Introduction
Carbapenem-resistant enterobacterales (CRE) are an urgent antibiotic resistance threat to public health. Patients at risk for these infections typically have healthcare exposures and/or exposure to antibiotics. Reference Guh, Bulens and Mu1 CRE can be transmitted patient-to-patient in healthcare environments, particularly in settings with insufficient infection control and hand hygiene practices. To date, there has been minimal evaluation of how social determinants of health (SDOH) impact risk for healthcare-associated infections such as CRE.
The Social Vulnerability Index (SVI) is a composite metric developed by the Centers for Disease Control and Prevention and the Agency for Toxic Substances and Disease Registry (CDC/ATSDR) and uses 16 U.S. census variables across four distinct themes, including socioeconomic status, household characteristics, racial and ethnic minority status, and housing type and transportation. 2 There is growing literature to suggest that SVI is associated with disparities in antimicrobial resistant infections. Reference Butler, Hranac and Johnston3,Reference Czaja, Beck and Orten4
A better understanding of the relationship between SVI and healthcare-associated infections such as CRE will allow for improved public health prevention by highlighting particular census tracts at risk and where resources are most needed to support health equity. We investigated census tract-level SVI and rates of CRE in the Denver metropolitan area.
Methods
We conducted an ecologic study of the approximately 2.8 million residents of the Denver Metropolitan area (Adams, Arapahoe, Denver, Douglas, and Jefferson Counties) in Colorado from 2016 to 2019. The outcome of interest was an incident case of CRE captured by population-based surveillance of the Emerging Infections Program Healthcare-Associated Infections Community Interface in Colorado. 5 A case of CRE was defined as carbapenem-resistant Escherichia coli, Enterobacter cloacae complex, or Klebsiella species isolated from a normally sterile site or urine from a resident of the surveillance area. Isolates had to be resistant to at least one carbapenem (imipenem, doripenem, meropenem, ertapenem). Incident infections were defined as the first case of each organism per patient in a 30-day period. 5 Laboratories reported CRE to the Colorado Department of Public Health and Environment according to statewide disease reporting requirements. Trained epidemiologists reviewed medical records of patients meeting the case definition. Patient residential addresses for cases were geocoded to 2020 census boundaries and counts aggregated by census tract. Cases were classified as hospital onset, community-onset healthcare-associated, and community-onset community-associated. Reference Grome, Grass and Duffy6
The exposure of interest was census tract SVI from 2020. Age and sex were covariates for standardization. Population denominators were taken from the 2016–2020 5-year American Community Survey and multiplied by four to estimate person-years. Reference Walker and Herman7
The unit of analysis was the census tract. We stratified census tract case counts by sex and age (0–19, 20–39, 40–59, and 60+ years) and calculated standardized morbidity ratios (SMRs) using the Denver metropolitan area population as the reference. A Moran’s I statistic indicated spatial autocorrelation of census tract rates. We, therefore, used Bayesian conditional autoregressive (CAR) models to conduct disease mapping and to estimate the association between SVI and CRE. Reference Butler, Hranac and Johnston3,Reference Czaja, Beck and Orten4,Reference Czaja, Cockburn and Colborn8 These methods address biases associated with variability in census tract boundaries, small denominators within census tracts, and spatial autocorrelation. We used R statistical software v.2024.04.2 (R Core Team, 2024), SAS On Demand (SAS Institute Inc., 2023), and OpenBUGS v.3.2.3 rev 1012 for statistical analysis. We used TIGER/Line Shapefiles for 2020 for mapping. 9
Results
There were 664 census tracts for the five-county Denver metropolitan area, with a total of 11,195,108 person-years. A total of 659 census tracts (99.2%) had a non-zero population and were included in mapping and 658 (99.1%) had non-missing SVI values and were included in the CAR model.
There were 476 CRE cases from 2016 to 2019. Of these, 476 (100%) had documented sex and age available and 474 (99.6%) were geocoded. There were more cases in females than males (n = 305, 64.1% vs n = 171, 35.9%). Twenty-one cases were in persons aged 0–19 years (4.4%), 38 cases were in persons aged 20–39 years (8.0%), 80 cases were in persons aged 40–59 years (16.8%), and 337 cases were in persons over 60 years of age (70.8%). The crude rate of CRE was 4.23 cases per 100,000 person-years. Among all CRE cases, 61 were considered hospital-onset (12.8%), 285 were community-onset healthcare-associated (59.9%) and 130 were community-onset community-associated (27.3%).
The median standardized morbidity ratio (SMR) for 659 census tracts was 0 with a range of 0 to 11.55. There was significant spatial autocorrelation of SMRs among census tracts (Moran’s I = 0.06, P = .002). A map of smoothed rates of CRE infection demonstrated spatial clustering of census tracts with elevated rates (Figure 1). The median census tract overall SVI was 0.36 (range 0 to 0.99; n = 6 missing). For every 0.1 unit increase in census tract overall SVI, the rate of CRE increased by 10% (RR 1.10, 95% CI 1.06–1.14).
Social vulnerability index (A) and smoothed sex- and age-adjusted rate ratios for carbapenem-resistant Enterobacterales (B) in census tracts relative to the five-county catchment area—Adams, Arapahoe, Denver, Douglas, and Jefferson Counties, Colorado, 2016–2019.
Figure 1. Long description
Two maps display social vulnerability index and rate ratios for carbapenem-resistant Enterobacterales in census tracts relative to the five-county catchment area in Colorado, 2016–2019. The first map (A) shows the social vulnerability index (SVI) with varying shades of gray indicating different levels of vulnerability. The second map (B) illustrates the smoothed sex- and age-adjusted rate ratios for carbapenem-resistant Enterobacterales (CRE) with a gradient scale from light to dark gray representing different rate ratios. The maps highlight areas with higher social vulnerability and higher rate ratios, particularly in the central and northeastern parts of the catchment area. The maps are bordered by a thick black line, and a legend is provided to interpret the shades of gray. The maps are labeled with the title ’Social vulnerability index (A) and smoothed sex- and age-adjusted rate ratios for carbapenem-resistant Enterobacterales (B) in census tracts relative to the five-county catchment area—Adams, Arapahoe, Denver, Douglas, and Jefferson Counties, Colorado, 2016–2019.’
Discussion
We demonstrated a positive association between SVI and rates of CRE in Denver area census tracts. This finding is consistent with previous studies on the relationship between social vulnerability and other types of infections, such as candidemia, Clostridioides difficile and influenza. Reference Butler, Hranac and Johnston3,Reference Czaja, Beck and Orten4,Reference Czaja, Cockburn and Colborn8 However, the underlying mechanisms driving the relationship between SDOH factors and census tract level and healthcare-associated infections such as CRE are unclear. In the U.S., most infections caused by CRE are associated with healthcare exposures. Reference Guh, Bulens and Mu1 This study’s observed association between census tract-level social vulnerability and risk for CRE is likely a complex interplay of individual patient, healthcare facility, community, and system-level factors that interact pre- and post-hospitalization.
We hypothesize social vulnerability at the community level may influence susceptibility to CRE. For example, poverty contributes to a higher prevalence of chronic health conditions, and some underlying conditions such as chronic lung disease, metabolic disease, neurological disorders, and immunodeficiency are risk factors for CRE. Reference Bulens, Reses and Ansari10 Populations with these conditions may also be more likely to be admitted to healthcare facilities and to have healthcare exposures such as longer lengths of stay, long-term acute or nursing care, or medical device intervention such as mechanical ventilation, catheters, or other indwelling devices. Other factors may be important to consider, such as resources and characteristics of healthcare facilities serving the populations within census tracts, language barriers, and unconscious biases. While these are some potential hypotheses, further work is needed to understand the underlying pathways between SVI and risk for CRE.
There are some limitations to this analysis. First, associations were observed at the aggregate level and may not translate to individual risk (i.e., ecological fallacy). Second, varied laboratory testing practices may have led to an underestimate of the true number of infections in the catchment area. However, clinical laboratories are required to report CRE cases, and staff perform routine laboratory audits to capture missed cases. Third, we did not separately evaluate rates of community-associated CRE, though the majority of CRE cases were healthcare-associated. Strengths of our study include the use of robust population-based surveillance methods and use of SVI, which is a marker that allows us to identify health disparities and generate hypotheses.
The finding of an association between SVI and rates of CRE highlights the need to consider SDOH as root causes of healthcare-associated infections. Further work is needed to identify and evaluate mechanisms underlying the association between SVI and CRE rates. Healthcare epidemiologists and public health professionals should look beyond typical healthcare risk factors to individual, facility, and community factors that may increase risk for CRE prior to, or during, admission to a healthcare facility.
Acknowledgments
None.
Author contribution
The study was conceived of and designed by A.B. and C.A.C. A.B. completed all data analyses. J.D., K.H., and H.J. conducted data collection. All authors contributed substantially to manuscript revisions. A.B. takes responsibility for data integrity and accuracy of the data analysis.
Financial support
This work was supported by Grant/Cooperative Agreement Numbers NU50CK000483 and NU50CK000641 from the Centers for Disease Control and Prevention (CDC). A.B.’s contribution is supported by the National Institutes of Health (NIH)/National Center for Advancing Translational Sciences (NCATS) Colorado CTSA award T32 TR004367 and UM1 TR004399. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of CDC or NIH.
Competing interests
None.
