Contribution of Governance and Socioeconomic Factors to the P. aeruginosa MDR in Europe

Abstract

This work aims to explain the behavior of the multi-drug resistance (MDR) percentage of Pseudomonas aeruginosa in Europe, through multivariate statistical analysis and machine learning validation, using data from the European Antimicrobial Resistance Surveillance System, the World Health Organization, and the World Bank. We ran a multidimensional data panel regression analysis and used machine learning techniques to validate a pooling panel data case. The results of our analysis showed that the most important variables explaining the MDR phenomena across European countries are governance variables, such as corruption control and the rule of law. The models proposed in this study showed the complexity of the antibiotic drugs resistance problem. The efforts controlling MDR P. aeruginosa, as a well-known Healthcare-Associated Infection (HCAI), should be focused on solving national governance problems that impact resource distribution, in addition to individual guidelines, such as promoting the appropriate use of antibiotics.

Keywords: corruption index; data panel; governance index; machine learning; multi-drug resistance.

Figure 1

Clusters using $k$-means method.

Figure 2

TWFE panel MDR-Pa model by geographical unit: country effects.

Figure 3

Machine learning using the XGBoost method. (a) Feature importance measured by SHAP values in the training dataset on the target variable MDR-Pa, respectively. (c) Feature importance measured by SHAP values in the testing dataset on the target variable MDR-Pa. (b–d) Impact of features for SHAP values for each feature for the XGBoost method in training (b) and testing (d) the dataset. Every observation is represented by one dot in each feature. The dot’s position on the x-axis represents the impact of that feature on the model’s prediction for the observation, and the dot’s color represents the value of that feature for the observation.

Figure 4

Machine learning using the random forest method. (a) Feature’s relevance/importance, measured by SHAP values of in the training dataset on the target variable MDR-Pa. (b) Feature importance measured by SHAP values in feature relevance of the testing dataset on the target variable MDR-Pa.