The quality of demographic variables, especially race and ethnicity, has suffered from inconsistencies and challenges in data collection for all types of data, not just health data. The same conditions that compromise disparities data in general, compromise disparities research on groups with MCC. Currently national surveys and databases lack standardization among the demographic variables collected, observer bias and inadequate and insensitive response categories can prevent minority populations from being accurately represented in data capture efforts. Analytical challenges also complicate disparities research in general (and therefore MCC research.) The challenges are described below.
Fortunately, as discussed later, a broad range of efforts are being put into place to standardize and improve data collection methods, and improve the overall quality of demographic data. The Affordable Care Act, for instance, called for the creation and use of uniform demographic variables in national surveys. While improved data collection methodologies will help researchers create a more accurate picture of the health challenges facing specific racial and ethnic groups in our nation, it is important to note the potential risks of improving coding of small subgroups of the population, and to ensure that as the methods for identifying and analyzing ever smaller populations improves, safeguards will be put in place to preserve the privacy of these individuals and shield them from potential discrimination.
5.1 Quality of Race and Ethnicity Variables
Accuracy and completeness of demographic information is a concern in studying disparities. Race and ethnicity variables, in particular, have suffered from inconsistent measurement over time, evolving definitions and categories, insufficiently sensitive categories, and a variety of data collection challenges. The first U.S. census in 1790 recognized three racial categories: whites, blacks (as three fifths a person) and Indians who paid taxes; an unbalanced and racially motivated classification scheme (Williams, 1999). Within the past decade, the Office of Management and Budget (OMB) has approved the use of increasing numbers of racial and ethnic categories up to the current standard of 14 racial and 5 ethnic categories for use in federal data collection initiatives (Cunningham 2012). Federal efforts to collect disparities data are also hindered by non-uniform data collection practices across states. Medicaid in particular lacks federal disparities data collection standards, resulting in a large range between states in the type and quality of disparities data collected. Even within individual states the use of different healthcare provider organizations leads to further variability in the disparities data that is collected (Byrd & Verdier, 2011).
The quality of race and ethnicity variables is a limitation of most federal and private databases. For example, the Medicare enrollment database (EDB) at CMS contains race/ethnicity variables that are highly specific (low false positive rate), but insensitive (low true positive rate) for categories other than white or black. In other words, race/ethnicity coding for white and black beneficiaries is considerably more accurate than other minority groups, such as Asian or American Indians (Waldo, 2005). The Hispanic ethnicity code in the EDB captures only one third of beneficiaries who identify as Hispanic, leading to significant underestimation. Overall, minority populations are more likely to be missing race/ethnicity information or have misclassified information, and those minorities who are misclassified are most often misclassified as white (Waldo, 2005; Williams, 1999). Other examples of databases that suffer from inadequate race/ethnicity coding include the National Ambulatory Medical Care Survey and Healthcare Cost & Utilization Project - Nationwide Inpatient Sample.
5.2 Analytical Challenges in Assessing Disparities
Comparing data across studies to look at trends can be thwarted by different aggregation schemes. For example, one study may examine prevalence by gender, race and age, while another looks at prevalence by age and race; making it difficult to interpret results. In addition, studies often use different definitions for variables. For example, researchers use different “cut offs” for age (<65 or >65...or 50–60, 60–70, etc.).
Researchers are only beginning to develop quality measures intended for disparities research. Weissman et. al. (2011) released a report outlining recommendations for the development of quality measures to monitor potential healthcare disparities from the National Quality Forum’s (NQF) 700 available quality measures. The report recommended a three-step process for identifying disparities-sensitive quality measures: 1) Assess the NQF’s quality measures using disparities-sensitive principles, 2) Apply new criteria for disparities sensitivity for quality measures that do not stratify data by race/ethnicity, or other disparities variables, and 3) develop new disparities specific measures (pg. 7).
There are challenges in obtaining state and local data to for intervention research at the local level, as well.