In general, our analysis indicated that total and health expenditures, different aspects of quality governance, scores of political rights and civil liberties were all inversely related to the proportions of AMR across European countries. These findings add to the currently limited knowledge about the complex contribution of anthropological and socioeconomic factors to AMR. In particular, we observed that the governance was the most contributing factor to antibiotic consumption and AMR among all domains under investigation.

Several indicators of governance (i.e., voice and accountability, government effectiveness, regulatory quality, rule of law, and control of corruption) and freedom (i.e., political rights and civil liberties scores), in fact, showed similar inverse correlation with AMR levels in the bivariate analysis. Higher levels of governance indicators were also associated with lower antibiotic consumption. These findings were in agreement with those reported by previous studies at national and international levels [9, 13]. For example, it has been proposed how corruption is one of the main contributing factors to AMR [9, 21, 22].

Total GDP and health expenditure per capita were also inversely correlated with AMR proportions of different combinations between pathogens and antibiotics. However, the effect of country’s expenditures on AMR levels deserves further examination, because economic indicators may reflect other socioeconomic features. Our results from bivariate analysis, in fact, were partially in contrast with the global analysis carried out by Collignon and colleagues, which showed how high GDP per capita was associated with higher AMR levels [13]. Although this association was not significant in the multivariable analysis, the authors referred to the well-known association between high GDP per capita and high antibiotic consumption in support of their claim [23]. Our analysis of European countries, however, did not find any correlation between GPD per capita and antibiotic consumption. For this reason, our hypothesis was that more directly relevant variables, such as health expenditure, may explain the beneficial effects of GDP on AMR levels observed in our study. The response to AMR, in fact, would benefit from higher health expenditure. In this context, it is also worth mentioning that earlier studies of European data showed that levels of AMR rose with increasing private health spending [9, 13]. However, this indicator could conceivably reflect the business volume of the private health sector, provided by individual health professionals and healthcare companies. According to previous data [24], there may be differences between the public and private health sectors in terms of regulation and antibiotic consumption; differences that would help explain the observed relationship between private health spending and AMR levels.

Overall, our analysis confirms AMR as a multifaced issue that requires concerted and informed approaches. To disentangle the multifactorial contribution to antibiotic use and AMR, we developed five indexes and an aggregate measure of AMR. Multivariable analysis showed that higher indexes of governance and health were significantly associated with lower aggregate levels of AMR. This was true even when considering the detrimental effect of antibiotic consumption on AMR. Among all indexes, the one related to governance was also associated with lower antibiotic consumption in the community. Given the dual action of the governance index, we evaluated the mediating effect of antibiotic consumption on AMR levels. Mediation analysis is increasingly being applied in many fields of research, including epidemiological research [18], and in our case can be used to investigate the antibiotic consumption as a mediator of the relationship between governance and AMR. Surprisingly, the mediation analysis estimated that only a third of the reduction in AMR can be attributed to the decrease in antibiotic consumption associated with increased governance index. This means that there may be other factors other than antibiotic use volumes that explain how governance quality affects AMR [7]. It is undeniable that antibiotic use is one of the main contributing factors to AMR, as also proven by our analysis. Nevertheless, uncontrolled spread of resistant pathogens, defined by Collignon and colleagues as "contagion" [7], might also be significant. In fact, one could expect higher levels of AMR in countries with conditions that favor the spread of resistant pathogens in the healthcare setting and in the community. A number of factors could contribute to this problem, including low hygiene awareness, weak governance systems, and poor sanitation management in all sectors (e.g., health facilities, houses, water treatment plants, and the food supply chain). This hypothesis is also supported by the observation of high AMR levels in countries with low antibiotic consumption, but weak governments, low incomes, and poor infrastructures [9]. In addition, the inefficiency of governance may contribute to a lack of monitoring of the appropriateness of antibiotic prescriptions [21, 22].

With so many factors contributing to AMR, conventional methods of data analysis are often inadequate to address the issue. Hence, we grouped countries by socioeconomics, governance, and health characteristics using a multivariate approach. When we combined countries into different clusters, antibiotic consumption and AMR levels were the highest in countries with the lowest health, governance, and freedom indexes (i.e., Bulgaria, Hungary, Poland, and Romania). This evidence increased our confidence in the robustness of the findings.

Our analysis had several limitations to be discussed. First, it was limited to European countries and data for one year. Although a broader approach could provide meaningful evidence at global level, there was the need for using data that were almost complete and consistent between countries. Just considering one year, in our analysis there were some countries with incomplete data on AMR, especially for some combinations between pathogen and antibiotic (e.g., Acinetobacter spp. resistance to fluoroquinolones, aminoglycosides, carbapenems). Missing data were imputed with the average values obtained from countries with available data. Although this was considered the best way to solve the issue, we were unable to identify and manage factors that determined the presence of missing values. The hypothesis that countries reporting data have the best surveillance systems suggests that imputing missing values might have introduced a confounding effect. However, the reasons behind the presence of missing values can be various and not always related to the effectiveness of surveillance systems or to the development of the country. For instance, countries with missing data for Acinetobacter spp. were Estonia, Iceland, Luxembourg, and Malta, which are heterogeneous in terms of socio-demographic and organizational features. In the near future, it would be interesting to replicate our analyses taking into account other factors determining the effectiveness of surveillance systems and/or comparing countries from other regions of the world. Additionally, even when complete, AMR data might be underestimated or overestimated without representative surveillance systems. The same applies, presumably, to data on antibiotic consumption, which might be underestimated due to the volume of antibiotics purchased through other sales channels. For the same reasons, Collignon and colleagues' analysis [13] covered just over seventy countries on a global scale and aggregated data from different years and sources to obtain a representative and almost complete dataset.

Second, we summarized the overall set of indicators using aggregate indexes that captured domains that may contribute to antibiotic consumption and AMR. The choice of creating ad-hoc aggregate indexes was motivated by the presence of many possible contributing indicators, which were also interrelated to each other. It was not our intention, in fact, to provide evidence on specific indicators, but rather to evaluate what are the main domains associated with antibiotic consumption and AMR. Moreover, the methodology behind the creation of aggregate indexes from available indicators was borrowed from Collignon and colleagues [13]. There is no doubt that other composite indicators could be used to summarize sociodemographic and economic characteristics of European countries (e.g., the Socio-demographic Index), however, we decided to create ad-hoc indexes reflecting the specific demographic, health, economic, governance, and freedom domains. Third, additional factors that might contribute to antibiotic use and AMR were not considered. For instance, we could not use data on infrastructures, education, healthcare access, sanitation, and public awareness on antibiotic consumption and AMR. Further, we did not take into account data on antibiotic consumption in hospitals and in the animal sector, which account for the largest part of total antibiotic consumption, nor we did examine antibiotic stewardship and infection prevention protocols. For instance, a country might report low antibiotic consumption in the human sector – whether in hospital or in the community – but high consumption in the animal sector. All these factors not considered in our analyses, or at least some of them, might explain the remaining part of the effect of governance on AMR. For this reason, future studies should be encouraged to approach the problem from a One Health perspective.