D3

Data Sources and Data Collection for Social Risk Factors

In its first report, Accounting for Social Risk Factors in Medicare Payment: Identifying Social Risk Factors (NASEM, 2016a), the committee presented a conceptual framework illustrating the primary hypothesized pathways by which five social risk factors—socioeconomic position (SEP); race, ethnicity, and cultural context; gender; social relationships; and residential and community context—and health literacy may influence health outcomes of Medicare beneficiaries (NASEM, 2016a). In its third report, Accounting for Social Risk Factors in Medicare Payment: Criteria, Factors, and Methods (NASEM, 2016b) (see Appendix C), the committee expanded the framework to include specific indicators, or ways to measure, the social risk factors. The committee also developed five criteria for selecting social risk factors that could be accounted for in Medicare quality measurement and payment programs and applied them to the social risk factor indicators. Based on this activity, the committee concluded that the following indicators could be included in Medicare quality measurement and payment programs in either the short or long term:

• income
• wealth
• education
• dual eligibility
• race and ethnicity
• language
• nativity
• acculturation

• sexual orientation and gender identity
• marital/partnership status
• living alone
• social support
• neighborhood deprivation
• housing stability and quality
• urbanicity
• other environmental measures of residential and community context

For each social risk factor, the committee identified data sources in the categories described in Chapter 2—new and existing sources of Centers for Medicare & Medicaid Services (CMS) data, data sources from providers and health plans, and alternative government data sources—with the aim to be more inclusive. The committee’s review of data sources considered sources that CMS could use in the short and long term. The committee notes that it has not been asked to recommend whether the CMS should include social risk factor adjustments in its public reporting and payment programs. The recommendations in this report indicate things CMS should do if it decides to move toward accounting for social risk factors.

To assess the advantages and disadvantages of specific data sources for specific social risk factor indicators, the committee identified three characteristics to consider: (1) collection burden, (2) accuracy, and (3) clinical utility. Collection burden describes the resources including clinician and administrative time, financial costs, and other effort required to collect and store data through any given source. This burden can be carried by individual patients or enrollees responding to questions about their social risk, as well as providers (including organizations, individual providers, and nonclinical staff) who collect data, and CMS itself. When considering collection burden, particularly where there are substantial barriers to data collection (such as high cost), CMS may weigh an important tradeoff to further guide its selection of any given indicator or social risk factor. In some cases, data collection may be burdensome, but the indicator has high predictive value with respect to the performance indicator(s) of interest. In these instances, it may be important to include the indicator despite the burden of data collection. However, in other cases, early pilot testing or modeling of a social risk factor indicator in a multivariable model may suggest only marginal gains. Where there is high burden and only marginal gains, CMS may choose not to include the indicator in quality measurement and payment. For example, if collecting accurate data on wealth is highly burdensome to CMS, providers, and Medicare beneficiaries, and it does not substantively contribute to adjustments to performance scores when other measures of socioeconomic position like income and education are already accounted for, CMS could choose not to also include wealth. Because lit-

erature does not (and cannot) indicate whether all social risk factors related to performance indicators used in value-based payment (VBP) must be individually accounted for to accurately adjust payment and quality measures, these are questions for the Office of the Assistant Secretary of Planning and Evaluation and CMS to test empirically.

Conclusion 1: If there are substantial barriers to collecting social risk factor data (such as high cost) and/or if early pilot testing or modeling in a multivariable model suggests only marginal gains from including any given indicator in any method of accounting for social risk factors in Medicare performance measurement and payment, inclusion of that social risk factor may not be warranted.

Accuracy refers to the degree that a given measure captures the construct that measure represents. In this report, this characteristic also captures related constructs important for data quality, such as validity, reliability, and completeness. In particular, the committee considered the extent to which standardized, validated, and reliable measures and data collection methods for a given social risk factor indicator are available and consistently used. Standardization is important to ensure valid comparisons across reporting units and settings. Accuracy should be assessed with respect to the specific purpose of accounting for social risk factors in Medicare quality measurement and payment. In other words, the level of accuracy needed should be assessed with reference to the level of accuracy required for a specific method of accounting for social risk factors.

Clinical utility describes whether providers can use information on a social risk factor in the management and treatment of that patient (IOM, 2014). Thus, this characteristic pertains specifically to data that plans and providers could collect such as through an electronic health record (EHR) or at enrollment in a health plan. If intervening on or otherwise addressing a social risk factor is beyond the purview of health care providers or can only be done at substantial cost, clinicians may be reluctant to collect data out of concern that patients would expect them to provide services they do not have the capacity to offer. The committee notes that its focus is on social risk factors important for use in Medicare quality measurement and payment. The EHR will include information on social and behavioral risk factors important to the clinical encounter but that would not be relevant for application to Medicare performance measurement and payment. The committee sees no conflict between the conclusions and recommendations in this report and those in the 2014 Institute of Medicine (IOM) report on capturing social and behavioral domains and measures in EHRs (IOM, 2014).

The committee also considered whether an indicator is relatively stable or changes over time. The distinction between relatively stable or changes

over time is not binary, but rather describes a spectrum. Some factors, such as race, ethnicity, and nativity, would not logically change over time, while other factors, such as income (especially when measured using lifetime earnings), wealth, and language, could potentially change over time, but such change is likely to be relatively slow. These factors are relatively stable. Other factors are likely to change more rapidly. For example, a Medicare beneficiary’s marital status could change rapidly owing to the loss of a spouse.¹

To weigh the trade-offs between, and identify priorities among, the potential data sources for each individual social risk factor indicator, the committee identified several guiding principles.

Recommendation 1: The committee recommends the Centers for Medicare & Medicaid Services (CMS) use five guiding principles when choosing data sources for specific indicators of social risk to be used in Medicare performance measurement and payment. These guiding principles are as follows:

• CMS should first use data it already has.
• CMS should second look for opportunities to use existing data collected by other government agencies (including elsewhere in the Department of Health and Human Services).
• To the extent that a social risk factor is relatively stable, CMS should examine the feasibility of collecting additional data at the time of enrollment in Medicare.
• Where social risk factors change over time and have clinical utility, requiring data collection through electronic health records or other types of provider reporting may be the best approach.
• For social risk factors that reflect a person’s context or environment, existing data sources that can be used to develop area-level measures should be considered.

Once the committee identified potential data sources for each of the social risk factor indicators, the committee assessed each potential data source in terms of the three characteristics (collection burden, accuracy, and clinical utility) and identified the relative advantages and disadvantages of each source. It then weighed the trade-offs for each source to identify preferences and priorities and develop proposed data collection strategies.

___________________

¹ The committee distinguishes this characteristic of change over time from modifiability as described in its third report. Because all of the indicators included in this report met all of the selection criteria, including the criterion that a social risk factor not be modifiable through provider actions, they are all considered unmodifiable. Although modifiable factors are also subject to change over time, modifiability is defined in terms of provider actions whereas change over time can occur regardless of provider action (NASEM, 2016b).

Based on the committee’s review and assessment of potential data sources for each of the social risk factor indicators, the committee identified the following categories of data that CMS could use for inclusion in Medicare quality measurement and payment:

1. Data sources exist that could be used in the short and long term.
2. Data sources with some limitations exist that could be used in the short term, and CMS should conduct research on new or improved data collection strategies in the long term. These include indicators for which
   1. CMS has some existing data that could be used in the short term, but CMS should research ways to improve accuracy and data collection in the long term.
   2. Area-level measures could be used in the short term, but CMS should research standardized measurement and data collection for the long term.
3. Measures and data collection methods exist, but data sources have considerable limitations and more research is needed to accurately collect data in the long term.
4. Some measures exist, but more research is needed on the effect of the social risk factor indicator on health care outcomes of Medicare beneficiary and on methods to accurately collect data for the Medicare population.

The subsequent sections describe the data sources for individual social risk factor indicators, organized by these categories of data availability. Each section begins with a committee recommendation; supporting text follows immediately. Table D3-1, near the end of this chapter, summarizes the information. The chapter closes with general considerations for any approach to collecting social risk factor data for use in Medicare quality measurement and payment programs.

DATA SOURCES FOR SOCIAL RISK FACTORS

Recommendation 2: The committee recommends that the Centers for Medicare & Medicaid Services use existing data on dual eligibility, nativity, and urbanicity/rurality in Medicare performance measurement and payment.

Dual Eligibility

For the Medicare population, Medicaid eligibility—also referred to as dual (Medicare and Medicaid) eligibility—is an indicator of insurance

status that can be used as a proxy measure of SEP. Because it captures elements of SEP such as income and wealth and also health insurance, and thus elements of health status, dual eligibility is an imperfect proxy of SEP that can be considered a broader measure of health-related resource availability that captures medical need (NASEM, 2016b). CMS administers both Medicare and Medicaid programs, and therefore already possess existing data on dual eligibility among Medicare beneficiaries. This includes graded data on full or partial eligibility and is the most reliable source of available data. Thus, following the committee’s guiding principle for CMS to first use data it already has, CMS should use its existing data on dual eligibility.

Nativity

Nativity refers to country of origin and measures can capture a specific country of origin or a dichotomous variable comparing foreign-born to U.S.-born individuals (NASEM, 2016b). CMS does not currently collect nativity data, nor is nativity routinely captured in EHRs. However, Medicare beneficiaries’ place of birth could be collected either by CMS or via EHRs with relatively little burden to patients, providers and plans, or CMS. Nativity is a stable social risk factor, which supports one-time collection by CMS to reduce burden, but nativity also has clinical utility, which supports collection through EHRs. Indeed, the 2014 IOM report on capturing social and behavioral domains and measures advocated including country of birth in EHRs because of its clinical utility and the relatively low collection burden (IOM, 2014). The Social Security Administration (SSA) collects place of birth including city and state or foreign country, such as on applications for a Social Security card (SSA, 2011) or at enrollment for Social Security benefits (SSA, n.d.), and it maintains place-of-birth data in its Numident file (McNabb et al., 2009). These data could be paired with Medicare beneficiary records. Because data exist in SSA records that could be linked to Medicare beneficiary records, CMS should use this available source of data.

Urbanicity/Rurality

Urbanicity/rurality describes where a place falls on the spectrum from urban to rural (NASEM, 2016b). Urbanicity/rurality can be a patient/enrollee or provider characteristic, and a patient’s urbanicity/rurality may differ importantly from his or her provider’s urbanicity/rurality—for example, when rural patients receive care from urban hospitals. For the purpose of inclusion in Medicare performance measurement and payment, urbanicity/rurality of a beneficiary’s place of residence is likely to be a more salient indicator of his or her social risk factors. Although urbanicity/rurality is conceptually continuous, it can be measured dichotomously

(i.e., urban or rural), trichotomously (i.e., urban, suburban, rural), or on a graded spectrum (e.g., percent urban) (NASEM, 2016b). Because urbanicity/rurality represents a beneficiary’s residential and community context, an area-level measure based on the beneficiary’s place of residence is appropriate.

A Medicare beneficiary’s place of residence is available in Medicare administrative records and is also likely to be captured in administrative or EHR data by providers and plans. Following the principle for CMS to first use its existing data, CMS should use beneficiaries’ residential address in its administrative records. The committee notes that although Medicare beneficiaries are required to select a single primary place of residence (CMS, 2016), some beneficiaries may have more than one residence (such as those who move seasonally), and methods that account for patient urbanicity/rurality in performance measures and payment may misclassify some patients receiving care near their secondary residences.

Beginning with the 2010 Census, the U.S. Census Bureau used a trichotomous measure to classify census tracts and/or census blocks (U.S. Census Bureau, 2015). Urban areas are defined as regions with 50,000 or more people, urban clusters are regions with at least 2,500 and fewer than 50,000 people, and rural characterizes all areas not included in either urban classification (U.S. Census Bureau, 2015). For both urban classifications, at least 1,500 persons must live outside of an institutional setting (U.S. Census Bureau, 2015). Because an area-level measure of urbanicity/rurality is appropriate and a trichotomous classification of census tract-/block-level urbanicity/rurality is available through the Census Bureau, this available measure should be used based on a Medicare beneficiary’s residential address in the Medicare record.

Recommendation 3: Data for individual measures of race and ethnicity, language, and marital/partnership status and for area-level measures of income, education, and neighborhood deprivation are currently available and the committee recommends that the Centers for Medicare & Medicaid Services (CMS) use them for performance measurement and payment applications in the short term. However, owing to limitations in these data, CMS should research ways to improve accuracy and collection of individual-level measures of race and ethnicity, language, marital/partnership status, income, and education, as well as an area-level measure of neighborhood deprivation for use in the future.

Race and Ethnicity

Race and ethnicity are social categories that represent dimensions of a society’s stratification system by which resources, risks, and rewards are

distributed (NASEM, 2016b). Race and ethnicity are conceptually distinct, albeit related, constructs that are typically identified through self-reported categories. Medicare currently maintains race and ethnicity data in its administrative records (Filice and Joynt, 2016). Current Medicare surveys and administrative records capture self-reported race and ethnicity using categories that adhere to federal standards issued by the White House Office of Management and Budget (OMB) (Filice and Joynt, 2016; OMB, 1995). However, race and ethnicity information for older beneficiaries who enrolled in Medicare prior to when these standards were issued and implemented may reflect outdated racial and ethnic classifications (Filice and Joynt, 2016; Zaslavsky et al., 2012). Some of these records were updated to improve accuracy using a survey of select beneficiaries in the 1990s (Zaslavsky et al., 2012), and methods also exist to impute race and ethnicity to improve accuracy where self-report is unavailable (Bonito et al., 2008; Elliott et al., 2009; Filice and Joynt, 2016; Grundmeier et al., 2015). EHRs are also likely to capture race and ethnicity data. To that end, Stage 2 meaningful use standards included capturing race and ethnicity using categories that adhere to OMB standards as a part of its measure of recording demographics (CMS, 2012). Race and ethnicity also have clinical utility social risk factors and were included in the 2014 IOM report on capturing social and behavioral domains and measures. Because race and ethnicity are relatively stable factors for which Medicare already has data, CMS should use available self-report and imputed race and ethnicity data in its existing records and existing methods in the short term. However, the committee acknowledges some limitations with regard to lack of standardization in current measurement and collection, and less accuracy for older age groups. Thus, over the long term, CMS should also continue to collect self-reported race and ethnicity data following the OMB standards and to work on standardizing measures and methods across the various self-report mechanisms it oversees—including administrative forms, Medicare sample surveys, and provider and plan reporting requirements.

Language

Language as a social risk factor typically represents language barriers, such as speaking a primary language that is not English, having limited English proficiency, or otherwise needing interpreter services (NASEM, 2016b). CMS currently maintains some data on preferred language, which has high specificity, but poor sensitivity. Additionally, in its Strategic Language Access Plan, CMS included having the CMS Civil Rights Agency Liaison examine the feasibility of including collection of language preferences to existing CMS surveys as well as ways to standardize data collection on existing and future surveys (CMS, 2014). Providers and plans could also

collect language data, because it is clinically useful for providers and plans to provide tailored care, such as providing health information in languages other than English or providing language interpreter services. Indeed, to provide such services, many health plans collect and maintain language data (Lawson et al., 2011; Nerenz et al., 2013a,b). Similarly, providers may voluntarily collect and maintain language data in adherence to national standards, such as those put forth by the CMS Office of Minority Health (CMS Office of Minority Health, 2016) and the HHS Office for Civil Rights (HHS, 2016). Capturing preferred language using the Library of Congress language codes was also included in the Stage 2 meaningful use regulations as part of the measure of recording demographics (CMS, 2012). Area-level measures, such as those from the American Community Survey (ACS) and some imputation methods, are also available as individual-level proxies where individual-level data do not exist. Although much research on language and health care outcomes has focused on limited English proficiency rather than preferred language (NASEM, 2016a), following the principle that CMS should first use its existing data, in the short term, CMS should use its existing data on preferred language while acknowledging its limitations. In the long term, CMS should continue efforts to standardize measures and data collection methods. A 2009 IOM report provides guidance on standardization of race, ethnicity, and language data (IOM, 2009).

Marital/Partnership Status

Marital/partnership status is a foundational structural element of social relationships and an indicator of social support. Marital or partnership status can be assessed using dichotomous measures (i.e., whether someone is married or not, whether someone is partnered or lacks a partner) or using measures with more categories (e.g., also including single, widowed, and divorced) (NASEM, 2016b). CMS maintains data on marital status, because it is important for Social Security benefits, but CMS does not collect or maintain data on partnership. Providers, plans, and other federal government agencies also do not collect data on partnership. However, because partnership can change over time, especially among older adults, and has clinical utility, it could be collected through EHRs. If so, validated measures of partnership exist in the literature, but CMS would need to develop standardized measures and data collection methods for its own collection or provider/plan reporting requirements. An important consideration for the longer term are ongoing demographic shifts in family structure, including the decline in marriage rates and increases in cohabiting individuals and persons who never marry (Aughinbaugh et al., 2013; Liu and Umberson, 2008; Tamborini, 2007; Wang and Parker, 2014), as well as the federal Supreme Court ruling

making same-sex marriage legal nationally.² These are likely to change the relationship between marital/partnership status and health. Thus, regardless of the data source CMS chooses, it will be important for CMS to monitor the empirical association between marital/partnership status and health care outcomes and revisit assumptions about marital/partnership status as an indicator of social support over time. In the short term, CMS should use available data on marital status. In the long term, research is needed on measurement and data collection for partnership. In particular, CMS may want to examine whether including partnership in any method to account for social risk factors in Medicare quality measurement and/or payment that already includes marital status and living alone adds substantial additional precision and explanatory value. As described in Appendix D2, national surveys that can be linked to individual-level health care outcomes of Medicare beneficiaries could serve as a test bed for such an assessment.

Income

Individual income can affect health and health care outcomes directly as a means of purchasing health care and indirectly as a means of acquiring health promoting resources, such as better education, housing, and nutrition (Adler and Newman, 2002; Braveman et al., 2005). Measuring income is burdensome on respondents if self-reported because income can be sensitive to collect, which leads to high nonresponse rates. However, reliable methods exist to accurately collect income data (Moore and Welniak, 2000). Partly because of such available measures and data collection methods, income is the most commonly used measure of economic resources (Braveman et al., 2005).

CMS does not currently collect or maintain income data (Samson et al., 2016), nor do providers and plans collect income data through EHRs or otherwise. The SSA maintains several sources of individual-level income data, including lifetime earnings data and information on Medicare payroll taxes, as well as data on Supplemental Security Income (SSI) for those who are eligible (i.e., adults and children with disabilities who have limited income and assets, and adults age 65 and older without disabilities and who meet financial limits) (Olsen and Hudson, 2009; SSA, 2015). Lifetime earnings and SSI may be less precise measures of income. Lifetime earnings are capped at $118,500 annually, which effectively censors high incomes (SSA, 2016), and SSI may be only part of an individual’s income (SSA, 2015). By contrast, the maximum earnings cap for Medicare payroll taxes was eliminated in 1994, and thus income data based on Medicare taxable wages

___________________

²Obergefell et al. v. Hodges, Director, Ohio Department of Health, et al. 576 US (2015).

would capture more variation, especially among higher-income individuals (Olsen and Hudson, 2009).

Government agencies that collect premiums for Medicare Parts B and D (e.g., SSA, Railroad Retirement Board, and Office of Personnel Management) also have income data used for determining premium amounts. Specifically, the SSA makes an Income Related Monthly Adjustment Amounts (IRMAA) determination for Medicare beneficiaries enrolled in Medicare Part B and/or Part D, which are used to determine monthly premium amounts for beneficiaries with higher incomes (CMS, n.d.-b). For 2016, adjustments are made to incomes greater than $85,000 for individuals and $170,000 for married couples in increasing categories (CMS, n.d.-b). Although CMS currently receives monthly data on the number of beneficiaries who have different IRMAAs, it does not have individual income information.³ Were these government agencies to provide individual income data to CMS for use in Medicare quality measurement and payment, data that are more granular than the available income categories (all of which apply to higher incomes) would be most useful. Relatedly, eligibility for the Medicare Part D Low Income Subsidy requires having an income below 150 percent of the federal poverty level, and could be used as to measure of low and high income (CMS, 2009). However, as a dichotomous measure, it would capture less variation in and be a less precise measure of income. Moreover, it only applies to beneficiaries enrolled in a Part D plan, and thus would not capture incomes for many beneficiaries.

An area-level measure of income from the ACS such as median household income could also be used as a proxy for individual-level income. However, because individual income is the construct of interest and an area-level measure may capture genuine area- or group-level effects, an area-level proxy measure is therefore an imperfect proxy for the individual-level measure and may therefore be less preferable than a true individual-level measure. In the short term, CMS should use available area-level income data from the ACS as a proxy for individual income. In the longer term, CMS should explore the feasibility of linking to SSA income data from the uncapped Medicare payroll tax and/or develop standardized measurements and methods for new data collection.

Education

Education can affect health and health care outcomes directly by enabling individuals to access and understand health information and health

___________________

³ Personal communication, John D. Shatto (Centers for Medicare & Medicaid Services, Office of the Actuary) to Kathleen Stratton (National Academies of Sciences, Engineering, and Medicine staff), September 8, 2016.

care, to make decisions that promote health and reduce health risks, and to advocate for him- or herself in health care (Cutler and Lleras-Muney, 2006; IOM, 2014). Education also shapes future occupational and economic resources and therefore indirectly shapes health and health care outcomes through other indicators of SEP—employment, occupation, and income (Adler and Newman, 2002; IOM, 2014; NASEM, 2016a,b). Education can be measured using continuous or categorical years of schooling completed or credentials of formal schooling (e.g., high school diploma, college degree) to assess educational attainment (Braveman et al., 2005; IOM, 2014). Currently, CMS does not collect or maintain data on education, nor do providers and plans routinely collect it. Although some of the more comprehensive EHRs may capture educational attainment, standardized measures and data collection strategies are needed. To that end, the earlier IOM report on social and behavioral domains and measures for EHRs identified education as a clinically useful social risk factor and recommended its inclusion in EHR meaningful use standards. With respect to other government sources, area-level measures are available through the ACS. Thus, in the short term, CMS should use these available area-level measures as a proxy for individual education. In the long term, because education is relatively stable for Medicare beneficiaries, CMS should develop standardized measures and methods to collect education data.

Neighborhood Deprivation

In its third report, the committee concluded that a measure of neighborhood deprivation (i.e., a composite measure of neighborhood compositional characteristics) at the census tract level is likely to be a good proxy for a range of both individual and true area-level constructs relevant to performance indicators used in VBP (NASEM, 2016b). Relevant area-level constructs include compositional characteristics of communities such as dimensions of SEP (e.g., the proportion of racial and ethnic minority residents, foreign-born residents, single-parent households, households below the federal poverty level, and English language–proficient residents) as well as elements of residential environments including the physical or built environment (e.g., housing, walkability, transportation options, and availability of services—including health care services) and social environments (e.g., safety and violence, social disorder, the presence of social organizations, and social cohesion).

Because neighborhood deprivation captures a patient or beneficiary’s environment or residential context, an area-level measure based on the beneficiary’s residential address is appropriate. As described in the section on urbanicity/rurality, although residential addresses are available from providers, plans, and Medicare records, the latter is preferable, because

these are the data CMS already possesses. Neighborhood deprivation can be assessed using a single-item measure such as median household income or using a multi-item composite measure. Numerous neighborhood deprivation indexes comprising multiple items (e.g., median household income, percent of residents with a high school degree, percent of unemployed residents, percent of households with an income below the federal poverty level) have been developed (Oka, 2015), and data on these area-level measures are available through the ACS. As described in the previous chapter, because of small sample sizes, ACS data will need to be pooled across years. Because neighborhoods can change rapidly, where this occurs, data that are just a few years old may not accurately reflect the neighborhood at present. Another important limitation of existing neighborhood deprivation measures and indexes is that they have been developed, tested, and applied primarily to urban contexts. It is possible that area-level factors most relevant to health care outcomes differ for urban and rural areas. For example, concentrated disadvantage may be most salient in urban contexts; whereas, availability of and distance to health care resources may be more relevant constructs in rural settings (NASEM, 2016a).

Defining the appropriate geospatial unit across urban and rural settings presents an additional challenge. Because population density and the density of available resources varies substantially between urban and rural areas, the spatial scale that is relevant for various health-related processes may differ for urban areas and rural areas. For example, census tracts may be the most relevant area for measuring urban neighborhoods (as they are used to define urban areas in the Census Bureau’s 2010 classification, as described in the earlier section on urbanicity/rurality). Although most rural research is conducted at the county level (Isserman, 2005), most counties are likely to be too heterogeneous for county-level measures of neighborhood deprivation to be useful. To be meaningful for certain methods of accounting for social risk factors in Medicare quality measurement and payment, the geographic area should have sufficient variability with respect to provider and plan performance.

Despite the challenges described above (which pertain primarily to research on area effects) even imperfect area-level measures can be useful for the purposes of accounting for social risk factors in Medicare quality measurement and payment. This is because crude (and geographically mis-specified) area-level measures will still capture some variability in health-relevant, area-level constructs (social and physical environments) and may also serve as imperfect proxies for unavailable individual-level socioeconomic data (because of strong residential segregation by class). For these reasons, the committee recommends that CMS test a composite measure (such as an existing indicator from the literature) and a simple single-indicator item (such as median household income), contrast their

performance at the census tract level, and also weigh the benefits of the simplicity of a single indicator against the increased precision from a composite measure for use in the short term. To increase accuracy in the long term, CMS could conduct research on measurement and data collection such as measures to better capture neighborhood deprivation in rural areas, to identify an improved geospatial unit of analysis for rural settings, and to assess the performance of any given variable (single or composite) across multiple geographic areas.

Recommendation 4: Individual measures of wealth, living alone, and social support exist, but they are sufficiently limited to preclude use by the Centers for Medicare & Medicaid Services (CMS) in Medicare performance measurement and payment at this time. Therefore, the committee recommends that CMS research ways to accurately collect data on these indicators.

Wealth

Wealth represents total accumulated economic resources (assets) that, like income, can affect health and health care outcomes directly as a means of purchasing health care and indirectly as a means of acquiring health-promoting resources (Braveman et al., 2005; Deaton, 2002; NASEM, 2016a). Because wealth accumulates over time, it can also buffer the effects of rapid changes in income, such as those caused by unemployment or illness (Cubbin et al., 2011). Thus, wealth may capture more variation than income among older persons, and may therefore be a more sensitive indicator of SEP among Medicare beneficiaries (Allin et al., 2009). Collecting self-reported net worth is difficult because it is sensitive and because many individuals simply do not know the value of their net worth or what assets they have (Braveman et al., 2005; Eggleston and Klee, 2015). Nevertheless, some reliable and validated measures and data collection methods do exist. In particular, the Health and Retirement Study (HRS) has designed measures and methods to collect data on wealth that overcome traditional barriers to collecting wealth data such as concerns about privacy and imprecise knowledge (NIA et al., 2007). More specifically, the HRS captures both the amount and composition of assets as well as current and future benefits including government benefits (such as Social Security, Medicare, and Medicaid) and employer-based benefits (like pensions and health insurance), as well as the movement of assets (such as housing within families, gifts and bequests, and savings and spendings) over time from retirement until death (NIA et al., 2007). Some HRS data are linked to Medicare records (ResDAC, n.d.), and therefore are useful for examining the effect of wealth on health care outcomes. However, as described in Appendix D2,

because samples for the HRS are small, these data are unlikely to be useful for application in Medicare quality measurement or payment. Some studies have also used simplified or proxy measures of wealth, such as home or car ownership. Because of these types of measurement challenges, there is less empirical evidence on the association between wealth and health care outcomes compared to other indicators of SEP (Braveman et al., 2005).

Wealth data are not currently available through CMS, providers and plans, or other government agencies. Medicaid programs do require assets below a certain threshold for eligibility, and this asset threshold could be used to measure wealth dichotomously (i.e., high wealth above the threshold, and low wealth at or below the threshold). However, because Medicaid is administered at the state level, eligibility criteria, including this asset threshold, vary by state. Moreover, this measure of wealth would be at least partly captured through dual eligibility status, for which there is better and available existing data (as described in the earlier section on dual eligibility).

Because no data sources are available for use in the short term, CMS should conduct more research on both measurement and data collection methods by CMS or through EHRs. Because collecting accurate wealth data is known to be difficult and burdensome and because data collected through EHRs could be done via multiple modes, which could augment potential accuracy issues, EHRs may be less preferable to centralized collection by CMS. In particular, CMS may want to consider the empirical question of whether the addition of wealth data adds sufficient precision above and beyond income data, for which some data are already available and for which methods and measures exist to collect data with less burden to warrant additional data collection for inclusion in any method to account for social risk factors in Medicare quality measurement and payment. As described in Appendix D2, national surveys such as the HRS that can be linked to individual-level health care outcomes of Medicare beneficiaries could serve as a test bed for CMS to assess this question.

Living Alone

Living alone is a structural element of social relationships, which is typically an indicator of social isolation or loneliness in health research, and which is also likely to capture elements of social support (Berkman and Glass, 2000; Brummett et al., 2001; Cohen, 2004; Eng et al., 2002; House et al., 1988; Wilson et al., 2007). Living alone can be assessed with little burden using a dichotomous measure (living alone or with others) or more finely graded measures of household composition (i.e., living alone, with one other person, two other persons, and so on). CMS currently collects data on living arrangements for some patients in postacute settings, such as

through the Home Health Outcome Assessment Information Set (AHRQ, 2014; CMS, n.d.-a), and in the Medicare Current Beneficiary Survey (CMS, 2015a). Providers and plans do not currently collect data on living arrangements, nor is national data available through other government agencies. Thus, there are no data sources that could be used in the short term. However, for the long term, because living arrangements can change rapidly especially for older adults and because living alone has clinical utility, living alone may best be captured in the clinical setting, and CMS should develop standardized measures and methods for data collection through EHRs.

Social Support

Social support is a crucial function of social relationships and includes instrumental components (such as material and other practical supports) and emotional dimensions (such as through caring and concern). Instrumental social support can facilitate access to health-promoting resources (e.g., delivery of nutritious meals) and health care services (e.g., providing transportation to a doctor’s appointment) (Berkman and Glass, 2000). Emotional social support can positively affect health through psychosocial mechanisms such as by boosting self-efficacy to practice health-promoting behaviors like quitting smoking, and social support may also buffer negative effects of health risks (Berkman and Glass, 2000; IOM, 2014). Social support can also negatively affect health such as by causing distress through negative social interactions or because negative social influences promote risky health behaviors (Uchino, 2006).

Currently, no social support data are available within CMS, from providers and plans, or from other national data via other government agencies. Thus, there are no data sources that could be used in the short term. However, for the long term, because social support can change rapidly especially among older adults and because it has clinical utility, it may best be captured in the clinical setting, and CMS should develop standardized measures and methods for data collection through EHRs. In its 2014 report on capturing social and behavioral domains and measures through EHRs, the IOM recommended inclusion of social support and recommended measures (IOM, 2014). Such measurement and data collection methods could be refined, standardized, and added to the Office of the National Coordinator for Health Information Technology’s (ONC’s) meaningful use regulations or mandated through reporting requirements to CMS to ensure accurate data. Thus, CMS should develop standardized measures and methods for data collection through EHRs for the long term.

Recommendation 5: Area-level measures exist for housing, but they have limitations for use by the Centers for Medicare & Medicaid

Services (CMS) in Medicare performance measurement and payment at this time. The committee recommends that CMS research ways to accurately collect housing data, whether at an individual level or an area level.

Housing

Elements of housing that may influence health and health care outcomes include housing stability, homelessness, and quality and safety. Homelessness and housing instability, defined as a lack of access or threats to reasonable quality housing (Frederick et al., 2014), can be barriers to accessing health care and are associated with poorer physical and mental health and increased mortality (NASEM, 2016a). Poor quality or unsafe housing can expose individuals to such environmental hazards as lead, poor air quality, infectious disease, and poor sanitation, and can lead to injury (IOM, 2003a; NASEM, 2016a). Currently neither CMS nor providers and plans routinely collect housing information. Some more comprehensive EHRs may collect or link to data on housing (e.g., Gottlieb et al., 2015; ONC, n.d.). However, because housing can change over time and has clinical utility, housing information could be collected through EHRs. Some area-level measures of housing are available through the ACS and the Department of Housing and Urban Development (HUD). For example, ACS housing data capture physical characteristics (e.g., rooms, age, access to utilities) as well as housing costs, age, and value (U.S. Census Bureau, 2013) and the HUD Healthy Communities Index captures vacancy rates, housing costs, blood lead levels in children as an indicator of environmental hazards, and age of housing (San Diego Council of Governments, n.d.). Because some dimensions of housing reflect beneficiaries’ environment, an area-level measure could be appropriate. This measure would be based on a beneficiary’s residential address, which is collected by CMS, through EHRs, and by plans. However, following the principle to first use available existing data it possesses, the residential address in the Medicare record is preferred. Thus, in the short term, the committee recommends that CMS test area-level measures based on a beneficiary’s residential address in the Medicare record and contrast their performance. Because other elements of housing, in particular, physical characteristics, occur at the individual level, and these are likely to change over time, individual-level housing data could be collected through EHRs in the long term, but more research is needed on measurement and data collection methods.

Recommendation 6: The committee recommends that research be conducted on the effect of acculturation, sexual orientation and gender identity, and environmental measures of residential and community

context on health care outcomes of Medicare beneficiaries and on methods to accurately collect relevant data in the Medicare population.

Acculturation

Acculturation describes how much an individual adheres to the social norms, values, and practices of his or her own home country or ethnic group or to those of the United States (NASEM, 2016a). Acculturation is frequently assessed with language use. Additionally, because there is a strong interaction between acculturation and race and ethnicity, measures of acculturation frequently assess acculturation among specific subgroups (e.g., Hispanic immigrants) (HHS, 2014). For example, the Brief Acculturation Scale for Hispanics is a reliable, validated measure to assess acculturation among Hispanic Americans using four self-reported language use items (Mills et al., 2014). Duration in the United States is also used as a proxy for acculturation, because acculturation is expected to increase with the amount of time spent in the United States. Although there is evidence on the relationship between acculturation and health, evidence on the effect of acculturation and health care outcomes is not well established (Abraído-Lanza et al., 2006; IOM, 2014; NASEM, 2016a). Because more evidence is needed on the empirical association between acculturation and health care outcomes, CMS should revisit this indicator and its appropriate measurement when more evidence is available. However, because acculturation is often measured using preferred language, which is available to CMS in the short term, language data could capture elements of acculturation in addition to language itself.

Sexual Orientation and Gender Identity⁴

Sexual orientation captures individuals who identify as lesbian, gay, bisexual, queer, questioning, or otherwise nonconforming, and is typically defined with respect to three dimensions: attraction, behavior, and identity (IOM, 2011). Gender identity typically refers to individuals who identify as gender minorities, including those who identify as transgender, intersex, or otherwise nonconforming (IOM, 2011). Although some measures and best practices for data collection exist and CMS has included data collection of sexual orientation and gender identity in its Equity Plan for Improving

___________________

⁴ As described in the committee’s third report (NASEM, 2016b), normative gender categories (men and women) are strongly associated with health and health care outcomes, despite the fact the gender effects are difficult to separate from biological sex effects. Thus, normative gender is a strong candidate for inclusion in methods to account for social risk factors in Medicare quality measurement and payment programs. However, the committee notes that gender is already included as a risk factor in clinical risk adjustments in Medicare.

Quality in Medicare, there are currently no standards for measuring and collecting data on sexual orientation and gender identity (CMS Office of Minority Health, 2015). One limitation of existing measures of sexual orientation is that they frequently only capture one dimension of sexual orientation, and some individuals do not present consistently across the three dimensions (e.g., men who have sex with men but do not identify as gay) (IOM, 2011). Outside of CMS, some national health surveys, including the National Health and Nutrition Examination Survey (NHANES), National Health Interview Survey (NHIS), and Behavioral Risk Factor Surveillance System (BRFSS) do collect data on sexual orientation and gender identity. NHANES includes sexual behavior questions, while NHIS and BRFSS include items capturing sexual identity and gender identity (CDC, 2013, 2015, 2016). Providers and plans also do not typically collect sexual orientation and gender identity data. However, ONC added collection of sexual orientation and gender identity to its measure of recording demographics in its Stage 3 meaningful use regulations (CMS, 2015b). Importantly, this does not require providers to collect sexual orientation and gender identity data, but rather that their EHRs have the capacity to do so. Partly because of a lack of standardized measures, there is currently little evidence on the effect of sexual orientation and gender identity on health care outcomes (NASEM, 2016a,b). Because more evidence is needed on the empirical association between sexual orientation and gender identity and health care outcomes, CMS should revisit this indicator and its appropriate measurement when more evidence is available. In particular, for sexual orientation, CMS should take notice of which dimension or dimensions are most relevant for health care outcomes. At the same time, CMS should continue efforts to develop standardized measures and data collection strategies and to collect data.

Other Environmental Measures of Residential and Community Context

Other environmental measures of residential and community context capture elements of the physical or built environment such as housing, walkability, transportation options, and proximity to services (including health care and social services) as well as social environments such as safety and violence, social disorder or cohesion, economic and educational opportunities, and the presence of social organizations. Neighborhood environments can affect health through the distribution of health-relevant resources (e.g., access to recreational spaces, healthy foods, or health care services), by exposing residents to environmental hazards like air pollution, and by exposing residents to physical and social hazards such as discrimination and physical decay that negatively affect health through stress and other psychosocial processes (Diez Roux and Mair, 2010; IOM, 2003b). Thus,

there is a conceptual relationship between neighborhood environments and health care outcomes, but evidence is currently limited and environmental measures need to be tested further (NASEM, 2016a). Therefore, CMS should revisit such environmental measures and their appropriate measurement when more evidence is available.

Recommendation 7: The committee recommends that the Centers for Medicare & Medicaid Services collect information about relevant, relatively stable social risk factors, such as race and ethnicity, language, and at the time of enrollment.

Indicators for which data might best be captured through a revised enrollment form include race and ethnicity, language, and education. Should other methods, such as linking to data from the SSA or the Internal Revenue Service, prove too difficult or not produce accurate information on other indicators (e.g., income, race and ethnicity, and nativity), these could be considered for inclusion in the revised enrollment form.

Should research demonstrate an important explanatory effect of one or more of these indicators and a pilot test shows it is feasible, CMS could supplement the information collected at enrollment with a survey of current beneficiaries, whose information would not have been captured at the time of enrollment.

Table D3-2 summarizes the availability of data for social risk factor indicators that could be accounted for in Medicare payment programs.

GENERAL CONCLUSIONS

In addition to the specific guidance the committee proposed for collecting data for specific social risk factor indicators, the committee also identified several general conclusions for CMS in its overall approach to collecting data on social risk factors for use in Medicare quality measurement and payment.

Different data collection strategies for the same indicator may be warranted depending on the purpose or methods it is used for. Additionally, the advantages and disadvantages of any specific source should be considered in reference to the intended use. For example, risk adjusting health plan quality measures may require data from different sources compared to risk adjusting hospital quality measures, because, for example, social risk factors that affect the outcome or cost of a hospitalization likely differ from those that affect quality or total cost of care measures. Similarly, CMS may need data on social risk factors regardless of whether care is sought or not

when accounting for social risk factors in health plan or accountable care organization performance scores; whereas, for adjustment related to performance measures that are associated with a health care episode, it may make sense to have providers report. Thus, any indicator may require a multimodal approach to data collection. This may be particularly relevant for data collected through an EHR, because there is substantial variation in providers’ stage of EHR adoption, as well as in their capacities for health information exchange. However, this may also be true for other sources of data, where there are limitations to data from existing sources, where data would be collected in different settings (e.g., hospitals, clinical practices, in the home), and when data are collected by different types of individuals (e.g., clinicians and non-clinical staff). Moreover, the specific modes of data collection needed may change over time. Specifically for EHR data, needs for complementary modes may diminish with advances in EHR adoption and interoperability. An example of an existing multimodal approach is CMS’s strategy for collecting race and ethnicity data. Data from beneficiaries enrolled since the 1990s are collected via self-report, but for older beneficiaries for whom current categories collected through self-reported data are unavailable, CMS imputes race and ethnicity and also updates older data with newer self-reported data collected through surveys. Additionally, when CMS revised its race and ethnicity measures, it conducted a survey of certain Medicare beneficiaries to improve the accuracy of its data (Zaslavsky et al., 2012). In short, regardless of the source, research on how to accurately and reliably collect data across different modes and in different settings will be needed.

Conclusion 2: Different data collection strategies for the same social risk factor indicator may be warranted depending on the purpose or methods used to account for social risk factors in Medicare performance measurement and payment. Additionally, the advantages and disadvantages of any specific source should be considered in reference to the intended use.

Conclusion 3: Any specific social risk factor indicator may require a multi-modal approach to data collection.

Conclusion 4: Regardless of the source, research on how to accurately and reliably collect social risk factor data across different modes and in different settings will be needed.

TABLE D3-1 Potential Data Sources for Each Social Risk Factor Indicator, Their Advantages and Disadvantages, and the Committee’s Proposed Data Collection Strategy

TABLE D3-2 Summary of Data Availability for Social Risk Factor Indicators

SOCIAL RISK FACTOR		DATA AVAILABILITY
	Indicator	1	2	3	4
SEP
	Income
	Education
	Dual eligibility
	Wealth
Race, Ethnicity, and Cultural Context
	Race and ethnicity
	Language
	Nativity
	Acculturation
Gender
	Gender identity
	Sexual orientation
Social Relationships
	Marital/partnership status
	Living alone
	Social support
Residential and Community Context
	Neightborgood deprivation
	Urbanicity/rurality
	Housing
	Other environmental measures

Available for use now

Available for use now for some outcomes, but research needed for improved, furure use

Not sufficiently available now; research needed for improved, future use

Research needed to better understand relationship with health care outcomes and on how to best collect data

REFERENCES

CMS. 2015b. Medicare and Medicaid programs; electronic health record incentive program-stage 3 and modifications to meaningful use in 2015 through 2017; final rule. Federal Register 80(200):62761–62955.

CMS. 2016. CMS manual system: Pub 100-02 Medicare benefit policy. https://www.cms.gov/Regulations-and-Guidance/Guidance/Transmittals/Downloads/R208BP.pdf (accessed September 9, 2016).

CMS. n.d.-a. Home health patient tracking sheet. https://www.cms.gov/Medicare/QualityInitiatives-Patient-Assessment-Instruments/HomeHealthQualityInits/Downloads/OASIS-C2-Item-Set-Effective_1_1_17a.pdf (accessed August 12, 2016).

CMS. n.d.-b. Initial IRMAA determination. https://www.medicare.gov/forms-help-andresources/mail-about-medicare/irmaa-determination.html (accessed September 12, 2016).

CMS Office of Minority Health. 2015. The CMS equity plan for improving quality in Medicare. https://www.cms.gov/About-CMS/Agency-Information/OMH/OMH_Dwnld-CMS_EquityPlanforMedicare_090615.pdf (accessed August 11, 2016).

CMS Office of Minority Health. 2016. The national CLAS standards. http://minorityhealth.hhs.gov/omh/browse.aspx?lvl=2&lvlid=53 (accessed September 12, 2016).

Cohen, S. 2004. Social relationships and health. American Psychology 59(8):676–684.

Cubbin, C., C. Pollack, B. Flaherty, M. Hayward, A. Sania, D. Vallone, and P. Braveman. 2011. Assessing alternative measures of wealth in health research. American Journal of Public Health 101(5):939–947.

Cutler, D. M., and A. Lleras-Muney. 2006. Education and health: Evaluating theories and evidence. Cambridge, MA: National Bureau of Economic Research.

Deaton, A. 2002. Policy implications of the gradient of health and wealth. Health Affairs (Millwood) 21(2):13–30.

Diez Roux, A. V., and C. Mair. 2010. Neighborhoods and health. Annals of the New York Academy of Sciences 1186:125–145.

Eggleston, J. S., and M. A. Klee. 2015. Reassessing wealth data quality in the survey of income and program participation. Proceedings of the 2015 Federal Committee on Statistical Methodology (FCSM) Research Conference.

Elliott, M. N., A. M. Haviland, D. E. Kanouse, K. Hambarsoomian, and R. D. Hays. 2009. Adjusting for subgroup differences in extreme response tendency in ratings of health care: Impact on disparity estimates. Health Services Research 44(2 Pt 1):542–561.

Eng, P. M., E. B. Rimm, G. Fitzmaurice, and I. Kawachi. 2002. Social ties and change in social ties in relation to subsequent total and cause-specific mortality and coronary heart disease incidence in men. American Journal of Epidemiology 155(8):700–709.

Filice, C. E., and K. E. Joynt. 2016. Examining race and ethnicity information in Medicare administrative data. Medical Care [Epub ahead of print].

Frederick, T. J., M. Chwalek, J. Hughes, J. Karabanow, and S. Kidd. 2014. How stable is stable? Defining and measuring housing stability. Journal of Community Psychology 42(8):964–979.

Gottlieb, L. M., K. J. Tirozzi, R. Manchanda, A. R. Burns, and M. T. Sandel. 2015. Moving electronic medical records upstream: Incorporating social determinants of health. American Journal of Preventative Medicine 48(2):215–218.

Grundmeier, R. W., L. Song, M. J. Ramos, A. G. Fiks, M. N. Elliott, A. Fremont, W. Pace, R. C. Wasserman, and R. Localio. 2015. Imputing missing race/ethnicity in pediatric electronic health records: Reducing bias with use of U.S. Census location and surname data. Health Services Research 50(4):946–960.

HHS (Department of Health and Human Services). 2014. Improving cultural competence: A treatment improvement protocol. Rockville, MD: U.S. Substance Abuse and Mental Health Services Administration.

HHS. 2016. Guidance to federal financial assistance recipients regarding Title VI prohibition against national origin discrimination affecting limited English proficient persons. http://www.hhs.gov/civil-rights/for-individuals/special-topics/limited-english-proficiency/guidance-federal-financial-assistance-recipients-title-VI (accessed September 12, 2016).

House, J. S., K. R. Landis, and D. Umberson. 1988. Social relationships and health. Science 241(4865):540–545.

IOM (Institute of Medicine). 2003a. The future of the public’s health in the 21st century. Washington, DC: The National Academies Press.

IOM. 2003b. Unequal treatment: Confronting racial and ethnic disparities in health care. Washington, DC: The National Academies Press.

IOM. 2009. Race, ethnicity, and language data: Standardization for health care quality improvement. Washington, DC: The National Academies Press.

IOM. 2011. The health of lesbian, gay, bisexual, and transgender people: Building a foundation for better understanding. Washington, DC: The National Academies Press.

IOM. 2014. Capturing social and behavioral domains and measures in electronic health records: Phase 2. Washington, DC: The National Academies Press.

Isserman, A. M. 2005. In the national interest: Defining rural and urban correctly in research and public policy. International Regional Science Review 28(4):465–499.

Lawson, E. H., R. Carreón, G. Veselovskiy, and J. J. Escarce. 2011. Collection of language data and services provided by health plans. American Journal of Managed Care 17(12):e479–e487.

Liu, H., and D. J. Umberson. 2008. The times they are a changin’: Marital status and health differentials from 1972 to 2003. Journal of Health and Social Behavior 49(3):239–253.

McNabb, J., D. Timmons, J. Song, and C. Puckett. 2009. Uses of administrative data at the Social Security Administration. Social Security Bulletin 69(1).

Mills, S. D., V. L. Malcarne, R. S. Fox, and G. R. Sadler. 2014. Psychometric evaluation of the brief acculturation scale for Hispanics. Hispanic Journal of Behavioral Sciences 36(2):164–174.

Moore, J. C., and E. J. Welniak. 2000. Income measurement error in surveys: A review. Journal of Official Statistics 16(4):331.

NASEM (National Academies of Sciences, Engineering, and Medicine). 2016a. Accounting for social risk factors in Medicare payment: Identifying social risk factors. Washington, DC: The National Academies Press.

NASEM. 2016b. Accounting for social risk factors in Medicare payment: Criteria, factors, and methods. Washington, DC: The National Academies Press.

Nerenz, D. R., R. Carreón, and G. Veselovskiy. 2013a. Race, ethnicity, and language data collection by health plans: Findings from 2010 American Health Insurance Plans Foundation-Robert Wood Johnson Foundation survey. Journal of Health Care for the Poor and Underserved 24(4):1769–1783.

Nerenz, D. R., G. M. Veselovskiy, and R. Carreón. 2013b. Collection of data on race/ethnicity and language proficiency of providers. American Journal of Managed Care 19(12):e408–e414.

NIA (National Institute on Aging), NIH (National Institutes of Health), and HHS (Department of Health and Human Services). 2007. Growing older in America: The Health and Retirement Study. http://hrsonline.isr.umich.edu/sitedocs/databook/HRS_Text_WEB_Ch3.pdf (accessed September 19, 2016).

Oka, M. 2015. Measuring a neighborhood affluence-deprivation continuum in urban settings: Descriptive findings from four US cities. Demographic Research 32(54):1469–1486.

Olsen, A., and R. Hudson. 2009. Social Security Administration’s master earnings file: Background information. Social Security Bulletin 69(3).

OMB (White House Office of Management and Budget). 1995. Standards for the classification of federal data on race and ethnicity. https://www.whitehouse.gov/omb/fedreg_race-ethnicity (accessed April 21, 2016).

ONC (Office of the National Coordinator on Health Information Technology). n.d. CHP participant community: All Chicago making homelessness history. https://www.healthit.gov/sites/default/files/all_chicago_making_homelessness_history-final.pdf (accessed August 15, 2016).

ResDAC (Research Data Assistance Center). n.d. Health and Retirement Survey—Medicare linked data. https://www.resdac.org/cms-data/files/hrs-medicare (accessed August 4, 2016).

Samson, L. W., K. Finegold, A. Ahmed, M. Jensen, C. E. Filice, and K. E. Joynt. 2016. Examining measures of income and poverty in medicare administrative data. Medical Care [Epub ahead of print].

San Diego Council of Governments. n.d. HCI domains and indicators. http://hci-sandiego.sandag.org/indicators (accessed August 11, 2016).

SSA (Social Security Administration). 2011. Application for a Social Security card. https://www.ssa.gov/forms/ss-5.pdf (accessed August 8, 2016).

SSA. 2015. Supplemental Security Income. https://www.ssa.gov/pubs/EN-05-11000.pdf (accessed August 9, 2016).

SSA. 2016. If you are self employed. https://www.ssa.gov/pubs/EN-05-10022.pdf (accessed August 10, 2016).

SSA. n.d. Checklist for online Medicare, retirement, and spouses applications. https://www.ssa.gov/hlp/isba/10/isba-checklist.pdf (accessed August 11, 2016).

Tamborini, C. R. 2007. The never-married in old age: Projections and concerns for the near future. Social Security Bulletin 67:25.

Uchino, B. N. 2006. Social support and health: A review of physiological processes potentially underlying links to disease outcomes. Journal of Behavioral Medicine 29(4):377–387.

U.S. Census Bureau. 2013. American Community Survey information guide. https://www.census.gov/content/dam/Census/programs-surveys/acs/about/ACS_Information_Guide.pdf (accessed August 11, 2016).

U.S. Census Bureau. 2015. 2010 census urban and rural classification and urban area criteria. https://www.census.gov/geo/reference/ua/urban-rural-2010.html (accessed August 8, 2016).

Wang, W., and K. C. Parker. 2014. Record share of Americans have never married: As values, economics and gender patterns change. Washington, DC: Pew Research Center, Social & Demographic Trends Project.

Wilson, R. S., K. R. Krueger, S. E. Arnold, J. A. Schneider, J. F. Kelly, L. L. Barnes, Y. Tang, and D. A. Bennett. 2007. Loneliness and risk of Alzheimer disease. Archives of General Psychiatry 64(2):234–240.

Zaslavsky, A. M., J. Z. Ayanian, and L. B. Zaborski. 2012. The validity of race and ethnicity in enrollment data for Medicare beneficiaries. Health Services Research 47(3 Pt 2):1300–1321.