Dynamic association of antimicrobial resistance in urinary isolates of Escherichia coli and Klebsiella pneumoniae between primary care and hospital settings in the Netherlands (2008-2020): a population-based study

Abstract

Background: It is unclear whether changes in antimicrobial resistance (AMR) in primary care influence AMR in hospital settings. Therefore, we investigated the dynamic association of AMR between primary care and hospitals.

Methods: We studied resistance percentages of Escherichia coli and Klebsiella pneumoniae isolates to co-amoxiclav, ciprofloxacin, fosfomycin, nitrofurantoin and trimethoprim submitted by primary care, hospital outpatient and hospital inpatient settings to the Dutch National AMR surveillance network (ISIS-AR) from 2008 to 2020. For each bacterium-antibiotic combination, we first conducted multivariable logistic regressions to calculate AMR odds ratios (ORs) by month and healthcare setting, adjusted for patient-related factors and a time term. Second, multiple time series analysis was done using vector autoregressive models including the (log) ORs for each bacterium-antibiotic combination. Models were interpreted by impulse response functions and Granger-causality tests.

Findings: The main AMR association was unidirectional from primary care to hospital settings with Granger-causality p-values between <0.0001 and 0.029. Depending on the bacterium-antibiotic combination, a 1% increase of AMR in E. coli and K. pneumoniae in primary care leads to an increase of AMR in hospital settings ranging from 0.10% to 0.40%. For ciprofloxacin resistance in K. pneumoniae, we found significant bidirectional associations between all healthcare settings with Granger-causality p-values between <0.0001 and 0.0075.

Interpretation: For the majority of bacterium-antibiotic combinations, the main AMR association was from primary care to hospital settings. These results underscore the importance of antibiotic stewardship at the community level.

Funding: ISIS-AR is supported by the Ministry of Health, Welfare and Sport of the Netherlands and the first author by the Central University of Ecuador to follow a PhD program in Erasmus MC.

Keywords: Antimicrobial resistance; General practitioners; Granger causality; Health care settings; Multiple time series.

Fig. 1

Antimicrobial resistance time trends in E. coli by healthcare settings and antibiotic type in the Netherlands. AMR series for ciprofloxacin, nitrofurantoin and trimethoprim are from 2008 to 2020. AMR series for co-amoxiclav are from 2008 to 2015, while AMR series for fosfomycin are from 2014 to 2020. Adjusted time series (OR of Time) with their 95% CIs (shadows) were calculated by multiple logistic regression analysis adjusted for patient-related factors and calendar time (left y-axis). As reference, the horizontal dotted line shows an OR = 1. Original time series (%) were calculated as the proportion of resistant isolates among the total number of tested isolates for each antibiotic of interest (right y-axis).

Fig. 2

Antimicrobial resistance time trends in K. pneumoniae by healthcare settings and antibiotic type in the Netherlands. AMR series for ciprofloxacin, nitrofurantoin and trimethoprim are from 2008 to 2020. AMR series for co-amoxiclav are from 2008 to 2015, while AMR series for fosfomycin are from 2014 to 2020. Adjusted time series (OR of Time) with their 95% CIs (shadows) were calculated by multiple logistic regression analysis adjusted for patient related factors and calendar time (left y-axis). As reference, the horizontal dotted line shows an OR = 1. Original time series (%) were calculated as the proportion of resistant isolates among the total number of tested isolates for each antibiotic of interest (right y-axis).

Fig. 3

Impulse response functions of AMR in E. coli between primary care (PC) hospital outpatient (HO) and hospital inpatient (HI) settings. Solid lines are the estimated impulse responses of AMR after a shock consisting of a 1% relative increase of AMR in each healthcare setting, and shadows are the 95% confidence intervals determined by bootstrapping 1000 repetitions. The y-axis shows the magnitude of the response proportional to the shock. The x-axis shows the 12-month time period over which the response was traced out.

Fig. 4

Impulse response functions of AMR in K. pneumoniae between primary care (PC) hospital outpatient (HO) and hospital inpatient (HI) settings. Solid lines are the estimated impulse responses of AMR after a shock consisting of a 1% relative increase of AMR in each healthcare setting, and shadows are the 95% confidence intervals determined by bootstrapping 1000 repetitions. The y-axis shows the magnitude of the response proportional to the shock. The x-axis shows the 12-month time period over which the response was traced out.