^{1}

^{2}

^{3}

^{4}

^{5}

^{6}

^{3}

The authors have declared that no competing interests exist.

Conceived and designed the experiments: PC PA SS. Performed the experiments: PA FK. Analyzed the data: PC PA SS FK. Contributed reagents/materials/analysis tools: PA. Wrote the paper: PC PA SS FK. Found and used econometric procedures to analyze data: PA FK.

To determine how important governmental, social, and economic factors are in driving antibiotic resistance compared to the factors usually considered the main driving factors—antibiotic usage and levels of economic development.

A retrospective multivariate analysis of the variation of antibiotic resistance in Europe in terms of human antibiotic usage, private health care expenditure, tertiary education, the level of economic advancement (per capita GDP), and quality of governance (corruption). The model was estimated using a panel data set involving 7 common human bloodstream isolates and covering 28 European countries for the period 1998–2010.

Only 28% of the total variation in antibiotic resistance among countries is attributable to variation in antibiotic usage. If time effects are included the explanatory power increases to 33%. However when the control of corruption indicator is included as an additional variable, 63% of the total variation in antibiotic resistance is now explained by the regression. The complete multivariate regression only accomplishes an additional 7% in terms of goodness of fit, indicating that corruption is the main socioeconomic factor that explains antibiotic resistance. The income level of a country appeared to have no effect on resistance rates in the multivariate analysis. The estimated impact of corruption was statistically significant (p< 0.01). The coefficient indicates that an improvement of one unit in the corruption indicator is associated with a reduction in antibiotic resistance by approximately 0.7 units. The estimated coefficient of private health expenditure showed that one unit reduction is associated with a 0.2 unit decrease in antibiotic resistance.

These findings support the hypothesis that poor governance and corruption contributes to levels of antibiotic resistance and correlate better than antibiotic usage volumes with resistance rates. We conclude that addressing corruption and improving governance will lead to a reduction in antibiotic resistance.

Antibiotic resistance is a growing international problem. This has led to increasing numbers of serious infections that are very difficult, or sometimes impossible to treat. Increasing resistance involves nearly all bacteria that infect people, including very common ones such as

Antibiotic resistance develops and spreads wherever antibiotics are used, not only in medical facilities but also in the community. Poor infection control, poor water sanitation and poor hygiene all facilitate the spread of resistant bacteria from person to person. The majority of antibiotic usage worldwide is in food animals [

The general perception of antibiotic resistance is that it is almost entirely related to the amounts of antibiotics used, not only in the broad sense of comparative usage by different countries but also in individuals [

We postulate that a fuller understanding of antibiotic resistance requires a unified approach which combines not only the amount of antibiotic usage but also other factors that may impact on the quality of antibiotic usage, control and the spread of resistant bacteria at national levels. We therefore undertook a multivariate analysis of determinants of antibiotic resistance in bacteria causing bloodstream infections throughout Europe.

For the empirical analysis, we used a newly-constructed panel data set covering 28 countries in Europe over the period 1998–2010 (full supporting information are available on the web;

The multivariate regression model used for explaining antibiotic resistance is,

The explanatory variables, with the expected sign of the regression coefficient in brackets, are

μ_{i} Country-specific and time-invariant fixed factors.

γ_{t} Time Effects (i.e. binary variables for each year except for the first year) to capture time varying common shocks or global trends, and

ε_{i,t} is the idiosyncratic error term that captures all other determinants of ABR

Europe was chosen for estimating the model because it is the only region where good data for all of our parameters are available across multiple countries. Our panel data set covered 28 countries in Europe over the period 1998–2010. The countries covered are Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Norway, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, and United Kingdom. The country- and time-coverage of the analysis are dictated by the nature of availability of data on the dependent variables (ABR). These data on antimicrobial resistance were compiled from EARS-Net Database of the

In this study, we looked at 25 pathogen/antibiotic combinations which are grouped under seven pathogen classes. The pathogen classes and the relevant antibiotics are listed as follows:

^{rd} gen cephalosporins, aminoglycosides, carbapenems and fluoroquinolones.

^{rd} gen cephalosporins, aminoglycosides, aminopenicillins, carbapenems and fluoroquinolones.

Data for ‘S.

Data on antibiotic usage are from

In this paper, we specifically focus on the quality of governance in comparison to antibiotic usage as determinants of antibiotic resistance, while controlling for the other possible determinants. We measure the quality of governance in terms of the control of corruption, which is the most ubiquitous manifestation of governance quality in policy debates [

The indicator governance (

Variables | Mean | Minimum | Maximum | Countries | Observations |
---|---|---|---|---|---|

ABR | 17.03 (7.87) | 0 | 38.96 | 28 | 247 |

ABU | 19.61 (6.36) | 9.7 | 45.2 | 28 | 266 |

GOV | 3.94 (1.20) | 2 | 6 | 28 | 304 |

PHE | 2.03 (0.68) | 0.43 | 4.01 | 28 | 308 |

PGDP | 26405(11739) | 6533 | 74114 | 28 | 308 |

TED | 54.86 (17.76) | 9.80 | 95.07 | 28 | 307 |

AGR | 6.52 (4.87) | 1.1 | 26.2 | 28 | 308 |

Note: Standard Deviations are reported in parentheses.

The model is initially estimated using the Pooled Ordinary Least Squares (POLS) method, which estimates the equation excluding the country—specific fixed factors (μ_{i}). However, ignoring country-specific fixed effects could cause omitted variable bias, because country-specific and time-invariant factors might be correlated with both ABR and the explanatory set. These factors include geography, historical experience, legal origin, ethno linguistic fragmentation and culture. We therefore re-estimated the model using the fixed effects (FE) technique. This technique involves estimating the equation after demeaning the variables to purge possible country-specific fixed effects in order to isolate the ‘within country’ relationship.

While the FE estimation technique addresses possible endogeneity arising from ignoring country specific fixed effects, it does not address two other sources of endogeneity—potential reverse causality from the dependent variable to the explanatory variables, and possible error in the measurement of the variables. For instance it is plausible that higher ABR can cause overuse of antibiotics rather than causality running in the direction we have assumed. In addition it is obvious that the construction of our variables (especially ABR, ABU and GOV which are the main variables of interest) may suffer from a measurement error. Both types of endogeneity lead to biased and inconsistent estimates for the coefficients. When the explanatory variables are potentially endogenous, we need to be cautious in inferring causality from the estimated coefficients.

We therefore use the system Generalized Method of Moments (system GMM) technique as a robustness check. This technique consistently estimates the equation even in the presence of endogenous explanatory variables and measurement error, and thus allows us to be more confident in inferring causality [

System GMM generates a set of internal instruments for the right hand side variables which are uncorrelated with the error term. This set consists of the appropriate number of lags of the right hand side variables as instruments in the differenced version of the original equation and the first differences (and the appropriate number of lagged first differences) as instruments in the original level equation. The validity of the instrument set can be tested through a Hansen test of over-identifying restrictions. However, this test is weakened if the number of instruments is large relative to the number of observations, and works best if the number of time periods in the panel is small [

Note that the unit of measurement used varies among the variables employed in our analysis:

Our results support the crucial role of governance in determining antibiotic resistance across these countries. When we use the simplest of our estimations, only 28 percent of the total variation in antibiotic resistance (adjusted R^{2} = 0.28) is attributable to variation in antibiotic usage in people. When the time effects are also included to take into account possible shocks or global trends (not captured by the other explanatory variables) impacting antibiotic resistance uniformly across all countries, the regression explains 33 percent of the total variation in antibiotic resistance. Once the control of corruption indicator is included as an additional explanatory variable, 63 percent of the total variation in antibiotic usage is explained by the regression. The complete multivariate regression only accomplishes an additional 7 percentage point in terms of goodness of fit, indicating that corruption is the main socioeconomic factor that explains antibiotic resistance. The income level of a country appeared to have no effect on resistance rates when we also controlled for the quality of governance in the multivariate analysis.

To aid interpretation of econometric estimates, the pair-wise correlation matrix for the variables is reported in

ABR | ABU | GOV | PHE | PGNI | TED | AGR | |
---|---|---|---|---|---|---|---|

ABR | 1.00 | ||||||

ABU | 0.53 | 1.00 | |||||

GOV | -0.71 | -0.28 | 1.00 | ||||

PHE | 0.47 | 0.35 | -0.18 | 1.00 | |||

PGDP | -0.39 | 0.05 | 0.52 | -0.23 | 1.00 | ||

TED | -0.11 | -0.17 | 0.02 | 0.15 | -0.07 | 1.00 | |

AGR | 0.47 | 0.23 | -0.38 | 0.28 | -0.59 | 0.11 | 1.00 |

The simple correlation of each explanatory variable with antibiotic resistance is consistent with our expectations for the expected signs of the estimated coefficients. Governance is negatively correlated with

Figs.

The estimation results are reported in

(1) | (2) | (3) | (4) | (5) | (6) | (7) | (8) | (9) | |
---|---|---|---|---|---|---|---|---|---|

Independent Variables | OLS | OLS | OLS | OLS | FE | FE | SGMM | SGMM | SGMM |

ABU | 0.51(0.12) |
0.36(0.09) |
0.29(0.09) |
0.25(0.14) |
0.16(0.11) | 0.12(0.11) | 0.08(0.32) | 0.32(0.07) |
0.07(0.11) |

GOV | -0.56(0.07) |
-0.37(0.10) |
-0.25(0.12) |
-0.22(0.12) |
-0.55(0.11) |
-0.65(0.24) |
|||

PHE | 0.20 (0.06) |
-0.08(0.16) | 0.22(0.07) |
||||||

PGDP | -0.17(0.08) |
-0.02(0.45) | 0.06(0.16) | ||||||

TED | -0.11(0.04) |
-0.01(0.18) | -0.07(0.07) | ||||||

AGR | 0.13(0.17) | -0.54(0.54) | -0.02(0.15) | ||||||

ABR_{t-2} |
0.31(0.10) |
0.18(0.14) | 0.15(0.16) | ||||||

Countries | 28 | 28 | 28 | 28 | 28 | 28 | 28 | 28 | 28 |

Observations | 242 | 242 | 242 | 242 | 242 | 242 | 99 | 99 | 99 |

T.E included | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |

p-value of ‘ABU = -Cor’ | 0.15 | 0.61 | 0.65 | 0.61 | 0.08 | <0.01 | |||

Adjusted R^{2} |
0.33 | 0.63 | 0.70 | 0.80 | 0.81 | 0.81 | |||

R^{2}(within) |
0.23 | 0.27 | 0.28 | ||||||

Instruments | 28 | 31 | 35 | ||||||

Sargan test p-value | 0.21 | 0.58 | 0.71 | ||||||

Hansen J test p-value | 0.47 | 0.60 | 0.79 | ||||||

AR(2) test p-value | 0.01 | 0.50 | 0.56 | ||||||

Wald chi-sq statistic | 74.36 | 227.99 | 845.02 | ||||||

Wald chi-sq p-value | <0.01 | <0.01 | <0.01 |

Notes:

The standardized (beta) regression coefficients are reported in the table. TE refers to the set of time dummy variables i.e. the time effects. The estimated coefficient for the constant term and TE are not reported. ‘ABU = -Cor’ refers to the F test for equality of the magnitude for the coefficient of ^{2} value of the regression in Column (1) without including the time effects is 0.28. The R^{2} in Columns (4), (5) and (6) refers to the coefficient of determination from estimation of the equivalent Least Squares Dummy Variable Model (LSDV).

For System GMM, Windmeijer-Corrected Robust standard errors from the two-step GMM estimation are reported in parenthesis. Observations at 2 year intervals used from 1998–2010. Orthogonal forward deviations are used to purge fixed effects. The main explanatory variables (

*Significant at the 10% level

**Significant at the 5% level

***Significant at the 1% level

Focusing on the complete multivariate regression (Column 3), the coefficients of all variables have the expected sign and, except that of the employment share of agriculture, are statistically significant at the one percent or five percent level. In terms of magnitudes of the coefficients, the governance variable has the greatest association with antibiotic resistance; a one unit improvement in the indicator is estimated to lower antibiotic resistance by 0.4 units. A similar reduction in antibiotic usage only reduces resistance by 0.3 units. As for the other variables which are statistically significant, a one unit increase in growth in tertiary enrollment brings about a reduction of 0.1 units in

The FE results reported in Columns [^{2}: adding additional variables into the specification practically makes no difference, indicating the central role of fixed country characteristics such as culture and geography. Interestingly, when fixed country characteristics are controlled for, governance is still a statistically significant determinant of antibiotic resistance. The estimated coefficients suggest that, in the within country context, a one unit improvement in the corruption score is associated with a 0.2 unit reduction in ABR.

The system GMM estimates are given in Columns 7–9. The results suggest a causal relationship between our variables of interest after addressing all possible sources of endogeneity. The diagnostics indicate that the model has been adequately estimated. The p-value of the Hansen test of over-identifying restrictions indicates that the instrument set, which identifies the exogenous variation in the explanatory variables, is valid. Moreover, the p-value is not unrealistically high, which would have suggested an over-fitting bias. The p-value of the AR (2) test also indicates that there is no second order serial correlation in the residuals, a necessary condition for consistent estimation. An unrealistically high p-value is a symptom of instrument proliferation in which case the Hansen test is weakened [

Control of corruption remains the most important determinant of antibiotic resistance according to the system GMM results. The coefficient of

The results of our empirical analysis show that factors other than antibiotic usage are potentially very important in explaining the different levels of resistant bacteria seen in different countries (Figs.

Private health expenditure was also an important factor. The higher the percentage of private health expenditure in a country, the larger was the degree of antibiotic resistance. The reason for this is not clear. We postulate that when healthcare is being delivered predominantly in the private sector, there are less controls and supervision of what is being done. This then may mean that there are fewer controls on broad-spectrum agents, the length of time of drug therapy and the volumes used.

We postulate that when Quality of Governance is poor, then there are likely to be less effective controls of antibiotic use (not only in people but in the animal sector). Thus, not only will more antibiotic resistant bacteria develop but the spread of these resistant bacteria will also be easier. This is because there will be less supervision and enforcement of laws that cover issues, not only of human medicine, but also of food and water safety. Reducing antimicrobial resistance requires a policy mix aimed at lowering antibiotic usage in people and, perhaps even more importantly, developing better controls on corruption.

Surprisingly, our results suggest that antibiotic resistance in the previous period does not affect present resistance. This implies that previously high levels of antimicrobial resistance in a country are not insurmountable barriers to future improvements.

The data we have shown only pertain to Europe. This is because this is only region where good data are available from multiple countries. Europe has data not only on the issues of government effectiveness and rule of law etc, but more importantly on the magnitude and extent of antimicrobial usage and resistance levels in serious bloodstream infections caused by common bacteria such as S. aureus and E. coli. We suspect that these same issues will pertain worldwide (e.g. corruption). In developing countries, they are likely to be much more of a factor, and quality of governance will likely explain why antibiotic resistance is so much higher in these countries than in most countries in Europe. The fact that these data are publicly available for Europe, suggest that issues of governance are also much better in Europe than in most of the rest of the world.

To our knowledge, this is the first cross-national examination of the determinants of antibiotic resistance. Our results, based on a systematic econometric analysis using a panel data set consisting of 29 European countries over the time period 1998–2010, suggest that governance is an even more important determinant of antibiotic resistance than antibiotic usage in people. The estimated impact of governance is large and highly statistically significant, using state-of-the-art econometric methodologies that pay particular attention to issues of endogeneity. The results are robust to the use of alternative estimation methods.

There are, however, limitations in our study. One of our main explanatory variables, “quality of governance” (

We acknowledge that the issue of antibiotic resistance and how it develops and spreads is complex and our results may not be generalizable to other settings. Antibiotic resistance involves factors both in the human and agricultural sectors and the environment [

Our findings have important policy implications. Just as a sizable literature has convincingly demonstrated that the quality of institutions (governance) is the fundamental determinant of economic growth—our results suggest that improving governance could be similarly fundamental in confronting the issue of antibiotic resistance. Improved governance at the national level is likely to imply better practices in the health sector, including controls and supervision of antibiotic use. It is also likely to lead to healthier outcomes in a whole array of other related areas that are intertwined with the issue of antibiotic resistance, such as the agriculture sector, food and water safety.

(XLSX)

We would like to thank Dr. Ashwin Swaminathan for his helpful advice after critically reviewing this manuscript.