Persistent geographic variations in availability and quality of nursing home care in the United States: 1996 to 2016

Nursing home data consist of the name and characteristics of each Medicaid/Medicare-certified nursing home from 1996 to 2016, including the number of Medicaid/Medicare-certified beds, number of residents in these beds, and various nursing home characteristics. The 1996 data were the most complete and available earliest data at the national level, publicly available from Nursing Home Compare (https://​healthdata.​gov/​dataset/​nursing-home-compare-data). To ensure that the analysis was based on a relatively long period and accounted for the fact that nursing homes may have consolidated during the 21-year study period, we required all nursing homes to have at least 7 years of data for inclusion in the study. This restriction ensured that each nursing home contributed at least 33% of data points over the study period. The completed 5-star rating data were available from 2011, and we required nursing homes to have at least 5 years of data to be included in the analysis. This restriction ensures that each nursing home was evaluated based on a relatively long period.

Our primary outcomes were the availability of all nursing home care and availability of 5-star nursing home care. The availability of all nursing home care was defined as the number of Medicaid/Medicare-certified nursing home beds available for a given county per 100,000 population aged ≥65 years, regardless of nursing home star rating. We selected the age group ≥65 years because the majority of nursing home residents are in this range. We obtained the estimated county-specific population data from the annual American Community Survey (https://​www.​census.​gov/​programs-surveys/​acs/​about.​html), conducted by the U.S. Census Bureau. The availability of 5-star nursing home care was defined as the number of Medicaid/Medicare-certified nursing home beds available for a given county per 100,000 population aged ≥65 years, restricted to 5-star nursing homes. CMS uses its Five-Star Quality Rating system [13] to measure nursing home performance. The rating system includes 3 indicators: health inspections, staffing hours, and quality measures (online Additional file 1: Text S1); each of these indicators has its own star rating, ranging from 1 to 5, with 5 stars representing the top performance level for that indicator. The composite of these ratings provides an overall rating, ranging from 1 to 5 stars, which was used in this study. Additionally, we examined the trend in the utilization of nursing home beds, defined as the rate of the occupied Medicaid/Medicare-certified beds among the total Medicaid/Medicare-certified beds.

Nursing home characteristics included ownership (not-for-profit, yes/no), nursing home location (inside a hospital, yes/no), years of Medicaid or Medicare certification, nursing home size (certified beds <25th, 25th–75th, and > 75th percentiles), change in ownership (changed in last 12 months, yes/no), nursing home geographic location (rural, yes/no), and resident council (both resident and family, yes/no). County characteristics included Consumer Price Index-adjusted median income, proportions of non-Hispanic white, non-Hispanic-black, Hispanic, female, aged 65 years or older, under national poverty level, Supplemental Nutrition Assistance Program participants, households without a car and low access to stores, and seniors with low income and low access to stores. Health risk factors and lifestyle included prevalence of age-adjusted adult diabetes and obesity, number of recreation or fitness facilities per 1000 population, number of fast-food restaurants per 1000 population, and physical inactivity. All information was available from the U.S. Census Bureau and the Centers for Disease Control and Prevention websites. We also included the 2013 county-specific average age-sex-race-adjusted nursing home Medicare reimbursement per Medicare beneficiary residing in a given county, available from the Dartmouth Atlas (http://​www.​dartmouthatlas.​org/​tools/​downloads.​aspx#spending).

To assess geographic variation in the availability of all nursing home care, we calculated county-specific availability for each county and then mapped the availability in 1996 and 2016, shading counties with a gradient from red to green (lowest availability in red to the highest in green). We divided states into 9 census divisions based on the U.S. census definition [27]. To assess whether geographic variation in availability persisted over time, we calculated a Pearson correlation coefficient of the availability between 1996 and 2016, weighted by county-specific population. We also fitted a mixed model with a Poisson link function and county-specific random intercepts to model the number of certified beds, regardless of nursing home rating, as a function of an ordinal time variable (time = 0 for 1996 and time = 21 for 2016), to assess the trend in availability of all nursing home care from 1996 through 2016, adjusted for county-specific characteristics. The risk ratio of the time variable represents the annual change in availability. We repeated these analyses to assess the geographic variation and trend in the availability of 5-star nursing home care, with restricted the data from 2011 to 2016.

To assess the relationship between availability of all nursing home care and availability of 5-star nursing home care, we modeled the availability of all nursing home care as a function of 5-star nursing home care, adjusted for county-specific characteristics and time and stratified by regions. To identify county-specific factors associated with the availability of all nursing home care, we restricted the study sample to 2016, the most recent year of the study period, and fitted a mixed model to regress the availability of all nursing home care as a function of county-specific characteristics. Using the 2016 data, we also assessed the nursing home characteristics associated with the 5-star rating (yes/no). We included a spherical covariance structure in all models to account for spatial autocorrelation and differences between counties and nursing homes. The county-specific population aged ≥65 years was used as an offset in models with a Poisson link function. Analyses were conducted using SAS version 9.4 64-bit (SAS Institute Inc., Cary, North Carolina).

The study sample included 17,875 unique Medicaid/Medicare-certified nursing homes over the 21-year study period. In 2016, the median (inter-quartile-range [IQR]) number of Medicaid/Medicare-certified years of a nursing home in business was 26 (20–36), 30.4% of nursing homes were not-for-profit ownership, 4.4% were in a rural area, and 93.4% were certified by both Medicaid and Medicare (online Additional file 1: Table S1).

Nationwide, the number (standard deviation [SD]) of Medicaid/Medicare-certified beds per 100,000 population aged ≥65 years declined from 4882 (931) in 1996 to 3480 (912) in 2016 (Fig. 1; online Additional file 1: Figure S1, top panel). Considerable geographic variation in the overall availability of nursing home care was observed. The county-specific number of Medicaid/Medicare-certified beds, per 100,000 population aged ≥65 years, ranged from 1049 to 7900 (difference 6851) in 1996 and from 70 to 6725 (difference 6655) in 2016 (Fig. 2). Availability in the West and East regions was lower than in the Central regions in both 1996 and 2016, and was higher in the West-South Central region (Fig. 2). The weighted Pearson correlation coefficient of county-specific availabilities between 1996 and 2016 was 0.78 (95% Confidence Interval (CI) 0.77–0.79), indicating that the geographic pattern of availability persisted over time. The county-specific characteristics adjusted annual decline in availability of all nursing home care, regardless of star rating, was 0.5% (95% CI 0.44–0.55). This decline was consistent across all census divisions but was most pronounced in the Pacific division, from 4064 beds per 100,000 population aged ≥65 years in 1996 to 2111 in 2016 (difference 1953), and least pronounced in the West-Central division, from 5296 per 100,000 population aged ≥65 years in 1996 to 4538 in 2016 (difference 758; Fig. 3, top panel). At the state level and in 2016, the 5 states with the highest mean (SD) availability of 5-star nursing home care were Texas (4340 [50]), Missouri (4263 [89]), Louisiana (4259 [593]), Oklahoma (4175 [107]), and Iowa (4114 [58] (Fig. 4, top panel). The 5 states with lowest availability were Hawaii (1500 [435]), Oregon (1613 [59]), Arizona (1748 [195]), California [1822 [61]), and Washington (2139 [98] (Fig. 4, top panel).

Between 2011 and 2016, the proportion of 5-star nursing homes increased from 16.2 to 24.0% (p < 0.01 for trend; online Additional file 1: Figure S2). The number (SD) of certified beds provided by 5-star nursing homes increased from 658 (303) in 2011 to 895 (661) in 2016, per 100,000 population aged ≥65 years (Fig. 5). Considerable geographic variation in the availability of 5-star nursing home care was observed. The county-specific number of Medicaid/Medicare-certified beds, per 100,000 population aged ≥65 years, provided by 5-star nursing homes ranged from 0 to 2749 (difference 2749) in 2011 and from 0 to 3278 (difference 3278) in 2016 (Fig. 5). The weighted Pearson correlation coefficient of the availabilities in 5-star nursing home care between 2011 and 2016 was 0.54 (95% CI 0.51–0.56), indicating moderately persistent geographic variation in availability of 5-star nursing home care. The county-specific characteristics adjusted annual increase in the availability of 5-star nursing home care was 9.5% (95% CI 8.41–10.64). At the state level and in 2016, the 5 states with the highest mean (SD) availability of 5-star nursing home care were North Dakota (2387 [88]), South Dakota (2124 [59]), Nebraska (2029 [61]), Iowa (1907 [27]), and Minnesota (1581 [33] (Fig. 4, bottom panel). The 5 states with lowest availability were Oregon (344 [10]), West Virginia (399 [7]), North Carolina (400 [5]), Texas (431 [5]), and Arizona (450 [48]) (Fig. 4, bottom panel).

The relationship between availabilities of all nursing home, regardless of star rating, and 5-star nursing home care varied by regions (online Additional file 1: Figure S3). Accounting for county characteristics and time, the relationship between all and 5-star nursing home care was not significant for the East-North Central, Middle-Atlantic, Mountain, New England, and Pacific regions, while the South-Atlantic region was positive and the West-North Central was negative. In 2016, the weighted Pearson correlation coefficients for county-specific availabilities between all nursing home and 5-star nursing home care was 0.33 (95% CI 0.30–0.36), indicating that the availability of care and quality of care do not align. In 2016, only one state (Iowa) remained in the top 5 states with highest availability of nursing home care, regardless of star rating, as well as in the top 5 states with the highest availability of 5-star nursing home care.

County geographic location is associated with availability of nursing home care. The availability for counties in the West South Central and West North Central regions was higher than for counties in other regions (online Additional file 1: Table S2). Counties with greater inactivity, more fast-food restaurants, more rural areas, and higher proportions of non-Hispanic white and non-Hispanic black populations were more likely to have higher availability of all nursing home care. Counties with higher income, higher proportions of people with diabetes mellitus, Supplemental Nutrition Assistance Program participants, and women were more likely to have lower availability of all nursing home care (online Additional file 1: Table S2). The age-sex-race-adjusted Medicare reimbursement dollars paid to a nursing home was associated with an increase in the availability of nursing home care (rate ratio 1.05, 95% CI 1.03–1.07). Nursing homes with a council joined by family and residents, with certified beds in the <75th percentile of the national average (130 beds in 2016) and with a high utilization of beds, were more likely to be 5-star. Nursing homes with for-profit ownership, located in a rural area, and in which ownership changed in the last 12 months were unlikely to be 5-star (online Additional file 1: Figure S4).

Between 1996 and 2016, the utilization of certified beds declined from 78.5 to 72.2% (Fig. 1; online Additional file 1: Figure S1, bottom panels). This decline was consistent across all census divisions, but was most pronounced in the Mountain division, from 86.7% in 1996 to 71.8% in 2016 (difference 14.9 percentage points), and less in the South-Atlantic division, from 89.9% in 1996 to 85.7% in 2016 (difference 4.2 percentage points; Fig. 3, bottom panel).