Quantifying antibiotic impact on within-patient dynamics of extended-spectrum beta-lactamase resistance

Niehus, Rene

van Kleef, Esther

Yin, Mo

Turlej-Rogacka, Agata

Lammens, Christine

Carmeli, Yehuda

Goossens, Herman

Tacconelli, Evelina

Carevic, Biljana

Malhotra-Kumar, Surbhi

Cooper, Ben S.

rniehus@hsph.harvard.edu

Published Mar 31, 2020 on Dryad. https://doi.org/10.5061/dryad.8vf034t

Cite this dataset

Niehus, Rene et al. (2020). Quantifying antibiotic impact on within-patient dynamics of extended-spectrum beta-lactamase resistance [Dataset]. Dryad. https://doi.org/10.5061/dryad.8vf034t

Abstract

Antibiotic-induced perturbation of the human gut flora is expected to play an important role in mediating the relationship between antibiotic use and the population prevalence of antibiotic resistance in bacteria, but little is known about how antibiotics affect within-host resistance dynamics. Here we develop a data-driven model of the within-host dynamics of extended-spectrum beta-lactamase (ESBL) producing Enterobacteriaceae. We use bla_CTX-M (the most widespread ESBL gene family) and 16S rRNA (a proxy for bacterial load) abundance data from 833 rectal swabs from 133 ESBL-positive patients followed up in a prospective study in three European hospitals. We find that cefuroxime and ceftriaxone are associated with increased bla_CTX-M abundance during treatment (21% and 10% daily increase, respectively), while treatment with meropenem, piperacillin-tazobactam, and oral ciprofloxacin is associated with decreased bla_CTX-M (8% daily decrea se for all). The model predicts that typical antibiotic exposures can have substantial long-term effects on bla_CTX-M carriage duration.

Usage notes

CTX-M and 16S qPCR measurements

File contains for all patients and all sampled timepoints the quantitative PCR (qPCR) results targeting both 16S and CTX-M ("type"-column). The measurements are given as qPCR ct values ("ct") and as adjusted quantity (adjusted for dilutions of certain samples) ("quantity_adj"). Additional columns give means (on log-scale) across replicate qPCR runs ("xlog_mean_swabtype") and respective standard deviation ("xlog_sd_swabtype"). Patient age, sex, and country of the hospital are also given.

qpcr_CTXM_16S.csv

Antibiotic treatment

File gives for all patients and all days of observation the antibiotic treatment as a binary indicator (0: not given, 1: given). It is also indicated on which days a rectal swab was obtained ("sample_take") and for rows where a sample was taken, the mean 16S and CTX-M qPCR measurement (on log-scale) is given ("m_16S", "m_CTX_M").

abx_treatment.csv

R script to infer dynamic parameters from real and simulated data

Script 1) loads empirical data 2) extracts structural info from that data 3) simulates fake data with identical structure 4) runs the stan model 5) plots the results

qPCR_real_and_sim_inference.R

Data in Rdata format

Contains several dataframes: da5 # includes all observed days for all patients (3993x74), xdf_s # includes every qPCR result (mostly duplicates) for all samples and the two genetic targets (CTX-M,16S) (3361x99), sw_df # includes all swabs and gives previous and next swab indicators, my_abxs_new_order # give the list of antibiotics that I want to inlude, mspline # the smooth spline linking qCPR mean and error (spline fitted to the data)

dfs_qPCR.Rdata

Bayesian dynamic model written in stan

Model as defined in Equation (4) in the paper written in stan (Carpenter et al., 2017)

qpcr02_09_exp_growth.stan

Location

Europe