We would like to thank Thomas Heister and Klaus Kaier for their efforts and time (presumably close to 9 months) that they have taken to look at our paper and present their comments for us to consider. We agree with many of their comments. This includes that “Increasing antimicrobial resistance is a worldwide societal concern” and that “antimicrobial resistance is a multifaceted problem entrenched in the entire process of medical service delivery with its complex structures.” We also agree that our paper was an “attempt to add to this growing body of research by arguing that corruption in a society is a major driver of resistance” and that “If this holds, it would be an important message to policy makers around the world concerned with fighting resistance that dealing with corruption could alleviate this issue.”

However we don’t agree with their interpretation of our methods and findings. Nor do we agree with many of their assertions and conclusions. We do not agree with their view that governance issues are unimportant in the area of antimicrobial resistance. We also disagree with their view that our work “entirely ignores many important factors that previous research has shown to significantly influence resistance levels.” In our paper many of these issues were extensively discussed.

We believe that one important reason that Heister and Kaier have reached their contrary viewpoint is that their view on antibiotic resistance appears to be mainly from just an ICU and hospital perspective. However, as is now widely acknowledged, the antibiotic resistance problem is better looked at from a “One Health” approach. We agree that what is done in the hospital sector is very important - both what volumes of antibiotics are used plus how effective infection control procedures are to prevent spread of resistant bacteria. However we can’t ignore that in people over 90% of antibiotic are used in the community rather than in hospital. While some bacteria and their resistance determinants seem to be mainly associated with hospitals, this is not true for many other common bacteria where the community is the source for most infections and resistance determinants (e.g. pneumococcus and E.coli). Even with MRSA, which was considered previously to be almost all hospital associated, now in many countries, infections by MRSA is predominantly caused by strains that originate and spread in the community and then are introduced into hospitals.

Also very importantly, the majority of antibiotics are not used in the human sector. They are mainly used in agriculture - in the US over 80% by volume. These include “critically important” antibiotics such as fluoroquinolones and 3rd/4th generation cephalosporins, on which we often place restrictions in hospitals because of resistance concerns. Resistance is driven by usage volumes and the spread of resistant bacteria or resistance genes. All sectors are inter-related and resistant bacteria move from sector to sector.

Our data showed that the level of resistance in countries correlated with the control of corruption in those countries. While Heister and Kaier look at this from their hospital and ICU perspective, if one takes a broader look at this issue, such a relationship is not surprising. They say “It is, however, difficult to see a causal relation between corruption and resistance as the former describes the “abuse of entrusted power for private gain”(https://www.transparency....). However they have only partly quoted from the reference they give. In the reference they quote, the importance of controlling corruption at the local level (including in hospitals) is commented on as an example.

The levels of resistance seen in bacteria in any country will be a result of the volume of antibiotics used and how easy it is for resistant bacteria to spread. Whenever there are issues with “governance” (in hospitals, in the local community and in a country as a whole) then there will most likely result in more resistance being seen. For many healthcare related issues, it is well established that poor governance results in worse patients outcomes. We are not sure why Heister and Kaier have difficulty believing this would also be the case for antibiotic resistance. If there is poor governance (of which control of corruption is one variable where data is available for countries) there is a high chance that people will more likely bend or break rules. This includes the use of over-the-counter antibiotics. Hence antibiotic usage may be higher than what official figures show, people will use banned or broad spectrum antibiotics in food animals when this is not permitted. Or factories or hospitals may be more lax about the disposal of antibiotic-contaminated waste into environmental waterways, with subsequent exposure of environmental bacteria to durable antibiotics. Resistant bacteria recycle to people and animals via foods and water. If appropriate standards for food and water are not regulated or enforced, more resistant bacteria spread. Within hospitals, lack of control of corruption means some products may not be available or substandard products are only supplied. Infection control will then be more expensive to implement and/or there will less ability of essential supplies to make sure appropriate infection control practises are able to be followed. There are numerous other examples we could give to show how poor governance (of which corruption is one measurable parameter) will likely result in more resistant bacteria developing and being easier to spread to other people.

Heister and Kaier imply that our methodology was inappropriate and thus our results and conclusions are therefore invalid. We don’t agree. They base this, among other issues, on sample sizes, whether appropriate weightings were given to data, and perceived limitations of the fixed effects (FE) estimation method. We used very large sample sizes based on data made available by ECDC. The data involved more than 700,000 bacterial isolates causing blood stream infections, 25 pathogen/antibiotic combinations grouped under seven pathogen classes and data from 28 European countries. Despite the large sample size it is not surprising that some “cells” might have low numbers. We agree that there are various ways and complexities by which this data could be looked at. However our methodology did as would be expected, clearly show there was a relationship in these countries with antibiotic resistance and their antibiotic usage data. Do Heister and Kaier suggest this is also an invalid result? The logic of their argument is that if we have shown such a relationship using our methodology, such a relationship is invalid. We don’t think many would agree with that proposition.

In experimental runs we also used two alternative indicators of governance (rules of law and government effectiveness) in pooled OLS estimation. The results were comparable. (We were not able to use these two variables in FE and system GMM estimation because of data gaps - see Page 4 of our paper).

The sole focus of their critique of our empirical evidence is on perceived limitation of the fixed effect estimator. They have completely ignored the results based on the system GMM estimator, which we used as a powerful robustness check. The system GMM estimator take care of potential reverse causality from the dependent variable to explanatory variables, and more importantly, possible errors in the measurement of variables (which Heister & Kaier have repeatedly emphasised in their comment). Our inferences remain remarkably robust to the use of system GMM; in fact, the coefficient of GOV is much larger in magnitude with a higher level of statistical significance in SGMM regressions compared that in the FE regression.

Relating to their critique of the FE results, it is true that GOV has little variability for Norway, Finland and Luxemburg. However, this variable has sufficient variability across the 28 countries and over time in most countries to warrant the use of FE estimator to control for unobserved country heterogeneity (See Table 2).
We reiterate that we believe there are many obvious reasons that one can see why poor governance (including “control of corruption”), would be associated with higher levels of antimicrobial resistance. We believe that the methodology we have employed is appropriate and clearly shows this association is present along with ten expected association of resistance and antibiotic usage volumes.

Heister and Kaier have obviously spent many months examining our data and our paper. We are preparing a much more detailed response to the many issues they have raised. However given the extensive nature of their comments it will take a reasonable amount of time for us to adequately prepare and reference our response. We are uploading the preliminary comments above because we believe that many of their assertions cannot go for a long period unchallenged. Our more detailed response will follow.