Identifying multivariate vulnerability of nursing home facilities throughout the southeastern United States

https://doi.org/10.1016/j.ijdrr.2019.101106Get rights and content

Abstract

To identify nursing home vulnerability attributable to location using a triangulated approach that includes historic natural hazards, community vulnerability and nursing home attributes, we use an inductive-hierarchical vulnerability index construction model. Principal components analysis (PCA) is used for two inductive models of community (CLI) and natural hazard (HLI) vulnerability. Analytical hierarchy process (AHP) is used to determine weights, according to expert ranks, for a hierarchical model of nursing home facility level vulnerability (NHLI). These three sub-indices are combined using an equal weights hierarchical approach to create a multivariate nursing home vulnerability index (MNHVI). Hazard level vulnerability is predominantly attributable to storm surge, minor hurricanes, and inland flooding. Drivers of community level vulnerability were found to be poverty and minority population, age, income and housing, Hispanic population, family status, employment type and female gender, and nursing home population. Nursing home vulnerability is found to be higher for tracts and counties that house nursing home residents with decreased or limited mobility. The clusters throughout the study area that were identified as the most vulnerable for the MNHVI are predominantly attributable to their geographic location along the coastline. The mapped outputs can provide nursing homes with an easily distributable form of visual and quantitative information to share with emergency management agencies, family members or representatives of residents in nursing homes. This study can also assist administrators in risk assessment, development of policies and procedures, communication planning, and personnel training to comply with emergency preparedness regulations.

Introduction

Over the last decade, there has been an increase in the number of billion-dollar weather disasters. In 2017, NOAA reported fifteen different billion-dollar natural disasters in the United States alone. This increase has resulted in a shift in disaster preparation methods, mitigation strategies, and emergency response programs [1,2]. Presently, healthcare facilities are required to be involved in collaborative preparation with their community partners using an all-hazards approach [3] and infrastructure situated in flood prone locations are required to purchase flood insurance [4]. In response to these disasters and policies, the hazards and vulnerability research fields have grown to address vulnerability to extreme weather events and disaster management policies in the U.S [[5], [6], [7], [8], [9]]. Findings from this research indicates that communities and their residents are vulnerable to a variety of natural hazards and some populations, including older adults, experience more harm than their younger peers [6,10,11]. Previous research has shown the need to geographically identify medically vulnerable older adult populations [12]. Nursing homes, and the older adult residents within them, are considered medically vulnerable and therefore more susceptible to the impacts of natural hazards [[13], [14], [15]]. In a study of post-Katrina harm, 30 days post-Katrina, nursing home residents experienced an additional 277 deaths and 872 hospitalizations. At 90-days, 579 deaths and 544 additional hospitalizations were observed in this demographic [16]. Other studies found that almost one half of the deaths following Hurricane Katrina were adults aged 75 and older [10,13], and 12% of the fatalities from Katrina and Rita combined were in nursing homes [10].

Natural hazards research has been a longstanding tradition in the field of geography that crosses the social and ecological divide. This inherent interdisciplinary approach to research provides an opportunity to investigate physical processes, human populations and demographics, social-ecological vulnerability, and the spatial distributions of these phenomenon [6,7,9,[16], [17], [18], [19], [20], [21], [22]]. Research on natural hazards combine literature from a multiplicity of social and natural sciences, health and human services, public safety, public policy, and information technology [9,23]. As more disciplines become involved in the social and physical responses to natural disasters, research approaches encompass more than only theoretical models and include approaches that emphasize the importance of the intersection of geophysical conditions, social systems, and vulnerable demographics such as nursing home residents and older adults [5,24,25].

Index creation studies have been used to identify vulnerability of various sorts critical, monetary, social, ecological, institutional, infrastructural, individual, and of communities [22,23]. Previous research has focused on the identification of a social systems vulnerability to specific natural hazards such as, hurricanes [6], flooding [17,22], and wildfire [26]. Considering vulnerability to specific hazards allows a more detailed analysis including specific caveats of that hazard. This is appropriate in some situations, but in other scenarios an understanding of vulnerability through an all hazards approach is necessary. Consequently, other studies have taken a multi-hazard approach to identifying vulnerability of social systems to multiple climatic and socially-sensitive hazards [17,19,[27], [28], [29]]; Shirley et al., 2012. The seminal work of Cutter et al. (2012) established the Social Vulnerability Index (SoVI), which has since become the most well-known index for vulnerability assessment at the sub-national level. The ability to quantitatively assign vulnerability measures make the SoVI a relatively simple method of visually and numerically conveying the complex underlying processes [30]. Due to higher concentrations of citizens living in areas considered at-risk for hazards (i.e., coastal and flood-prone riverine areas), understanding the social characteristics of these populations is becoming increasingly important for disaster risk management [7,17]. The SoVI uses the characteristics of social groups within a region to quantitatively determine their potential hazard vulnerability, preparedness, response, and recovery at a specific point in time [19].

Previous studies have identified older adults as being more vulnerable to natural disasters for many reasons, including post-disaster psychological stress, inability to comply with evacuation procedures, decreased cognitive abilities, limitations of mobility, vision/hearing impairments, and fewer economic resources, which can reduce willingness or ability to evacuate [7,16,31]. These frailties associated with physical and psychological impairments have been noted to increase the probability of death of nursing home residents during an evacuation. Post-hazard hospitalization and mortality can be observed with a lag period due to the increased physical and psychological vulnerabilities directly related to evacuation and indirectly related to the hazard occurrence [16]. Despite the risk of evacuating this demographic, post Hurricane Katrina public policy, requires evacuation for these at-risk facilities. Few studies, however, have examined the spatial and institutional vulnerability of nursing homes in relation to social and natural hazard vulnerability, which is an important first step in allocating resources and increasing public awareness. This study identifies the vulnerability of each nursing home according to its spatial location in reference to historic natural hazard occurrences, surrounding community characteristics, facility demographics, staffing, and resident quality indicators aggregated to the facility, census tract, and county levels.

To accurately conceptualize vulnerability according to natural hazard frequency, community characteristics, and nursing home data; three conceptual frameworks outlined by Fussel [56] were considered. Initially, a deterministic conceptualization was used for a Hazards Level Index containing variables that consider the frequency of natural disasters for census units since natural hazards cannot be avoided by human development and progress and are therefore unavoidable. A socio-ecological framework was employed for a Community Level Index, which contains United States Census variables at two scales of analysis (i.e. County and Census Tract) and the Nursing Home Level Index, which considers variables relating nursing home residents and facilities. A socio-ecological framework was employed for the CLI and NHLI. The socio-ecological concept was most appropriate for these indices since they both consider human behavior, perception, and physical/social conditions. The theory behind the MNHVI considers a mechanistic approach to vulnerability research where the implementation of technological advancements is believed to assist in the reduction of vulnerability. The only conceptual approach outlined in Fussel [56] which is not considered in this article is the political/economical approach, known as the structural theory. This structural theory considers the ideology that political structure influences vulnerability more than nature, technology, or society. The researchers determined this approach to be outside the scope of this analysis and instead emphasized the deterministic, the socio-ecological, and the mechanistic frameworks.

The study area is composed of ten states within the southeastern United States: Alabama, Florida, Georgia, Kentucky, Mississippi, North Carolina, South Carolina, Tennessee, Virginia, and West Virginia. The study area was selected due to the proximity to the gulf and atlantic coasts, where tropical cyclone and other extreme weather events are common (e.g., tornadoes, flash flooding, etc.).

Data were collected at the nursing home facility level (n = 2824) and census tract level (n = 16,284) to assess fine-scale patterns of vulnerability. Map outputs were aggregated to the county level (n = 924) to aid visual interpretation when necessary.

Previous vulnerability index research uses one of three structural approaches deductive, hierarchical, and inductive to quantitatively identify vulnerability [30]. The inductive approach has historically been the most commonly used (e.g. Refs. [12,32,33]; and was employed for the Community Level Index (CLI), the Hazard Level Index (HLI), and as a baseline comparison for the Nursing Home Level Index (NHLI).

The NHLI measured the inherent vulnerability of the institutionalized population and the resources available to support the residents (i.e., staffing and organization type) at the facility and census tract levels. The HLI, CLI, and NHLI, were created using 15 natural hazard variables for the HLI (Table 1), 23 socio-economic variables (Table 1), as recommended by the Cutters SoVI model, for the CLI, and 20 variables for the NHLI (Table 1). Data sources for the HLI include the Homeland Infrastructure Foundation Level Database (HIFLD), the National Oceanic and Atmospheric Administration (NOAA), and the Federal Emergency Management Agency (FEMA). For HLI data, which were not initially aggregated to the county or tract levels, a spatial join calculated the frequency of events (i.e., number of tornadoes experienced by each census tract) for each administrative unit. The SLOSH model, which is represented as a raster data set, was converted to vector data by calculating percentage of the area the census unit (tract or county) intersecting with the SLOSH polygons. All spatial analysis was conducted in ArcMap 10.6 [34].

Data sources for the CLI included the U.S. Census 2015 American Community Survey (ACS) 5-year estimates and the 2010 U S. Census, and variables were selected according to the SoVI model [35]. NHLI variables were selected from the Centers for Medicare & Medicaid Services, Nursing Home Compare Minimum Data Set (MDS), which provides an assessment of functional, emotional, cognitive, and disease status for all long-term residents (i.e. residents who have stayed in the nursing home for 100 days or more) within each institution. As well as the public use staffing files and type of ownership (i.e., Staffing and Organization Medicare & Medicaid Services, Nursing Home Compare data sets). Appropriate vaccinations denote the facility includes preventive measures in the care provided as an indication of the quality of care. Staffing hours and organization type have been associated with quality of care and resident outcomes (ie. lower staffing hours and for-profit ownership have been linked to poorer quality of care) [[36], [37], [38]]. Staffing hours available in the CMS public use data are registered nurses (RN) licensed practice nurses (LPN) and certified nursing assistants (CNA); physician extenders are not in this data set.

All variables for each subindex were analyzed using Principal Components Analysis (PCA) to reduce multicollinearity within the data. Data were normalized using z-score standardization and Pearson's correlation was used to assess correlation between variables. A KMO & Bartlett's test of sphericity was implemented to quantitatively establish which variables were suitable for use within a component [39]. Upon completing the Pearson's correlation, factor analysis allowed examination into which variables, and subsequent factor components, were responsible for the largest proportion of variance.

To assist in interpreting variable impact for component selection, a varimax rotation was implemented to assess which variable explains the highest portion of variance within each component [33]. Components were selected according to the Kaiser criterion rather than parallel analysis, which removed components related to nursing homes [33,40].

Analytical hierarchy process (AHP) was used for the NHLI to address concerns pertaining to the validity of factor selection from the NHLI PCA [[41], [42], [43]]. AHP allowed for an expert choice weighting scheme to be created for the 20 individual NHLI variables that could be combined in a hierarchical design. A convenience sample of experts who have/had worked in long-term care facilities in either research or occupational capacities ranked the nursing home variables (e.g., administrator, regional vice president).

All individuals hold/held positions of leadership in this field. There were five total experts chosen for this portion of the analysis. Each expert was emailed the list of variables with the instructions to rank them on their criticality (9 being most critical, 1 being least critical) during a disaster and the facility's evacuation or shelter in place. The results from the survey were averaged to determine the importance of each variable and a pairwise comparison matrix was used to evaluate each variable compared to one another (Saaty 1980). The largest eigenvalue in the comparison matrix is isolated and placed into the Consistency Index (CI) formula where P = the largest eigenvalue and n = the size of the matrix:CI = (P-n) / (n - 1)

The CI value is then compared to the Random Consistency Index (RI) value which is given by Saaty (1980) to determine if the weights calculated are appropriate. This process is completed by calculating the Consistency Ratio (CR):CR = CI / RI

If the CR value is found to be < 0.1 then the variable weights which were calculated from the comparison matrix are considered appropriate and can be confidently implemented. Results from both NHLI outputs (PCA and AHP) were compared to the existing literature on organizational theory [44,45] to determine which output was more appropriate to be retained for the final index.

The resultant indices created from the two inductive designs (e.g., CLI, HLI) were combined with the hierarchical AHP index (NHLI) using a equal weights hierarchical approach to create a final multivariate nursing home vulnerability index (MNHVI) (Fig. 1). All index scores were standardized before combination for the MNHVI. The hierarchical approach provides well-defined theoretical organization and reduced statistical complexity compared to the inductive approach [5,46]. The three indices were weighted according to the recommended equal weights aggregation scheme [30,33]. This inductive-hierarchical approach to a vulnerability assessment allows for a statistically robust factor selection through the inductive approach while retaining an organized theoretical design implicit in the hierarchical approach. The multivariate nursing home vulnerability index displays, both visually and quantitatively, the locations found to be most at-risk to natural disasters, socioeconomic conditions, and institution level factors. To determine the correlation and clustering of each value with itself across the study area, Global Moran's I and Anselin Local Moran's I were employed to test for spatial autocorrelation.

Section snippets

Hazard Level Index (HLI)

The HLI included 15 natural hazard variables and after the PCA yielded 2 components that collectively explained 65.06% of the total variance (Table 2). The two components were named according to the variables with the highest loading values within each factor. Component 1 was titled Storm Surge (43.77%) and component 2 was titled Minor Hurricanes and Flooding (21.29%).

As expected, the majority of the Very High and High vulnerable tracts are along the coast where potential for hurricanes and

Discussion

This study identifies nursing home vulnerability attributable to location using a triangulated approach that included natural hazards, community vulnerability and nursing home attributes. Vulnerability reduction is a core element of managing disaster risk and has been identified as the most important prerequisite for minimizing the destruction of physical structures and harm to their inhabitants [47,48]. Vulnerability within the U.S. disproportionately falls on older adults, who often

Acknowledgments

The researchers thank Lauren Anderson who contributed to data collection and statistical analysis. This project was made possible by the University Research Council and the Sustaining Collaborative Opportunities for Research and Education (SCORE) initiative.

References (56)

  • J. Chakraborty et al.

    Population evacuation: assessing spatial variability in geophysical risk and social vulnerability to natural hazards

    Nat. Hazards Rev.

    (2005)
  • S.L. Cutter et al.

    Moral hazard, social catastrophe: the changing face of vulnerability along the hurricane coasts

    Ann. Am. Acad. Pol. Soc. Sci.

    (2006)
  • S.L. Cutter et al.

    Temporal and spatial changes in social vulnerability to natural hazards

    Proc. Natl. Acad. Sci. Unit. States Am.

    (2008)
  • B.E. Flanagan et al.

    A social vulnerability index for disaster management

    J. Homel. Secur. Emerg. Manag.

    (2011)
  • J. Brunkard et al.

    Hurricane katrina deaths, Louisiana, 2005

    Disaster Med. Public Health Prep.

    (2008)
  • D. Dosa et al.

    Effects of hurricane katrina on nursing facility resident mortality, hospitalization, and functional decline

    Disaster Med. Public Health Prep.

    (2010)
  • E. Hames et al.

    A gis approach to identifying socially and medically vulnerable older adult populations in south Florida

    Gerontol.

    (2016)
  • L. Brown et al.

    Issues in mental health for older adults during disasters

    Generations

    (2007)
  • S.B. Laditka et al.

    Protecting nursing home residents during emergencies or disasters: an exploratory study from South Carolina

    Prehospital Disaster Med.

    (2007)
  • S.B. Laditka et al.

    Providing shelter to nursing home evacuees in disasters: lessons from hurricane katrina

    Am. J. Public Health

    (2008)
  • D. Dosa et al.

    To evacuate or shelter in place: implications of universal hurricane evacuation policies on nursing home residents

    J. Am. Med. Dir. Assoc.

    (2012)
  • W.N. Adger et al.

    Social-ecological resilience to coastal disasters

    Science

    (2005)
  • M.K. Andrew et al.

    Social vulnerability, frailty and mortality in elderly people

    PLoS One

    (2008)
  • C.T. Emrich et al.

    Social vulnerability to climate sensitive hazards in the southern United States

    Weather, Climate, and Society

    (2011)
  • J. Evans

    Mapping the vulnerability of older persons to disasters

    Int. J. Older People Nurs.

    (2010)
  • J. Perdikaris et al.

    A methodology for undertaking vulnerability assessments of flood susceptible communities

    International journal of safety and security engineering

    (2011)
  • G.A. Tobin et al.

    Natural hazards and technology: vulnerability, risk, and community response in hazardous environments

  • K. Hyer et al.

    Establishing and refining hurricane response systems for long-term care facilities

    Health Aff.

    (2006)
  • Cited by (5)

    • Mapping community-level determinants of COVID-19 transmission in nursing homes: A multi-scale approach

      2021, Science of the Total Environment
      Citation Excerpt :

      The population density was calculated per square mile and used as a proxy for rurality from this data using the 2010 Census. These variables were selected based on a prior literature on the underlying social vulnerability in a community setting (e.g., SoVI, Cutter et al., 2003) and the potential impact to nursing home populations in the context of a disaster (Wilson et al., 2019; Wilson et al., 2020). Exploratory spatial data analysis examined spatial autocorrelation of COVID-19 cases.

    View full text