Disparities in post–acute rehabilitation care for joint replacement

Authors


Abstract

Objective

To determine the extent to which demographic and geographic disparities exist in the use of post–acute rehabilitation care (PARC) for joint replacement.

Methods

We conducted a cross-sectional analysis of 2 years (2005 and 2006) of population-based hospital discharge data from 392 hospitals in 4 states (Arizona, Florida, New Jersey, and Wisconsin). A total of 164,875 individuals who were age ≥45 years, admitted to the hospital for a hip or knee joint replacement, and who survived their inpatient stay were identified. Three dichotomous dependent variables were examined: 1) discharge to home versus institution (i.e., skilled nursing facility [SNF] or inpatient rehabilitation facility [IRF]), 2) discharge to home with versus without home health (HH), and 3) discharge to an SNF versus an IRF. Multilevel logistic regression analyses were conducted to identify demographic and geographic disparities in PARC use, controlling for illness severity/comorbidities, hospital characteristics, and PARC supply. Interactions among race, socioeconomic, and geographic variables were explored.

Results

Considering PARC as a continuum from more to less intensive care in regard to hours of rehabilitation per day (e.g., IRF→SNF→HH→no HH), the uninsured received less intensive care in all 3 models. Individuals receiving Medicaid and those of lower socioeconomic status received less intensive care in the HH versus no HH and SNF versus IRF models. Individuals living in rural areas received less intensive care in the institution versus home and HH versus no HH models. The effect of race was modified by insurance and by state. In most instances, minorities received less intensive care. PARC use varied by hospital.

Conclusion

Efforts to further understand the reasons behind these disparities and their effect on outcomes are needed.

INTRODUCTION

Although racial, socioeconomic, and geographic disparities in the use of joint replacement procedures in the US have been well documented (1–3), little is known about the extent of disparities in the use of post–acute rehabilitation care (PARC) following joint replacement. PARC is provided by therapists and other rehabilitation professionals to functionally impaired individuals following an acute hospitalization and has been shown to be effective in improving function and accelerating recovery following joint replacement (4, 5). PARC is typically delivered in a skilled nursing facility (SNF), inpatient rehabilitation facility (IRF), the patient's home, or in an outpatient setting.

Few studies have specifically examined disparities in the use of PARC following joint replacement. In a multivariate analysis of Medicare data, FitzGerald et al reported significant regional variation in the use of home health (HH) for Medicare beneficiaries following joint replacement, controlling for patient, institutional, and PARC supply characteristics (6). Data from Canada also indicate geographic and sex differences in the use of HH versus institutional care following joint replacement (7, 8). In a descriptive analysis of Medicare data, Buntin reported that nonwhites were more likely to use IRF care, while whites were more likely to use SNF care following hip or knee joint replacement (9).

Ottenbacher et al analyzed 1994–1998 data from the Uniform Data System for Medical Rehabilitation (UDSMR) on patients who received PARC in IRFs after lower extremity joint replacement (10). These authors reported that a lower percentage of Hispanic patients received this type of care relative to white, black, and Asian patients. Insurance status did not moderate this relationship. They also reported that blacks were more likely to receive outpatient or HH therapy following discharge from the IRF. One limitation of their study was that only patients discharged to IRFs and to facilities participating in the UDSMR were examined. Furthermore, this study was conducted prior to Medicare's implementation of prospective payment systems (PPS) in all post–acute care settings (10). Some studies suggest that the use of PARC has decreased in response to PPS, and that these reductions are differentially larger for certain subgroups (e.g., the elderly, women, patients receiving state assistance) (9, 11–13).

The objectives of our study were to use hospital discharge data to determine the extent to which demographic, socioeconomic, and geographic disparities exist in the use of PARC following lower extremity joint replacement, as well as to identify factors that may contribute to these disparities. This study extends previous research by using population-based data on both Medicare and non-Medicare patients in the years following implementation of PPS in post–acute care and by examining the use of different types of PARC following discharge from the acute care setting.

MATERIALS AND METHODS

Research design and data sources.

We conducted a cross-sectional analysis of 2 years of population-based hospital discharge data (2005 and 2006) from short-term acute care hospitals in 4 demographically and geographically diverse states (Arizona, Florida, New Jersey, and Wisconsin). Records on patients admitted for a hip or knee joint replacement were identified based on International Classification of Diseases, Ninth Revision (ICD-9) codes. Because community and hospital factors affect utilization of post–acute care, we merged these discharge data with hospital, zip code, and county-level data to create the final analytic data set.

State Inpatient Databases (SIDs).

Our primary source of data was the SIDs, part of the Healthcare Cost and Utilization Project sponsored by the Agency for Healthcare Research and Quality. The SIDs contain discharge abstracts from the universe of all inpatient stays at short-term general hospitals in participating states (n = 38). A core set of clinical (e.g., ICD-9-Clinical Modification [CM] diagnosis and procedure codes) and nonclinical data elements (e.g., insurance, age, sex, discharge disposition) are included and coded in a uniform format across all SIDs. SIDs from some states also include other data elements (e.g., race, ethnicity, patient zip code). We selected Arizona, Florida, New Jersey, and Wisconsin as the study sample based on availability of data elements critical to our objectives and to obtain representation from all 4 US census regions.

Hospital-level data.

Hospital characteristics were obtained from multiple sources, including the American Hospital Association (AHA) Annual Survey Database (2005) and the Centers for Medicare and Medicaid Services (CMS) Provider of Services (POS) Files and Hospital Cost Reports. The AHA database, produced annually, includes information on the organizational structure of the hospital, facilities, and services available, utilization of services, and staffing. CMS POS files are created quarterly from the Online Survey and Certification Reporting System database and characterize various types of institutional providers, including hospitals. The POS files (last quarter of 2005) were used to obtain data on rehabilitation staffing (i.e., physical and occupational therapists [PTs and OTs, respectively]). Hospital cost report data contain provider characteristics and utilization data. The hospital cost report data from 2005 were used to determine whether the hospitals in our sample had an inpatient rehabilitation department/facility and/or an affiliated HH agency.

Zip code and county-level data.

We used the 2005 Demographic Update of the Census 2000 data, conducted annually by Claritas (14), to obtain zip code–level data on the socioeconomic status (SES) of our sample subjects. Data were linked based on the zip code of the patient's residence. We used the National Center for Health Workforce Analysis 2005 Area Resource File (15) to obtain county-level measures of PARC supply based on county residence of the patient.

Sample and data set creation.

Our analysis was limited to individuals age ≥45 years, who had a hip or knee joint replacement procedure performed in a short-term, acute care hospital in one of the study states. Joint replacement procedures were identified based on primary ICD-9-CM procedure codes (81.51 for total hip replacement; 81.52 for partial hip replacement; 81.53 for revision of hip replacement; 81.54 for total knee replacement; and 81.55 for revision of knee replacement). The creation of our final sample is outlined in Figure 1. We excluded individuals who died during their inpatient stay, were transferred to hospice care, or who were transferred to another short-term facility. We also excluded individuals who had a primary diagnosis of hip fracture, were missing race data, or who resided outside the study states. Our final sample consisted of 164,875 records from 392 hospitals. These data were merged with the hospital, zip code, and county- level data to create our final analytic data set.

Figure 1.

Creation of final sample. SNF = skilled nursing facility; IRF = inpatient rehabilitation facility; AZ = Arizona; FL = Florida; NJ = New Jersey; WI = Wisconsin.

Conceptual model for analyses.

The conceptual model for our analyses is presented in Figure 2. We hypothesized that a number of factors at the individual, hospital, community, and state levels influence the post–acute care of an individual following joint replacement. Figure 2 also illustrates our hypothesis that individual and hospital factors, which are more proximal to the patient, have a stronger impact on use of post–acute care. The conceptual model guided our selection of variables to include in the analyses. Previous literature and the availability of variables in the data sets also affected our final selection of variables.

Figure 2.

Conceptual model.

Dependent variables.

We created 3 dichotomous dependent variables for our analyses: 1) whether the subject was discharged to home or to an institution, 2) whether the subjects discharged home received HH, and 3) whether subjects discharged to an institution were discharged to an SNF or an IRF. These variables were created from the DISPUB92 variable in the SIDs, which is based on the patient status data element in the UB-92 claim form.

Independent variables.

Our primary variables of interest were race, sex, SES, urban/rural status, and state. We coded the subject's race as white, black, or Hispanic. Race and ethnicity are coded as 1 data element in the SIDs, with ethnicity taking precedence over race (16). Therefore, white and black are non-Hispanic white and non-Hispanic black, respectively. Due to small sample sizes, records with another or missing race were excluded (Figure 1). The patient's SES was represented by 2 variables: insurance (uninsured, Medicaid, or Medicare/private) and median household income in the patient's zip code, a proxy for the patient's income (17). We followed the National Center for Health Statistics urban/rural classification scheme and categorized the counties in which the patients resided as large metropolitan (central or fringe counties with a population ≥1 million), medium/small metropolitan (counties with populations of 50,000–999,999), or micropolitan/nonmicropolitan (populations <50,000) (18). Finally, we included indicator variables for the state in which the patient resided.

To account for clinical factors that may influence discharge, we included the following control variables: length of stay, type of replacement (hip or knee), whether the procedure was a revision, whether the procedure was routine/elective versus an emergency, and illness severity and mortality risk based on the All Patient Refined Diagnostic Related Groups (APR-DRG) scoring system (19). The APR-DRG measures are based on secondary diagnoses and interactions between these diagnoses and age, principal diagnosis, and selected procedures (19). Each measure has 4 levels (minor, moderate, major, and extreme). Based on the distribution of our data, we collapsed severity into 3 categories (minor, moderate, or major/extreme) and the mortality into 2 categories (minor and moderate to extreme). APR-DRG measures have been shown to correlate with the functional abilities of individuals following hip or knee joint replacement (20). Comorbidity measures, derived from the list of 29 measures by Elixhauser et al (21), were also included. We limited the comorbidities to those that had support for inclusion based on the literature (22, 23) and that were significantly related to our outcomes in preliminary analyses. We included 7 specific comorbidity indicators (chronic pulmonary disease, congestive heart failure, diabetes mellitus, neurologic disorder, obesity, osteoarthritis, and peripheral vascular disease), as well as an indicator for individuals with ≥3 comorbidities. Descriptive data on the characteristics of the sample are presented in Table 1.

Table 1. Demographic, geographic, and clinical characteristics of the study sample (n = 164,875)*
 Entire sampleHome (n = 89,648, 54.6%)Institution (n = 75,227, 45.6%)
Home (42.1%)HH (57.9%)IRF (35.1%)SNF (64.9%)
  • *

    Values are the percentage unless otherwise indicated. HH = home health; IRF = inpatient rehabilitation facility; SNF = skilled nursing facility; APR-DRG = All Patient Refined Diagnostic Related Groups.

  • By discharge status. All comparisons significantly different (P < 0.01) unless indicated.

  • Based on patient zip code.

  • §

    Based on patient county of residence.

  • Based on 2006 data only. In 2005, the Florida State Inpatient Database did not distinguish between IRF and SNF discharges.

  • #

    No difference in proportion discharged to institution versus home.

  • **

    No difference in proportion discharged to home with HH versus no HH.

  • ††

    No difference in proportion discharged to IRF versus SNF.

Demographic/geographic     
 Women60.352.954.066.469.0
 Race     
  White90.893.991.986.890.0
  Black4.72.83.97.25.7
  Hispanic4.53.44.26.14.5
 Age, mean ± SD years68.1 ± 10.264.6 ± 9.765.7 ± 9.670.3 ± 9.971.8 ± 9.7
 Insurance     
  Private or Medicare97.896.697.798.598.1
  Medicaid1.52.41.41.21.4
  None/self-pay0.71.00.90.30.5
 Median HH income     
  Highest quartile25.723.222.940.829.0
  Quartile 327.731.728.024.526.7
  Quartile 226.929.827.718.425.9
  Lowest quartile19.715.421.416.329.0
 Patient residence§     
  Micropolitan, nonmetro area12.524.58.34.612.5
  Medium/small metro area36.438.042.528.830.1
  Large metro area51.137.549.366.657.4
Patient state     
  Arizona16.018.721.713.313.4
  Florida46.825.864.220.837.6
  New Jersey15.74.54.953.026.6
  Wisconsin21.651.19.212.922.4
Clinical     
 Routine admission93.493.895.296.490.2
 Length of stay, mean ± SD days3.8 ± 2.03.5 ± 1.83.7 ± 1.73.8 ± 2.14.0 ± 2.3
 Replacement type     
  Knee67.572.367.466.664.6
  Hip32.527.732.633.435.5
 Revision procedure4.1#4.0**4.2**3.2††3.4††
 APR-DRG severity measure     
  Minor42.250.245.336.737.8
  Moderate46.241.445.049.750.0
  Major/extreme11.68.49.713.612.2
 APR-DRG mortality risk     
  Minor80.188.385.574.172.0
  Moderate-extreme19.911.814.525.928.0
 Comorbidities     
  Obesity12.3#14.410.713.712.5
  Diabetes mellitus16.914.5**14.8**19.120.2
  Osteoarthritis93.293.8**93.6**93.793.0
  Chronic pulmonary disease14.813.4**13.8**15.6††16.2††
  Neurologic disorder2.41.6**1.7**3.4††3.4††
  Peripheral vascular disease2.41.71.93.3††3.3††
  Congestive heart failure2.61.6**1.5**3.44.1
  ≥3 comorbidities26.319.522.431.433.0

We included several hospital variables as proxies for the quality of care. These included joint procedure volume (i.e., number of procedures averaged across the 2 years) (24, 25), whether the hospital had a major medical school affiliation (26, 27), registered nurse (RN) full-time equivalents (FTEs) per 100 admissions (28, 29), and PT and OT FTEs per 1,000 admissions. We reasoned that the latter variable may be important since the input of the OT and PT and their time spent with the patient may impact the appropriateness of the discharge location. We also included a variable to indicate the for-profit status of the hospital since literature suggests that patient outcomes and incentives at for-profit hospitals differ from those at not-for-profit hospitals (30–32). Finally, to control for PARC availability (33, 34), we included variables to indicate whether the hospital maintained a separate nursing home, HH agency, or inpatient rehabilitative department/facility. We also created county-level measures of PARC supply: the number of PTs and OTs, HH agencies, SNF beds, and IRFs, each standardized to the county population. Approximately 75% of our study variables had complete data. The remaining variables were missing few observations (<0.5% of the records).

Statistical analysis.

We used a generalized linear mixed-model approach with a logit link to account for the correlation of patients within hospital and to examine the associations between our independent variables and PARC use (35). This approach also allowed us to assess the random effects of hospital characteristics on PARC use. We estimated 3 separate mixed-effects logistic regression models, one for each of our 3 dichotomous outcome variables (discharge to institution versus home; use of HH versus no HH, conditional on discharge home; and use of SNF versus IRF, conditional on discharge to an institution). Because SIDs in Florida did not distinguish between the use of SNFs and IRFs in 2005, Florida data from this year were excluded from the SNF versus IRF analysis. Records with missing data on one or more variables were excluded from the analyses.

For each outcome, we constructed a model with a maximum number of level 1 variables (independent variables/fixed effects) as determined by our conceptual model, content knowledge, preliminary descriptive analyses, and available covariates (36). Although several level 2 (hospital) variables were available to explain the variability in PARC use by hospital, we encountered nonconvergence problems when using combinations of level 2 variables. Only hospital-specific random intercepts proved stable enough to estimate variability in PARC use by hospital.

The first level of our models included all independent variables at the patient, hospital, county, and state levels. The second level included random intercepts for the hospitals to account for the combined effect of all omitted hospital-specific factors that contribute to variation in use of PARC. We explored the use of a 3-level model (patient within hospital within hospital county), but our models did not converge because a majority of the counties had only 1 or 2 hospitals. We chose to treat the state as a fixed effect because we did not want to generalize our findings beyond the 4 states we examined.

We explored interactions between the level 1 race, socioeconomic, and geographic variables in all models by first stratifying on these variables to determine potential interactions and then expanding the main effects model by adding interaction terms. All analyses were conducted using SAS statistical software, version 9.2.

The level 1 fixed effects in our models are conditional on the random intercepts and pertain to within-hospital comparisons. To quantify the heterogeneity across hospitals, we calculated the median odds ratio (MOR) (37, 38). An MOR of 1 indicates no variation between clusters; the larger the MOR, the greater the variation in hospital intercepts.

RESULTS

Our sample was predominantly female and white with a mean age of 68.1 years (Table 1). The sample was generally healthy, with the majority having minor to moderate severity, low mortality risk, and <3 comorbidities. Knee replacements were more common than hip replacements.

Fifty-five percent of the sample was discharged home. Of those discharged home, 58% received HH. Of those discharged to an institution, 65% received SNF care. Demographic, geographic, and clinical differences were apparent in the descriptive analyses by PARC use. Tables 2–4 present the results of the multilevel analyses on factors associated with PARC use, controlling for illness severity, length of stay, and comorbidities.

Table 2. Mixed-effects logistic regression analysis, institution versus home (n = 163,900)*
Independent variableORP95% CI
  • *

    Home is base category and controlling for disease severity, comorbidities, length of stay, and clinical variables; n = <164,875 because records with one or more missing variables were excluded from the analysis. Random-effects variance, hospital: estimate = 0.7491, SE = 0.06258, P < 0.0001, median odds ratio (OR) = 2.28. 95% CI = 95% confidence interval; PT = physical therapist; OT = occupational therapist; FTE = full-time equivalent; RN = registered nurse; PARC = post–acute rehabilitation care; IRF = inpatient rehabilitation facility; HHA = home health agency; SNF = skilled nursing facility.

  • Age divided by 10 to assist in interpretation of OR.

  • Median household income for patient residence zip code.

Sociodemographic   
 Race/insurance   
  White/Medicare/private (reference)1.00
  Black/Medicare/private1.69< 0.00011.59–1.80
  Hispanic/Medicare/private1.28< 0.00011.20–1.38
  White/Medicaid1.85< 0.00011.63–2.11
  Black/Medicaid0.46< 0.00010.35–0.60
  Hispanic/Medicaid0.57< 0.00010.44–0.73
  White/uninsured0.47< 0.00010.39–0.57
  Black/uninsured0.06< 0.00010.03–0.13
  Hispanic/uninsured0.27< 0.00010.16–0.45
 Female2.02< 0.00011.96–2.07
 Age/102.42< 0.00012.38–2.45
 Median income   
  Highest quartile (reference)1.00
  Quartile 31.12< 0.00011.07–1.16
  Quartile 21.22< 0.00011.17–1.27
  Quartile 11.33< 0.00011.27–1.39
Geographic, metro status   
  Micropolitan/rural (reference)1.00
  Medium1.060.08790.99–1.13
  Large1.27< 0.00011.17–1.37
 State   
  Wisconsin (reference)1.00
  Arizona1.150.45630.79–1.69
  Florida2.28< 0.00011.67–3.10
  New Jersey15.22< 0.000110.77–21.48
Hospital characteristics   
 Joint replacement volume/1000.950.00130.91–0.98
 PT and OT FTEs/1,000 admissions1.040.32120.96–1.12
 RN FTEs/100 admissions0.930.03920.87–1.00
 Major medical school affiliation1.220.0840.97–1.53
 For-profit hospital0.950.7110.74–1.22
PARC supply at hospital   
 Has nursing home unit1.170.34440.84–1.63
 Has IRF0.790.04350.63–0.99
 Has HHA0.970.78090.78–1.20
PARC supply in patient residence county   
 PTs and OTs/10,0001.01< 0.00011.01–1.02
 HHAs/100,0001.000.7130.99–1.01
 SNF beds/1,000 age ≥65 years1.05< 0.00011.03–1.07
 IRFs/100,000 age ≥65 years1.000.3601.00–1.01
Table 3. Mixed-effects logistic regression analysis, HH versus home, conditional on discharge home (n = 89,076)*
 ORP95% CI
  • *

    Home is base category and controlling for disease severity, comorbidities, length of stay, and clinical variables; n = <89,648 because records with one or more missing variables were excluded. Random-effects variance, hospital: estimate = 2.1991, SE = 0.1841, P < 0.0001, median odds ratio (OR) = 4.1146. HH = home health; 95% CI = 95% confidence interval; PT = physical therapist; OT = occupational therapist; FTE = full-time equivalent; RN = registered nurse; PARC = post–acute rehabilitation care; IRF = inpatient rehabilitation facility; HHA = HH agency; SNF = skilled nursing facility.

  • Age divided by 10 to assist with interpretation of odds ratio.

  • Median household income for patient resident zip code.

Sociodemographic   
 Race/state   
  White/Wisconsin (reference)1.00
  White/Arizona6.92< 0.00013.61–13.28
  White/Florida25.66< 0.000114.99–43.91
  White/New Jersey7.43< 0.00014.06–13.60
  Black/Arizona0.700.0630.48–1.02
  Black/Florida0.590.0000.45–0.78
  Black/New Jersey0.650.0230.45–0.94
  Black/Wisconsin1.81< 0.00011.43–2.29
  Hispanic/Arizona0.620.0160.42–0.91
  Hispanic/Florida0.640.0280.43–0.95
  Hispanic/New Jersey0.550.0220.33–0.92
  Hispanic/Wisconsin1.500.0311.04–2.17
 Female1.14< 0.00011.09–1.18
 Age/101.20< 0.00011.18–1.23
 Insurance1.00
  Medicare/private (reference)   
  Uninsured0.43< 0.00010.36–0.52
  Medicaid0.59< 0.00010.51–0.67
 Median income1.00
  Highest quartile (reference)   
  Quartile 30.940.0300.89–0.99
  Quartile 20.970.2940.91–1.03
  Quartile 10.930.0500.87–1.00
Geographic, metro status   
  Micropolitan/rural (reference)1.00
  Medium1.27< 0.00011.14–1.40
  Large1.40< 0.00011.24–1.57
Hospital characteristics   
 Joint replacement volume/1001.15< 0.00011.08–1.22
 PT and OT FTEs/1,000 admissions0.900.1400.79–1.03
 RN FTEs/100 admissions1.030.5010.94–1.13
 Major medical school affiliation0.860.4580.59–1.27
 For-profit hospital1.530.0521.00–2.34
PARC supply at hospital   
 Has a nursing home unit0.690.2110.38–1.24
 Has an IRF0.810.3070.55–1.21
 Has an HHA1.330.1310.92–1.93
PARC supply in patient residence county   
 PTs and OTs/10,0001.02< 0.00011.01–1.03
 HHAs/100,0001.010.0861.00–1.03
 SNF beds/1,000 age ≥65 years1.010.4430.98–1.04
 IRFs/100,000 age ≥65 years1.010.0941.00–1.03
Table 4. Mixed effects of logistic regression analysis, SNF versus IRF, conditional on discharge to an institution (n = 60,117)*
 ORP95% CI
  • *

    IRF is base category and controlling for disease severity, comorbidities, length of stay, and clinical variables. Florida 2005 data were excluded from analysis as well as records with one or more missing variables. Random-effects variance, hospital: estimate = 4.8863, SE = 0.4461, P < 0.0001, median odds ratio (OR) = 8.24. SNF = skilled nursing facility; IRF = inpatient rehabilitation facility; 95% CI = 95% confidence interval; PT = physical therapist; OT = occupational therapist; FTE = full-time equivalent; RN = registered nurse; PARC = post–acute rehabilitation care; HHA = home health agency.

  • Age divided by 10 to assist with interpretation of OR.

  • Median household income for patient resident zip code.

Sociodemographic   
 Race/state   
  White/Wisconsin (reference)1.00
  White/Arizona0.470.1470.17–1.31
  White/Florida1.160.7360.48–2.79
  White/New Jersey0.06< 0.00010.02–0.15
  Black/Arizona3.34< 0.00012.04–5.45
  Black/Florida1.560.0131.10–2.22
  Black/New Jersey1.94< 0.00011.40–2.69
  Black/Wisconsin0.620.0030.46–0.85
  Hispanic/Arizona3.070.0001.70–5.58
  Hispanic/Florida1.140.6610.63–2.06
  Hispanic/New Jersey1.370.2880.77–2.44
  Hispanic/Wisconsin0.830.5070.47–1.45
 Female1.15< 0.00011.09–1.20
 Age/101.06< 0.00011.03–1.08
 Insurance1.00  
  Medicare/private (reference)   
  Uninsured1.210.0501.00–1.47
  Medicaid1.580.0111.11–2.26
 Median income1.00  
  Highest quartile (reference)   
  Quartile 30.950.0940.89–1.01
  Quartile 21.000.9470.93–1.08
  Quartile 11.090.0371.01–1.18
Geographic, metro status   
  Micropolitan/rural (reference)1.00
  Medium1.250.0051.07–1.47
  Large1.52< 0.00011.28–1.81
Hospital characteristics   
 Joint replacement volume/1000.970.5600.89–1.06
 PT and OT FTEs/1,000 admissions1.210.1040.96–1.52
 RN FTEs/100 admissions0.960.5700.84–1.10
 Major medical school affiliation0.530.0370.30–0.96
 For-profit hospital0.400.0090.20–0.79
PARC supply at hospital   
 Has a nursing home unit1.880.1920.73–4.86
 Has an IRF0.09< 0.00010.05–0.17
 Has an HHA2.330.0051.29–4.24
PARC supply in patient residence county   
 PTs and OTs/10,0000.980.0020.97–0.99
 HHAs/100,0000.980.0810.95–1.00
 SNF beds/1,000 age ≥65 years1.030.1000.99–1.07
 IRFs/100,000 age ≥65 years0.96< 0.00010.94–0.98

Use of institutional care.

The results of our final model, which include interaction terms for race by insurance status, are shown in Table 2. Minorities receiving Medicare or private insurance and whites receiving Medicaid were more likely to receive institutional care (relative to whites with Medicare or private insurance), while minorities receiving Medicaid and individuals who were uninsured were less likely. Blacks who were uninsured were the least likely to receive institutional care with an OR of 0.06. In regard to other sociodemographic variables, women, older individuals, and individuals living in communities with lower income were more likely to receive institutional care.

Use of institutional care varied by geographic, hospital, and supply variables. Use of institutional care was more likely for individuals living in metropolitan areas. State effects were large with institutional use much more likely in New Jersey relative to the other states, and in Florida relative to Arizona and Wisconsin. The volume of joint procedures at the hospital and RN staffing ratios were positively associated with the use of institutional care, although the latter relationship was weak. The use of institutional care was also positively associated with the availability of SNF beds and supply of PTs and OTs.

The MOR was 2.28, indicating considerable heterogeneity across hospitals in the propensity for individuals with similar characteristics to receive institutional care. The proportion of the unobserved variation in the use of institutional care attributable to unobserved (i.e., unmeasured) hospital characteristics was 19% (0.7491/[0.7491 + π2/3] = 0.19).

Use of HH versus no HH, conditional on discharge home.

The effect of race on the use of HH was modified by state (Table 3). Minorities in Wisconsin were more likely to receive HH relative to whites in Wisconsin. Minorities in the other 3 states were less likely to receive HH relative to whites in these states. State effects for HH use were large for whites in Arizona, Florida, and New Jersey, relative to whites in Wisconsin.

In regard to other sociodemographic characteristics, women, older individuals, and individuals receiving Medicaid or uninsured (relative to individuals with Medicare or private insurance) were less likely to receive HH. Individuals living in metropolitan areas were more likely to receive HH. The only hospital and supply variables positively associated with HH use were joint volume and availability of PTs/OTs, respectively.

The MOR was 4.11, indicating considerable heterogeneity across hospitals in the propensity for individuals with similar characteristics to use HH. The proportion of the unobserved variation in use of HH attributable to unobserved hospital characteristics was 40%.

Use of SNF versus IRF, conditional on discharge to an institution.

The effect of race on the use of SNF versus IRF care was modified by state. Minorities in Arizona and New Jersey were more likely to use SNF care relative to whites in these states. Blacks in Wisconsin, however, were more likely to use IRF care relative to whites and Hispanics in the state. Women, older individuals, individuals receiving Medicaid or who were uninsured, individuals with lower income, and individuals living in metropolitan areas were more likely to use SNF versus IRF care. For-profit hospital status and being a hospital with a major medical school affiliation were associated with the use of IRF care, although the latter association was weak. IRF availability at the hospital and county levels were strong predictors of IRF use. The supply of PTs and OTs in the county was also positively associated with IRF use. The MOR was 8.24, and the proportion of the unobserved variation in the use of SNF versus IRF care attributable to unobserved hospital characteristics was 60%.

DISCUSSION

Considering PARC as a continuum ranging from more to less intensive care in regard to number of hours of rehabilitation per day (i.e., IRF, SNF, HH, no HH) (11, 39, 40), we identified several demographic and geographic differences in PARC use after controlling for illness severity, comorbidities, and PARC supply. In many instances, minorities received less intensive care. In the institution versus home model, minorities who were uninsured or receiving Medicaid were less likely to receive institutional care. This effect modification was particularly large for race and Medicaid status. In the HH/no HH model, minorities in Arizona, Florida, and New Jersey were less likely to receive HH relative to whites in these states. In the SNF/IRF model, minorities in Arizona and New Jersey were less likely to receive IRF care.

Another consistent finding was that individuals of lower SES generally received less intensive care. In the HH/no HH and SNF/IRF models, individuals who were uninsured, receiving Medicaid, or who had a lower SES were more likely to receive less intensive care. In the institution versus home model, however, lower SES and being white and receiving Medicaid were associated with the receipt of institutional care. Minorities receiving Medicare or with private insurance were also more likely to receive institutional care. Several studies have reported that minorities and individuals with lower SES have poorer function preoperatively (likely due to delaying the procedure) and are more at risk for postoperative complications following joint replacement (41–45). The findings of these studies may be one explanation for those instances when minorities and individuals of lower SES received more intensive care. The findings of these studies are also notable in regard to the instances when minorities and individuals with lower SES received less intensive care.

While individuals living in rural areas were more likely to receive less intensive care in the institution/home and HH/no HH models, they were more likely to receive IRF care in the SNF/IRF model. One possible explanation for the latter finding is that individuals in urban areas, where both SNFs and IRFs are available and where SNFs outnumber IRFs, choose the facility closest to their home. A study by Buntin et al supports this explanation, since they found that the distance to provider (SNF or IRF) was a strong predictor of where individuals went for rehabilitation following joint replacement (33).

Other consistent findings across the models were that women were more likely to be discharged to the more intensive settings. While not all of the state databases contained marital status information, 2 of them did. In a subgroup analysis of these 2 states, we found that being married/having a partner diminished the effect of being female, but did not eliminate it. Studies indicate that women present for joint replacement at an older age and with greater functional limitations (46, 47).

Large state effects were also present in all the models. Of particular note were the large state effects for New Jersey. Individuals living in New Jersey were much more likely to use institutional care, and whites in New Jersey were much more likely to use IRF care. Some of this was likely driven by the size of the state and the dispersion of IRFs (n = 19) throughout, making an IRF within “driving distance” for most residents. Although we controlled for the availability of IRFs, our measure was somewhat crude (IRFs per county population) and did not take into account the number of beds in the facility. Despite the imprecision of some of our PARC supply measures, many were significantly associated with PARC use in our models.

Our analyses of the random effects indicated significant variation in the use of PARC across hospitals and showed that unmeasured hospital characteristics accounted for 20–60% of the unexplained variation in PARC use. Future research should identify the source of this variation; local physician practice/preferences, patient preferences, and other unobserved patient and hospital characteristics could be important drivers of this variation. Few of the level 1 hospital variables were significant. We did find that the volume of joint procedures and RN staffing ratios were positively associated with discharge home. Both of these hospital variables are considered proxies for hospital quality and have been found to be associated with better patient outcomes. (24, 25, 29, 48). Our finding of an association between for-profit hospital status and the use of IRF over SNF care is unclear and needs further exploration.

Our study has notable limitations. We examined data from only 4 states and found patterns of PARC use varied considerably by state. Our findings may not be generalizable to other states. A significant limitation is the fact that we did not have any direct measures of the patient's functional status. We used multiple measures of illness severity/comorbidities as proxies for functional status based on the literature (20, 22, 49, 50). Although almost all of these were significant in the expected directions, we cannot be certain that we accounted for all of the variation in functional status. Our models also lacked direct measures of patient preferences and provider characteristics. Finally, we only examined the first site for PARC. PARC is often received in more than one setting over the course of several months.

There is a great deal of controversy over the appropriate PARC setting(s) and the extent that patient outcomes differ among them (51). Although we found variation in the use of PARC, we did not examine how this relates to outcomes. Studies by DeJong et al suggest that IRFs are moderately superior for short-term outcomes following joint replacement, but these effects are diminished over the longer term as individuals use additional PARC (39, 40). Coulter et al found no differences in patient outcomes for patients receiving PARC in an outpatient group setting versus home following joint replacement (52). In our study, we were unable to determine if individuals who did not receive HH received outpatient care. Not receiving HH may not be indicative of a disparity (i.e., lower quality of care) if individuals received outpatient care in a timely manner.

This study adds to the limited recent literature on disparities in PARC for joint replacement and serves as a point of departure for future studies. After controlling for illness severity/comorbidities, hospital characteristics, and community-level factors, demographic and geographic differences in PARC following joint replacement remained. Some of these differences suggest racial, socioeconomic, and geographic disparities in PARC. Because the literature suggests that minorities and individuals of lower SES delay joint replacement and have poorer preoperative and postoperative status, efforts to further understand the reasons behind these disparities and their effect on outcomes of care are needed.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Freburger had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Freburger, Holmes, Cutchin, Edwards.

Acquisition of data. Freburger.

Analysis and interpretation of data. Freburger, Holmes, Ku, Heatwole-Shank, Edwards.

Ancillary