Background: Statistically significant variation in antimicrobial resistance (AMR) occurs between hospitals, within hospitals, and over time. Whilst case mix and antimicrobial use contribute, the impact of cross-transmission on these fluctuations is not well understood. We investigated the utility of applying a statistical algorithm to identify outbreaks of carbapenem-resistant infections across three critical care units in a multi-centre teaching hospital network serving a population of 2 million in London, UK.

Methods & Materials: We applied a negative binomial regression model which accounts for seasonality and linear trends, as described by Noufaily et al., to routinely collected microbiology data (fiscal years 2009-2015 for two units, 2012-2015 for the third) for carbapenem-resistant Pseudomonas spp. and Enterobacteriaceae (CRE). The first two years of data for each unit was used to train the algorithm. Exceedances (i.e. weeks with possible outbreaks) were validated by antibiogram comparison (as a proxy-indicator of strain similarity), against hospital infection control reports, and where available through genotypic typing.

Results: Across the three units, 154 CRE (from 3640 Enterobacteriaceae) were identified. The algorithm identified 17 exceedance weeks, in 11 multi-week clusters. In four of these clusters (three K. pneumoniae, one E. coli) organisms shared identical antibiograms; typing was available for one K. pneumoniae cluster, indicating clonal NDM cross-transmission, and this was the only outbreak (of the 11 clusters) identified in hospital infection control reports. Among 786 carbapenem-resistant Pseudomonas spp. (from 2378 isolated), 27 exceedance weeks were detected, in 15 multi-week clusters. Organisms in eight clusters shared identical antibiograms. No typing was available and none of the clusters had been identified in hospital infection control reports. No additional outbreaks of CRE or carbapenem-resistant Pseudomonas spp. were identified through routine surveillance or in hospital infection control reports.

Conclusion: The rise of carbapenem resistant organisms necessitates low-cost, easy-to-use surveillance mechanisms to aid early identification of outbreaks, particularly in critical care. Our data suggests such outbreaks may be more common than previously thought, and may be going undetected by current surveillance systems. Application of the Noufaily algorithm to routinely collected microbiology data provides a valid mechanism to better target limited hospital epidemiology, infection control, and diagnostics resources.