Introduction
There have been few evaluations of health disparities related to healthcare-associated infections (HAIs). Reference Chen, Khazanchi, Bearman and Marcelin1 HAIs are often attributed to exposures after admission to a healthcare setting. However, social determinants of health (SDOH) associated with location of residence may drive HAI disparities.
We evaluated disparities in rates of candida bloodstream infection (candidemia). Candidemia is a HAI estimated to affect 23,000 persons in the U.S. in 2017 and associated with a 15% seven-day all-cause mortality. Reference Tsay, Mu and Williams2 Established risk factors include age, severe illness, surgeries or immunocompromising conditions, antibiotic or steroid use, and intravascular catheters. Reference Kullberg and Arendrug3 U.S. population-based surveillance indicated higher candidemia rates among Black compared to White persons. Reference Tsay, Mu and Williams2 However, variables related to SDOH are not readily available from surveillance data. We therefore used the social vulnerability index as a measure of community SDOH. 4 Our objective was to assess an association between SDOH and candidemia rates to motivate more research into social drivers of HAI disparities and outreach to improve health equity.
Methods
We conducted an ecological study of the 664 census tracts within Adams, Arapahoe, Denver, Douglas, and Jefferson Counties, Colorado, total population 2,826,127, 5 for the years 2021 and 2022. The census tract was the unit of analysis.
The outcome of interest was census tract rate of candidemia. An incident case of candidemia was defined as the first isolation of Candida species from a blood culture within a 30-day period. Colorado Department of Public Health and Environment (CDPHE) conducts candidemia surveillance as part of the Centers for Disease Control and Prevention’s Emerging Infections Program. 6 Laboratories in the five-county Denver metropolitan area are required to report candidemia to CDPHE according to state regulation. Testing was ordered at the discretion of the healthcare provider. CDPHE geocoded cases to the census tract of residence.
The primary explanatory variable was the CDC and Agency for Toxic Substances and Disease Registry Social Vulnerability Index (SVI) from 2022 and its four themes: socioeconomic status, household characteristics, race and ethnic minority status, and housing type and transportation. 4 SVI themes are composed of 16 U.S. census variables from the 5-year American Community Survey (ACS). SVI is measured on a scale of 0–1, with 1 being the most vulnerable. While SVI was developed for public health emergency planning, it contains variables known to be associated with health disparities related to communicable diseases, including HAI. Reference Butler, Hranac and Johnston7
We took population denominators for the catchment area and its census tracts from the 2018–2022 American Community Survey. Census tract boundaries were taken from the 2020 U.S. Census TIGER/Line files. 5
Statistical analyses included smoothing methods that address rate instability due to small numbers within areal units and spatial autocorrelation, with age- and sex- adjustment to address population variation. First, we counted cases by sex and age (0–19, 20–39, 40–59, or ≥60 years) for census tracts and the overall catchment area and calculated sex- and age-adjusted standardized morbidity ratios (SMRs) for each census tract, using the catchment area as the reference. Global spatial autocorrelation of census tract candidemia rates was confirmed using Moran’s I statistic (P = 0.01). Reference Elliott, Wakefield, Best and Briggs8 We therefore used Bayesian conditional autoregressive models to map smoothed rate ratios (RRs) of candidemia relative to the five-county catchment area rate (essentially a smoothed SMR) and to estimate the RR and 95% credible intervals (CIs) associated with a 0.1 unit increase in SVI and each SVI theme. Reference Elliott, Wakefield, Best and Briggs8 Analyses were conducted using SAS 9.4, RStudio 2023.06.0, and OpenBUGS 3.2.3 software. The project was consistent with existing protocols for public health surveillance under our disease control authority and did not require institutional review board approval.
Results
Of 664 catchment area census tracts, 658 (99%) had a non-zero population and were included in disease mapping (median [range] pop. 4,070 [51, 15,052]). The SVI analysis included 657 (99%) census tracts with non-missing values for SVI or SVI themes other than the race and ethnicity minority status theme, which included 658 (99%).
There were 339 candidemia cases, of which 321 (95%) were geocoded and included in the analysis. Of these, 264 (82%) were hospital-onset, meaning the candidemia diagnosis was made from a sample collected greater than three days after admission, 26 (8%) were community-onset with healthcare exposures in the preceding 90 days, and 31 (10%) were community-onset with no healthcare exposures. 6 Twenty-eight (8.7%) cases were in a long-term care facility prior to hospitalization or diagnosis. Species of candida included C. albicans (128, 40%), C. glabrata (113, 35%), C. parapsilosis (38, 12%), other (37, 11.5%), and multiple species (5, 1.5%). There were no cases of C. auris. There were 178 (55%) cases in men. Counts by age strata were as follows: 11 cases in 0–19-year-olds, 40 cases in 20–39-year-olds, 89 cases in 40–59-year-olds, and 181 cases in persons 60 years and older. The median (range) census tract case counts and SMRs were 0 (0, 6) and 0 (0, 20.13), respectively. The median (range) value of SVI was 0.46 (0.00, 1.00) and the distributions of SVI themes were similar (Figure 1A).
Social vulnerability index (A) and smoothed sex- and age-adjusted rate ratios (RR) for candidemia relative to the five-county catchment area (B) in census tracts—Adams, Arapahoe, Denver, Douglas, and Jefferson Counties, Colorado, 2021–2022. The median (range) RR estimate for the 598 mapped census tracts was 0.90 (0.48, 7.92).
Disease mapping indicated clusters of census tracts with higher-than-expected rates of candidemia relative to the catchment area after adjusting for sex and age (Figure 1B). For every 0.1 unit increase in census tract SVI, there was a 15% increase in the rate of candidemia (RR 1.15, 95% CI, 1.10, 1.20) after adjusting for sex and age. There were similar modest associations between each SVI theme and census tract candidemia rates (Table 1).
Rate ratio for candidemia associated with a 0.1 unit increase in census tract social vulnerability index—Adams, Arapahoe, Denver, Douglas, and Jefferson Counties, 2021–2022
95% CI, 95% credible interval.
a Adjusted for sex and age.
Discussion
We demonstrated clustering of census tracts with higher-than-expected rates of candidemia and a modest association with SVI. Our findings add to a growing body of evidence that SDOH tied to location of residence contribute to HAIs, including candidemia. Reference Chen, Khazanchi, Bearman and Marcelin1,Reference Tsay, Mu and Williams2,Reference Butler, Hranac and Johnston7,Reference Smith, Falvey and Grace9 The association between SVI and candidemia rates raises questions about the mechanism of association.
The root causes of health disparities in candidemia are likely multifactorial. Reference Marcelin, Hicks, Evans, Wiley, Kalu and Abdul-Mutakabbir10 Possible mechanisms may include community-level, hospital-level, and individual-level factors that incur risk pre- and post-hospitalization. For example, poorer baseline health may contribute to higher rates of hospitalization, sicker patients, or patients requiring surgeries, antibiotics, or medical devices after admission. Unconscious biases of providers or patients may affect care. Reference Chen, Khazanchi, Bearman and Marcelin1 The research agenda should consider a framework that incorporates SDOH. Reference Chen, Khazanchi, Bearman and Marcelin1,Reference Smith, Falvey and Grace9 Identifying root causes of candidemia disparities will guide attempts by public health and healthcare providers to improve health equity.
There are limitations to this analysis. First, this was an ecological study by design, and the results should be interpreted at the census tract rather than individual level. Second, SVI is an index with multiple components that we used as an initial indicator of the existence of health disparities. An evaluation of mechanisms of association would require more detailed data at the individual-, healthcare facility-, and community-level. Major strengths of this analysis include the use of population-based surveillance for candidemia, disease mapping with identification of disproportionately affected census tracts, and the use of spatial statistical methods.
We found disparities in rates of candidemia associated with census tract SVI. Healthcare and public health epidemiologists should investigate the mechanisms by which SVI and component SDOHs may contribute to rate disparities to identify prevention targets.
Acknowledgements
Devra M. Barter, MSc, Colorado Department of Public Health and Environment.
Financial support
The project described was supported by Grant/Cooperative Agreement Number NU50CK000641-02-00 from the Centers for Disease Control and Prevention (CDC). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of CDC.
Competing interests
CAC, RC, and HJ report funding from the Centers for Disease Control and Prevention. AB, AGA, and RC report funding from the National Institutes of Health. AGA and RC report funding from the Colorado Department of Public Health and Environment. AGA reports funding from the Health Resources and Services Administration. AGA is a voluntary commissioner for the Denver Commission on Aging. RC reports funding from the American Diabetes Association. RC reports ownership in the American Academy of Epidemiology, Inc.
