Getting to the root: Examining within and between home health agency inequities in functional improvement

Abstract

Objective

To quantify racial, ethnic, and income‐based disparities in home health (HH) patients' functional improvement within and between HH agencies (HHAs).

Data Sources

2016–2017 Outcome and Assessment Information Set, Medicare Beneficiary Summary File, and Census data.

Data Collection/Extraction Methods

Not Applicable.

Study Design

We use multinomial‐logit analyses with and without HHA fixed effects. The outcome is a mutually exclusive five‐category outcome: (1) any functional improvement, (2) no functional improvement, (3) death while a patient, (4) transfer to an inpatient setting, and (5) continuing HH as of December 31, 2017. The adjusted outcome rates are calculated by race, ethnicity, and income level using predictive margins.

Principal Findings

Of the 3+ million Medicare beneficiaries with a HH start‐of‐care assessment in 2016, 77% experienced functional improvement at discharge, 8% were discharged without functional improvement, 0.6% died, 2% were transferred to an inpatient setting, and 12% continued using HH. Adjusting for individual‐level characteristics, Black, Hispanic, American Indian/Alaska Native (AIAN), and low‐income HH patients were all more likely to be discharged without functional improvement (1.3 pp [95% CI: 1.1, 1.5], 1.5 pp [95% CI: 0.8, 2.1], 1.2 pp [95% CI: 0.6, 1.8], 0.7 pp [95% CI:0.5, 0.8], respectively) compared to White and higher income patients. After including HHA fixed effects, the differences for Black, Hispanic, and AIAN HH patients were mitigated. However, income‐based disparities persisted within HHAs. Black‐White, Hispanic‐White, and AIAN‐White disparities were largely driven by between‐HHA differences, whereas income‐based disparities were mostly due to within‐HHA differences, and Asian American/Pacific Islander patients did not experience any observable disparities.

Conclusions

Both within‐ and between‐HHA differences contribute to the overall disparities in functional improvement. Mitigating functional improvement inequities will require a diverse set of culturally appropriate and socially conscious interventions. Improving the quality of HHAs that serve more marginalized patients and incentivizing improved equity within HHAs are approaches that are imperative for ameliorating outcomes.

What is known on this topic

• For Medicare home health patients, within home health agency inequities exist in hospital readmissions and emergency department use, where patients are treated differently by race, ethnicity, and income.

• Racial, ethnic, and income‐based inequities in functional improvement also exist for Medicare home health patients, but we do not know if that is driven by within‐agency or between‐agency differences.

• Most of what we know about functional improvement disparities in the home health setting comes from one region of the country.

What this study adds

• Disparities in functional improvement exist across the U.S.

• Low‐income patients have less functional improvement than their higher income counterparts within the same home health agencies.

• Functional improvement inequities for Black, Hispanic, and American Indian/Alaska Native patients are mostly attributable to agency characteristics, which emphasizes the need to improve access to higher quality agencies.

1. INTRODUCTION

One of the primary responsibilities of home health (HH) is to improve the functional status of HH patients. Home health care is the provision of skilled care services in the home. Skilled services include nursing, physical and occupational therapy, speech‐language therapy, and medical social services. These services are ordered by a healthcare provider, and HH staff develop a plan of care that is implemented to facilitate the patients' physical, mental, and social well‐being. The goal of intermittent HH services is to enable the patient to regain or maintain life at home through the improvement or maintenance of functional ability.

Approximately 20% of Medicare HH patients are from minoritized groups, over 1/2 live below 200% of the Federal Poverty Line, over 80% have 3+ chronic conditions, and almost 1/3 need assistance with activities of daily living. ¹ , ² In fact, HH patients on average have greater assistance needs for activities of daily living (e.g., bathing, eating, transferring, etc.) than do nursing home residents. ¹ As the demand for HH continues to grow, it is increasingly important to ensure high‐quality care and functional improvement for all HH patients. ³

There have been only a few studies that focus on racial disparities in functional improvements among home health agency (HHA) patients, and the findings have been mixed. ⁴ , ⁵ , ⁶ , ⁷ Among the first studies in this area, Peng et al. found no differences in functional status by race or ethnicity in one non‐profit HHA in the Northeast. ⁴ Conversely, using a small sample of HH patients nationwide, Brega et al. found evidence of functional improvement differences between racial and ethnic groups, especially between Black and non‐Hispanic White HH patients. ⁵ Similarly, Chase et al. found that Black, Hispanic, and Medicare/Medicaid dually eligible HH patients in New York experienced significantly less functional improvement than non‐Hispanic White HH patients, even after adjusting for various covariates. ⁶ Most recently, Wang et al. also found that being Black was associated with less functional improvement at discharge as compared to White HH patients in New York. ⁷ While providing valuable insights into disparities that exist within HH, this prior work is limited in that it mostly focuses on services delivered in only one geographic region (i.e., Northeast), does not examine differences in death, transfers, or continued use of services, and only examines disparities in care delivered within agencies as opposed to also considering disparities between or across agencies.

Two prior studies have examined the question of within‐ and between‐HHA disparities. ⁸ , ⁹ A study examining disparities in the patient‐reported experiences of care found that 14 of 19 statistically significant quality outcomes were associated with within‐HHA racial disparities where racially minoritized patients had a different experience of care than their White counterparts. ⁹ The second study identified within‐HHA BlackWhite and income‐based disparities in 30‐ and 60‐day hospital readmissions and emergency department use. ⁸ However, compared to functional improvement, disparities in hospital readmissions and emergency department use may be considered more distal from the quality of care received by the HHA, prompting our focus on functional improvement.

While the previous two studies document within‐HHA disparities, as it relates to between‐HHA disparities, prior research suggests that there are racial, ethnic, and income inequities in receiving care from high‐quality HHAs. ¹⁰ , ¹¹ In addition, various HHA characteristics such as agency racial and Medicaid‐patient composition, ownership, size, staffing, and rurality have been shown to be associated with quality differences between HHAs. ⁸ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ Given this research, it is important to understand how disparities in outcomes may be a function of differences in agencies from whom older patients receive their care. Knowing whether within‐ or between‐HHA differences contribute to disparities can help shape interventions and policies to achieve health equity. ¹⁸ , ¹⁹

“Disparities” and “inequities” are often used interchangeably to describe racial, ethnic, and socioeconomic differences in care. However, these words have different connotations. ²⁰ , ²¹ Disparities are specific differences that further disadvantage historically marginalized/disadvantaged populations as compared to more privileged populations. ²⁰ , ²¹ Inequities occur when disparities are driven by unjust and avoidable circumstances such as racism or discrimination. ²⁰ , ²¹ Race is socially constructed, and not biological; as such, the racial disparities that have been noted in functional improvement, along with those in access and quality of care, are fundamentally driven by racism. ²² Similarly, socioeconomic disparities observed may be driven by factors of income‐based discrimination. ²² , ²³ Racism and discrimination happen at both interpersonal and structural levels. ²⁴ , ²⁵ Within‐HHA disparities associated with differences in treatment by race, ethnicity, and/or income would be the result of interpersonal racism or discrimination (i.e., prejudice and bias leading to differential treatment). Conversely, between‐HHA disparities associated with differences in access and outcomes by race, ethnicity, and/or income would be the result of structural racism (i.e., laws, policies, and practices that create and reify racial inequities) that leads to unequal access to high‐quality care. ²³

The overall objective of this study is to examine racial, ethnic, and income‐based disparities in HH patient functional improvement to determine what portion of the overall disparities are attributable to differences within an HHA (i.e., interpersonal factors) versus differences between HHAs (i.e., structural factors). Addressing this objective is vital because solutions and interventions to promote health equity in HH will be different depending on the source of the disparity (i.e., differences in treatment of individual patients or differences in access to high‐quality care). We hypothesize that: (1) disparities in functional improvement will be driven by both between‐ and within‐HHA factors; and (2) between‐HHA differences will contribute to a large portion of the overall disparities given that there are racial, ethnic, and income‐based disparities in access to high‐quality HHAs. ¹⁰

2. METHODS

2.1. Study design

This study estimates racial, ethnic, and income‐based differences in functional improvement rates among patients successfully discharged from HH using a retrospective cohort design.

2.2. Data

Data come from the Centers for Medicare and Medicaid Services (CMS) Chronic Conditions Warehouse 2016 and 2017 Medicare Beneficiary Summary File (MBSF), the 2016 and 2017 Outcome and Assessment Information Set (OASIS), the 2015 Social Deprivation Index (SDI), ²⁶ and the 2015 American Community Survey (ACS) 5‐year estimates.

The MBSF contains patients' demographic characteristics, enrollment information, and current address ZIP codes. Medicare‐certified HHAs are required to submit OASIS assessments for all Medicare beneficiaries receiving skilled HH services. We used the OASIS to identify individual HH patients and the HHA serving them, as well as other patient‐level information (e.g., race/ethnicity, health status, living arrangements, and social support). These data are linked to the MBSF using the patient/beneficiary ID number.

Last, we used the publicly available Social Deprivation Index, which uses ACS 5‐year estimates to collate information on place‐based economic disadvantage. ²⁶ Neighborhood racial composition is collated from the ACS 5‐year estimates available through data.census.gov. The neighborhood details are described further, below. These data are linked to the beneficiary‐level data using the ZIP Code Tabulation Areas.

2.3. Study sample

Our sample consists of Black, Hispanic (any race), Asian American/Pacific Islander (AAPI), American Indian/Alaska Native (AIAN), and White Medicare‐enrolled HH patients aged 65 years and older who had a start‐of‐care assessment in 2016. Because we were interested in functional improvement, we excluded HH patients without any functional limitations reported on their start‐of‐care assessment (n = 24,237). We also exclude HH patients residing in congregate housing at the start of care (e.g., assisted living n = 478,208) due to potential overlap with facility‐based long‐term care (e.g., continuing care retirement communities) that may limit choice in HHA. Our analytic sample consists of 3,298,276 patients who were followed from their first 2016 start‐of‐care assessment until their first transfer or discharge assessment, or December 31, 2017, if they did not have a transfer or discharge assessment.

2.4. Variables

While functional improvement is our primary outcome of interest, we must account for potential differences in the likelihood of being discharged between racial/ethnic and income groups that may impact their opportunity to functionally improve. In other words, if patients die or are transferred at discharge, they may not have had the opportunity to experience functional improvement, likewise, if patients remain as patients at the end of our study period, we would be unable to observe functional improvement. Therefore, we created a mutually exclusive five‐category outcome variable to account for these competing risks, operationalized as: (1) discharged with any functional improvement, (2) discharged without functional improvement, (3) death while still a patient of the HHA, (4) transfer with discharge from the HHA to an inpatient facility, and (5) continuing care at the end of the study period (December 31, 2017). Categories 3–5 allow us to carefully consider the patients and assessments that would be censored and account for the other potential outcomes/competing risks of a HH episode.

Functional improvement is calculated using the overall improvement in the composite Activities of Daily Living (ADL) score. ⁵ , ⁶ , ²⁷ We calculate the composite ADL score using the 9 ADLs available in the OASIS ²⁸ : grooming, dressing lower body, dressing upper body, bathing, toileting (transfer and hygiene), transferring, ambulation, and eating. To calculate our ADL measure, we use a corrected Likert approach where each individual ADL is divided by the highest possible value for that ADL, allowing all of the individual ADLs to be on the same scale, ranging from zero, limited assistance needed, to one, full assistance needed. ²⁷ We then sum all of the individual ADLs and create a score ranging from 0 to 9, where 0 would indicate that the HH patient required no assistance with any of the ADLs and 9 indicates extensive assistance needed for all ADLs.

To quantify the overall change in ADL functional improvement, we calculate the difference between the ADL composite score on the start‐of‐care and discharge assessments. Positive scores indicate improvement, negative scores indicate a decline, and scores of 0 indicate no change in function. HH patients with scores of 0 or less are grouped into the “no improvement” category, and everyone with scores above 0 is categorized as “any improvement.” The “any improvement” category captures patients who fully regain independence as well as those who partially regain some independence. For example, patients who move from a score of 9 on their start‐of‐care assessment to a 5.5 on their discharge assessment are captured in “any improvement,” along with those who go from 9 to 0.

The independent variables of interest are measured at the patient level and describe the patients' race, ethnicity, and socioeconomic status. To identify the race and ethnicity of HH patients, we use the self‐reported race/ethnicity variables in the OASIS, which is considered the gold standard. ²⁹ All racial/ethnic groups are mutually exclusive. A patients' low‐income status is determined by dual enrollment in Medicare and Medicaid and participation in Medicare Part‐D low‐income cost‐sharing subsidy (LIS) at the time of HH initiation. We use the Part‐D LIS to capture more potentially low‐income patients, as the LIS has more generous eligibility than Medicaid, does not vary by state, and therefore allows for a more uniform and potentially sensitive measure of low‐income status.

A number of covariates are also included in the study based on the disablement conceptual model (see Table 1). ³² These variables include risk factors (e.g., age, sex, and SDI), intra‐individual factors (i.e., behavioral risk factors), extra‐individual factors (e.g., Medicare Advantage status, ³³ living alone, and caregiver assistance), pathology (i.e., prognosis), impairment (e.g., vision impairment, surgical wound, and incontinence), functional limitations (e.g., pain, shortness of breath, and cognitive impairment), and disability/prior function. ⁶ We also control for referral source as a proxy for patient risk and need at the start of care. ⁶ , ³⁴ The values for all patient‐level control variables come from the start‐of‐care assessment.

TABLE 1.

Conceptual and operational definition of covariates as informed by prior studies and the disablement model.

Covariates	Type	Description (data source)
Risk factors
Sex	Dichotomous	Male/Female (MBSF)
Age	Continuous	(MBSF)
Proportion of Black residents in neighborhood centile score	Ordinal	Ranging from 1 to 100, neighborhoods in the 100th centile represent those with the greatest percentage of Black residents (ACS).
Proportion of Hispanic residents in neighborhood centile score	Ordinal	Ranging from 1 to 100, neighborhoods in the 100th centile represent those with the greatest percentage of Latine residents (ACS).
Social deprivation index (SDI) score 26	Ordinal	The SDI score is a centile that divides the ordered set of sociodemographic measures into 100 parts, making the SDI score easily interpretable by way of an underlying scale. The SDI score is calculated from a composite of the percent of the population: (1) living below 100% FPL; (2) 25 years of age or more with less than 12 years of education; (3) non‐employed; (4) unemployed; (5) living in renter‐occupied housing; (6) living in crowded housing units; (7) without a car; (8) single‐parent households with dependents <18 years. The higher the SDI score, the more socioeconomically disadvantaged the neighborhood. An SDI score of 100 means that the neighborhood is the most deprived, while a score of 1 makes it the least deprived. (ACS and Robert Graham Center)
Intra‐individual factors
Behavioral risk factors	Categorical	Total number of behavioral risk factors (i.e., smoking, obesity, alcohol dependency, and drug dependency). 0 = no behavioral risks; 1 = 1 behavioral risk; 2 = 2+ behavioral risks. (OASIS Items: M1036 1–4)
Pathology
Prognosis	Categorical	Patients' overall status, stability, and risk for serious complications and death. 0 = No heightened risks, patient is stable; 1 = Temporarily facing heightened health risks but will likely return to “0”; 2 = Heightened health risks with low likelihood have becoming stable and possible increased risk for death. (OASIS Item: M1034; original OASIS codes 2 & 3 are collapsed into category 2 [high risk])
Impairment
Vision impairment	Dichotomous	Yes/No (OASIS Item: M1200)
Hearing impairment	Dichotomous	Yes/No (OASIS Item: M1210)
Has a surgical wound	Dichotomous	Yes/No (OASIS Item: M1340)
Urinary incontinence	Dichotomous	Yes/No (OASIS Item: M1610)
Bowel incontinence	Dichotomous	Yes/No (OASIS Item: M1620)
Functional limitations
Interfering pain	Dichotomous	Yes/No for patient having pain that interferes with activity or movement at any frequency. (OASIS Item: M1242; original OASIS codes 0 & 1 = No and code 2–4 = Yes)
Shortness of breath	Categorical	Indicates when patients experience shortness of breath. 0 = Never; 1 = With minimal to moderate exertion; 2 = With heavy exertion. (OASIS Item: M1400; original OASIS codes 2–4 = With minimal to moderate exertion and code 1 = With heavy exertion)
Cognitive impairment	Dichotomous	Yes/No for any cognitive impairment if the patient requires prompting, assistance, or are totally dependent on care providers due to disturbances (codes 1–4) (OASIS Item: M1700)
Confusion	Dichotomous	Yes/No for confusion at any time (codes 1–4) (OASIS Item: M1710)
Cognitive, behavioral, or psychiatric symptoms	Categorical	Total cognitive, behavioral and psychiatric symptoms (i.e., verbal disruption, memory deficit, impaired decision making, physical aggression, delusional/hallucinatory/paranoid) present at least once/week categorized as 0 = No Symptoms; 1 = 1 Symptom; 2 = 2+ Symptoms. (OASIS Item: M1740)
Disability
Prior function	Ordinal	Patients' total dependence with self‐care, ambulation, transfer, and household tasks prior to current illness, exacerbation, or injury leading to home care episode summed. Ranges from 0–8 where 0 indicates full independence and 8 indicates full dependence in all areas. (OASIS Item: M1900)
Extra‐individual factors
Medicare advantage status 30	Dichotomous	Yes/No if ever on Medicare Advantage during the course of the episode. (MBSF)
Caregiver activities of daily living (ADL) assistance	Dichotomous	Yes/No for having a caregiver that provides assistance with Activities of Daily Living (ADLs) (OASIS Item: M2100 code 1)
Frequency of caregiver ADL assistance	Categorical	Categories for how often the patient receives ADL or IADL assistance from any caregiver: 0 = No assistance received; 1 = Daily Assistance; 2 = Weekly Assistance; 3 = Less than Weekly Assistance. (OASIS Item: M2110)
Living alone 31	Dichotomous	Yes/No if living alone (OASIS Item: M1100; Codes 1–5 = Living Alone)
Other factors
Referral source 5	Categorical	Categorizes where HH patients are referred or discharged from in the past 14 days as: 0 = No discharge/referral code; 1 = Acute hospitalization; 2 = Post‐Acute Inpatient Stay; 3 = Other. (OASIS Item: M1000)

Abbreviations: ACS, American Community Survey; HH, home health; MBSF, Medicare Beneficiary Summary File; OASIS, Outcome and Assessment Information Set.

2.5. Analytic approach

Summary statistics are calculated for patient characteristics by race, ethnicity, and income level. We conduct two separate analyses. First, we estimate the relationship between patient characteristics and the outcome using a multinomial logit model adjusting for all aforementioned covariates. Second, to determine within‐HHA disparities, we estimate another multinomial logit model adjusting for all covariates and using HHA fixed effects. We used a Mundlak hybrid model to account for HHA differences by inserting HHA‐level means of all model covariates and clustering standard errors on the HHA to approximate fixed effects. ³⁵ , ³⁶ The approximated fixed effects account for both observed and unobserved HHA characteristics.

To determine the differences/disparities attributable to individual characteristics, net of the HHA effects, we compare the models with and without HHA fixed effects. Comparing the estimates across the two models (with and without fixed effects) allows us to quantify the observed differences/disparities within and between HHAs. Though our focus is on functional improvements, we estimated probabilities for all categories of the multinomial outcome that allow us to understand the role of different competing outcomes. Thus, the summation of estimated differences in probabilities of the five categories is equal to zero. However, we only quantify disparities for discharged HH patients with or without functional improvement, our main outcome of interest. It should be noted that disparities can be calculated for death, but differences in transfer and continuing in care are not conceptualized as categorically “good” or “bad” outcomes.

The Brown University Institutional Review Board reviewed and approved this study. All data cleaning and analyses were completed using STATA 16. ³⁷

3. RESULTS

Of the 3,298,376 Medicare patients who started HH in 2016, 79% were White, 12% were Black, approximately 7% were Hispanic, 2% were AAPI, <1% were AIAN, and 27% were low‐income (Table 2). Overall, 77% of patients had functional improvement, while 8% had no functional improvement upon discharge. The percentage of patients with functional improvement was 79% for White and higher income patients, 72% for Black and AIAN patients, 70% for Hispanic patients, 76% for AAPI patients, and 71% for low‐income HH patients. Ten percent of Black, AAPI, AIAN, and low‐income patients, and 13% of Hispanic patients were discharged without functional improvement, compared to 7% of White and higher income patients. The percentage of patients who died was highest among Black, AAPI, and AIAN HH patients. The percentage of patients who were transferred at discharge was lowest for White and higher income patients. Asian American/Pacific Islander and higher income patients were the least likely to continue using HH at the end of the study period. Among HH patients who were discharged, there was an average two‐point improvement in the overall function score. Functional improvement was highest for White and higher income patients. Other individual‐level covariates are displayed by race and income in Table 2 and by outcome in Table S1.

TABLE 2.

Descriptive characteristics of home health patients starting care in 2016, by race, ethnicity, and income (N = 3,298,376).

	Overall	White	Black	Hispanic	AAPI	AIAN	Higher income	Low‐income
	3,298,376	2,599,951	382,223	224,488	79,843	11,871	2,386,613	911,763
Outcome, %
Any functional improvement	77.1	78.5	72.1	69.9	76.3	71.8	79.3	71.4
No functional improvement	8.1	7.3	10.1	12.8	10.3	9.5	7.3	10.2
Death	0.6	0.6	0.8	0.5	0.5	0.9	0.6	0.6
Transfer with discharge	2.2	2.0	2.8	2.9	2.1	3.2	1.8	3.0
Continued use–no discharge	12.0	11.6	14.2	13.9	10.8	14.6	11.0	14.8
Amount of functional improvement, mean (SD)	2.2 (1.5)	2.3 (1.5)	2.1 (1.6)	1.9 (1.5)	2.0 (1.5)	2.1 (1.6)	2.3 (1.5)	2.0 (1.5)
Race, %
White	78.8	–	–	–	–	–	87.7	55.5
Black	11.6	–	–	–	–	–	7.8	21.5
Hispanic	6.8	–	–	–	–	–	3.0	16.7
AAPI	2.4	–	–	–	–	–	1.2	5.7
AIAN	0.4	–	–	–	–	–	0.3	0.5
Low‐income‐Eligible, %	27.6	19.5	51.2	67.9	65.6	41.5	–	–
Medicare advantage, %	30.5	28.5	39.1	41.2	26.7	21.3	28.5	35.8
Female, %	61.1	60.4	65.6	62.2	61.1	60	57.7	70
Average age, mean (SD)	79.2 (8.4)	79.5 (8.4)	77.6 (8.4)	78.6 (8.2)	79.9 (8.1)	77.4 (8.1)	79.7 (8.3)	78.1 (8.4)
Behavioral risks, %
No behavioral risks	63.8	63.0	60.7	71.3	84.9	62.2	65.4	59.8
1 behavioral risk	30.4	30.9	33.0	25.4	13.9	30.2	29.3	33.3
2+ behavioral risks	5.8	6.1	6.3	3.4	1.3	7.6	5.3	7.0
Prognosis, %
No heightened risk	7.1	7.0	7.1	7.9	6.2	8.6	7.3	6.6
Temporary risk	56.7	56.0	56.4	62.7	63.2	55.0	56.8	56.4
High risk	36.2	36.9	36.5	29.3	30.6	36.4	35.9	36.9
Vision impairment, %	27.2	24.5	33.5	43.9	37.3	32.4	23.8	36.0
Hearing impairment, %	39.8	41.0	31.8	40.5	40.6	40.8	39.3	41.2
Presence of a surgical wound, %	28.4	30.3	21.0	21.2	21.4	28.2	32.6	17.3
Urinary incontinence, %	56.2	55.7	59.3	58.7	52.9	54.7	53.4	63.6
Bowel incontinence, %	14.1	13.4	17.4	16.1	16.5	17.2	12.9	17.4
Interfering pain, %	76.9	77.0	75.5	78.5	75.1	76.0	76.1	78.8
Shortness of breath, %
None	24.3	24.2	23.5	26.0	28.2	25.8	26.1	19.8
With minimal‐ moderate exertion	50.3	50.2	51.6	50.9	44.2	50.8	47.6	57.2
With heavy exertion	25.4	25.6	24.9	23.1	27.6	23.4	26.3	23.0
Cognitive impairment, %	46.4	44.8	50.2	54.3	56.0	47.9	43.3	54.4
Confusion, %	53.9	52.7	57.6	60.0	60.2	55.2	51.0	61.5
Cognitive, behavioral, & psychiatric symptoms, %
No symptoms	73.5	74.2	70.9	70.5	72.0	73.4	75.1	69.3
1 symptom	17.0	16.4	18.9	19.5	17.5	18.0	15.8	20.0
2+ symptoms	9.5	9.3	10.3	10.0	10.4	8.6	9.1	10.7
Prior function, mean (SD)	2.6 (2.3)	2.5 (2.3)	3.20 (2.3)	3.4 (2.3)	3.2 (2.3)	2.8 (2.3)	2.4 (2.3)	3.2 (2.3)
Has ADL assistance, %	52.5	51.8	54.3	58.2	52.6	52.5	52.8	51.9
ADL assistance frequency, %
Daily	84.3	84.0	82.6	89.0	89.9	83.3	85.5	81.3
Weekly	12.3	12.6	13.6	8.4	8.2	12.7	11.6	14.4
Less than weekly	1.7	1.7	1.9	1.1	1.0	1.9	1.5	2.1
None	1.7	1.7	1.9	1.5	1.0	2.1	1.4	2.2
Living alone, %	28.1	29.2	27.7	20.1	15.2	23.9	26.2	32.9
Neighborhood percent Black, mean (SD)	12.8 (18.9)	8.7 (12.5)	43.5 (28.6)	9.3 (13.2)	9.2 (11.8)	9.8 (15.4)	11.2 (17.0)	16.9 (22.6)
Neighborhood percent Hispanic, mean (SD)	14.9 (19.3)	11.3 (13.7)	14.2 (17.4)	54.6 (29.1)	26.0 (20.2)	14.6 (17.0)	12.5 (15.8)	21.3 (25.3)
SDI Score, mean (SD)	49.7 (28.2)	44.0 (26.1)	72.8 (24.6)	73.1 (24.6)	56.8 (29.6)	56.5 (26.8)	44.6 (27.1)	62.8 (27.0)
Referral source, %
Acute hospitalization	41.9	43.2	38.3	35.0	38.0	43.1	44.6	35.0
PAC inpatient stay	25.3	26.7	21.7	16.9	18.9	23.3	26.5	22.0
Other	0.3	0.3	0.3	0.4	0.4	0.4	0.3	0.3
Missing source	32.5	29.8	39.6	47.7	42.7	33.1	28.6	42.7

Note: N = 3,298,376. All differences between groups are statistically significant (p < 0.001) according to Chi‐square tests. Data are derived from the 2016 start‐of‐care assessments. Low‐Income identified a beneficiary as having dual enrollment in Medicare and Medicaid and/or participation in the Medicare Part‐D low‐income cost‐sharing subsidy. Neighborhood is defined by the ZIP Code Tabulation Area. See Table 1 for all the other variable descriptions.

Abbreviations: AAPI, Asian American/Pacific Islander; ADL, Activities of Daily Living; AIAN, American Indian/Alaska Native; PAC, Post‐Acute Care; SD, standard deviation; SDI, Social Deprivation Index.

3.1. Discharged with any functional improvement (Figure 1A)

FIGURE 1.

Racial, ethnic, and income‐based differences in “any” and “no” functional improvement, overall and within home health agencies. (A) Differences in the adjusted rates of any functional improvement, overall and within home health agencies, by race, ethnicity, and income. (B) Differences in the adjusted rates of no functional improvement, overall and within home health agencies, by race, ethnicity, and income. N = 3,298,376. Analysis uses a multinomial logit model with and without HHA fixed effects and adjusts for all covariates listed in Table 2. The Stata MARGINS command was used to calculate the predicted percentage of the outcome. Income differences refer to differences between higher income and low‐income patients. Low‐Income identified a beneficiary as having dual enrollment in Medicare and Medicaid and/or participation in Medicare Part‐D low‐income cost‐sharing subsidy. AAPI, Asian American/Pacific Islander. AIAN, American Indian/Alaska Native. HHA, Home Health Agency. FEs, Fixed Effects. [Color figure can be viewed at wileyonlinelibrary.com]

After controlling for various patient‐level characteristics, the overall Black‐White disparity in any functional improvement was −1.8 percentage points (pp; 95% CI: −2.2, −1.4), and within HHAs, it was −0.1 pp (95% CI: −0.3, 0.07), meaning that 93% of the Black‐White disparity in any functional improvement was explained by between‐HHA differences. The overall Hispanic‐White disparity in any functional improvement was −0.9 pp (95% CI: −1.7, −0.1), and within HHAs, there was no disparity (1.4 pp; 95% CI: 1.0, 1.7), as such between‐HHA differences primarily explained the overall Hispanic‐White disparities. The overall AIAN‐White disparity in any functional improvement was −3.9 pp (95% CI: −5.1, −2.8), and within HHAs, it was −2.6 pp (95% CI: −3.6, −1.6), meaning that 35% of the AIAN‐White disparity was driven by between‐HHA differences. The overall income‐based disparity in any functional improvement was −3.5 pp (95% CI: −3.7, −3.2), and within HHAs, it was −3.2 pp (95% CI: −3.4, −3.1), meaning that only 8% of the income‐based disparity in any functional improvement was explained by between‐HHA differences. Conversely, AAPI patients had a significantly higher adjusted rate of functional improvement as compared to White patients, both overall (2.9 pp, 95% CI: 2.1, 3.6) and within HHAs (2.8 pp, 95% CI: 2.4, 3.2). Thus, between‐HHA differences only contributed to 2% of AAPI‐White differences in any functional improvement.

3.2. Discharged with no functional improvement (Figure 1B)

After controlling for various patient‐level characteristics, the overall Black‐White disparity in no functional improvement was 1.3 pp (95% CI: 1.1, 1.5), and within HHAs, it was 0.5 pp (95% CI: 0.3, 0.6), meaning that 64% of the Black‐White disparity in no functional improvement at discharge was explained by between‐HHA differences. The overall Hispanic‐White disparity in no functional improvement was 1.5 pp (95% CI: 0.8, 2.1), and within HHAs, it was 0.2 pp (95% CI: 0.02, 0.4), meaning that 86% of the disparity in having no functional improvement could be explained by between‐HHA differences. The overall AIAN‐White disparity in no functional improvement was 1.2 pp (95% CI: 0.6, 1.8), and within HHAs, it was 0.5 pp (95% CI: −0.1, 1.06), meaning that 60% of the disparity in having no functional improvement could be explained by between‐HHA differences. The overall income‐based disparity in no functional improvement was 0.7 pp (95% CI: 0.5, 0.8), and within HHAs, it was 0.6 pp (95% CI: 0.5, 0.7), meaning that only 15% of the income‐based disparity in no functional improvement was related to between‐HHA disparities. For AAPI HH patients, there were no observable disparities in discharging from HH without functional improvement.

Other adjusted results for competing states (death, transfer, and continuing care) are included in Table S2. Main regression results can be found in Table 3 and full regression results can be found in Tables S3 and S4.

TABLE 3.

Multinomial regression results of main effects for models with and without home health agency fixed effects.

	Overall effects model without HHA fixed effects					Within‐HHA effects model with HHA fixed effects
Variable	RRR	SE	95% CI LL	95% CI UL	p	RRR	SE	95% CI LL	95% CI UL	p
Any functional improvement (base outcome)
No functional improvement
White (reference group)
Black	1.21	0.02	1.17	1.24	<0.001	1.06	0.01	1.04	1.08	<0.001
Hispanic	1.21	0.05	1.12	1.31	<0.001	1.00	0.01	0.98	1.03	0.832
AAPI	0.99	0.03	0.93	1.06	0.873	0.94	0.02	0.91	0.97	<0.005
AIAN	1.24	0.05	1.14	1.34	<0.001	1.11	0.05	1.03	1.21	<0.05
Higher income (reference group)
Low‐income	1.15	0.01	1.13	1.18	<0.001	1.14	0.01	1.12	1.15	<0.001
Death during home health
White (reference group)
Black	1.19	0.03	1.13	1.26	<0.001	1.13	0.03	1.07	1.20	<0.001
Hispanic	0.81	0.03	0.75	0.88	<0.001	0.79	0.03	0.74	0.85	<0.001
AAPI	0.81	0.05	0.71	0.91	<0.005	0.86	0.05	0.77	0.96	<0.01
AIAN	1.45	0.16	1.17	1.79	<0.005	1.37	0.15	1.11	1.69	<0.01
Higher income (reference group)
Low‐income	0.97	0.02	0.93	1.01	0.210	0.96	0.02	0.93	1.00	0.053
Transfer with discharge from home health
White (reference group)
Black	1.13	0.02	1.08	1.17	<0.001	0.98	0.02	0.94	1.01	0.121
Hispanic	0.99	0.03	0.93	1.05	0.707	0.86	0.02	0.81	0.90	<0.001
AAPI	0.72	0.03	0.66	0.79	<0.001	0.67	0.02	0.62	0.71	<0.001
AIAN	1.47	0.12	1.25	1.73	<0.001	1.29	0.09	1.12	1.49	<0.001
Higher income (reference group)
Low‐income	1.43	0.02	1.40	1.47	<0.001	1.39	0.02	1.36	1.42	<0.001
Continued use of home health
White (reference group)
Black	1.05	0.01	1.02	1.08	<0.005	0.97	0.01	0.96	0.99	<0.001
Hispanic	0.98	0.03	0.92	1.05	0.530	0.88	0.01	0.86	0.91	<0.001
AAPI	0.75	0.02	0.71	0.80	<0.001	0.80	0.02	0.77	0.83	<0.001
AIAN	1.22	0.05	1.14	1.32	<0.001	1.17	0.04	1.08	1.25	<0.001
Higher income (reference group)
Low‐income	1.26	0.01	1.24	1.28	<0.001	1.25	0.01	1.23	1.27	<0.001

Note: N = 3,298,376. Analysis uses a multinomial logit model with and without Home Health Agency (HHA) fixed effects and adjusts for all covariates listed in Table 2. White patients are the reference group for all other racial/ethnic groups, and higher income patients are the reference group for low‐income patients. Low‐Income identified a beneficiary as having dual enrollment in Medicare and Medicaid and/or participation in Medicare Part‐D low‐income cost‐sharing subsidy.

Abbreviations: AAPI, Asian American/Pacific Islander; AIAN, American Indian/Alaska Native; HHA, Home Health Agency; CI, Confidence Interval; LL, Lower Limit; P, p‐value; RRR, Relative Risk Ratio; SE, Standard Error; UL, Upper Limit.

4. DISCUSSION

Our study contributes to the limited literature around HH inequities and suggests the mechanisms giving rise to HH disparities in functional improvement vary by race, ethnicity, and income. Disparities in the lack of functional improvement exist both overall and within HHAs. We find that there are significant Black‐White, Hispanic‐White, AIAN‐White, and income‐based overall and within‐HHA disparities in the lack of functional improvement. However, the overall disparities for Black, Hispanic, and AIAN patients were mostly attributable to the HHAs used (i.e., between‐HHA differences). Income‐based disparities in the lack of functional improvement were more related to within‐HHA differences in treatment for lower income patients.

The overall disparities in functional improvement may be inequities caused by racism and income‐based discrimination. ²² The overall inequities driven by between‐HHA differences may in part reflect structural racism and discrimination in access to higher quality HHAs for patients, ¹⁰ , ¹¹ , ²³ while the within‐HHA disparities may reflect differences in how patients are treated by the HHA or interpersonal racism and discrimination within HHAs. ²⁴

Our study provides insight into why the literature may be mixed concerning racial disparities in functional improvement. The earliest study in this area found no racial/ethnic differences but only examined disparities within one HHA in a Northeast urban area, ⁴ in parallel, we illustrate that some of the overall racial and ethnic disparities are attenuated within HHAs. Another study only found differences affecting Black HH patients, ⁵ but was limited to a small sample of HHAs. Our work best aligns with the findings from the two most recent studies by Chase et al. and Wang et al, ⁶ , ⁷ who find racial and ethnic disparities in functional improvement. Our study further contributes to the literature by including all HH patients in our model—as opposed to solely focusing on patients with a discharge assessment—thus allowing for the observed differential rate of mortality, transfers, and censoring among marginalized patients, compared to patients with more privilege.

When accounting for all possible outcomes, we find that most of the Black‐White disparities in functional improvement are attributable to HHA characteristics. When we controlled for which HHA served patients, the Black‐White differences in any functional improvement were no longer statistically significant. This finding is consistent with other research indicating that Black HH patients are accessing HHAs whose patients experience poorer outcomes. ⁸ , ¹⁰

Conversely, Hispanic patients have greater functional improvement than their White counterparts within the same HHA. This finding is consistent with findings of other studies, ⁵ , ⁷ but also inconsistent with some literature. ⁴ , ⁶ For example, Brega et al. found that Hispanic patients were significantly more likely than White patients to improve in the “transferring” ADL measure but less likely than White patients to improve on other individual ADL measures. ⁵ In addition, Wang et al. recently reported that Hispanic patients without dementia had greater functional improvement than all other racial groups. ⁷ We also found that Hispanic patients have higher rates of discharging without functional improvement which aligns with recent research findings by Chase et al. showing a lower functional improvement for Hispanic and Black HH post‐acute patients. ⁶ Our findings indicate that the observed ethnic differences in functional improvement are greatly influenced by HHA characteristics and therefore differences in the HHAs used by Hispanic patients. The contradictory findings within the Hispanic population may be related to health status or caregiver support characteristics that we do not have in this study and are an issue for further study.

Findings on the AIAN HH patient population are new, only one other study by Wang et al. included this group but included AAPI and AIAN patients together. Unsurprisingly, due to a history fraught with racism, AIAN patients were more likely to experience no functional improvement than their white counterparts, and most of that disparity was related to between‐HHA differences, which aligns with the decreased access to high‐quality HH experienced by AIAN older adults. In contrast to existing literature, AAPI patients had greater functional improvement than their white counterparts. This is an issue for further study, that will require disaggregating the AAPI population.

For low‐income HH patients, we do not find that a large share of the disparity in functional improvement is attributed to the quality and types of HHAs that they are using. Conversely, it seems that much of the income‐based disparity is related to differences in treatment within HHAs; however, because we are unable to adequately adjust for other clinical factors, our results may be influenced by lower income HH patients being unable to functionally improve and more likely to maintain their functioning due to fewer resources or other challenges. Future work should consider controlling for or stratifying analyses by various diagnoses that may make functional improvement difficult, in addition to examining Medicaid HH patients specifically who are more likely to be focused on maintaining function as opposed to improving.

4.1. Mitigating between‐HHA inequities

Our findings suggest that the HHA from which patients receive care explains much of the between‐group differences observed in prior literature. ⁵ , ⁶ , ⁷ Black, Hispanic, and AIAN HH patients experienced disparities in functional improvement that were largely related to the HHA used. When Black‐White, Hispanic‐White, and AIAN‐White disparities were examined within HHAs, the overall magnitude of the disparities was greatly reduced. These findings are supported by prior work that illustrates inequities in access to high‐quality HHAs for Black, Hispanic, and AIAN HH patients when compared to White HH patients. ⁸ , ¹⁰ However, our findings also contradict those of Maddox et al. who found that Black‐White disparities in 30‐ and 60‐day readmissions and emergency department use were related to within‐HHA differences, rather than differences in the agencies from which patients received care, ⁸ which is why we chose to study an outcome that is more proximate to HHA services. To have an impact on health equity in functional improvement, interventions and policies should focus on addressing structural/institutional racism by improving access to higher quality HHAs ¹⁰ or targeting HHAs with large proportions of minoritized patients for quality improvement initiatives.

There are three potential approaches to mitigating between‐HHA functional improvement disparities. First, the Centers for Medicare and Medicaid Services (CMS) could incentivize high‐quality HHAs to serve marginalized HH patients, as the agency is working to do through the “Better Care for Dually Eligible People” initiative. ³⁸ , ³⁹ However, increasing financial payments does not directly correspond with improving expertise on supporting more vulnerable patients and/or mitigating disparities. ⁴⁰ Future work should explore what characteristics of HHAs are associated with the greatest functional improvement and the lowest disparities. Alternatively, CMS could target the HHAs that are already serving more marginalized patients for quality improvement initiatives focused on functional improvement since those HHAs tend to have lower quality. ⁸ Last, CMS could also target marginalized patients to increase their awareness and use of the CMS quality information, ⁴¹ , ⁴² , ⁴³ , ⁴⁴ which may in turn increase their use of higher quality HHAs.

4.2. Mitigating within‐HHA Inequities

Our findings also indicate that even within the same HHA, there are racial, ethnic, and income‐based disparities in functional improvement. This implies that within a given HHA, marginalized patients are being treated differently than their higher income and/or White counterparts. This reflects the presence of interpersonal discrimination and racism within HHAs. ²⁵ We propose three potential approaches that may help to mitigate within‐HHA disparities. First, CMS could report inequities as a measure of HHA quality. ⁴⁵ , ⁴⁶ This will help reward HHAs for mitigating disparities within their agency. Second, HHAs should develop targeted and socially conscious caregiver training programs to better engage and encourage caregivers to support ADL improvement. ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ Education and training programs and resources should be culturally appropriate and socially conscious programs that take into account the possible limitations in the built and social environment. ⁵¹ , ⁵² , ⁵³ Last, HHAs may be able to augment pain management, wound care, and physical and occupational therapy for marginalized populations. Research shows that proper management of pain and wound care, and increased therapy could promote ADL improvement, and if the HHA increased their focus on providing these services to marginalized patients, this could help mitigate disparities. ⁶

This study is not without limitations. We only examine functional improvement among those who start and are discharged from the HHA during our study period. Then operationalizing functional improvement as a dichotomous indicator within our discharge group means that we cannot account for the magnitude of functional changes by groups as is done in other studies. ⁶ However, our choice to examine any improvement versus no improvement was done to minimize floor/ceiling effects; additional work incorporating a clinically meaningful degree of change would be beneficial. ⁵⁴ In addition, we use a composite ADL measure; future work could disaggregate this measure to examine individual measures of ADL, as prior research has suggested that improvement in specific ADLs has varied between racial groups. ⁵ Relatedly, we do not control for the number or type of HH visits, which may impact our outcome directly. However, data on visits are only available through claims data and should be explored further in future work. Furthermore, our racial and ethnic groups are mutually exclusive, meaning that we may not be appropriately assigning people to their preferred race or ethnicity; for example, people who identify as Black and Hispanic are referred to as Hispanic only. In addition, we include AAPI and AIAN patients, and due to a small sample size, results must be interpreted with caution. More data are needed to disaggregate these groups, and improved methods are required to calculate stable estimates. Notedly, this study does not take an intersectional approach ⁵⁵ ; however, there are important joint effects of race and other factors (such as income and health) that may be important to consider, such as the interactive effect between race and dementia. ⁷ , ⁵⁶ As such, future research should consider other intersections of identity, income, and health status. Last, our data predate the advent of the 2020 HH Patient‐Driven Groupings Model, ⁵⁷ which places a lower value on therapy services and community entry. Having fewer therapy services and being that marginalized HH patients who tend to be community referrals, disparities in functional improvement may be exacerbated by the new payment model. ⁵⁶ , ⁵⁸

5. CONCLUSION

There are inequities that reflect structural differences in access to quality HHAs and within‐HHA disparities that reflect differences in how patients are treated. Mitigating the inequities described here requires that we acknowledge how differently these inequities are created across groups. Rarely are inequities eliminated with a broad approach to quality improvement; in fact, broad quality improvement may exacerbate disparities, and targeted approaches are needed. ¹¹ , ⁴⁰ , ⁴⁶ , ⁵⁹

Supporting information

Table S1. Home health patient characteristics by main outcome variable.

Table S2. Overall, Within, and Between Agency Differences in Functional Improvement for Individual Home Health Patients by Race, Ethnicity, and Income, Comparing the Inclusion of Home Health Agency Characteristics.

Table S3. Multinomial Logit Model without HHA Fixed Effects.

Table S4. Multinomial Logit Model with HHA Fixed Effects.

Fashaw‐Walters SA, Rahman M, Jarrín OF, et al. Getting to the root: Examining within and between home health agency inequities in functional improvement. Health Serv Res. 2024;59(2):e14194. doi: 10.1111/1475-6773.14194