Exploring Factors Contributing to Race Differences in Post-Stroke Disability

Abstract

Background and Purpose:

Cross sectional analyses have found large race differences in post-stroke disability, yet these analyses do not account for pre-stroke disability, hospitalization factors, post-acute care, transitions, or mortality. In this context, we explore mortality, nursing home placement and disability in a longitudinal analyses of older stroke survivors who survived at least 90 days post-stroke.

Methods:

A prospective cohort of Black or White stroke survivors from the National Health and Aging Trends Study (2009–2016) linked to Medicare were used. Disability was assessed during in-person interviews with validated scales (0–7). We used cox proportional hazards models to separately assess mortality and nursing home admission adjusting for age, sex, sociodemographics (marital status, education, income, insurance status, social network size), comorbiditites, hospitalization factors, post-acute care, and 90-day readmissions. To estimate racial differences in disability, we used a multi-level linear regression model initially adjusting for age and sex and then compared to a model adjusted for sociodemographics, comorbiditites, hospitalization factors, post-acute care, and 90-day readmissions.

Results:

There were 282 stroke survivors of which 76 (12.6%) were Black. There were no race differences in long term mortality (HR for Black=1.2, 95% CI 0.7–2.2, p=0.5) or nursing home placement (HR for Black=0.7, 95% CI 0.2–2.4, p=0.5). The largest race differences in disability were observed immediately pre-stroke, estimated age and sex adjusted activity limitations were (2.6 [2.0–3.2] in Blacks vs.1.4 [1.0–1.8] in whites, mean diff 1.2 [0.5–1.9], p<0.001) and immediately post-stroke (2.6 [2.0–3.3] in Blacks vs. 1.7 [1.2–2.1] in whites, mean diff 1.0 [0.2–1.7], p<0.01). Full adjustment did not substantially change the associations between race and disability.

Conclusions:

Race differences in nursing home placement, long-term mortality, sociodemographics, comorbidities, hospitalization factors, post-acute care, and readmissions are unlikely to be large contributors to race differences in post-stroke disability. Further research is needed to understand the drivers of race differences in post-stroke disability.

Non-Hispanic Blacks/African Americans (Blacks) have the highest stroke prevalence of any US racial/ethnic group and the prevalence is expected to increase by over 25% by 2030.^{1, 2} In the US, national studies have shown that community-dwelling Black stroke survivors experience more disability and receive more hours of caregiving than non-Hispanic White (White) stroke survivors.^3–6

A number of non-exclusive explanations may account for these cross-sectional findings in community-dwelling populations. First, patient factors such as age, comorbidities, socioeconomics or other factors that are not yet understood may contribute to race difference via pre-stroke disability or access to care.^{7, 8} Second, characteristics of the acute hospitalization such as length of stay, number of procedures, recurrent stroke or hospital readmissions may disproportionately negatively impact Black stroke patients.^9–12 Third, Blacks may have less recovery or accumulate more disability in the years following stroke than Whites after stroke. For example, such differences may arise if there are differences in the location, intensity, complications or transitions of post-acute care¹³ or; Fourth, nursing home residence, which may impact over 10% of stroke survivors,¹⁴ could lead to selection bias in prior community-based samples; if Blacks are less likely to live in a nursing home this could result in over-representation of severely disabled black stroke survivors in community-dwelling samples. Finally, cross-sectional studies of stroke survivors cannot account for differences in post-stroke mortality.¹⁵ If among severely disabled stroke patients, Blacks are more likely to survive than Whites, Blacks would appear to have greater disability than Whites in cross-sectional samples because they would be relatively over-represented compared to Whites.

To explore these hypotheses, we first evaluated race differences in mortality, nursing home placement and disability in a national sample of stroke patients. The drivers of race differences in post-stroke disability are likely to be multi-factorial with complex interplay between factors. Thus, to begin to generate hypotheses and select factors for further study, we explored whether sociodemographics, comorbidities, hospitalization factors, post-acute care, and 90-day readmissions were possible contributors of race differences in post stroke disability. We combined the strengths of a nationally representative ongoing cohort of older adults that annually collects detailed measures of disability with Medicare claims for the stroke hospitalization and post-acute care settings. This comprehensive approach mitigates several limitations of prior research that has been restricted to a single setting,¹³ conducted in selected populations such as stroke survivors with a caregiver,¹⁶ limited exploration of stroke hospitalization or post-acute care factors,^{7, 8, 13, 17} or with limited temporal granularity (i.e. evaluated every 2 years) which may miss the critical acute recovery period.^{7, 8}

Methods

Data Source

The data that support the findings of this study may be made available from the corresponding author upon reasonable request and obtaining appropriate approvals.

This is a prospective longitudinal cohort analysis of the National Health and Aging Trends Study (NHATS) from 2011–2016 linked to Medicare from 2009–2015. NHATS is a nationally representative ongoing survey of adults over the age of 65 designed to enhance the understanding of trends and trajectories of disability among older Americans.¹⁸ After informed consent, NHATS conducts in-person interviews annually, including detailed assessments of cognition and disability, the disability assessment protocol and the systematic identification of the setting which older adults live, including interviews with facilities are key features of NHATS.¹⁹ NHATS has an over 85% follow-up rate.¹⁸ NHATS was sampled from the Medicare enrollment file and for this analysis NHATS participants with Medicare fee-for service were included. This study was approved by the University of Michigan Institutional Research Board.

Study Population and Outcomes

Our study population included all NHATS participants hospitalized with a primary or secondary diagnosis of ischemic stroke or intracerebral haemorrhage (ICD-9-CM 431, 433.x1, 434.x1, 436; ICD-10-CM I61.xx, I63.xxx, I67.82). We excluded individuals with stroke who reported race/ethnicity other than Black or White, did not complete any disability assessments, or died within 90 days (figure 1). The rationale for excluding early mortality was that prior data has characterized the early mortality difference after stroke and, thus, we wanted to focus on longer term mortality.²⁰ Date of stroke was ascertained as the admission date of a stroke hospitalization and the first assessment post-stroke was defined as immediate post-stroke with pre-and post-stroke time defined relative to this date. Stroke survivors would contribute pre- and post-stroke data or solely post-stroke data if their stroke occurred prior to the start of NHATS.

Figure 1.

Cohort Extraction

To address the limitations of prior cross-sectional work, which are unable to account for mortality and oftentimes exclude nursing home placement, we included long term mortality, nursing home placement and post-stroke disability as outcomes. Mortality was ascertained from the Medicare Beneficiary Summary File (2011 to 2015) and the NHATS tracker file (2016). Nursing home admission was defined in NHATS and any nursing home resident with measures of disability was also included in the disability analysis.¹⁸ NHATS collects in-person assessments of activity limitations.¹⁸ Disability was defined as the receipt of help for any of 7 self-care and mobility activities; eating, bathing/showering, using toilet, dressing, going outside, getting around in home and getting out of bed (0–7).¹⁹

Covariates: Sociodemographic and comorbidities

Sociodemographics were collected from NHATS, including sex, race, annual income (<$12,102, $12,102–21,000, $21,001–34,409, $34,410–60,000, >$60,000), education, marital status, insurance status and social network size. Age at time of stroke was obtained from Medicare Inpatient Claims. We performed multiple imputation (5) for the missing data on income (9%), marital status (9%), and education (10%), using logistic regression based on specified predictors with complete data, i.e. race, sex, and age at time of stroke. Rubin’s rules were used to combine the coefficient estimates. Comorbidities were constructed as the count of conditions (0–26).²¹

Covariates: Stroke and Stroke Hospitalization Care

Patient characteristics during the index hospitalization include length of stay and Intracerebral Haemorrhage. We created two acute hospitalization indexes. A complication index (0–7) of conditions that were not present at admission including: Pneumonia, Urinary Tract Infection, Sepsis, DVT/PE, Myocardial Infarction/Coronary Artery Disease, Dysrhythmia, Congestive Heart Failure.²² We also defined a hospital procedures index (0–8) including: IV Tissue Plasminogen Activator (tPA), gastrostomy, hemicraniectomy, ventriculostomy, tracheostomy, intubation, hemodialysis, and cardiopulmonary resuscitation.²³

Covariates: Post-acute care/readmissions

Medicare claims data are based on the place of service. By combining data from the Inpatient (including acute hospitalization and inpatient rehabilitation facilities (IRF)), skilled nursing facilities (SNF), home health, outpatient and hospice files, we captured all claims for each post-acute care setting. These files were used to determine minutes of physical therapy, occupational therapy and speech and language therapy and number of transitions for 90 days post-stroke.²⁴ Among stroke patients discharged to institutional post-acute care, inpatient rehabilitation facility or skilled nursing facility, first discharge location, length of stay and complications defined similar to hospital complications with the addition of accidental falls were assessed.²² We also determined 90 day hospital readmissions overall and due to recurrent stroke.

Statistical Analysis

Descriptive statistics were used to explore racial differences in demographics, comorbidities, hospitalization, post-acute care and readmissions. To explore the role of long term mortality and nursing home admissions to race differences in post-stroke disability, survival probabilities and nursing home admission were explored by race with Kaplan-Meier method. We then used the cox proportional hazards regressions to model the adjusted associations between race and mortality and nursing home admission after adjusting for sociodemographics, comorbidities, hospitalization factors, post-acute care, and 90-day readmissions.

Given the large number of possible factors contributing to post-stroke disability, the possibility of complex interactions between these factors, and the lack of a prior understanding of post-stroke disability, the number of plausible disability model specifications is enormous. Consequently, given our hypothesis-generating goal, we opted for a simple model specification with a primary goal of identifying whether any individual factors appeared to be important predictors. We estimated average marginal effects using the age and sex adjusted model estimates to characterize the activity limitations by race across each discrete time period. To estimate, in this population, whether there were race-differences in function across each year before, immediately after and post-stroke, Wald tests were used to compare the average marginal effects across each year and by race. Then to explore the role of patient, hospital and post-acute care factors, we added sociodemographics, comorbidities, hospitalization factors, post-acute care, and 90-day readmissions to the multi-level linear regression model. We did not build models or conduct a formal mediation analysis as our goal was to determine whether race differences persisted after full adjustment rather than identify the specific factors that may be contributing to race differences given the lack of prior research identifying possible drivers. We performed a sensitivity analysis using a poisson model and found similar results, thus we present our linear model as the primary analysis here. We also performed a sensitivity analysis limiting our study population to those with a primary ICD stroke code and found similar results. Statistical analysis was completed using SAS 9.4 (SAS Institute, Cary, NC) and Stata 14.

Results

There were 282 stroke survivors, representing about 750,000 stroke survivors who survived at least 90 days after their stroke. Black stroke survivors comprised 12.6% (76 stroke survivors) of the sample (Table 1). The mean years of pre-stroke follow up for Blacks was 1.8 years and 2.1 for Whites, and post-stroke was 3.1 for Blacks and 3.0 for Whites. Compared to White stroke survivors, Blacks were younger (75.2 (95% CI 74.0, 76.4) vs. 78.6 (95% CI 77.2, 80.0), p<0.01)) and on average had larger social networks (4.9 (95% CI, 4.4, 5.4) vs. 3.5 (95% CI 3.2, 3.7), p<0.01). Blacks had less income (31.9% vs. 14.6%, p<0.01 less than $12,102 annual income) and were less likely to have a supplemental insurance (56.5% vs. 77.7%, p<0.01) than Whites. There was a trend toward Black stroke survivors having a longer length of stay (6.4 days vs. 5.1 days, p=0.06) and more likely to be discharged to an IRF rather than a SNF than Whites (p=0.07).

Table 1:

Sociodemographics, comorbidities, hospitalization factors, post-acute care, and 90-day readmissions of stroke patients by race

	White (obs=206; survey weighted=679,962)	Blacks (obs=76; survey weighted=98,314)	P Value
Sociodemographics and Comorbidities
Age at Stroke, Mean (95% CI)	78.6 (77.2, 80.0)	75.2 (74.0, 76.4)	<0.01
Men, %	43.3	43.6	0.95
Married, %	45.1	33.3	0.04
College Graduate and above, %	16.4	9.7	0.11
Income, %			<0.01
Less than $12,102	14.6	31.9
$12,102-$21,000	21.0	30.6
$21,001-$34,409	23.9	20.7
$34,410-$60,000	26.0	14.8
$60,001+	14.4	2.2
Insurance Status, %			<0.01
Medicare only	11.1	12.5
Medicaid	11.2	31.0
Any Other Supplemental Insurance	77.7	56.5
Social Network, Mean (95% CI)	3.5 (3.2, 3.7)	4.9 (4.4, 5.4)	<0.01
Comorbidities, Mean (95% CI	7.0 (6.4, 7.5)	6.7 (5.9, 7.4)	0.54
Stroke Characteristics
Intracerebral Haemorrhage, %	4.4	6.0	0.53
Length of Stay, Mean (95% CI)	5.1 (4.7, 5.5)	6.4 (5.4, 7.3)	0.06
Stroke Hospitalization Care
Complications, Mean (95% CI)	0.07 (0.03, 0.11)	0.14(0.06, 0.22)	0.39
Procedures, Mean (95% CI)	0.08 (0.03, 0.13)	0.14 (0.05, 0.22)	0.27
Postacute Care
First Institutional Rehabilitation, %			0.07
None	47.9	43.6
IRF	21.0	35.6
SNF	31.1	20.8
Complications at Institutional Rehabilitation, Mean (95% CI)	0.01 (0.00, 0.03)	0.01 (−0.01, 0.03)	0.94
90-day Therapy Treatments in Minutes, Mean (95% CI)	1885.9 (1555.0, 2216.7)	3046.1 (2582.1, 3510.1)	0.11
90-day Transitions, Mean (95% CI)	2.2 (2.0, 2.4)	2.2 (1.9, 2.6)	0.88
Readmission
90-day Readmission for Non-stroke condition	20.8	19.6	0.83
90-day Readmission for Stroke	3.1	7.4	0.19

Race Differences in Mortality and Nursing Home Placement

There were no race differences in mortality after 90 days unadjusted (supplemental figure 1) or after full adjustment (Hazard Ratio for Black=1.2, 95% CI 0.7–2.2, p=0.5 supplemental table 1). Similarly, there were no race differences in unadjusted nursing home placement (supplemental figure 2) or after full adjustment (HR for Black=0.7, 95% CI 0.2–2.4, p=0.5, supplemental table 2).

Racial Differences in Disability

Disability varied by race (Table 2, Supplemental table 1). These differences were largest immediately pre-stroke (2.6 [2.0–3.2] in Blacks vs.1.4 [1.0–1.8] in Whites, mean diff 1.2 [0.5–1.9], p<0.001) and immediately post-stroke (2.6 [2.0–3.3] in Blacks vs. 1.7 [1.2–2.1] in Whites, mean diff 1.0 [0.2–1.7], p<0.01) and persisted one year after stroke (2.8 [2.0–3.5] vs. 1.9 [1.4–2.4], p=0.03).

Table 2:

Acquisition of New Activity Limitations Across the Stroke Continuum: Results from the age and sex adjusted and fully adjusted multi-level linear regression model

Parameter Estimates (5 Imputations)
	Adjusted for age and sex				Fully adjusted
	Estimate	95% CI		P value	Estimate	95% CI		P value
Intercept	−4.94	−7.64	−2.23	<0.01	−2.27	−5.19	0.66	0.13
Race and Time Interaction
Black	1.21	0.43	1.99	<0.01	1.33	0.56	2.09	<0.01
White*Pre-stroke	0.30	0.18	0.43	<0.01	0.32	0.19	0.44	<0.01
Black*Pre-stroke	0.70	0.32	1.09	<0.01	0.71	0.33	1.09	<0.01
White*Immediately Post-Stroke	0.24	0.02	0.47	0.03	0.29	0.08	0.51	0.01
Black*Immediately Post-Stroke	0.12	−0.31	0.54	0.59	0.05	−0.36	0.47	0.80
White*Post-stroke	0.28	0.19	0.37	<0.01	0.29	0.21	0.38	<0.01
Black*Post-stroke	0.05	−0.16	0.25	0.65	0.00	−0.21	0.20	0.97
Age at Stroke	0.08	0.05	0.12	<0.01	0.05	0.01	0.08	0.01
Men	−0.24	−0.75	0.26	0.35	−0.25	−0.74	0.24	0.32
Married					0.22	−0.31	0.76	0.42
College Graduate and above					0.24	−0.46	0.94	0.50
Income: Less than $12,102					0.41	−0.65	1.47	0.45
Income: $12,102-$21,000					0.54	−0.41	1.49	0.27
Income: $21,001-$34,409					0.67	−0.21	1.54	0.13
Income: $34,410-$60,000					0.93	0.06	1.81	0.04
Insurance: Medicaid					−0.35	−1.41	0.71	0.52
Insurance: Any Other Supplemental Insurance					−1.19	−1.98	−0.39	0.00
Social Network					−0.03	−0.13	0.07	0.54
Nursing Home Admission Prior to Stroke					0.93	−0.80	2.65	0.29
Comorbidities					0.03	−0.04	0.09	0.41
Intracerebral Hemorrhage					0.20	−0.80	1.20	0.70
Hospital Length of Stay					−0.05	−0.12	0.03	0.20
Hospital Complications					−0.29	−1.15	0.57	0.51
Hospital procedures					−0.11	−0.78	0.56	0.75
Type of the First Rehabilitation: inpatient rehabilitation facility					0.47	−0.31	1.25	0.24
Type of the First Rehabilitation: skilled nursing facility					1.81	1.12	2.49	<0.01
Institutional Rehabilitation complications					0.78	−1.19	2.74	0.44
90-day Post-stroke Therapy Treatments					0.00	0.00	0.00	0.17
90-day Post-stroke Transitions					−0.06	−0.38	0.25	0.69
90-day Post-stroke Readmission for Non-stroke condition					0.27	−0.62	1.15	0.56
90-day Post-stroke Readmission for Stroke					0.40	−0.87	1.67	0.54

^*Covariance parameter for age and sex adjusted model 3.4, for fully adjusted model 2.5

In fully adjusted models, there was little change in the associations between race and disability (Table 2). Predictors of increased disability include older age at time of stroke (0.05 activity limitations/year, 95% CI 0.01–0.08), high income (0.93 activity limitations/ income category, 95% CI 0.06–1.81) and discharge to a SNF (1.81 activity limitations, 95% CI 1.12–2.49). Supplemental insurance was associated with less disability (−1.19 activity limitations, 95% CI −1.98- - 0.39).

Discussion

In a longitudinal, national study of older American long-term stroke survivors, we found race differences in post-stroke disability. The magnitude of these differences is substantial — with Blacks facing about one additional activity limitation compared to whites immediately post-stroke. Patient, hospitalization, post-acute care and readmissions had little impact on the associations of race differences and disability. Additionally, we found no large race differences in mortality after 90 days or nursing home placement. Our findings do not support some of our hypotheses regarding contributors to race differences in post-stroke disability. Namely, major race differences in: 1) patient, hospital and post-acute care factors; 2) more rapid accumulation of disability after stroke amongst Blacks; 3) nursing home placement leading to selection bias; and 4) long-term mortality are unlikely to account for the observed race differences in post-stroke disability.

Our findings suggest that race differences are largest immediately before and post-stroke, but do not identify specific factors that make large contributions to race differences in post-stroke disability. Notably, sociodemographics (age, sex, marital status, education, income, insurance status, social network size), comorbidities, hospitalization factors, post-acute care, and 90-day readmissions had little impact on the race-disability associations. While our statistical power to find small effects is limited, our data suggests that the factors included in our models are not, individually or collectively, large contributors to race differences in post-stroke disability. Consequently, it is likely that the factors driving race differences are not included in our models. Hospital factors such as rates of endovascular treatment, quality of care, or hospital procedure practices could be considered.^{10, 12, 25, 26} However, with the exception of race differences in life-sustaining treatment, prior work has suggested that there are not large race differences in these factors as currently measured.^{10, 27, 28} Furthermore given the overall low prevalence of life sustaining treatments among stroke patients, life sustaining treatments, may contribute to, but are unlikely to be the sole drivers of the large observed race difference in post stroke disability. In the post stroke recovery period, a number of alternate possibilities could also be considered, including differential responsiveness to rehabilitation or differential quality of rehabilitation. Another possibility is that among stroke patients with severe disability, Black stroke patients may be more likely to survive than Whites. While many stroke survivors experience motor recovery,²⁹ a subset of stroke patients, typically with extensive corticospinal tract damage, do not recover.^{30, 31} Blacks may be over-represented in this group, a hypothesis supported by their decreased short term mortality,²⁸ which could lead to less response to rehabilitation and recovery.¹³

Overall, the role of pre-stroke disability is uncertain. We found that Blacks have greater disability than Whites immediately pre-stroke. Our findings should be interpreted cautiously, however, given the small sample size and our exclusion of stroke patients who died within 90 days of their stroke. Prior work that has explicitly focused on this question in larger samples has diverged.^{7, 8} Additional work exploring the role of pre-stroke disability is needed.

The pattern of racial differences we observed diverges somewhat from the South London Stroke Registry (SLSR). The SLSR found that Black Carribeans and Black Africans, have lower 30 day and long-term post-stroke mortality compared to Whites.³² However, contrary to our findings, in a population of over 3000 stroke survivors followed over 10 years, there were no race differences in activities of daily living and White stroke survivors had more independent activities of daily living limitations than Black stroke survivors in the SLSR.³³ These findings point to an overall advantage for Blacks in south London suggesting that race differences in disability may be mitigated by England’s stroke systems of care, universal healthcare system, greater social care system, or cultural differences such as trust in the healthcare system or differences in preferences for intensity of care.

Our study has limitations. In order to include older adults prior to stroke, we necessarily have a relatively small sample size. This results in limited statistical power to make inferences about racial differences in disability — particularly many years before or after stroke- and limits the evaluation of individual mediators. Another limitation is survivorship bias, we cannot exclude that there may be differential study follow up by stroke severity in NHATS. For example, if Black stroke survivors with disability were more likely than White stroke survivors of the same level of disability to participate in the study, this could lead to the observed results. Third, while we included many confounders of the association between race and disability, there are possible confounders that are not available. One such measure is stroke severity. While we included hospital procedures, length of stay, we do not have a true measure of stroke severity at the time of hospitalization. This limitation is unlikely to influence conclusions about racial differences as stroke severity differs minimally by race.³⁴ Fourth, stroke diagnosis relied on administrative records, which have good post predictive value, and should not be impacted by race.^{35, 36} Finally, our findings are limited to older adults and cannot be extrapolated to American stroke survivors under the age of 65.

Summary

We conclude that among older Americans, race differences in nursing home placement, long-term mortality, sociodemographics, comorbidities, hospitalization factors, post-acute care, and readmissions are unlikely to be large contributors to race differences in post-stroke disability. Given our limited sample size, we cannot exclude that these factors may be small contributors. Further research is needed to understand the drivers of race differences in post-stroke disability.