Medicaid and Nursing Home Choice: Why do Duals End up in Low-Quality Facilities?

Abstract

We provide empirical evidence on the relative importance of specific observable factors that can explain why individuals enrolled in both Medicare and Medicaid (duals) are concentrated in lower-quality nursing homes, relative to those not on Medicaid. Descriptive results show that duals are 9.7 percentage points more likely than non-duals to be admitted to a low-quality (1–2 stars) nursing home. Using the Blinder-Oaxaca decomposition approach in a multivariate framework, we find that 35.4 percent of the difference in admission to low-quality nursing homes can be explained by differences in the distribution of observable characteristics. Differences in education and distance to high-quality nursing homes are important drivers, as are health status and race. Our findings highlight the need for creative policy solutions targeting the modifiable factors to reduce the disparity.

Introduction

Individuals enrolled in both Medicare and Medicaid, known as “duals”, are often physically and cognitively impaired, tend to be disproportionately from racial or ethnic minority groups, and have a higher use of health care services and costs. In 2011, duals accounted for 14% of Medicaid enrollment but 33% of Medicaid expenditures and 20% of Medicare enrollment but 35% of Medicare expenditures (Medicare Payment Advisory Commission, 2016). Nursing home residents are a particularly important segment of the duals, as they are dependent on Medicaid for the entirety of their long-term nursing home stay costs (with Medicare paying only for acute-care events such as hospitalizations and post-acute care).

It is well established that duals receive lower-quality nursing home care than non-duals (Rahman, Gozalo, et al., 2014; Rahman, Grabowski, Gozalo, Thomas, & Mor, 2014). Furthermore, evidence supports that disparities are largely driven by differences in where people receive care, as opposed to discrimination within facilities (Grabowski, Gruber, & Angelelli, 2008; Konetzka & Werner, 2009; Troyer, 2004). Existing studies have shown that patients on Medicaid go to lower-quality nursing homes, but evidence on the relative contributions of different drivers of this disparity is lacking.

Prior studies have suggested several potential reasons why duals are more likely to enter lower-quality nursing homes, including education, health needs, and lack of supply of high-quality nursing homes nearby. Lack of education may be a barrier to understanding health care information (Williams et al., 1995), and perceptions about quality differ by insurance type (Jewett & Hibbard, 1996). One study found that less educated people are likely to be admitted to poor quality nursing homes (Angelelli, Grabowski, & Mor, 2006). Recent qualitative studies have found that blacks or those with lower levels of education were less likely to use nursing home quality information available in Nursing Home Compare (NHC), a web portal that publishes quality information and 5-star ratings for nursing homes (Schapira, Shea, Duey, Kleiman, & Werner, 2016), and people’s choices of nursing homes are often limited by the availability of specialized services, distance from family, and Medicaid bed availability (Konetzka & Perraillon, 2016).

A limited supply of high-quality nursing homes in poorer neighborhoods is likely to be a particularly important driver of disparities (Perraillon, Brauner, & Konetzka, 2017). Several studies have documented that individuals are likely to select facilities that are close to their residence (Pesis-Katz et al., 2013; Shugarman & Brown, 2006; Zwanziger, Mukamel, & Indridason, 2002). Public reporting of quality may have worsened the disparity by payment source as nursing homes are segregated by quality, with higher quality facilities located mostly in affluent areas (Konetzka, Grabowski, Perraillon, & Werner, 2015). Several studies have shown that racial and ethnic minorities are more likely to enter facilities with low quality (Angelelli et al., 2006; Grabowski, 2004; Smith, Feng, Fennell, Zinn, & Mor, 2007). Research has also shown that disparities in access to health care services are lower when blacks and whites are living in the same neighborhoods (LaVeist, Pollack, Thorpe, Fesahazion, & Gaskin, 2011).

In this study, we sought to quantify the relative importance of resident and provider characteristics associated with the disparity in the quality of nursing homes to which duals and non-duals are admitted. Using nursing home quality ratings data as reported in NHC for the year 2009 combined with resident characteristics, we employ the Blinder-Oaxaca approach to decompose the difference in observed nursing home quality between duals and non-duals. Although prior studies underscore the notion that factors like education and distance are associated with nursing home choice, we extend these prior studies by quantifying how much of the difference in the quality of nursing homes selected by duals vs. non-duals can be attributed to the difference in these and other observable characteristics. By identifying the relative influence of important factors that contribute to the disparity, our findings can help policymakers design appropriate policies to reduce disparities in the quality of nursing homes chosen by duals and non-duals.

Conceptual Framework

Selection of a nursing home is a complex process, involving not only the preferences and constraints faced by the care recipient, but also often depending on input and referrals from families, hospitals, and other health care providers. Choices may be adequate or may be constrained such that there is effectively no choice. Ultimately, a number of observed and unobserved factors determine where individuals go for nursing home care. We conceptualize an individual’s preference over particular nursing homes to be a function of multiple factors: patient’s health needs; expected quality of life and comfort with one’s surroundings; the quality of the nursing home; and distance from home. For individuals paying privately, price is also a factor. Preferences may be influenced by information from care providers, Nursing Home Compare report cards, and recommendations from family and friends. In translating these preferences into a “choice” or admission to a particular facility, preferences are maximized subject to supply-side constraints: availability of a bed and the willingness of the facility to accept a new resident of one’s insurance type and health status.

All else being equal, we assume that individuals prefer high-quality nursing homes. However, the preference for a high-quality facility is constrained by other individual characteristics. We conceptualize education as a proxy for the ability to seek out information on access and quality of health care, suggesting that educated individuals may select high-quality facilities. Since duals tend to have less formal education than non-duals, we hypothesize that educational differences will contribute substantially to the disparity in high-quality nursing home choice between duals and non-duals.

Similarly, we view distance to high-quality facilities as a proxy for adequate supply of high-quality facilities near one’s neighborhood. Well-established patterns of segregation of the supply and quality of facilities, mirroring residential segregation, implies that duals may have to travel farther from their home to get to high-quality facilities. Because existing studies have documented a strong preference for a facility close to one’s residence, we expect a negative relationship between distance and admission to high-quality facilities.

Finally, prior research suggests that individuals seek care where most of the current residents are of the prospective resident’s race (Rahman & Foster, 2015). Preferences for race concordance coupled with the lack of high-quality facilities close to the residence may imply that minorities may be less likely to be admitted to high-quality facilities. The direction of the relationship between health status and nursing home choice is less obvious. On one hand, high-quality facilities may be attractive to sicker individuals as they can signal their ability to care for sicker individuals (e.g., high-quality facilities may be able to maintain dementia care units). And yet, high-quality facilities may prefer non-duals who tend to be less sick and whose private-pay prices for a longer stay are typically higher than the reimbursements from Medicaid for duals. Thus, the ability of sicker individuals, especially if they are duals, to select high-quality facilities may be limited.

Methods

Data

We use 2009 data from several sources: Nursing home Minimum Data Set (MDS) 2.0,NHC, Medicare Beneficiary Summary files, Online Survey, Certification and Reporting (OSCAR), and ZIP-code level income data. We selected the year 2009 for a number of reasons. First, 2009 was the first year after the 5-star ratings for nursing homes were made public in December of 2008, before nursing homes had a chance to substantially improve their scores. Second, several studies have questioned the reliability of the improvement in nursing home quality after the release of 5-star rating system (Ryskina, Konetzka, & Werner, 2018; Sharma, Konetzka, & Smieliauskas, 2017). By 2011, a large proportion of nursing homes were rated as 4 or 5 stars. Using recent data would reduce the variability in quality among nursing homes, with the added concern that improvements in quality ratings may not reflect true changes in the quality of care. In contrast, the factors that we consider, such as distance, education, etc. have remained relatively unchanged in the nursing home population. Moreover, education, which is one of our key variables, was no longer collected when CMS implemented the MDS 3.0 in 2010. Thus, nursing home quality in 2009 presents us with the best data for our research question.

MDS data contain demographic and clinical information collected at regular intervals for all residents in Medicare and/or Medicaid certified nursing homes including information on sex, education, marital status, race, age at time of admission, and clinical characteristics. Following standard practice, we classify all residents with a 5-day Medicare admission assessment as those entering for a post-acute stay and all others for a long-term stay (He & Konetzka, 2015).

We use the overall star rating from Nursing Home Compare archives to classify nursing homes as low quality (1–2 stars) or high quality (4–5 stars). Nursing Home Compare includes three domains of quality that determine the 5-star rating system: clinical quality, staffing, and health inspections. Clinical quality measures are derived from MDS data provided by nursing homes and include outcomes like percent of residents with falls, pressure ulcers, delirium, and urinary tract infection. Self-reported staffing measures are based on case-mix-adjusted measures of total nursing hours per resident day and RN hours per resident day. Health inspections are based on the scope and severity of deficiencies found during Medicare/Medicaid recertification surveys and the number of repeat visits needed to confirm the correction of deficiencies. These surveys represent a thorough assessment of whether a nursing home is in compliance with regulations including medication management, environment, and quality of life. In December of 2008, CMS combined three quality domains into a composite 5-star rating, with health inspections carrying the most weight. The 5-star rating is the only composite measure of quality readily available nationally.

We use the Medicare Beneficiary Summary files to define dual status. Nursing home ZIP codes and occupancy rates at the time of inspection come from the OSCAR database, a compilation of facility-level characteristics collected by surveyors during the inspections. Finally, we use publicly available IRS tax returns data to derive average ZIP-code level income (IRS, 2016). Income levels are linked to MDS using the ZIP code of each resident’s prior primary residence.

Sample

Since our methods are complex and time-consuming with large samples, we draw a 20% random sample of all 2009 admissions for both chronic and post-acute care to Medicare-certified, freestanding nursing homes. We include only the first admission for each individual in 2009. We include individuals who selected facilities within 20 miles from the home ZIP code for urban areas and 40 miles for rural areas. From the remaining 516,565 individuals, we exclude 40,536 individuals who were younger than 65. After further excluding individuals with missing data on education (31,759) and other variables (30,597), we have data on 413,673 individuals. Our final analysis has 409,881 individuals with access to both low-quality (1–2 stars) and non-low-quality (3–5 stars) facilities within 20 miles for urban and 40 miles for rural areas.

Dependent Variables

Our dependent variable is a binary variable indicating whether an individual was admitted to a low-quality (1–2 star) nursing home.

Independent Variables

Duals:

The main resident-level characteristic of interest is a dichotomous variable indicating whether a resident is dually enrolled (“full duals”) in Medicare and Medicaid, identified using the dual status code of 02, 04, or 08 in the Medicare Beneficiary Summary files (Virnig, Skellan, O’Donnell, & Kane, 2011).

Education:

Using the MDS data, we classify education into three categories: less than a high school degree, high school graduate, and college graduate.

Differential Distance:

The ability and willingness of individuals to select a particular nursing home depends partly on the distance of the nursing home from their residence. People prefer to stay closer to their family and home (Konetzka & Perraillon, 2016; Pesis-Katz et al., 2013; Shugarman & Brown, 2006; Zwanziger et al., 2002). Since individuals might travel farther for a better quality facility, we construct a measure of distance (“differential distance”) that allows us to assess how far an individual has to travel for a better quality facility.

We use MapQuest API to obtain the driving distance between the centroids of ZIP codes of residents and all nursing homes within 20 miles for urban areas and 40 miles for rural areas. We define differential distance as the driving distance from a patient’s ZIP code to the nearest facility of a given quality type minus the driving distance to the nearest facility of all other quality types. For example, if the nearest non-low quality facility (3–5-star) is 10 miles away and nearest low-quality (1–2-star) facility is 5 miles away for a given resident, the differential distance between non-low quality facilities vs. low-quality facility for this resident would be 5 miles; it implies that this resident would have to travel 5 additional miles to get to a higher quality facility. A negative measure for differential distance would imply that a higher quality facility is the nearest facility for the resident.

Differential Occupancy Measures:

To capture the availability and willingness of facilities of different quality to accept residents with dual status, we calculate differential occupancy measures using the strategy applied earlier for differential distance. For instance, if the nearest non-low quality facility has Medicaid occupancy of 60% and nearest low-quality facility has Medicaid occupancy of 40% for a given resident, the differential Medicaid occupancy between non-low quality facilities vs. low-quality facility for this resident would be 20%. Data limitations preclude a true assessment of Medicaid bed availability, but these occupancy measures serve as a proxy for the availability and willingness of facilities with different quality types to accept residents with dual status.

Demographics:

The demographic variables included in our analysis are age at admission, sex, race, marital status, and urban/rural residence using Rural-Urban Commuting Area (RUCA) Codes (WWAMI Rural Health Research Center, 2017). RUCA incorporates population density, commuting patterns, and proximity to urban areas (Meilleur et al., 2013).

Health characteristics:

A patient’s admission to a particular nursing home is determined in part by his or her health condition, as some nursing homes specialize in particular needs. MDS 2.0 contains a rich array of relevant clinical characteristics including RUG-ADL scores (a scale for functional impairment), the cognitive performance scale (CPS) (Morris et al., 1994), and comorbidities. Although we model each of these as a separate variable in the regressions, we summarize the disparity explained by all health-related variables into one category.

Statistical Analysis

First, we summarize patient demographics, health characteristics, and differential distance and occupancy variables across all individuals as well as by dual status. We report counts and proportions for binary variables and means and standard deviations for continuous variables. Next, using Blinder-Oaxaca decomposition techniques (Blinder, 1973; Oaxaca, 1973), we estimate the contribution of each characteristic to the difference in admissions to low-quality nursing homes between duals and non-duals. The Blinder-Oaxaca decomposition technique has been used extensively in the health care literature (Bowblis & McHone, 2013; Grabowski & McGuire, 2009; Van de Poel & Speybroeck, 2009; Wagstaff, O’Donnell, Van Doorslaer, & Lindelow, 2007).

Blinder-Oaxaca decomposition techniques are based on multivariate regressions of the outcome variable on all observed potential determinants of the outcome (Jann, 2008). The outcome variable, Y, is a binary nursing home choice variable equal to 1 if an individual selected a low-quality facility. Let X represent a vector of independent variables including differential distance and education. We also have two groups of residents – duals and non-duals; with the indicator variable Duals equaling 1 if a resident is dually enrolled. β_pooled represents coefficients from a pooled regression that includes both duals and non-duals. Using a linear probability model, we can write the equation as follows:

$$Y_i=\delta Dual{s}_i+{\beta }_{\text{pooled }}{X}_i^{\prime }+{\epsilon }_i$$ (1)

The difference in the proportion of individuals selecting a low-quality nursing home between duals and non-duals is the difference in the expected values obtained using the estimated non-discriminatory coefficients from a pooled regression as the reference coefficients (Neumark, 1988). We can write the two-fold decomposition as follows (Jann, 2008):

$$Difference=E\left(Y_{duals}\right)-E\left(Y_{nonduals}\right)={\left[E\left(X_{duals}\right)-E\left(X_{nonduals}\right)\right]}^{\prime }{\beta }_{pooled}+\left[E{\left(X_{nonduals}\right)}^{\prime }\left({\beta }_{nonduals}-{\beta }_{pooled}\right)+E{\left(X_{duals}\right)}^{\prime }\left({\beta }_{pooled}-{\beta }_{{}_{duals}}\right)\right]$$ (2)

Using this approach, we disaggregate the difference in nursing home choice between duals and non-duals into the part that is attributable to the differences in prevalence of characteristics (“characteristics effect”) and the part attributable to the differences in the coefficients on those characteristics (“coefficients effect”). The characteristics effect (explained portion) is often the focus in studies of disparities because it is the amount of the disparity that could be eliminated if both groups had the same (average) characteristics. In contrast, the coefficient effects may capture the potential impact of unobserved variables, making the interpretation of the “coefficient effect” difficult. Our estimate of interest in equation 2, the “characteristics effect”, depends only on the differences in average characteristics (Table 1) and coefficients from the pooled regression model (Appendix Table A).

Table 1:

Summary Statistics by Dual Status

	Duals	Non-duals	Total
Number of observations	110,839	299,042	409,881
Demographic variables (n, %)
Female	76,779 (69.3%)	191,860 (64.2%)	268,639 (65.5%)
Education
No high school	38,910 (35.1%)	48,175 (16.1%)	87,085 (21.2%)
High school	64,315 (58.0%)	201,747 (67.5%)	266,062 (64.9%)
College graduate	7,614 (6.9%)	49,120 (16.4%)	56,734 (13.8%)
Marital status
Never married	14,714 (13.3%)	18,832 (6.3%)	33,546 (8.2%)
Married	22,602 (20.4%)	107,739 (36.0%)	130,341 (31.8%)
Widow/separated/divorced	73,523 (66.3%)	172,471 (57.7%)	245,994 (60.0%)
Race
American Indian/Alaskan Native	448 (0.4%)	576 (0.2%)	1,024 (0.3%)
Asian/Pacific Islander	3,526 (3.2%)	2,317 (0.8%)	5,843 (1.4%)
Non-hispanic black	18,617 (16.8%)	18,699 (6.3%)	37,316 (9.1%)
Hispanic	9,158 (8.3%)	5,766 (1.9%)	14,924 (3.6%)
Non-hispanic white	79,090 (71.4%)	271,684 (90.9%)	350,774 (85.6%)
Age
Age<75	30,546 (27.6%)	58,964 (19.7%)	89,510 (21.8%)
76<=Age<80	19,301 (17.4%)	52,094 (17.4%)	71,395 (17.4%)
81<=Age<=85	23,275 (21.0%)	72,844 (24.4%)	96,119 (23.5%)
86<=Age<89	17,718 (16.0%)	57,646 (19.3%)	75,364 (18.4%)
Age>=90	19,999 (18.0%)	57,494 (19.2%)	77,493 (18.9%)
Urban residence	83,994 (75.8%)	242,894 (81.2%)	326,888 (79.8%)
Average ZIP-code income ($; mean, SD)	49,163 (25,868)	57,162 (33,931)	54,999 (32,149)
Selected closest facility	53,058 (47.9%)	127,724 (42.7%)	180,782 (44.1%)
Selected health variables (n, %)
RUG-ADL score (mean, SD)	13.35 (3.9)	13.02 (3.6)	13.11 (3.7)
Cognitive performance scale (CPS)
0<=CPS<=2	57,356 (51.7%)	214,137 (71.6%)	271,493 (66.2%)
3<=CPS<=4	41,503 (37.4%)	70,478 (23.6%)	111,981 (27.3%)
5<=CPS<=6	11,980 (10.8%)	14,427 (4.8%)	26,407 (6.4%)
Alzheimers	9,098 (8.2%)	15,858 (5.3%)	24,956 (6.1%)
Other dementia	23,863 (21.5%)	48,415 (16.2%)	72,278 (17.6%)
Diabetes	45,462 (41.0%)	91,468 (30.6%)	136,930 (33.4%)
Peripheral vascular disease	14,012 (12.6%)	26,882 (9.0%)	40,894 (10.0%)
Aphasia	3,927 (3.5%)	6,377 (2.1%)	10,304 (2.5%)
Stroke	22,186 (20.0%)	39,749 (13.3%)	61,935 (15.1%)
Hemiplegia	9,082 (8.2%)	13,929 (4.7%)	23,011 (5.6%)
Parkinson’s disease	3,943 (3.6%)	10,463 (3.5%)	14,406 (3.5%)
Seizure disorder	4,941 (4.5%)	8,312 (2.8%)	13,253 (3.2%)
TIA	2,088 (1.9%)	7,069 (2.4%)	9,157 (2.2%)
Schizophrenia	3,969 (3.6%)	1,373 (0.5%)	5,342 (1.3%)
Asthma	5,222 (4.7%)	11,662 (3.9%)	16,884 (4.1%)
Chronic obstructive pulmonary disease	29,355 (26.5%)	61,770 (20.7%)	91,125 (22.2%)
Anemia	22,641 (20.4%)	65,695 (22.0%)	88,336 (21.6%)
Cancer	6,915 (6.2%)	26,738 (8.9%)	33,653 (8.2%)
Renal failure	9,065 (8.2%)	23,138 (7.7%)	32,203 (7.9%)
Access variables (mean, SD)
Differential distance (non-low quality to low-quality)	−0.6 (7.7)	−0.9 (6.9)	−0.9 (7.1)
Differential occupancy (non-low quality to low-quality)	−1.7 (20.9)	−1.5 (21.3)	−1.5 (21.2)
Differential Medicaid occupancy (non-low quality to low-quality)	−10.3 (28.9)	−11.8 (29.1)	−11.4 (29.1)

Notes:

^a.Differential distance is the distance to the closest facility of one quality type minus the distance to the closest facility of all other quality types.

^b.Differential occupancy is the minimum occupancy percentage at the closest facility with one quality type minus the minimum occupancy percentage at the closest facility of all other quality types. Total occupancy percentage equals 100 × (total number of residents/total beds in the facility) whereas Medicaid occupancy percentage equals 100 × (total Medicaid residents/total beds in the facility).

In addition to the above analyses on the full sample, we conduct several subgroup analyses. First, post-acute care residents and chronic-care residents differ on some of the key factors that may affect the decision to select a given nursing home: distance, length of stay, access to quality information, and out of pocket payments. A recent report found that location and access to transportation were critical factors in nursing home choice for post-acute care suggesting that distance might play a pivotal role (Levine & Ramos-Callan, 2019). Some evidence exists that the choices of nursing homes for post-acute care patients may depend on the discharging hospital and its relationship to specific nursing homes (Shield, Winblad, McHugh, Gadbois, & Tyler). Similarly, admission patterns for the elderly can differ for rural versus urban location (Cohen & Bulanda, 2016) and the importance of distance may be different for rural and urban areas. We create four distinct groups (urban post-acute care, urban non-post-acute care, rural post-acute care, and rural non-post-acute care) to analyze these two factors simultaneously.

Robustness Checks

We conduct several robustness checks to determine if our findings are sensitive to choices made in our analyses. First, we limit our analysis to residents in markets with the entire range of star ratings (at least one nursing home of each rating) to increase homogeneity in potential access to nursing homes of different quality. Residents may be more likely to travel farther from home for a meaningful increment in quality. Second, we employ a non-linear decomposition approach described by Fairlie (Fairlie, 2005) and use logit models to estimate the contribution of different characteristics to the disparity in admissions to low-quality facilities. Although more accurate given our dichotomous outcome, the non-linear version is less tractable than our preferred linear version. Finally, we evaluate the factors contributing to disparity in admissions to high-quality nursing homes. We define a facility to be a high-quality facility if it had 4–5 stars.

Results

We describe sample characteristics in Table 1, stratified by dual status. Approximately 27% of the 409,881 residents in our sample are duals. Duals are more likely to be minorities (blacks and Hispanics), less likely to have a high school degree, and less likely to be married than non-duals. A higher proportion of duals tends to have various comorbidities including Alzheimer’s, other dementia, and diabetes. Finally, 47.9% of duals and 42.7% of non-duals selected the facility closest to their residence.

Table 2 summarizes the unadjusted differences in the probability of admission to low-quality nursing homes for duals vs. non-duals. In the full sample, duals are 9.7 percentage points more likely than non-duals to be admitted to low-quality facilities. Our findings show that 35.4 percent of the difference in admission to low-quality homes can be explained by observable differences in characteristics of duals and non-duals. Duals are 7.7–10.8 percentage points more likely to be admitted to low-quality facilities in the subgroup analyses.

Table 2:

Disparity in Admission to Low-Quality Nursing Homes - Linear Decomposition Analysis

Patient groups	Duals	Non-duals	Difference*	Difference explained by differences in characteristics
Overall sample	50.7%	40.9%	9.7%	35.4%
Urban residents
Post-acute care	52.2%	41.5%	10.8%	41.7%
Non-post-acute care	49.8%	40.1%	9.7%	21.3%
Rural residents
Post-acute care	48.6%	40.6%	8.0%	38.0%
Non-post-acute care	45.7%	38.0%	7.7%	27.5%

Note:

^a.All the differences are significantly different from zero at 5% level.

Figure 1 shows the proportion of the disparity in admission to low-quality facilities between duals vs. non-duals attributable to characteristics of the populations obtained using differences in the prevalence of a characteristic (such as education) in each population (Table 1) and the estimated effect size of that characteristic on the choice of a facility (Appendix Table A). Most of the variables have expected coefficients; duals (vs. non-duals), residents living in low-income areas (vs. high-income areas), and blacks (vs. non-Hispanic whites) are more likely to be admitted to low-quality facilities. At the same time, there are some surprising relationships. Hispanics are less likely to be admitted to low-quality facilities than whites. Patients with certain health conditions like hemiplegia are less likely to be admitted to low-quality facilities; this may be because there are no low-quality facilities nearby that are able to cater to their needs. Using the coefficients and differences in the prevalence of characteristics, we can obtain the magnitude of disparity explained by differences in characteristics. In Figure 1, difference in education explains 4.9% of the disparity in low-quality nursing home choice: fewer duals have higher education and higher education is associated with a lower likelihood of being admitted to low-quality facilities. Thus, differences in education increase the disparity in admission to low-quality facilities. It is also possible that some factors narrow the disparity. In the case of gender, these two components combine to narrow the disparity in low-quality facility choice (hence, the negative contribution): duals are more likely to be women and women are less likely to be admitted to low-quality facilities. Health-related variables explain about 11.7% of the difference in admission to low-quality facilities. Of the explained difference in admission to low-quality facilities, about one-third is attributable to differences in education (4.9%) and distance (7.6%) combined. The results from the non-linear decomposition using the Fairlie approach shown in Figure 2 are consistent with our main findings.

Figure 1: Determinants of Nursing Home Quality Disparity between Duals and Non-duals – Linear Decomposition Analysis.

Note: As described in Equation 2, the percentage of disparity explained by the differences in characteristics is obtained using two components: differences in the prevalence of a characteristic in each population and the effect of that attribute on the choice of a nursing home. For example, in Figure 1 above, difference in education explains 4.9% of the disparity in low-quality (1–2 star) nursing home choice: fewer duals have higher education (Table 1) and higher education is associated with a lower likelihood of being admitted to low-quality nursing homes (Appendix Table A). In the case of gender, these two components combine to narrow the disparity in low-quality nursing home choice (hence, the negative percent explained): duals are more likely to be women and women are less likely to be admitted to low-quality nursing homes.

Figure 2: Determinants of Nursing Home Quality Disparity between Duals and Non-duals – Non-Linear Fairlie Decomposition Analysis.

Note: In the non-linear decomposition above, observed characteristics explain about 37.1% of the 9.7 percentage point difference in admission to low-quality facilities. Education and distance play a meaningful role in explaining the difference in admission to low-quality facilities for duals vs. non-duals.

Appendix Figure A shows results across subgroups. Observable characteristics explain a much higher percentage of the difference in admission to a low-quality facility in the post-acute care (PAC) group than in the non-PAC group in both urban and rural areas. Differences in educational attainment contribute less to the disparity in admissions to low-quality nursing homes in the rural areas than in the urban areas irrespective of whether the admission is a post-acute care admission. Distance is particularly important in rural areas, accounting for 16.7% of the total disparity in PAC admission to low-quality facilities. Results from our robustness checks are largely consistent with our main findings (Appendix Figures B & C). However, the role of education is much lower (1%) and the role of racial composition much higher (8.4%) in driving the disparity in admission to high-quality facilities, relative to the low-quality analysis.

Discussion

In this study, we find that a substantially higher proportion of duals are admitted to lower quality nursing homes compared with non-duals. A sizable portion of the disparity can be explained by differences in observable characteristics between duals and non-duals, where education and distance to facility, along with health-related factors, are key drivers.

Our findings are consistent with existing literature on nursing home supply. Studies have found that high-quality facilities tend to be located in affluent areas (Konetzka et al., 2015; Perraillon, Konetzka, He, & Werner). Because duals tend to reside in relatively less affluent areas, our finding that distance is an important determinant of the disparity in nursing home quality between duals and non-duals is not surprising. Similarly, duals are likely to reside in facilities with a higher share of Medicaid residents but facilities that have a higher share of Medicaid residents generally have higher deficiencies (Mor, Zinn, Angelelli, Teno, & Miller, 2004), implying lower quality.

The importance of education in driving the disparity in the quality of nursing homes between duals and non-duals points to several potential policy interventions. Education allows individuals to access and utilize quality information to make optimal decisions about where to seek care. However, decreasing disparities in formal education is a long-run process. Furthermore, there is evidence that individuals seek care where most residents are of the prospective resident’s race (Rahman & Foster, 2015). So, even with decreased disparities in formal education, preferences for race concordance may still lead to disparities in where people seek care. Still, interpreting education as a proxy for the ability to find and process quality information, policymakers may focus on measures that may compensate for disparities in formal education, such as awareness campaigns and assistance with nursing home choice. Because some people are not aware of the sources of quality information or do not trust it (Konetzka & Perraillon, 2016), awareness programs and efforts to increase usability of quality information can be beneficial. Education may also proxy for socioeconomic status, and addressing barriers associated with low education (for example, language) can help reduce the disparity.

The importance of distance (or supply) in nursing home choice suggests a different set of policy options. Our results show that a substantial portion of nursing home residents were admitted to the facility closest to their prior home. Policy interventions to increase the use of quality information may induce individuals to travel farther for a high-quality facility, but given strong preferences for proximity, this path is likely to have limited impact on the disparity. Instead of spending resources on inducing individuals to travel farther from home, our findings underscore the need to shift resources to meaningful quality improvement in low-quality facilities.

Finally, health-related factors contribute a substantial portion of the disparity in admission to low-quality facilities between duals and non-duals, but the policy implications are less straightforward. Although efforts can be made to nudge high-risk patients to select high-quality nursing homes, some of the decisions around the selection of a low-quality facility may be due to patient need – for example, a particular facility may not have the staff to cater to patients with dementia. This brings us to our broader policy suggestion that it is important to improve low-quality nursing homes in underserved areas in addition to incentivizing individuals to select high-quality nursing homes.

The results from our study are subject to several limitations. First, decomposition techniques are meant to provide the relative importance of different observable characteristics in explaining the disparity, but efforts to narrow the disparity via factors like education and distance will work only to the extent that they have a causal relationship. Second, although the differences in observable characteristics explain an important percentage of the disparity in access to low-quality facilities, a substantial percentage of the disparity remains unexplained. There may be unobserved factors that contribute to the disparity. However, it is not unusual that observable factors explain about half or less of the disparity (Bowblis & McHone, 2013). Third, differences in reimbursement rates may limit the ability of nursing homes to provide quality care even if duals have a choice. For instance, Medicaid rates are lower than Medicare or private-pay rates for nursing homes. We investigated this issue indirectly through Medicaid occupancy, but did not find a substantial contribution of Medicaid occupancy on the disparity. Fourth, it is possible that the contribution of education to the disparity in admission to low-quality facilities is underestimated in our study because awareness of the five-star system was still low in 2009. However, educated individuals may also be better at seeking out quality information about nursing homes from other sources. Finally, we focused on individual characteristics associated with nursing home selection in this study but the decision to select a particular nursing home is complex and may involve family members and other care providers.

Although existing studies have documented a disparity in admission to low-quality nursing homes between duals and non-duals, our study fills an important gap by quantifying the relative importance of observable characteristics to this disparity. Our finding that distance and education are meaningful drivers of the disparity points to several distinct but important policy directions. Efforts to increase awareness and usability of quality information may compensate for generally lower education levels among duals, but preferences for facilities in close proximity to family and friends may make the information less salient. Moreover, patient need may determine where individuals can go for nursing home care. Thus, quality improvement in nursing homes located in poorer neighborhoods is still essential. Policy initiatives to encourage quality improvement could do a better job of taking geography into account, directing resources to the neighborhoods where few alternatives exist. Finally, it is important to note that while the reasons for disparities in access to high-quality nursing homes at times seem self-evident, our findings indicate that much of the disparity is still unexplained. Even if policies to increase use of quality information and availability of high-quality nursing homes in poorer neighborhoods were well funded and implemented, the disparity would likely be reduced and not eliminated. The need for policymakers and researchers to unearth additional strategies to reduce disparities will continue.