The potential health impact from AMR has prompted researchers to determine the main drivers for the development and spread of AMR. Several studies have assessed the effect of socioeconomic factors, including income level, corruption and infrastructure, on promoting AMR [8,9,10]. Our study adds to this body of research, while also drawing on water and sanitation quality data from the WHO/UNICEF JMP to investigate how these factors contribute to AMR, both irrespective of country income as well as after stratification according to income level. Our study also supports the links between AMR and water and sanitation issues, as underlined in the recent WHO/FAO/OIE technical brief on water, sanitation, hygiene and waste management to prevent infections and reduce the spread of AMR [28].
An increase in unsafely managed drinking water was significantly associated with an increase in blood culture QREC (%) when analysed with univariate regression in the combined model and the middle-income model, although it lost its significance in the multivariable models. We suggest that there was too little variance in the data to draw the conclusion that unsafely managed drinking water contributed independently to blood culture QREC (%) levels.
In the univariate analysis, we found that as the proportion of unsafely managed sanitation increased, so too did the levels of blood culture QREC (%). Poorer quality sanitation results in an environmental breeding ground for antibiotic resistance and thus explains the positive relationship with blood culture QREC (%) [29]. Increased unsafely managed sanitation was significantly associated with an increase in blood culture QREC (%) in the univariate model including only middle-income countries, but not the model including only high-income countries. One can infer that there is a fairly similar and high level of sanitation quality in high-income countries, whereas in middle-income countries the variance is greater, and it shows its significance. This finding agrees with the recently published study by Collignon et. al. which found a positive correlation between infrastructure (i.e. access to sanitation), and AMR levels [10]. In the multivariable models, unsafely managed sanitation loses its significant association with blood culture QREC (%), possibly because there are too few data points among lower-income countries where one expects that sanitation may play a larger role in AMR.
Human fluoroquinolone consumption had a positive association with blood culture QREC (%) level in the multivariable model including all countries, and when analysed in high-income countries only, but not for middle-income countries. A vast body of research shows that high consumption of antimicrobials is a main driver of AMR. Most of these studies, however, are performed in high-income settings, and have not included possible socioeconomic and environmental factors. Collignon et. al. on the other hand did include these additional factors and found that antibiotic use was not significant in any of the multivariable models [10]. Their findings, and ours, suggest that, especially in middle-income countries, and possibly also in low-income countries, other variables are also at play.
An increase in GNI per capita was significantly associated with an increase in blood culture QREC (%) in the univariate analyses in both the combined and high-income countries only analyses, however, it lost its significance in the multivariable models. This positive association is consistent with the findings of Collignon et. al. who explained it by the fact that wealthier countries use more antibiotics [10]. The result differs, however, from the study performed by Alvarez-Uria et. al. where they found that income was inversely related to AMR, although no other variables were considered in this study [8].
The corruption perceptions index (the higher the score in the index, the lower the corruption) remained strongly negatively associated with blood culture QREC (%) across all models, being the only significant variable (p value < 0·05) in all three univariate analyses. It also remained significant in the multivariable models that analysed all countries and in the high-income countries only, although it lost its significance in the middle-income multivariable model, likely due to too few observations. It is probable that countries with more corruption will have fewer antibiotic stewardship programmes and less stringent policies on the disposal of pharmaceuticals, as well as worse infrastructure and water and sanitation services.
Previous research indicates that the transfer of resistant genes occurs at an optimum temperature of 30 °C meaning that countries with higher average temperatures are more likely to produce optimal conditions for bacterial transfer [7]. In our study, this was true in the univariate analyses, as the annual average temperature of a country increased, so too did the level of blood culture QREC (%). However, it lost its significance in the multivariable models. It is also possible that average temperature is a proxy for other socioeconomic factors. Although not the direct causal mechanism, higher temperatures have previously been associated with lower levels of income and reduced quality of water and sanitation [30, 31]. One of the ongoing concerns of climate change is the risk that increased temperatures will have on vector-borne diseases. As temperatures rise, so too will the incidence of vector-borne diseases, thus resulting in increased use of antibiotics, and thus, potentially, an increase in AMR [32].
Unlike other studies, we included livestock and crop production index, which reflect the growth in agricultural production relative to a baseline period. We found them to be positively associated with blood culture QREC (%) in univariate and multivariable analyses of all countries combined but showed no significance when analysed in the high and middle-income models only. Often policy makers focus only on the misuse of antimicrobials amongst the human population, but antimicrobials are also used extensively in the agricultural, farming and aquaculture industry. As demonstrated above, as the quantity of livestock and agricultural production increases in a country, there is an associated increase in blood culture QREC (%). This could be as a result of increased use of antimicrobials in these industries and subsequent run-off into the environment.
Strengths and limitations
The major downfall of our study was the quantity of data available and the non-randomness of some of the missing data. The data we used was also cross-sectional which prevented us from capturing relationships which take place with a time lag. Data collection is a costly process and one that requires a stable economy and political state. Consequently, developed countries are over-represented, and indeed, there was only one low-income country with data on blood culture QREC (%). While we can draw relatively certain conclusions on high-income countries, it is possible that there are different drivers in low-income countries for which we have little data, and even within the middle-income group for which we probably had insufficient data. The collection of national data on AMR via the WHO GLASS project is a relatively new endeavour and one that requires large input from governments and health departments. Several countries have enrolled in GLASS, but it will take many years before it becomes routine data collection internationally.
Our study only looked at blood culture QREC (%) as a marker of AMR. It has been shown that pathogens have different responses to antimicrobial use and it is possible that by focusing on a single drug-pathogen combination that the results may have limited generalisability [33]. It has also been demonstrated that intense and repeated use of antibiotics has a stronger association with AMR that extensive low-intensity use [34]. It was out of the scope of this study to include the distribution and intensity of antimicrobial use, but we are aware that this may be another potential source of bias. Bacteria also display either disjoint or concurrent resistance and if we had looked at other drug-pathogen combinations, we may have found that the effect of a variable on one drug-pathogen combination might have differed to the effect on another combination and, indeed, the overall levels of AMR.
In addition, while we included diverse variables, it is possible that some statistical errors of endogeneity exist. There is potential for reverse causality where factors associated with AMR also influence prescribing practices or where the level of AMR influences the independent variables. It is also possible that some factors that are drivers for AMR were omitted or unobserved or that some of the included variables exhibit simultaneity. For example, it is possible that countries with higher income have both less corruption, greater access to antimicrobials and improved sanitation and water infrastructure. Indeed, it is unlikely that corruption itself causes higher levels of AMR, this association likely reflects a range of socioeconomic factors which are correlated with perceived corruption. We tried to account for potential multi-collinearity by calculating the Variance Inflation Factor (< 6) which demonstrated the likelihood of the analysed variables being correlated with each other. When performing multivariate analysis, there is always a balance between potentially omitting relevant drivers or including too many, resulting in simultaneity or reversed causality, and we acknowledge this as a limitation.
This aside, our study analysed a large number of socioeconomic and environmental factors associated with AMR. Our study is the first we know to perform multivariable analyses where we stratified drivers of AMR by different levels of income. Although there is uncertainty about details in our results, they did indicate there might be different main drivers of AMR in different economic settings, pointing to the need for differentiated policies to fight AMR. There has been a call for environmental regulation and monitoring to be included in AMR action plans, although a lack of understanding of the drivers and pathways of AMR and the differences between these drivers in different settings has resulted in an inability to include many of them in policy until now [35]. As other research has demonstrated, AMR action plans also need to bridge the worlds of antibiotic use in health care, agriculture and aquaculture [36]. Our research emphasises that there cannot be a single action plan that meets the needs of all countries globally and that AMR action plans need to take into account different socioeconomic, geospatial and environmental factors while acknowledging gaps in the surveillance capacity of lower economic settings.
Future research
We have identified gaps in the research, namely, in the amount of data available on AMR in low-income countries. We recommend steps to improve AMR surveillance in low-income countries. If such data becomes available, we suggest running further multivariable analyses of AMR drivers in low-income countries in order to guide political prioritization of action against AMR.