The Persistence of Antibiotic Resistance in Observational Studies: Is It Really Due to Differences in Sub-Populations Rather than Antibiotic Use?

Background: The carriage of resistant bacteria and prior antimicrobial treatment are related, but in an individual, this diminishes over time. To better manage antimicrobial resistance risks, it is crucial that we better untangle any lasting impact of antibiotic use compared to other factors. This understanding is essential for informing antimicrobial stewardship programs and to better manage other important factors that likely contribute to persistently higher rates of antimicrobial resistance in different populations. The true association between antibiotic use and resistance is likely to be significantly overestimated due to the confounding influence of varying infection risk patterns within populations. Though missing explanatory covariates are a well-known cause of falsely interpreted statistical findings, how the problem manifests in this context has a particular and interpretable structure. This issue does not appear to have been previously addressed with clarity. To be more easily understood, a simple model is used to demonstrate this. Results: In our theoretical model case study, when we exclude an effect of past antibiotic usage, clinical history alone can predict future resistance patterns. Heterogeneity in infection risk and antibiotic resistance carriage rates, along with consequently observed antimicrobial treatment, often suffice to predict a pattern of resistance that mimics what is assumed to be caused by genuine biologically driven resistance by the associated use of antibiotics. The biological impact and/or lasting effects of antibiotics are not necessary for this prediction. Conclusions: Antimicrobial stewardship policies and future research must directly address how much of the apparent persistence of resistant bacteria results from biological consequences of antibiotic use compared to pure statistical confounding arising due to heterogeneous risks in community infection patterns.