Assessment of Healthcare Facility Network Analysis to Inform Facility Enrollment into Regional Antimicrobial Resistance Prevention Collaboratives

Abstract

Background. Evidence suggests that any hospital intervention that prevents transmission of bacteria should have a larger impact on resistant strains if they transmit more in hospitals than sensitive strains and thereby have 'more to lose' from any hospital intervention (Lipsitch PNAS 2000). We aimed to identify whether observed declines in lengths of stay (LOS) and patient census can bias studies of hospital interventions, hypothesizing that such declines would decrease transmission in the hospital by reducing infectious patients' LOS per hospitalization and also decreasing the frequency with which infectious and susceptible people are hospitalized.Methods. We used a compartmental model of C. difficile transmission in a hospital and its catchment area by adapting a previously published model (van Kleef Wellcome Open Research 2017). Resistant and sensitive strains were included. We chose parameters to match data from a large study of CDI at hospital admission and its nosocomial spread (Loo NEJM 2011). We simulated decreases in average LOS from 6 to 5 days and decreases in average patient census by 15% as seen in England from 2006 to 2013. We used the simulated data to calculate rate ratios measuring changes in CDI due to resistant vs. sensitive strains. We plotted the simulated rate ratios and compared them to CDI trends in the UK reported by Dingle et al (Lancet ID 2017) for comparison.Results. We found that declines in LOS and patient census would be expected to produce large drops in CDI caused by resistant strains in the hospital and community, but no significant changes in CDI due to sensitive strains, Figure 1.Conclusion. We showed that changes in LOS and patient census, even in the absence of any stewardship or infection control interventions, can cause differential changes in disease rates for resistant vs. sensitive strains of bacteria. Analyses of HAI interventions should account for changes in LOS and patient census, especially when long time periods are considered (e.g., ≥5 years).