Statistical Approaches to Assessing Health and Healthcare Disparities

Abstract

Determining whether racial and ethnic disparities exist for a health-related outcome requires first specifying how outcomes will be measured and disparities calculated. We explain and contrast two common approaches for quantifying racial/ethnic disparities in health, with an applied example from nursing research. Data from a national for-profit chain of nursing homes in the US were analyzed to estimate racial/ethnic disparities in incidence of pressure ulcer within 90 days of nursing home admission. Two approaches were used and then compared: logistic regression and Peters-Belson. Advantages and disadvantages of each approach are given. Logistic regression can be used to quantify disparities as the odds of the outcome for one group relative to another. Peters-Belson can be used to quantify an overall disparity between groups as a risk difference and also provides the proportion of that disparity that is explained by available risk factors. Extensions to continuous outcomes, to survival outcomes, and to clustered data are outlined. Both logistic regression and Peters-Belson are easily implementable and interpretable and provide information on the predictors associated with the outcome. These disparity estimation methods have different interpretations, assumptions, strengths, and weaknesses, of which the researcher should be aware when planning an analytic approach.

Racial and ethnic disparities in the healing of pressure ulcers present at nursing home admission

“…Health disparities were analyzed using the Peters-Belson method (Eberly et al, 2015; Graubard, Sowmya Rao, & Gastwirth, 2005; Rao, Graubard, Breen, & Gastwirth, 2004). The Peters-Belson method tests whether observed outcomes of a disadvantaged group (presumed to be the various groups of non-White minority NH admissions in this study) differ from their expected outcomes.…”

“…When the characteristics of a minority group are used in the model developed for Whites to predict the healing of pressure ulcers, it is as if we are examining a hypothetical population of Whites with the same clinical and care characteristics as the minority residents. The Peters-Belson method used in this analysis enables a quantification of the percentage of the observed disparity that is explained and unexplained by the available predictors (Eberly et al, 2015; Graubard et al, 2005; Rao et al, 2004). The explained disparity estimates how close the estimated outcomes of those hypothetical Whites would be to the observed White outcomes.…”

Prevention of Incontinence-Associated Skin Damage in Nursing Homes

“…The practitioner orders provide a unique resource enabling investigation of the understudied issues of disparities and predictors of preventing IASD in nursing homes. The Peters-Belson approach is an innovative method for analyzing racial-ethnic disparities, and its advantages have been reviewed elsewhere (Eberly et al, 2015). …”

“…The regression coefficients from the model for Whites were then applied to each racial or ethnic minority resident group separately in other logistic regression models. The expected and observed proportions of residents receiving IASD prevention for each minority group were then compared using a one-sample two-sided log-rank test (Eberly et al, 2015). A significant difference represented the disparity in IASD prevention unexplained by predictors in the model suggesting disparity due to race or ethnicity (Eberly et al, 2015; Graubard et al, 2005).…”

“…The expected and observed proportions of residents receiving IASD prevention for each minority group were then compared using a one-sample two-sided log-rank test (Eberly et al, 2015). A significant difference represented the disparity in IASD prevention unexplained by predictors in the model suggesting disparity due to race or ethnicity (Eberly et al, 2015; Graubard et al, 2005). As residents were clustered within nursing homes, the nursing homes of Whites being modeled included the minority group of interest to control for unmeasured nursing home confounders and heterogeneity in variability.…”