 2019. Resistance rate (%) is indicated by colors in the scale below the maps. Grey countries indicate missing value for 2019. White countries are not included in the analysis.

From 2006 to 2019, ABR showed different temporal trends between drug-bug pairs (appendix pp 14-18). The heat map in figure 2, displaying amplitude of trends, highlights that ABR temporal trends do not specifically cluster by world’s region, but rather cluster by species or resistances. Noticeably, *E. coli* pairs clustered together, suggesting similarities in their temporal trends, but no clear worldwide increasing or decreasing temporal pattern was observed: 46% of countries exhibited increasing trends for FR-*Ec*; 40% for APR-*Ec* and 50% for 3GCR-*Ec* (appendix pp 14-18). No global patterns could be observed neither for 3GCR-*Kp*, VR*-E* and PR*-Sp* with respectively 63%, 57% and 40% of countries exhibiting increasing trends. By opposition, a massive number of countries with increasing ABR trends were found for drug-bug pairs associated with carbapenems resistance: 80% of countries had increasing trends for CR-*Kp*; 78% for CR-*Pa*; and 85% for CR-*Ab* (appendix pp 14-18).

![Figure 2.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2022/02/09/2022.02.09.22270726/F2.medium.gif)

[Figure 2.](http://medrxiv.org/content/early/2022/02/09/2022.02.09.22270726/F2)

Figure 2. Antibiotic resistance temporal trends (2006-2019) by country and drug-bug pair.
The heat map shows temporal trends, estimated from weighted linear temporal regressions (appendix pp 14-18), representing the ABR rate change per year for each country and drug-bug pair. Temporal trends are colored from blue (highly decreasing trend) to red (highly increasing trend) as in the scale on the left. A grey cell represents a missing value for a specific country and drug-bug pair. Clustering was performed on both rows and columns (complete linkage method for hierarchical clustering, using Euclidean distance). Countries were categorized into world’s regions based on World Bank indicators and drug-bug pairs were categorized into bacterial species, as indicated in the figure keys on the left and right respectively.

From the multivariable analyses, we could jointly assess the association between ABR rates and the proposed co-variables. Results are summarized in table 1 and models’ fits are presented in appendix pp 19-23. We found that key factors associated with ABR varied greatly between drug-bug pairs but were more similar within same bacterial species. First, antibiotic sales were only significantly associated in three drug-bug pairs: FR-*Ec* rates were positively associated with quinolones sales; CR-*Ab* rates were also positively correlated with carbapenems sales but inversely correlated with global antibiotic sales; finally, PR-*Sp* resistance was positively associated with global antibiotic sales. Second, meteorological factors were mostly found to be associated with Enterobacterales drug-bug pairs: all *E. coli* resistances were significantly associated with average temperature and extreme climatic events; average temperature was also strongly associated with CR-*Kp*. PR-*Sp* rates were inversely associated with average relative humidity. Third, we found that the GHS index was inversely associated with ABR rates in all *E. coli* and K. *pneumoniae* drug-bug pairs, as well as in *P. aeruginosa* and *A. baumannii* pairs. CR-*Ab* was the only pair positively associated with the GDP. This agrees with the univariate analysis, in which most drug-bug pairs were significantly inversely associated with GDP whereas CR-*Ab* was the only pair where the relationship was positive (appendix pp 24-28). Finally, tourist departures were inversely associated with ABR rates in two pairs, APR-*Ec* and CR-*Pa*. VR-*E* was not significantly associated with any of the proposed factors in the multivariable analysis. Of interest, population density, rainfall and tourism arrivals were not selected for any drug-bug pairs. Univariate analyses results are summarized in appendix pp 24-28.

View this table:
[Table 1.](http://medrxiv.org/content/early/2022/02/09/2022.02.09.22270726/T1)

Table 1. Estimated coefficients from the mixed-effect negative binomial model for each drug-bug pair.
* n represent the number of observations; C, the number of countries and Y, the number of years.

** M\_null represent, the intercept-only model and M\_multi, the final multivariable model (including only statistically significant variables after backward selection, p-value < 5%). Coefficients for the fixed effects are reported on the exponential form. Spatial and temporal random effects’ variances are reported, respectively ![Graphic][2]</img> and ![Graphic][3]</img>. AIC are reported for all models.

For all drug-bug pairs, we found higher heterogeneity in ABR rates between countries than between years (table 1): in reference M_null models, spatial RE variance was high (ranging from 0·03 for APR-*Ec* to 1·92 for CR-*Kp*) in comparison with temporal RE variance (0·0003 for APR-*Ec* to 0·5321 for CR-*Kp*). For all drug-bug pairs, the M_multi model had lower AIC compared with M_null (table 1) suggesting that multivariable model was better at explaining the data. The introduction of covariables led to a high RE variance reduction in *E. coli* resistances, 3GCR-*Kp* and CR-*Pa* whereas it was smaller for the other pairs.

The resulting unexplained spatial RE variances after introduction of explanatory variables in final M_multi models are shown in figure 3. We noticed that *E. coli* pairs, 3GCR-*Kp* and CR-*Pa* exhibited low spatial RE variance compared to the other pairs. Moreover, we observed that some country-drug-bug trios systematically exhibited outlying spatial RE estimates compared to the observed average ABR rate in ATLAS. For example, spatial RE estimate was systematically high for Mexico (for all *E. coli* pairs and PR-*Sp*) and South Korea (for FR-*Ec*, 3GCR-*Ec*, CR-Ab and PR-*Sp*) and low for Japan (in APR-*Ec*, 3GCR-*Kp* and CR-*Ab*) and Czech Republic (CR-*Kp*, CR-*Ab* and PR-*Sp*) in figure 3. This suggests that the proposed co-variables are not sufficient to explain the observed spatial heterogeneity in ABR rates (appendix pp 35-39).

![Figure 3.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2022/02/09/2022.02.09.22270726/F3.medium.gif)

[Figure 3.](http://medrxiv.org/content/early/2022/02/09/2022.02.09.22270726/F3)

Figure 3. Spatial random effects distribution from final multivariable models for each drug-bug pair.
Violin plots of spatial random effects resulting from final multivariable (M_multi) models by drug-bug pair, colored by bacterial species. Random effects’ estimates from Czech Republic (Europe and Central Asia), Japan, South Korea (East Asia and Pacific) and Mexico (Latin America and Caribbean) are highlighted based on their world’s regions and corresponding flags are shown.

Sensitivity analysis on the choice of isolates threshold and on the exclusion of countries for which no IQVIA antibiotic sales data was available led to comparable results and similar interpretations to those presented above, suggesting no major impact of our assumptions (appendix pp 29-33). Regarding the analysis restricted to blood isolates only, it represented a reduction in the isolates sample size: 24% of all *E. coli* and VR-*E* isolates came from blood, and down to 11% for CR-*Pa* (appendix p 4). They resulted in overall consistent estimates compared with the main analysis (see appendix p 34 for more details on sensitivity analyses results). We noted that GDP was not significant anymore for CR-*Ab* resistance and association between temperature and Enterobacterales pairs was found in fewer pairs.

## Discussion

Using longitudinal data provided by the surveillance program ATLAS, we analyzed antibiotic resistance (ABR) spatial and temporal patterns as well as ABR determinants for nine drug-bug pairs of clinical relevance. Our results confirmed that worldwide ABR dynamics were highly drug-bug pair dependent, both spatially and temporally. We found that key factors varied greatly between drug-bug pairs, but some similarities existed between bacteria of the same species (table 1). However, after hypothesis driven investigation of factors, high unexplained country-level variance remained in most of the drug-bug pairs.

The ATLAS data exhibited strong heterogeneities in spatial and temporal ABR patterns across drug-bug pairs. In 2019, the highest median ABR rates were observed in APR-*Ec* and CR-*Ab*. Aminopenicillins are broad-spectrum penicillins, the most used antibiotics class worldwide.21 Between 2000 and 2015, consumption of broad-spectrum penicillins in low- and middle-income countries (LMICs) doubled, reaching levels observed in high-income countries (HICs).21 In contrast, consumption of carbapenems - a last resort antibiotics class - increased in all countries but levels were still much lower in LMICs compared to HICs. This could explain the large inter-country variation observed in CR-*Ab* rates in 2019.21 Moreover, we found that carbapenem-resistant Gram-negative bacteria were on the rise globally in ATLAS in agreement with previous studies.22,23

The antibiotic consumption-resistance association strongly depended on the considered drug-bug pair. We found that only FR-*Ec* and CR-*Ab* were significantly associated with sales of their selecting antibiotics in the multivariable analysis, namely quinolones and carbapenems. PR-*Sp* was significantly associated with global antibiotic sales. Surprisingly, none of the 3GC resistances were associated to any antibiotic sales, possibly reflecting decreasing cephalosporins consumption in HICs.21 While antibiotics consumption exerts a selective pressure and favors the emergence of resistant bacteria, our results suggested that dissemination and persistence of such resistant bacteria might contribute more widely to the observed ABR heterogeneities worldwide than consumption itself. Significant signals were found in all drug-bug pairs for factors including meteorological factors, extreme climatic events, the health system quality, wealth, or tourism. Thus, the spread of resistant bacteria could influence ABR rates in populations where antibiotic use is low. Olesen et al. indeed showed that spillover can happen at large geographical scales, weakening observed global antibiotic consumption-resistance associations.24

Warm temperatures were found to be significantly associated with ABR rates only for Enterobacterales, namely *E. coli* and *K. pneumoniae* in the multivariable analysis. This is consistent with previous ecological studies from smaller datasets in the United States and in Europe.7–9 Here we show that this association is highly drug-bug pair dependent, as the association does not hold true for other bacterial species. In addition, extreme climatic events were only positively associated in *E. coli*. This result suggests that natural disasters such as flooding, rather than rainfall itself, potentially cause ABR spread by disrupting infrastructures and creating sanitation failures.25 Examples from the literature report elevated numbers of resistant infections in natural disasters survivors or unprecedented levels of ABR genes in water following hurricanes.26,27 This association might be especially strong in orofecally-transmitted bacteria such as *E. coli*.

High GHS index, used as a proxy for the health system quality, was significantly associated with decreased ABR rates in most drug-bug pairs. This result highlights the crucial role of hygiene and infection control measures for containing resistance, especially in hospital settings. They are consistent with the findings from Collignon et al. which showed that aggregated resistance in *E. coli, K. pneumoniae* and *S. aureus* were significantly inversely corelated with infrastructures and health expenditure.4 In our analysis, we showed that the GHS index was not associated in VR-*E* nor in PR-*Sp*, the only two investigated Gram-positive bacteria. For VR-*E*, this could potentially reflect discrepancies in nosocomial infection control approaches regarding this specific pathogen across countries of similar socio-economic background.28 For PR-*Sp*, this could be explained by the fact that *S. pneumoniae* is mostly acquired in the community. Overall, these findings highlight the importance of testing the validity of global ABR associations in different clinically relevant bacterial resistances.

Carbapenem-resistant *A. baumannii* was the only pair positively associated with the GDP per capita. From the univariate analysis, we found a large reduction in temporal RE variance indicating that the individual contribution of GDP was mostly temporal. This result is consistent with other findings that showed a fastest increase in *A. baumannii* resistances in OECD countries compared to other countries, where OECD countries have a higher GDP per capita. 23

After introduction of significant explanatory variables in the final models, spatial RE variance exhibited strong heterogeneity across drug-bug pairs. For Enterobacterales pairs, such as FR-*Ec*, RE variance was low suggesting that the proposed factors (quinolones sales, average temperature, extreme climatic events, GHS index) explained well ABR heterogeneities between countries. On the contrary, for VR-*E* or carbapenem resistances, RE variances remained high suggesting that the proposed factors could not explain most of ABR spatial heterogeneities. This result stresses the lack of knowledge about global factors favoring the dissemination of newly emerged resistances. In addition, specific countries such as Mexico, Japan, or South Korea, exhibited high RE estimates compared to other countries in ATLAS; and this was true across different drug-bug pairs. In South Korea, it is documented that primary care system is less established than in other OECD countries.29 Patients thus mostly seek care in secondary and tertiary care system (e.g. hospitals), where the per capita antibiotic exposure is higher. Additional factors, like healthcare seeking behaviors, but also antibiotic use in livestock or percentage of rural population, might then better explain these ABR rates for these outliers.

Some limitations should be mentioned in this study. First, the number of isolates tested in ATLAS was uneven across world’s regions and was especially low in Sub-Saharan Africa and might bias the associations we found. Second, data reported by participating hospitals might be partial representatives of true prevalence, leading to under- or overestimations especially in low- and middle-income countries (LMICs). It is currently challenging to obtain relevant representative data for LMICs and available data should be used in the absence of alternatives.30 Third, some choices regarding co-variables data can be discussed. For the meteorological factors, the capital city was used as a proxy for the entire country’s weather, which could be misleading for large countries. However, previous studies suggested that using either the capital city’s data or the country’s weighted temperature mean did not affect associations.8 Here, we used the IQVIA MIDAS database which currently represents one of the only source of harmonized data on global antibiotic sales21. However, this might be an underestimation of true antibiotic consumption as it did not include over-the-counter drugs. Data also represented both retail and hospital sales which represent different consumption behaviors. Finally, we found an inverse association with tourism-associated factors, while evidence of travel-related colonization by resistant bacteria and subsequent dissemination is abundant.6 ABR data in ATLAS did not represent resistance in carriage, but rather resistance in infections which could be less influenced by tourism.

Our findings showed that antibiotic resistance is a plural threat. Associations between antibiotic resistance rates and global factors, found in Enterobacterales for example, do not necessarily hold true in other bacterial species. Moreover, these known global factors cannot explain all heterogeneities observed worldwide. Our work highlighted the importance to study global antibiotic resistance in all drug-bug pairs of clinical relevance. From a global health perspective, strategies to tackle worldwide antibiotic resistance should thus be tailored accounting for the species, the resistance, and the epidemiological settings at stake.

## Supporting information

Appendix [[supplements/270726_file02.pdf]](pending:yes)

## Data Availability

The antibiotic resistance data can be visualized through the ATLAS website ([https://atlas-surveillance.com](https://atlas-surveillance.com)). Co-variables data are publicly available through the World Bank Data website ([https://data.worldbank.org](https://data.worldbank.org)); the Global Health Security index website ([https://www.ghsindex.org](https://www.ghsindex.org)); the Solar Radiation and Meteorological Data services website ([http://www.soda-pro.com](http://www.soda-pro.com)) for the MERRA-2 data; the ECDC website ([https://www.ecdc.europa.eu/en/antimicrobial-consumption/surveillance-and-disease-data/database](https://www.ecdc.europa.eu/en/antimicrobial-consumption/surveillance-and-disease-data/database)) for the ESAC-Net data. IQVIA MIDAS data are not publicly available. R codes to describe country-year longitudinal resistance data for an antibiotic-bacterium pair of interest and quantify its association with candidate factors are available here: [https://github.com/EveRah/drug\_bug\_pair\_NBMM](https://github.com/EveRah/drug_bug_pair_NBMM). 

## Contributors

ER, PG and LO participated in the initial statistical design and results interpretation. ER analyzed the data and wrote the manuscript. PG and LO were major contributors in the writing of the manuscript. LW contributed to the statistical design and to the manuscript review. DG participated in the manuscript review.

## Declaration of interests

ER’s work was funded by an independent research Pfizer Global Medical Grant (number 57504809). PG received consulting fees from Pfizer for an unrelated project. LW received consulting fees from Pfizer, HEVA and IQVIA for unrelated projects.

## Data sharing

The antibiotic resistance data can be visualized through the ATLAS website ([https://atlas-surveillance.com](https://atlas-surveillance.com)). Co-variables data are publicly available through the World Bank Data website ([https://data.worldbank.org](https://data.worldbank.org)); the Global Health Security index website ([https://www.ghsindex.org](https://www.ghsindex.org)); the Solar Radiation and Meteorological Data services website ([http://www.soda-pro.com](http://www.soda-pro.com)) for the MERRA-2 data; the ECDC website ([https://www.ecdc.europa.eu/en/antimicrobial-consumption/surveillance-and-disease-data/database](https://www.ecdc.europa.eu/en/antimicrobial-consumption/surveillance-and-disease-data/database)) for the ESAC-Net data. IQVIA MIDAS data are not publicly available.

R codes to describe country-year longitudinal resistance data for an antibiotic-bacterium pair of interest and quantify its association with candidate factors are available here: [https://github.com/EveRah/drug\_bug\_pair\_NBMM](https://github.com/EveRah/drug_bug_pair_NBMM).

## Acknowledgments

We thank the data providers for sharing helpful material with us: IQVIA for the antibiotic sales data and Pfizer for the isolates data from the ATLAS database.

*   Received February 9, 2022.
*   Revision received February 9, 2022.
*   Accepted February 9, 2022.


*   © 2022, Posted by Cold Spring Harbor Laboratory

This pre-print is available under a Creative Commons License (Attribution-NonCommercial-NoDerivs 4.0 International), CC BY-NC-ND 4.0, as described at [http://creativecommons.org/licenses/by-nc-nd/4.0/](http://creativecommons.org/licenses/by-nc-nd/4.0/)

## Bibliography

1.  W H O. Global action plan on antimicrobial resistance. 2015. [http://www.emro.who.int/health-topics/drug-resistance/global-action-plan.html](http://www.emro.who.int/health-topics/drug-resistance/global-action-plan.html) (Accessed Sept 20, 2021).
    
    

2.  Antimicrobial Resistance Collaborators. Global burden of bacterial antimicrobial resistance in 2019: a systematic analysis. Lancet 2022; published online Jan 18. DOI:10.1016/S0140-6736(21)02724-0.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S0140-6736(21)02724-0&link_type=DOI) 

3.  OECD. Stemming the superbug tide: just A few dollars more. OECD, 2018 DOI:10.1787/9789264307599-en.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1787/9789264307599-en&link_type=DOI) 

4.  Collignon P, Beggs JJ, Walsh TR, Gandra S, Laxminarayan R. Anthropological and socioeconomic factors contributing to global antimicrobial resistance: a univariate and multivariable analysis. Lancet Planet Health 2018; 2: e398–405.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S2542-5196(18)30186-4&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=30177008&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 

5.  Blommaert A, Marais C, Hens N, et al. Determinants of between-country differences in ambulatory antibiotic use and antibiotic resistance in Europe: a longitudinal observational study. J Antimicrob Chemother 2014; 69: 535–47.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/jac/dkt377&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=24080501&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 

6.  Frost I, Van Boeckel TP, Pires J, Craig J, Laxminarayan R. Global geographic trends in antimicrobial resistance: the role of international travel. J Travel Med 2019; 26. DOI:10.1093/jtm/taz036.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/jtm/taz036&link_type=DOI) 

7.  MacFadden DR, McGough SF, Fisman D, Santillana M, Brownstein JS. Antibiotic Resistance Increases with Local Temperature. Nat Clim Chang 2018; 8: 510–4.
    
    

8.  McGough SF, MacFadden DR, Hattab MW, Mølbak K, Santillana M. Rates of increase of antibiotic resistance and ambient temperature in Europe: a cross-national analysis of 28 countries between 2000 and 2016. Euro Surveill 2020; 25. DOI:10.2807/1560-7917.ES.2020.25.45.1900414.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.2807/1560-7917.ES.2020.25.45.1900414&link_type=DOI) 

9.  Kaba HEJ, Kuhlmann E, Scheithauer S. Thinking outside the box: Association of antimicrobial resistance with climate warming in Europe - A 30 country observational study. Int J Hyg Environ Health 2020; 223: 151–8.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.ijheh.2019.09.008&link_type=DOI) 

10. García-Rey C, Fenoll A, Aguilar L, Casal J. Effect of social and climatological factors on antimicrobial use and Streptococcus pneumoniae resistance in different provinces in Spain. J Antimicrob Chemother 2004; 54: 465–71.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1093/jac/dkh375&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=15269198&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000223372100028&link_type=ISI) 

11. Pfizer. Antimicrobial surveillance, Pfizer : the ATLAS program. [https://www.pfizer.com/science/therapeutic-areas/anti-infectives/antimicrobial-surveillance](https://www.pfizer.com/science/therapeutic-areas/anti-infectives/antimicrobial-surveillance) (Accessed Nov 21, 2020).
    
    

12. Tacconelli E, Carrara E, Savoldi A, et al. Discovery, research, and development of new antibiotics: the WHO priority list of antibiotic-resistant bacteria and tuberculosis. Lancet Infect Dis 2018; 18: 318–27.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/S1473-3099(17)30753-191&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 

13. EUCAST TEC on AST. Breakpoint tables for interpretation of MICs, version 9.0, 2019. Available at: [https://www.eucast.org/ast\_of\_bacteria/previous\_versions\_of\_documents/](https://www.eucast.org/ast\_of\_bacteria/previous_versions_of_documents/) (Accessed Sept 15, 2021).
    
    

14. IQVIA. IQVIA website. [https://www.iqvia.com](https://www.iqvia.com) (Accessed Nov 21, 2020).
    
    

15. Gelaro R, McCarty W, Suárez MJ, et al. The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). J Clim 2017; 30: 5419–54.
    
    

16. World Bank Data. [https://data.worldbank.org](https://data.worldbank.org) (Accessed Nov 22, 2020).
    
    

17. GHS website. [https://www.ghsindex.org](https://www.ghsindex.org) (Accessed Nov 22, 2020).
    
    

18. Bolker BM, Brooks ME, Clark CJ, et al. Generalized linear mixed models: a practical guide for ecology and evolution. Trends Ecol Evol (Amst) 2009; 24: 127–35.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.tree.2008.10.008&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19185386&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 

19. Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. J Stat Softw 2015; 67: 1–48.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.18637/jss.v067.i0&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=24227677&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 

20. R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria., 2020.
    
    

21. Klein EY, Van Boeckel TP, Martinez EM, et al. Global increase and geographic convergence in antibiotic consumption between 2000 and 2015. Proc Natl Acad Sci USA 2018; 115: E3463–70.
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoicG5hcyI7czo1OiJyZXNpZCI7czoxMjoiMTE1LzE1L0UzNDYzIjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjIvMDIvMDkvMjAyMi4wMi4wOS4yMjI3MDcyNi5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

22. Johnson AP, Woodford N. Global spread of antibiotic resistance: the example of New Delhi metallo-β-lactamase (NDM)-mediated carbapenem resistance. J Med Microbiol 2013; 62: 499–513.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1099/jmm.0.052555-0&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=23329317&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000318203100001&link_type=ISI) 

23. Xie R, Zhang XD, Zhao Q, Peng B, Zheng J. Analysis of global prevalence of antibiotic resistance in Acinetobacter baumannii infections disclosed a faster increase in OECD countries. Emerg Microbes Infect 2018; 7: 31.
    
    

24. Olesen SW, Lipsitch M, Grad YH. The role of “spillover” in antibiotic resistance. Proc Natl Acad Sci USA 2020; 117: 29063–8.
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoicG5hcyI7czo1OiJyZXNpZCI7czoxMjoiMTE3LzQ2LzI5MDYzIjtzOjQ6ImF0b20iO3M6NTA6Ii9tZWRyeGl2L2Vhcmx5LzIwMjIvMDIvMDkvMjAyMi4wMi4wOS4yMjI3MDcyNi5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

25. Ivers LC, Ryan ET. Infectious diseases of severe weather-related and flood-related natural disasters. Curr Opin Infect Dis 2006; 19: 408–14.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1097/01.qco.0000244044.85393.9e&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=16940862&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 

26. Uçkay I, Sax H, Harbarth S, Bernard L, Pittet D. Multi-resistant infections in repatriated patients after natural disasters: lessons learned from the 2004 tsunami for hospital infection control. J Hosp Infect 2008; 68: 1–8.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jhin.2007.10.018&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=18079019&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2022%2F02%2F09%2F2022.02.09.22270726.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000252803800001&link_type=ISI) 

27. Yu P, Zaleski A, Li Q, et al. Elevated Levels of Pathogenic Indicator Bacteria and Antibiotic Resistance Genes after Hurricane Harvey’s Flooding in Houston. Environ Sci Technol Lett 2018; 5: 481–6.
    
    

28. Isenman H, Michaels J, Fisher D. Global variances in infection control practices for vancomycin resistant Enterococcus - results of an electronic survey. Antimicrob Resist Infect Control 2016; 5: 41.
    
    

29. OECD. The public health system in Korea. 2020. [https://www.oecd-ilibrary.org/sites/6e005d47-en/index.html?itemId=/content/component/6e005d47-en](https://www.oecd-ilibrary.org/sites/6e005d47-en/index.html?itemId=/content/component/6e005d47-en) (Accessed Dec 13, 2021).
    
    

30. Ashley EA, Recht J, Chua A, et al. An inventory of supranational antimicrobial resistance surveillance networks involving low- and middle-income countries since 2000. J Antimicrob Chemother 2018; 73: 1737–49.

 [1]: /embed/graphic-1.gif
 [2]: T1/embed/inline-graphic-1.gif
 [3]: T1/embed/inline-graphic-2.gif