Association between Residential Segregation and Long-Term Acute Care Hospital Performance on Improvement in Function among Ventilated Patients

To the Editor:

One in five survivors of critical illness who require prolonged mechanical ventilation are discharged to long-term acute care hospitals (LTCHs) (1). Although disparities based on race and insurance have been described in LTCH use (2), studies have not evaluated equity in outcomes. Improvement in function (e.g., mobility) is a crucial recovery goal for patients requiring prolonged mechanical ventilation (3, 4). Research in other areas has suggested that Black patients disproportionately receive care in lower-performing facilities (5, 6). Whether LTCHs serving communities with more segregation achieve lower gains in functional outcomes is unknown.

In response to the Improving Medicare Post-Acute Care Transformation Act, the Centers for Medicare and Medicaid Services mandated reporting of change in mobility among ventilated patients as an LTCH quality measure (7). We sought to examine the association between the racial composition of the neighborhood and county of an LTCH and performance on the functional mobility improvement measure.

Methods

We used the LTCH Care Compare data set reporting information from 2018 to 2019. We merged facility data with the 2018 Agency of Healthcare Research and Quality Social Determinants of Health database at zip code and county levels. Description of the change in mobility measure has been previously reported (7). Briefly, ventilated patients were identified by facilities using manually entered codes for item O0150A on the Patient Assessment Form, collected as part of the LTCH Continuity Assessment Record and Evaluation data set. Mobility was assessed for eight activities (rolling, sitting-to-lying, lying-to-sitting, sitting-to-standing, bed-to-chair, toilet transfer, walking 50 ft with two turns, and walking 150 ft), on a scale of 1–6, with higher scores indicating higher functional status. Item scores were added to create a mobility score (range, 8–48). The difference between mobility scores at discharge and admission was averaged across qualifying patient stays to calculate facility-level average observed change in mobility score. The mean of observed change in mobility score across all patient stays nationally was used to calculate the national average change in mobility score. The average facility-level expected change in mobility score was then calculated using risk adjustment for patient factors including age; moderate to severe communication impairment; prior functioning in indoor ambulation; prior assistive device use; primary medical condition (categorized as chronic respiratory condition, acute-onset and chronic respiratory conditions, chronic cardiac condition, and other medical condition); stage 3, 4, or unstageable pressure ulcer; and comorbidities including severe and metastatic cancers, dialysis and stage 5 chronic kidney disease, diabetes mellitus, major lower limb amputation, stroke, hemiplegia or hemiparesis, dementia, paraplegia, incomplete tetraplegia, and other spinal cord disorder/injury. The difference between facility-level average observed and expected change in score was added to the national average change in mobility score to obtain the facility-level risk-adjusted average change in mobility score reported in the LTCH Care Compare data set.

Our primary outcome was the facility-level risk-adjusted average change in mobility score. We constructed multivariable linear regression models with the percentage of residents in the neighborhood reporting Black race to measure racial composition at the micro (zip code) level and dissimilarity index (DI) to measure segregation at the macro (county) level as primary predictors (8). Community racial composition and segregation were chosen as exposures of interest because they have been shown to reflect racial distribution within facilities (8, 9). DI is a measure of the unevenness with which Black and White individuals are distributed across census tracts in a county. The score, ranging from 0 (complete integration) to 100 (complete segregation), can be interpreted as the percentage of either Black or White residents that would have to move to different geographic areas to produce a distribution that matches that of the larger area (10).

We tested for significance of variation in primary exposures and outcome by region using one-way analysis of variance. In our multivariable model, we adjusted for facility bed size, ownership, region, rurality, and volume of ventilated patient stays. We did not adjust for neighborhood-level socioeconomic characteristics given their potential to mediate, and hence potentially yield inaccurate estimates of the observed effects of racial segregation, in the absence of a targeted mediation analysis (11).

Results

Of 365 LTCHs included in the Care Compare data set, 311 (85.2%) reported the change in mobility measure. The median number of LTCH beds was 49 (interquartile range [IQR], 35–70). Approximately half (51%) were in the South compared with 10% in the Northeast, 22% in the Midwest, and 16% in the West; 4% were in rural areas. Most LTCHs (70%) were for-profit. The median volume of ventilated patient stays was 83 (IQR, 49–144). Median risk-adjusted change in mobility score was 8.1 (IQR, 6.0–10.1), ranging from 11.1 in zip codes with the lowest percentage of Black individuals (0.04%) to 4.2 in those with the highest percentage of Black individuals (91.8%) and from 7.7 in counties with the lowest DI (18.5) to 4.2 in those with the highest DI (78.6). Both primary exposures varied significantly by region, with the median percentage of Black individuals in zip codes ranging from 4.6 (IQR, 2.2–8.7) in the West to 19.1 (IQR, 7.7–33.3) in the South (P < 0.0001) and the DI ranging from 46.4 (IQR, 41.5–57.1) in the West to 57.8 (IQR, 51.6–67.3) in the Northeast (P < 0.0001). The outcome measure, change in mobility score, was not significantly associated with region, with the median score ranging from 8.7 (IQR, 6.4–10.5) for LTCHs in the South to 7.3 (IQR, 5.3–9.1) in the Midwest (P = 0.14).

In the multivariable model at neighborhood level, each 10% increase in Black individuals was associated with a 0.4 lower change in mobility score (coefficient, −0.40; 95% confidence interval [CI], −0.58 to −0.22). At the county level, an increase in DI by 10 points was associated with a 0.4 lower mobility improvement score (coefficient, −0.43; 95% CI, −0.78 to −0.09). Predictive margins showed that an increase in the percentage of Black individuals from 10% to 80% and in DI from 0 to 80 was associated with a lower gain in mobility by one-third (Figures 1 and 2).

Figure 1.

Predictive margins for facility-level risk-adjusted improvement in mobility score against percentage of residents reporting Black race in the zip code of a long-term acute care hospital (LTCH) from the multivariable linear regression model. Covariates in the model included total number of beds, geographic region, rural location, ownership status of the facility, and volume of ventilated patients. LTCHs located in zip codes with 20% residents reporting Black race can be expected to have a change in mobility score closest to the national average (8.2) with consistent decline in scores at increasing proportion of Black residents in the neighborhood.

Figure 2.

Predictive margins for facility-level risk-adjusted improvement in mobility score against Dissimilarity Index of the county of a long-term acute care hospital (LTCH) from the multivariable linear regression model. Covariates in the model included total number of beds, geographic region, rural location, ownership status of the facility, and volume of ventilated patients. LTCHs located in a county with complete integration (Dissimilarity Index of 0) can be expected to have a change in mobility score of 10.5 as compared with 6.2 for one in a county with complete segregation (Dissimilarity Index of 100).

Discussion

LTCHs located in neighborhoods with a greater proportion of Black individuals and in segregated counties achieved less improvement in a multicomponent measure of mobility. Those serving communities with the highest proportion of Black individuals and highest levels of segregation accomplished a one-third lower gain in mobility than those with the lowest levels of these indices. This finding was independent of ownership status and volume of ventilated patients, known determinants of LTCH outcomes on mortality (12, 13).

Our findings are consistent with prior work demonstrating the association of community racial composition with quality of care in dialysis centers (6), acute care hospitals (14), and nursing homes (8). The proportion of Black individuals and the DI reflect racial segregation at different spatial levels, and the association of poor outcomes with both in our study highlights structural inequity in the performance of LTCHs in both minority-predominant and segregated neighborhoods. A potential explanation for lower performance of these facilities could be structural differences in leadership, staffing, equipment, and protocols known to influence LTCH outcomes (13). This could have significant implications for patients treated at LTCHs in predominantly Black and segregated neighborhoods, who can be expected to make lower gains in functional status at discharge.

Our study has several limitations. First, we did not have information about patient race to measure racial composition at the facility level. However, prior work has demonstrated high correlation between community and facility segregation (8), and the use of neighborhood indices is supported by the conceptual model of LTCH transfer in which clinicians and LTCHs are guided by proximity (1). Second, we could not assess patient factors beyond those included in the risk-adjustment model. It is possible that unmeasured differences in illness severity of patients admitted to LTCHs in Black-predominant neighborhoods could have influenced potential for functional improvement. Third, we could not account for characteristics of upstream, transferring acute-care hospitals that might result in differences in in-hospital treatment, influencing subsequent rehabilitation potential. A multilevel modeling approach, clustering patients by transferring hospitals, would add to our insight of how these hospital characteristics could be contributing to the observed variation in our outcome. Fourth, although the DI, one of our primary exposures, assesses segregation between Black and White individuals, it does not account for segregation between Hispanic individuals, those of other races, and non-Hispanic White individuals. It is possible that performance measures vary differently in the context of other racial/ethnic segregation indices and should be investigated in future studies.

In conclusion, our finding that LTCHs located in Black-predominant and segregated communities achieve less improvement in function for mechanically ventilated patients raises concerns about differences in quality of care delivered by LTCHs. Future research using a multilevel approach to investigate individual- and facility-level differences is needed to identify potential mechanisms including differences in resources or practices at the LTCHs or at the upstream, transferring hospitals to help develop effective interventions to eliminate disparities in functional improvement among the chronically critically ill.