Assessing the role of inter-facility patient transfer in the spread of carbapenemase-producing Enterobacteriaceae: the case of France between 2012 and 2015

Abstract

The spread of carbapenemase-producing Enterobacteriaceae (CPE) in healthcare settings is a major public health threat that has been associated with cross-border and local patient transfers between healthcare facilities. Since the impact of transfers on spread may vary, our study aimed to assess the contribution of a patient transfer network on CPE incidence and spread at a countrywide level, with a case study of France from 2012 to 2015. Our results suggest a transition in 2013 from a CPE epidemic sustained by internationally imported episodes to an epidemic sustained by local transmission events through patient transfers. Incident episodes tend to occur within close spatial distance of their potential infector. We also observe an increasing frequency of multiple spreading events, originating from a limited number of regional hubs. Consequently, coordinated prevention and infection control strategies should focus on transfers of carriers of CPE to reduce regional and inter-regional transmission.

Figure 1

Network of patient transfers between French counties and incident CPE episodes reported between 2012 and 2015. (A) The county network is comprised of 96 counties linked together by over 3,000 connections (grey lines). Patient transfer links were centralized to the county seat. For each county, the cumulative number of incident CPE episodes reported in healthcare facilities of the county from 2012 to 2015 is depicted by a circle. (B) A zoomed-in view of the Ile-de-France region. (C) A multi-level pie chart showing the proportion of incident CPE episodes for each year, along with their importation status and mechanisms of resistance.

Figure 2

Comparison of the distributions of network distances for observed vs. permutated data, 2012–2015. For each year, the density distribution of distances between incident episodes and their potential infector in the observed data is shown; as well as the density distribution of the mean distances computed based on 500 random permutations of the data, illustrating the distribution expected under the null hypothesis of independence between patient transfers and CPE incidence. Distances of zero correspond to linked episodes occurring in the same county. Distances of one correspond to a range of distances between zero and one; the same applies to distances of 2–20.

Figure 3

Inter and intra-county probable transmission events, 2013–2015. (A) For each county, the total number of pairwise episodes or “probable transmission events” for the 2013–2015 period were compared to the proportion of these events that occurred within the same county. A simple linear regression model with 95% confidence intervals is shown. Colors represent individual counties from largest number of total events (purple) to least (yellow). (B) The Senkey diagram shows the number of linked events between the counties of the potential infector and the counties of the incident episodes. To reduce noise, we limited our observations to only the inter-county links of ten counties with the highest number of total probable transmission events.

Figure 4

Multiple spreading events and number of potentially associated secondary incident episodes, 2013–2015. (A) The county network mapping the size or number of linked secondary events to one potential infector episode from 2013 to 2015. (B) A zoomed-in view of Ile-de-France counties including Paris (purple). (C) The distribution of the total number of transmission events and the size of the multiple spreading events with proportions per year.

Figure 5

Sensitivity analysis on the impact of the time window chosen to look for candidate transmitters: mean network distances between incident episodes and their closest potential infectors obtained for sliding 1-week time windows, 2012–2015. For each year, the mean network distance is plotted as a function of the first day of the 1-week time window, Window_n, for observed data and permuted data. For the permutations, 95% bands are also provided.