Limitations of estimating antibiotic resistance using German hospital consumption data - a comprehensive computational analysis

Abstract

For almost a century, antibiotics have played an important role in the treatment of infectious diseases. However, the efficacy of these very drugs is now threatened by the development of resistances, which pose major challenges to medical professionals and decision-makers. Thereby, the consumption of antibiotics in hospitals is an important driver that can be targeted directly. To illuminate the relation between consumption and resistance depicts a very important step in this procedure. With the help of comprehensive ecological and clinical data, we applied a variety of different computational approaches ranging from classical linear regression to artificial neural networks to analyze antibiotic resistance in Germany. These mathematical and statistical models demonstrate that the amount and particularly the structure of currently available data sets lead to contradictory results and do, therefore, not allow for profound conclusions. More effort and attention on both data collection and distribution is necessary to overcome this problem. In particular, our results suggest that at least monthly or quarterly antibiotic use and resistance data at the department and ward level for each hospital (including application route and type of specimen) are needed to reliably determine the extent to which antibiotic consumption influences resistance development.

Keywords: Acute care hospital; Antibiotic consumption; Antibiotic resistance; Antimicrobial stewardship; Computational modelling; Machine learning.

Fig. 1

Time course of antibiotic consumption. (a) Boxplots of total antibiotic consumption of all individual hospital departments and ward types for each quarter from 2015 to 2019. (b) Average proportion of total antibiotic consumption for each quarter and some selected antibiotic classes.

Fig. 2

Boxplots of the proportion of resistant E. coli to (a) ciprofloxacin and (b) cefotaxime of all regions for each quarter from 2015 to 2019 for German hospital department and ward types in internal ARS database. The boxplots in (c) and (d) show the results for the in-house UKD department and ward type data accordingly.