Thank Martin Luther that ciprofloxacin could cure your gonorrhoea? Ecological association between Protestantism and antimicrobial consumption in 30 European countries [version 2; peer review: 1 approved, 2 approved with reservations]

Background: Higher consumption of antimicrobials plays an important role in driving the higher prevalence of antimicrobial resistance in Southern compared to Northern Europe. Poor controls on corruption (CoC), high uncertainty avoidance (UA) and performance vs. cooperation orientation (POCO) of societies have been found to explain much of this higher consumption in Southern European countries. We hypothesized that these predictors were in turn influenced by the Protestant Reformation in the 16 th century onwards. Methods: We used structural equation modelling (SEM) to assess the relationships between country-level proportions being Protestant, CoC, UA, POCO and four markers of antimicrobial consumption in the community (all antibacterials, cephalosporin, macrolides and fluoroquinolones). Results: The proportion of a country that was Protestant was negatively correlated with the consumption of all antibacterials. SEM revealed that UA predicted all antibacterial consumption (direct effect coef. 0.15, 95% Confidence Interval [CI] 0.04-0.26). The proportion Protestant exerted an indirect effect on consumption (coef. -0.13, 95% CI -0.21- -0.05). This effect was mediated predominantly via its in the Reformation. These findings may explain the differential efficacy of similar antibiotic stewardship campaigns in Northern and Southern European populations.


Introduction
Countries in Southern Europe have been noted for some time to have a higher prevalence of antimicrobial resistance (AMR) than Northern European countries 1,2 . As an example, the prevalence of Neisseria gonorrhoeae resistance to ciprofloxacin varies over three fold between countries in Europe 3 .
The major determinant of these variations in AMR is the higher consumption of antimicrobials (AMC) in Southern European countries 3,4 . Fluoroquinolone consumption for example varies 6-fold between European countries and is associated with the prevalence of ciprofloxacin resistance in N. gonorrhoeae 3 . What is less clear is what the underlying reason is for the variations in AMC 4,5 .
Previous studies have found a range of cultural and structural factors underpin the large variations in the consumption of fluoroquinolone and other antibiotics between European countries and globally 5-10 . The act of prescribing an antimicrobial is highly social. Providing an antimicrobial represents the doctor's concern for the patient, legitimizes the patient's sick-role and reinforces the doctor's claim to expert knowledge 8,10-13 . These vary between countries as do patients perceptions of the need for antimicrobials to treat infections 8,11,13,14 . One study compared a Dutch and Belgian city, 60km apart, both Dutch speaking but the two cities being historically Protestant and Catholic, respectively 14 . Where the Dutch labelled their upper respiratory tract infections (URTI) as 'colds' or 'flu,' the Belgians labelled most episodes as 'bronchitis' and used more antimicrobials. In general, the researchers found that those from a Protestant background were more sceptical about using antimicrobials than those from Catholic backgrounds. This type of observation has led to some authors to speculate that there may be a negative association between the proportion of a country that is Protestant and antimicrobial consumption 2,12,15 . Others have contested this claim 11 , but no one has, to the best of our knowledge, formally tested this association. In this paper, we test the hypothesis that the proportion of a country's population that are reported to be Protestant is negatively associated with AMC in Europe. Furthermore, we assess if the pathway through which this operates is via cultural and structural factors associated with Protestantism and AMC.
Two cultural dimensions have consistently been found to explain variations in AMC both within Europe and worldwide 5-8,13,14,16 . These are Hofstede's uncertainty avoidance index (UA) and performance-orientation versus cooperation-orientation index (POCO) 5-8,13,14,16 . Both indices are negatively associated with both the proportion of a country that is Protestant 17 and AMC 5-8,16 . The UA index is a measure of the extent to which a society feels threatened by ambiguous or unknown situations. In high UA cultures, individuals feel discomfort and stress in unstructured situations that are novel 15 . The POCO index (also termed the masculinity index) provides a measure of how performanceoriented cultures are. In high POCO cultures ego needs, assertiveness, targets and success are emphasized, whereas cooperation-oriented cultures place more focus on caring for all members of society, including the weak 14 . An important reason not to take an antibiotic for an illness such as an URTI is that this will select for AMR -an adverse effect for the population at large. It has been argued that this low-POCO populations are more receptive to this population-benefit message than high POCO-populations 8,18 .
The key structural factor found to be associated with AMC is control of corruption (CoC) at country and regional levels 19 In this ecological study we use structural equation modelling to assess the association between countries' population proportions being protestant and AMC, modelling UA, POCO and CoC as potential mediating variables.

Data
Antibiotic consumption. Data from the European Surveillance of Antimicrobial Consumption (ESAC) were used as a measure of national general population-level antimicrobial drug consumption 23,24 . ESAC reports antimicrobial consumption as the number of defined daily doses per 1000 inhabitants (DID) following the World Health Organization guidelines 25 . In our study, we used four measures of country-specific antimicrobial drug use in ambulatory care: Total antibacterials for systemic use (ATC group J01), Cephalosporins/other Beta lactams (ATC group J01D), fluoroquinolones (ATC group J01MA), macrolides, lincosamides and streptogramins (ATC group J01F). Data was available from 1998 to 2018 and we used this data to calculate the peak consumption of each of these four classes of antimicrobial over this time period. All countries with available data were used in all the analyses. This data is available from ESAC without restrictions: https://www.ecdc.europa.eu/en/ antimicrobial-consumption/database/quality-indicators Percent protestant. The proportion of a national population that was protestant was sourced from the Pew Research Centre estimates for 2010: https://www.pewforum.org/2011/12/19/ table-christian-population-as-percentages-of-total-populationby-country/

Amendments from Version 1
The new version has been edited along the lines suggested by the reviewers. The main change we have made is to expand the limitations section.
Any further responses from the reviewers can be found at the end of the article UA and POCO. Individual scores for UA and POCO were obtained for each country from Hofstede Insights, freely available from (https://www.hofstede-insights.com/product/compare-countries/) is denoted as masculinity on the website.

Control of corruption.
The World Bank has provided indicators pertaining to six dimensions of governance since 1996. We used the dimension (control of corruption) that has been found to be most closely linked to AMC 12,16,17 : Control of corruption (CoC) is defined as the country-level extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as the capture of the state by elites. The index provides each country's score in units of standard normal distribution, ranging from approximately -2.5 (low CoC) to 2.5 (high CoC). The values used are average scores for the years 2013 to 2015, which we calculated from the original data, which was obtained from the following site: http://datatopics.worldbank. org/world-development-indicators/.

Data analysis
A correlation matrix was performed to investigate the relationship between the different variables hypothesized to be associated with AMC. This approach was complemented by scatterplots of the associations between percent Protestant and AMC. We used structural equation modelling (SEM) to analyse the factors predicting AMC. SEM provided a way to analyse and graphically represent the complex direct and indirect pathways between endogenous and exogenous variables. All variables were assessed for non-linearity. No transformation was necessary. The analyses were performed using the SEM-builder in STATA 16. A P-value of less than 0.05 was used as the threshold of statistical significance.

Results
Complete data was available for 29 of 30 countries with AMC data in the ESAC database (Table 1). Data for UA and POCO were missing for Cyprus. Peak total antibacterial consumption varied fourfold between 10.1 DID in the Netherlands and 40.4 DID in Greece (median 21.2 [IQR 16.7-23.9]; Table 2).

Analysis by antibiotic class
The SEM analysis found that the consumption of each class of antimicrobial was positively associated with UA and POCO ( Figure 2; Extended Data). Only in the case of POCO predicting fluoroquinolone consumption was this association not statistically significant ( Figure 2; Extended Data). Once again, the percent Protestant only exerted an indirect effect on AMC. This effect was mediated by UA and POCO both of which were negatively associated with percent Protestant. For each class of antimicrobial, the effect of UA explained the greatest proportion of variation in consumption (Extended Data). Overall the models explained 52% to 61% of the variation in consumption (Extended Data).

Discussion
Our results recapitulate those from other studies that UA and, to a lesser extent, POCO mediate a considerable proportion of the variation of AMC in Europe 6,8,10,16 . Whilst percent Protestant has little to no direct effect on AMC, our analysis found it has an indirect effect via its negative association with UA and POCO. The fact that this effect was similar for all 4 categories of antimicrobials investigated makes this finding more robust.
These results are compatible with the theory that the profound rupture in European society in the 16 th century induced by the Reformation may have had enduring effects that explain a portion of the contemporary variations in antimicrobial consumption between European countries. Our study provides evidence supportive of the thesis that this effect is mediated via Protestantism's effect on two cultural variables -UA and POCO.
The reason that predominantly Catholic countries have higher UA and POCO scores is not clear, but may be related to factors such as the rituals and certainty-of-Faith that have characterized Catholicism 11,14,17,26 . It has been argued that Protestant teaching provided less certainty-of-Faith, encouraged more discussion, discouraged rituals, promoted austerity/simplicity and placed the locus of control less in the priest or church but in each individual 12,17,26 . Protestant populations may therefore be more tolerant of uncertainty, have less faith in quick-fix solutions and be more amenable to discussions about therapeutic strategies not involving antibiotics 12,17 . Protestants have also been found to have more trust in the 'self-healing power of the body', which has in turn been found to be correlated with scepticism towards the use of antibiotics 14 . Both patients and doctors in historically Protestant, low-UA populations may therefore be more receptive to antibiotic stewardship messages that strongly discourage antibiotics for infections such as URTIs 2 . High-UA societies on the other hand, may be less receptive to stewardship messages due to the uncertainty of 'what if the URTI is caused by a bacterial infection?' 4,8,10 . There are a number of important limitations to this analysis. We should be extremely guarded about drawing causal inferences concerning processes hundreds of years ago based on contemporary data from a small selection of countries. We did not control for possible confounders in the association between proportion of the population Protestant and AMC. We did not control for socio-economic markers such as GDP/capita as previous analyses have found that these did not explain differences in AMC within Europe 16,18 . We also did not control for differences in environmental temperatures. Southern European countries tend  to be both hotter and more Catholic than Northern European countries. A previous ecological study has found that environmental temperature is associated with the prevalence of certain types of antimicrobial resistance 27 . Our sample size was also too small to justify controlling for a large range of confounders.
There are a number of fundamental problems with classifying countries by religion. To an important extent, countries have a fluid mix of particular religions and both the relative sizes of the religions and the nature of these religions vary over time 26 . There are also considerable differences within a religion such as differences between Catholicism in different countries and regions 17 . This problem is compounded by our percent-Protestant-variable which combines a heterologous group of Catholic, Orthodox and other groups into one non-Protestant category. This classification could, however, be defended since our hypothesis is that low AMC was a byproduct (spandrel) of the Protestant Reformation. This line of thought is strengthened by a European study that found that the percent of the population describing themselves as atheist as opposed to religious was strongly associated with lower AMC 9 . The study did not include a religious denomination variable but the authors noted evidence that secularization has been more pronounced in historically Protestant countries 26 and concluded that the lower AMC in these countries may be indirectly related to Protestantism. We considered reverse causation unlikely, but this cannot be excluded. Finally, the various dimensions of Hofstede's model have been criticized as being over-simplifications of cultural differences 28 .
A spandrel is an architectural term referring to the tapering triangular space formed by the intersection of two rounded arches at right angles 29 . Gould argued that "evolutionary biology needs such an explicit term (spandrels) for features arising as by-products, rather than adaptations, whatever their subsequent exaptive utility…Causes of historical origin must always be separated from current utilities; their conflation has seriously hampered the evolutionary analysis of form in the history of life" 29 . Previous analyses have found evidence of a range of spandrels exerting their effects hundreds of years later 30,31 . One example comes from Southern Africa, where differential HIV prevalence by ethnic group has been linked to distant historical processes. A number of colonial policies that were imposed on indigenous ethnic groups practicing polygamous partnering resulted in dense sexual networks that facilitated the spread of HIV in these groups hundreds of years later 30,32 . In contrast, Southern African ethnic groups from European origin have low sexual network connectivity and HIV prevalence. This low connectivity stems primarily from historical processes in Europe many centuries prior that resulted in forms of monogamous partnering being normative 30 . Appreciating this historical connection has been shown to have three major benefits. Firstly, it provides an explanation as to how dramatic differences in behaviour and disease outcome can emerge. Secondly, it provides clues to the high HIVprevalence populations as to how to tackle the underlying determinants of high prevalence. Thirdly, it does this in a nonjudgemental way. Contemporary populations cannot be held responsible for events and processes occurring centuries prior 30 .
Similar arguments could be made as to the relevance of the current analysis. It provides a possible deep historical explanation for how differences in AMC have emerged in Europe. It suggests that the lower AMC in predominantly Protestant countries could be explained by cultural differences that emerged in a process starting centuries ago. If this is correct, then this insight should generate greater understanding for how much harder antibiotic stewardship work is in non-Protestant countries. This is not an argument that stewardship is impossible or should not be attempted, but rather that campaigns might need to be more intense to achieve the same outcomes. It also provides further evidence that stewardship efforts need to be adapted to the local cultural context 6,7 . A concrete example of this would be to incorporate rapid diagnostic tests (that can remove uncertainty about bacterial infections) as a part of stewardship campaigns in high UA populations 7 . If evidence were to come to light of ways to decrease uncertainty avoidance and favour cooperationvs. performance-orientation these may also be considered as upstream interventions to reduce AMC.

Data availability
Underlying data All data underlying the results are available as part of the article and no additional source data are required.

Open Peer Review
I was intrigued by the title of this paper, as I suspect many other readers have been (or will be). In the back of my mind, I was asking: "Is this a tongue-in-cheek update of Emile Durkheim's (the 'father' of sociology) famous treatise on suicide, focusing now on antimicrobial consumption as a means to illustrate the concept of ecological fallacy?". In Durkheim's work (1897 1 ; as interpreted by Morgenstern (1995) 2 ), an ecological analysis of suicide rates by region of then-Prussia in Europe showed a 7.6-fold increase in rates of suicide comparing the highest proportion Protestant versus the lowest proportion Protestant (blame (not thank) that on Martin Luther?). That was when analyzed in the aggregate, but showed only a 2-fold increase in suicide rates when analyzed at the individual-level by Durkheim himself. Still increased, but greatly upwardly biased estimates for individuals based on their religious affiliations. Of course, the present paper on AMC appears to lack the data granularity to confirm or refute the religion of those individuals either prescribing or else receiving antibiotics. And this isn't necessary, especially given the limitations and caveats that Kenyon and Fatti have provided. The authors have been careful to describe their analyses and conclusion as being entirely ecological, and we can accept this. Studies of the aggregate are indeed crucially important for exploring differences in the aggregate, especially when allocating resources or seeking solutions to complex social problems. However the social aggregations and normative consequences at the societal level, it is clear that in some way the aggregation is something less or more than the sum of parts (now moving into atomistic fallacy...). There are countries within the dataset where there are substantive regional north/south differences in the distribution of Protestant versus 'other' (Catholic, generally) and so both Germany and The Netherlands might have warranted some discussion to this effect. If Prussia still existed, it would be a great microcosm in which to explore the phenomena in the current paper! In other words, if the phenomenon is to be believed, then there should be somewhat obvious differences in prescribing rates between Hamburg versus Munich (as one example). Likely insufficient for statistical analysis, but would serve as a nice confirmation or refutation vignette. In addition, Cyprus as a country poses more problems than simply missing data on UA and POCO; that is, since it is divided into a Turkish (largely Muslim) north and Greek (largely Orthodox Christian) south and complex governance and conflict issues likely pose problems in interpretation beyond those listed in the paper. And, it is no doubt an obvious caveat that a tourist from a Scandinavian country holidaying at a Mediterranean beach resort should probably not wait until returning home before receiving an antibiotic for a newly acquired case of gonorrhea.
I recognize the authors have added substantive discussion on the limitations of inference and I concur with those (e.g., temperature gradients, lack of control of other confounders, etc). It would be useful to also include reference to publicly available rates of bacterial infectious diseases per EU surveillance (or, at least summarize whether or not increased rates of AMC are related to reported disease (perhaps inaccurately).
In summary, I do think a nod to the most famous of historical papers that explored relations between a significant pubic health outcome (suicide) and religion is warranted, especially since that work is commonly used as an exemplar of the 'Ecological Fallacy', at least among epidemiologists. The authors have invoked the religious origins of a potent social upheaval/reformation as leading to differences in prescribing and use of a modern technology hundreds of years later. A brief stop in the 19th century to acknowledge the work of a French sociologist who made similar conclusions seems warranted.
One minor note: I was looking for IS in Figure 1 legend but couldn't find it. I imagine it is for Iceland based on Table  1 and % Protestant?
Rönnerstrand, and Lapuente show that within countries in Europe, there are marked variations by regions and presumably not likely influenced by the percent of that country's population that were protestant, as it also occurs within countries where there is little Protestantism e.g., Italy. What did correlate, however, were the local levels of corruption. This again is what these authors have shown in their data. So, in summary, the authors (Kenyon and Fatti) did show an association with Protestantism and antimicrobial consumption, but even though they only looked at limited parameters, their own data suggest religion itself plays little part in any cause and effect. It appears that lower consumption of antimicrobials is likely very much more associated with regions that have better governance and lower corruption rates, rate than their conclusion of cultural differences that emerged in the Reformation.
We thank the reviewer for their useful comments. We agree that we need to be very careful with the conclusions we draw from this study for all the reasons outlined by the reviewer. We acknowledge that the Collignon et al., paper referred to by the reviewer is one of the most definitive ecological analyses of the drivers of AMR globally. This is however different to our research questions which were: what are the determinants of antimicrobial consumption (AMC), is the percent of a countries population that is Protestant associated with AMC and if so is this association a direct or an indirect one (via UA, POCO or CoC)?
As the reviewer notes, we find that there is a weak positive association and the effect is indirect-acting via UA and POCO. One parsimonious explanation for these findings is that an upstream determinant of UA and POCO is the percent of the population that is Protestant. As we note in the discussion, while we cannot prove that this is the causal pathway, this is at least a plausible way to interpret the data. To the best of our knowledge, this is the first time that statistical evidence has been produced to back up the percent Protestant-POCO/UA-AMC pathway.
Once again we have endeavoured in our discussion to make it very clear that whilst we find some statistical support for this pathway, this does not constitute strong evidence that this pathway played any role in the genesis of differential levels of AMC in European countries.
To address the reviewer's valid concerns, we have added the Collignon et al. reference and the following text in the discussion (4th paragraph):