Analysis of multiple bacterial species and antibiotic classes reveals large variation in the association between seasonal antibiotic use and resistance

Abstract

Understanding how antibiotic use drives resistance is crucial for guiding effective strategies to limit the spread of resistance, but the use-resistance relationship across pathogens and antibiotics remains unclear. We applied sinusoidal models to evaluate the seasonal use-resistance relationship across 3 species (Staphylococcus aureus, Escherichia coli, and Klebsiella pneumoniae) and 5 antibiotic classes (penicillins, macrolides, quinolones, tetracyclines, and nitrofurans) in Boston, Massachusetts. Outpatient use of all 5 classes and resistance in inpatient and outpatient isolates in 9 of 15 species-antibiotic combinations showed statistically significant amplitudes of seasonality (false discovery rate (FDR) < 0.05). While seasonal peaks in use varied by class, resistance in all 9 species-antibiotic combinations peaked in the winter and spring. The correlations between seasonal use and resistance thus varied widely, with resistance to all antibiotic classes being most positively correlated with use of the winter peaking classes (penicillins and macrolides). These findings challenge the simple model of antibiotic use independently selecting for resistance and suggest that stewardship strategies will not be equally effective across all species and antibiotics. Rather, seasonal selection for resistance across multiple antibiotic classes may be dominated by use of the most highly prescribed antibiotic classes, penicillins and macrolides.

Fig 1. Seasonal patterns of antibiotic use by class.

(A) Average daily antibiotic claims per 10,000 people by calendar month in Boston, Massachusetts from 2011 to 2015. Lines indicate LOESS smoothing curves and shaded regions indicate 95% CIs. (B) Sinusoidal model fits for monthly prescribing rate. Points indicate monthly mean seasonal deviates in average daily antibiotic claims per 10,000 people by calendar month and error bars indicate the standard error of the mean. Lines indicate the point estimate for the amplitude and phase of the sinusoidal model. Shaded regions indicate the 95% CIs for the amplitude. Asterisks indicate the amplitude of seasonality is statistically significant (FDR < 0.05). FDR, false discovery rate. Underlying data are available at https://github.com/gradlab/use-resistance-seasonality/tree/master/figure_data/Fig1 [16].

Fig 2. Seasonality of antibiotic use and resistance by class in Staphylococcus aureus.

Solid lines indicate point estimates of the amplitude and phase from the best-fitting sinusoidal model of resistance (comparing 6- and 12-month periods) to each antibiotic, colored by class. Dashed gray lines indicate point estimates of the amplitude and phase of sinusoidal models for use of the corresponding antibiotic class. Shaded regions indicate the 95% CIs for the amplitude. Points indicate the monthly mean seasonal deviates in resistance, and error bars indicate the standard error of the mean. Asterisks indicate the amplitude of seasonality in resistance is statistically significant (FDR < 0.05). FDR, false discovery rate; MIC, minimum inhibitory concentration. Underlying data are available at https://github.com/gradlab/use-resistance-seasonality/tree/master/figure_data/Fig2 [16].

Fig 3. Amplitudes and phases of seasonality by species and antibiotic class.

(A) Comparison of amplitudes estimated from best-fitting sinusoidal models of resistance (comparing 6- and 12-month periods) across antibiotics in Staphylococcus aureus, Escherichia coli, and Klebsiella pneumoniae. Error bars indicate 95% CIs of the amplitude. Point color indicates the antibiotic class. (B) Comparison of phases of seasonality of use and resistance across species and antibiotic classes. Points indicate peak month(s) of seasonal resistance estimated by the best-fitting sinusoidal model (comparing 6- and 12-month periods) for each species–antibiotic combination, and error bars indicate the 95% CIs. Included are species–antibiotic combinations for which the amplitude of seasonality of resistance was statistically significant (FDR < 0.05). Vertical lines indicate the peak month(s) of seasonal use estimated by the best-fitting sinusoidal model (comparing 6- and 12-month periods) for each antibiotic class, and shaded regions indicate the 95% CIs. AMC, amoxicillin-clavulanate; AMP, ampicillin; CIP, ciprofloxacin; ERY, erythromycin; FDR, false discovery rate; NIT, nitrofurantoin; OXA, oxacillin; PEN, penicillin; TET, tetracycline. Underlying data are available at https://github.com/gradlab/use-resistance-seasonality/tree/master/figure_data/Fig3 [16].

Fig 4. Spearman correlations between seasonal use and resistance with 0 to 3 months lag in Staphylococcus aureus.

Spearman rank correlation coefficients were calculated between the monthly mean seasonal deviate in resistance (in log₂(MIC)) and the monthly mean seasonal deviate in use (in average daily claims per 10,000 people) with 0, 1, 2, or 3 months lag between use and resistance, for each pairwise combination of antibiotics and classes. Error bars indicate the 95% CIs. Colors indicate the use antibiotic class. CIP, ciprofloxacin; ERY, erythromycin; Mac, macrolide; MIC, minimum inhibitory concentration; Nit, nitrofuran; NIT, nitrofurantoin; OXA, oxacillin; Pen, penicillin; Qui, quinolone; Tet, tetracycline. Underlying data are available at https://github.com/gradlab/use-resistance-seasonality/tree/master/tables/correlations.csv [16].