Table 2.
Detailed Summary of the Measurement of Segregation via Formal Measures (Massey & Denton, 1988) Included in This Systematic Review
| Data sources† | |||||||
|---|---|---|---|---|---|---|---|
| Reference | Measure of segregation | Dimension of segregation* | Resident | Facility | Regional | Macro unit of analysis‡ | Description of segregation measure |
| Chang and colleagues (2012) | Dissimilarity index (DI; Massey & Denton, 1988) and Disparities quality index (DQI; Siegel et al., 2009) | Unevenness | Resident-level quality measure dataset provided by CMS; Healthcare Center (HD QIOSC) - MDS, Medicare enrollment data | NH Compare website | U.S. Census Data, CMS dataset (urban/rural) | MSA, rural/urban | The DI was selected to represent segregation via dissimilarity, while the DQI was selected to measure the extent of disparities in quality of care among a sub-population (in this case black residents) compared with the total population (all residents). |
| Davis and colleagues (2014) | Modified Thiel’s entropy index, Dissimilarity Index (DI) | Unevenness | MDS | OSCAR/ CASPER | Area Resource File (drawn from U.S. Census Data), Residential segregation data | Degree of residential segregation (DI via MSA bottom 25%, middle 50%, top 25%) | Thiel’s entropy index was used a dependent variable to represent NH racial/ethnic diversity in ordinal least squares regression modeling, comparing each ethnic group in an MSA vs in NHs. The DI (independent variable) was used to represent residential segregation within MSAs, while the Herfindahl and Hirshman Index (independent variable) was used to quantify market concentration in the model. |
| Feng, Lepore and colleagues (2011) | Gini Coefficient, Geographic Information System (GIS) | Unevenness (Gini Coefficient), clustering (GIS) | N/A | OSCAR/ CASPER | U.S. Census Data | State, MSA, zip codes | The Gini coefficient (ranging from low (0) – high (1) inequality, weighted by the total number of NHs to ever exist) was used to represent racial/ethnic inequalities in concentration of NH closures by facility-matched regional characteristics (proportion minority and poverty rate). GIS was used to illustrate clustering of NH closures across zip codes. |
| Miller and colleagues (2006) | Multilevel modeling | Racial composition (proxy measure), clustering | MDS, CMS claims data | OSCAR/ CASPER, NHCompare | Area Resource File (U.S. Census Data) | County | A multilevel modeling approach was used to assess the effect of the percentage of blacks in NHs (and in the respective NH facility counties) on use of physical restraint and antipsychotic drug use. Beta-binomial distribution (via William’s method) was used to calculate the facility-level quality indicator (QI) outcomes dependent on respective county. Stratified by race (black/white), GEE logistic regression models were used to calculate the effect of QIs given the racial composition of NHs and the county of the NH. An overdispersion parameter was used to address within facility correlation with individual outcomes and an independence correlation matrix was used with county clustering to address within county correlation in facility-level outcomes. |
| Smith and colleagues (2007) | Dissimilarity Index (DI) | Unevenness | MDS | OSCAR/ CASPER | U.S. Census data (MSA boundaries), MDS | MSA | The DI was used to measure racial segregation by indicating the proportion of residents needed to make a facility have an equal percentage of black and white residents. Measures of quality were calculated relative to the MSA region. |
| Strully (2011) | Dissimilarty Index (DI), racial composition | Unevenness, Racial composition (proxy measure) | U.S. National NH Survey | U.S. National NH Survey | U.S. National NH Survey | NH-level | The DI was calculated with both resident- and NH- level variables including racial composition (%black residents/ NH) to quantify the level of overall racial segregation in the sample. Logistic regression models with receipt of influenza vaccination as the outcome were run starting with individual level covariates and adding the percentage of black resident variables and then adding other facility characteristics (NH bed capacity, location, ownership). |
| Troyer and McAuley (2006) | Probit model | Unevenness | MEPS-NHC; MDS supplemented with medical record information | MEPS-NHC | Area Resource File | County | Probit model estimates were used to evaluate the role of characteristics relating to race on the gap between the percentage of NH blacks and whites with documentation of advanced directives. Probability (via marginal effect) of documentation by race was estimated with individual-, facility-, and regional-level characteristics. Characteristics included facility-level metropolitan or non-metropolitan locale and county-level racial composition, poverty level, proportion aged ≥65 years etc. The difference of two subsample models (for black and white residents) were broken down to determine the amount of the gap attributable to each specific characteristic. |
| Rahman and Foster (2015) | Multidimensional spatial segregation model | Unevenness, clustering | Medicare enrollment records; Medicaid Analytical eXtract (MAX); Medicare Part A Claims; MDS; choice sets (created by authors) | OSCAR/ CASPER, American Hospital Association data | U.S. Census Data | zip codes | A multidimensional spatial modeling approach was used to evaluate the role in race-based vs distance-based NH sorting, by taking into account residential racial composition and NH quality with weights for preferences of certain NH characteristics. The Haversine formula was used to calculate distance variables. Primary parameters of interest were racial composition (race-based sorting) and distance parameters (distance-based sorting) using zip codes of pre-NH admission residences and zip codes surrounding respective NHs. The importance of each parameter in the relationship between segregation and facility quality of care was tested using choice-based simulations and counterfactual modeling. Specifically, instrumental variables were used to account for the racial differences in NH preferences. |
Notes: NH = nursing home.
*Segregation dimensions based on (Massey and Denton 1988) include unevenness, exposure/isolation, concentration, clustering, centralization.
†Data Source abbreviations: CMS = Center for Medicare/ Medicaid, MDS = Minimum Data Set, OSCAR = Online Survey, Certification and Reporting, replaced in 2012 by CASPER = Certification and Survey Provider Enhanced Reporting, Residential segregation data from the Lewis Mumford Center for Comparative Urban and Regional Research from the State University of New York at Albany, MEPS-NHC = Medical Expenditure Panel Survey NH Component.
‡Macro unit of analysis abbreviations: MSA = metropolitan statistical area.