Fig. 2: Prevalence and determinants of longitudinal ESBL-E carriage.

a, ESBL prevalence stratified by the three study groups; inpatients exposed to antimicrobials (red), inpatients without antimicrobial exposure (blue), community members (green), showing sharp increase in prevalence following antimicrobial exposure. Prevalence is estimated using a LOESS non-parametric regression with 95% CI. Community members are censored on antimicrobial exposure or hospitalization and antimicrobial-unexposed inpatients on antimicrobial exposure. b, Simulated ESBL-E prevalence (with 95% CrI) using final fitted model for a hypothetical cohort of patients with initial ESBL-E colonization prevalence 50%, admitted to hospital for 7 d and exposed to 7, 2 or 0 d of antimicrobials, showing that there is little difference between 7 and 2 d. c–f, Posterior estimates of parameter values from final fitted model. Shaded grey areas shows 95% CrI and grey vertical line shows median parameter estimate. c,d, Hazard ratio of gain or loss of ESBL-E (expressed as natural logarithm) showing that antimicrobial exposure (abx) acts primarily to prolong carriage by reducing ESBL-E loss, whereas hospitalization (hosp) acts to increase both gain and loss, with a net effect to increase prevalence. e, Mean time in colonized/uncolonized states with all covariate values set to 0 (that is, no hospitalization or antimicrobial exposure). f, Half-life of effect of antimicrobial exposure, showing that antimicrobial exposure acts with a prolonged effect to prolong colonization.