Goals-of-Care Consultations Are Associated with Lower Costs and Less Acute Care Use among Propensity-Matched Cohorts of African Americans and Whites with Serious Illness

Abstract

Background: African Americans receive more aggressive end-of-life care than Whites. Little is known about how palliative care consultation to discuss goals-of-care (“PCC”) is associated with acute care utilization and costs by race.

Objective: To compare future acute care costs and utilization between propensity-matched cohorts of African Americans with and without PCC, and Whites with and without PCC.

Design: Secondary analysis of a retrospective cohort study.

Setting/Subjects: Thirty-five thousand one hundred and fifty-four African Americans and Whites age 18+ admitted for conditions other than childbirth or rehabilitation, who were not hospitalized at the end of the study, and did not die during index hospitalization (hospitalization during which the first PCC occurred).

Measurements: Accumulated mean acute care costs and utilization (30-day readmissions, future hospital days, future intensive care unit [ICU] admission, future number of ICU days) after discharge from index hospitalization.

Results: No significant difference between African Americans with or without PCC in mean future acute care costs ($11,651 vs. $15,050, p = 0.09), 30-day readmissions (p = 0.58), future hospital days (p = 0.34), future ICU admission (p = 0.25), or future ICU days (p = 0.30). There were significant differences between Whites with PCC and those without PCC in mean future acute care costs ($8,095 vs. $16,799, p < 0.001), 30-day readmissions (10.2% vs. 16.7%, p < 0.0001), and future days hospitalized (3.7 vs. 6.3 days, p < 0.0001).

Conclusions: PCC decreases future acute care costs and utilization in Whites and, directionally but not significantly, in African Americans. Research is needed to explain why utilization and cost disparities persist among African Americans despite PCC.

Introduction

Despite growing evidence that patient–provider discussions about goals-of-care (GOC) and end-of-life (EOL) issues are associated with patients receiving less aggressive treatment^1–11 and EOL care concordant with preferences,^12,13 African Americans in the United States are less likely than Whites to have these important conversations with health care providers.^14–19 This disparity suggests an unmet need that may result in lower quality care.^14,20 Because race does not predict EOL preferences^21,22 and communication-based interventions can affect preferences,^11,21 it is important to understand relationships between GOC discussions and the care that African Americans receive.²³

Communication disparities and differences in the effectiveness of GOC discussions¹⁹ may help explain why African Americans are more likely than Whites to prefer and receive intensive EOL care^{14,20,24–30} and less likely to receive hospice care.^31,32 Differences in acute care utilization contribute to cost disparities near EOL.³³ One large study of Medicare data found that medical costs were 32% higher for African Americans than Whites in the last six months of life and that about 40% of the higher costs was due to greater use of intensive procedures (e.g., mechanical ventilation) and intensive care unit (ICU) hospitalization.²⁴ African Americans are also more likely to die in hospitals compared with Whites,²⁴ adding to cost differences.

Unfortunately, high EOL costs are associated with worse quality death, underscoring the ethical need to understand racial/ethnic cost disparities and interventions such as palliative care consultation that mitigate disparities.^10,34

Inpatient palliative care consultations that address GOC (hereafter called “PCC”) mutually benefit patients and health systems: Patients who have GOC conversations seem to experience a higher quality of dying experience that is more consistent with their preferences and health systems incur lower costs.^4,10 Early evidence of urban academic medical centers suggests that race/ethnicity is not a factor in which patients receive PCC,^4,35–37 making it an intervention that may improve racial/ethnic disparities in EOL communication and care. Despite known benefits of PCC, it is unknown how PCC is associated with acute care utilization and costs across racial groups, particularly among African Americans.

The purpose of this secondary analysis is to compare future acute care costs and utilization (30-day readmissions, future hospital days, future ICU admission, and number of ICU days) between African Americans with serious illness who had PCC and a propensity-matched cohort of African Americans who did not, and between Whites with serious illness who had PCC and a propensity-matched cohort of Whites who did not.

Methods

Study design

This secondary analysis included pre-existing clinical, administrative, and cost data from a retrospective cohort study that found that PCC was associated with lower use of acute care and an average savings of more than $6,000 per patient, but it did not explore outcomes by race.⁴ Supplementary Medicaid data were pulled from electronic medical records. This study was approved by the institutional review board of the University of Pennsylvania and followed strict procedures for ensuring patient data privacy and security.

Our sample included 35,154 self-identified African American and White patients 18 years or older admitted to a 776-bed urban, academic medical center in the Northeast between July 1, 2014 and October 31, 2016 who were admitted for conditions other than childbirth or rehabilitation, were not hospitalized at the end of the study period, and did not die during index hospitalization (defined as the hospitalization during which the first PCC occurred). The medical center serves a diverse area composed of 46% African Americans, 36% Whites, 9% Asians, and 6% Hispanics³⁸ and receives transfer patients and referrals from surrounding suburban areas. More than half of the families in the center's main service area live in poverty.³⁸

The center's interdisciplinary palliative care team is well established and predominantly works as a consultation service.⁴ The ICU teams request one-third of all palliative care consultations.⁴ The center's palliative care registry, which features demographic and clinical information including reason for consultation, was used to identify all patients who received a consultation with palliative care during the study period. Only consultations involving GOC discussion were included (consultations to discuss pain management but not GOC were excluded). Our sample included 11,158 African Americans (PCC = 383; No PCC = 10,777) and 23,994 Whites (PCC = 814; No PCC = 23,180) (Table 1).

Table 1.

Description of Study Population

	Total n = 35,154	African Americans Non-PCC	African Americans with PCC	p Value	Whites Non-PCC	Whites with PCC	p Value
		n = 10,777	n = 383		n = 23,180	n = 814
Age (years)
18–39	5739 (16%)	2385 (22.1%)	19 (5.0%)	<0.0001	3280 (14.2%)	55 (6.8%)	<0.0001
40–45	2419 (7%)	980 (9.1%)	4 (1.0%)		1405 (6.1%)	30 (3.7%)
46–50	2642 (8%)	953 (8.8%)	21 (5.5%)		1639 (7.1%)	29 (3.6%)
51–55	3614 (10%)	1215 (11.3%)	45 (11.8%)		2294 (9.9%)	60 (7.4%)
56–60	4338 (12%)	1330 (12.3%)	55 (14.4%)		2854 (12.3%)	99 (12.2%)
61–65	4320 (12%)	1165 (10.8%)	53 (13.8%)		2987 (12.9%)	115 (14.1%)
66–70	4223 (12%)	955 (8.9%)	51 (13.3%)		3093 (13.3%)	124 (15.2%)
71–75	3199 (9%)	668 (6.2%)	41 (10.7%)		2386 (10.3%)	104 (12.8%)
>75	4660 (13%)	1126 (10.5%)	94 (24.5%)		3242 (14.0%)	198 (24.3%)
Gender
Male	17,286 (49%)	4543 (42.2%)	172 (44.9%)	0.2836	12,148 (52.4%)	423 (52.0%)	0.80
Female	17,868 (51%)	6234 (57.8%)	211 (55.1%)		11,032 (47.6%)	391 (48.0%)
Medicaid
Yes	4819 (14%)	3454 (32.0%)	83 (21.7%)	<0.0001	1243 (5.4%)	39 (4.8%)	0.48
No	30,335 (86%)	7323 (68.0%)	300 (78.3%)		21,937 (94.6%)	775 (95.2%)
Primary diagnosis
Cancer	6955 (19.8%)	1063 (9.9%)	103 (26.9%)	<0.0001	5472 (23.6%)	317 (38.9%)	<0.0001
Cardiovascular disorder/heart failure	6430 (18.3%)	1774 (16.5%)	76 (19.8%)		4444 (19.2%)	136 (16.7%)
Endocrine disorder	1942 (5.5%)	902 (8.4%)	8 (2.1%)		1014 (4.4%)	18 (2.2%)
GI disorder	4120 (11.7%)	1147 (10.6%)	29 (7.6%)		2872 (12.4%)	72 (8.9%)
Gynecologic or urologic disorder	2393 (6.8%)	1086 (10.1%)	17 (4.4%)		1267 (5.5%)	23 (2.8%)
Infectious disease and sepsis	2950 (8.4%)	1227 (11.4%)	63 (16.5%)		1565 (6.8%)	95 (11.7%)
Neurologic disorder	3508 (10%)	1204 (11.2%)	31 (8.1%)		2227 (9.6%)	46 (5.7%)
Respiratory disorder	1394 (4.0%)	493 (4.6%)	23 (6.0%)		819 (3.5%)	59 (7.3%)
Other	5462 (15.5%)	1881 (17.5%)	33 (8.6%)		3500 (15.1%)	48 (5.9%)
APR-DRG severity of illness
Minor	8044 (22.9%)	2343 (21.7%)	6 (1.6%)	<0.0001	5680 (24.5%)	15 (1.8%)	<0.0001
Moderate	13,876 (39.5%)	4446 (41.3%)	36 (9.4%)		9316 (40.2%)	78 (9.6%)
Major	9912 (28.2%)	3140 (29.1%)	167 (43.6%)		6270 (27.1%)	335 (41.2%)
Extreme	3322 (9.4%)	848 (7.9%)	174 (45.4%)		1914 (8.3%)	386 (47.4%)
APR-DRG risk of mortality
Minor	15,914 (45.3%)	5142 (47.7%)	11 (2.9%)	<0.0001	10,741 (46.3%)	20 (2.5%)	<0.0001
Moderate	10,112 (28.8%)	3072 (28.5%)	50 (13.1%)		6854 (29.6%)	136 (16.7%)
Major	6637 (18.9%)	1940 (18.0%)	179 (46.7%)		4177 (18.0%)	341 (41.9%)
Extreme	2491 (7.1%)	623 (5.8%)	143 (37.3%)		1408 (6.1%)	317 (38.9%)
Acute care hospitalization 30 days before index hospitalization
Yes	653 (1.9%)	127 (1.2%)	80 (20.9%)	<0.0001	240 (1.0%)	206 (25.3%)	<0.0001
No	34,501 (98.1%)	10,650 (98.8%)	303 (79.1%)		22,940 (999.0%)	608 (74.7%)
ICU care during index hospitalization
Yes	11,448 (32.6%)	2792 (25.9%)	197 (51.4%)	<0.0001	8025 (34.6%)	434 (53.3%)	<0.0001
No	23,706 (67.4%)	7985 (74.1%)	186 (48.6%)		15,155 (65.4%)	380 (46.7%)
ICU care >6 days during index hospitalization
Yes	2637 (7.5%)	624 (5.8%)	104 (27.2%)	<0.0001	1683 (7.3%)	226 (27.8%)	<0.0001
No	32,517 (92.5%)	10,153 (94.2%)	279 (72.8%)		21,497 (92.7%)	588 (72.2%)
Visited by oncology service during index hospitalization (first or second service)
Yes	2984 (8.5%)	456 (4.2%)	77 (20.1%)	<0.0001	2188 (9.4%)	263 (32.3%)	<0.0001
No	32,170 (91.5%)	10,321 (95.8%)	306 (79.9%)		20,992 (90.6%)	551 (67.7%)
DNR documented during index hospitalization
Yes	1425 (4.1%)	242 (2.2%)	172 (44.9%)	<0.0001	594 (2.6%)	417 (51.2%)	<0.0001
No	33,729 (95.9%)	10,535 (97.8%)	211 (55.1%)		22,586 (97.4%)	397 (48.8%)
Mean number of days hospitalized during index hospitalization (SD)		5.96 (7.63)	17.05 (19.5)	<0.0001	6.29 (7.67)	16.53 (19.06)	<0.0001
Median number of days hospitalized during index hospitalization (IQR)		4.0 (2.0–7.0)	10.0 (6.0–20.0)	<0.0001	4.0 (2.0–7.0)	10.0 (6.0–20.0)	<0.0001
Mean number of ICU days during index hospitalization (SD)		1.18 (4.26)	6.03 (12.70)	<0.0001	1.45 (4.00)	6.36 (13.78)	<0.0001
Median number of ICU days during index hospitalization (IQR)		0 (0–0)	1.0 (0–6.0)	<0.0001	0 (0–1.0)	1.0 (0–7.0)	<0.0001
Mean direct acute care costs during index hospitalization (SD)		$15,665 ($22,667)	$35,982 ($49,024)	<0.0001	$19,583 ($25,209)	$40,126 ($59,607)	<0.0001
Median direct acute care costs during index hospitalization (IQR)		$9,723 ($5,898–$15,641)	$18,578 ($10,811–$38,119)	<0.0001	$12,114 ($7,639–$21,373)	$20,017 ($10,455–$41,445)	<0.0001
Changed goals-of-care during PCC index hospitalization (% yes)		n/a	60.1%	—	n/a	67.7%	0.01

Significance tests for the percentage variables involved a Chi-squared test; significance tests for parametric continuous variables involved a t test; and significance tests for nonparametric continuous variables involved a Kruskal–Wallis nonparametric test of ranks.

Bold indicates a significant p value with significance taken at the 0.05 level.

APR-DRG, All Patient Refined–diagnosis-related group; ICU, intensive care unit; IQR, interquartile range; n/a, not applicable; PCC, palliative care consultation to discuss goals-of-care; SD, standard deviation.

Future acute care utilization and accumulated direct costs were tracked forward after discharge from the index hospital admission through the end of the study period by using the medical center's cost accounting system (McKesson Health Solutions, King of Prussia, PA).⁴ (Utilization and costs incurred during the index hospitalization are described in Table 1.) Direct costs represent the best estimate of the actual cost of providing hospital services, including nursing labor, other allied health professional labor, pharmaceuticals, supplies, procedures, and testing.⁴

Measures

Independent variable

The primary independent variable was PCC, as recorded in the palliative care registry according to patient self-identified racial group (African American, White). All hospital admissions during the study period were placed into one of the four groups: (1) African Americans with PCC, (2) African Americans without PCC, (3) Whites with PCC, and (4) Whites without PCC.

Outcomes

Future acute care costs

The primary outcome was future acute care costs, defined as accumulated mean acute care costs from all hospitalizations in the health system after the index hospitalization during the study period. We measured “future acute care costs” in two ways: (1) if a patient had any direct future acute care costs after discharge from index hospitalization and (2) total direct future acute care costs that a patient had during the study period after discharge from index hospitalization. Direct costs include hospital services such as physician and nursing labor, allied health professional labor, pharmaceuticals, supplies, procedures, testing, and emergency department (ED) visits that resulted in hospitalization. Direct costs provide the best estimate of actual costs, as they exclude indirect or fixed costs (i.e., overhead costs such as the cost of utilities) that cannot be reduced by avoiding future hospitalizations.⁴

The medical center's cost accounting system, McKesson Health Solutions (King of Prussia, PA), provided acute care costs data based on charges coded during each hospital encounter throughout the study period. Health care costs outside the health system, such as ED visits at local hospitals unaffiliated with the medical center, were not available and therefore not included, and they have been identified as a limitation.

Future health care utilization

Secondary outcomes included the number of future hospital days, number of future ICU days, any ICU care, and 30-day readmissions, each in the health system after discharge from index hospitalization.

Covariates and confounders

Covariates included age, gender, Medicaid status, acute care utilization 30 days before index hospitalization, acute care costs accumulated during index hospitalization to represent acute care utilization (defined as the impact of $1,000 to show the expected increase in getting PCC), and the following clinical variables: primary diagnosis at the time of index hospitalization discharge [based on diagnosis-related group (DRG)], All Patient Refined (APR)-DRG Severity of Illness (extent of physiological decompensation) at the time of index hospitalization discharge, APR-DRG Risk of Mortality (likelihood of dying) at the time of index hospitalization discharge, “any ICU care” during index hospitalization, ICU care greater than six days during index hospitalization, and oncology services during index hospitalization to account for patients with cancer who were admitted with primary diagnoses other than cancer, for example, major symptoms related to cancer treatment.⁴

The APR-DRG variables have four subcategories (minor, moderate, major, extreme) that are systematically assigned at the time of discharge based on diagnoses and procedures coded for billing during hospitalization.^4,39 A 2018 study found that APR-DRG Risk of Mortality and Severity of Illness are better predictors of in-hospital mortality than the Charlson Comorbidity Index.⁴⁰ Although health literacy is associated with health care utilization and costs,^41–43 it was unavailable for measurement in this dataset.

Statistical analysis

The sample size met requirements for a statistical power of 0.91 and an alpha level of 0.01. The parent study excluded 0.4% of patients in its sample due to missing data (<5% of patients), resulting in no missing data for our study. Descriptive statistics characterized all variables and described the sample (Table 1). Student's t tests used means and standard deviations to describe continuous variables. Chi-squared (χ²) tests described categorical variables as frequencies and percentages. For the PCC groups, whether or not the patient changed GOC preferences was also included.

A systematic process to balance the four PCC-race groups (African Americans with and without PCC, Whites with and without PCC) was followed. We created two stratified propensity score-matching models: one for African Americans then, separately, for Whites to account for possible inherent differences and nonrandom assignment of variables in the PCC and Non-PCC groups within each race. Before propensity scoring, patients in the PCC and Non-PCC groups for each race had different baseline characteristics (Table 1). To make PCC and Non-PCC groups similar for each race, we employed a two-step matching process.

First, we used logistic regression analysis to identify factors associated with the likelihood of receiving PCC (Table 2) by using variables of gender, age, Medicaid status, primary diagnosis, APR-DRG Risk of Morality and Severity of Illness, ICU care during index hospitalization, ICU care greater than six days during index hospitalization, index hospitalization acute care costs to represent utilization, and 30-day prior inpatient hospitalization. Results from the logistic model for African Americans and for Whites are presented in Table 2. The model for African Americans had a C statistic of 0.907 and the model for Whites had a C statistic of 0.901, indicating adequate model fit.

Table 2.

Logistic Regression Analysis of Likelihood of Receiving Palliative Care Consult

Parameter	African American patients			White patients
	Estimate	SE	p Value	Estimate	SE	p Value
Intercept	−2.0415	0.1688	<0.0001	−1.8617	0.1351	<0.0001
Age (years)
18–39	−0.858	0.212	<0.0001	−0.315	0.133	0.018
40–55	−0.189	0.128	0.139	−0.148	0.094	0.113
56–65	0.222	0.113	0.0497	0.119	0.077	0.123
66–75	0.077	0.126	0.544	−0.013	0.078	0.872
>75	0	—	—	0	—	—
Gender
Male	−0.0548	0.0593	0.356	−0.1562	0.0407	0.0001
Female	0	—		0	—
Medicaida
Yes	0.114	0.082	0.164	0.141	0.095	0.139
No	0	—		0	—
Primary diagnosis
Cancer	0	—		0	—	—
Cardiovascular disorder and heart failure	−0.100	0.137	0.467	−0.492	0.102	<0.0001
Endocrine disorder	−0.449	0.334	0.18	0.082	0.235	0.727
GI disorder	−0.011	0.203	0.956	0.066	0.130	0.612
Gynecologic or urologic disorder	−0.193	0.247	0.436	−0.260	0.215	0.2259
Infectious disease and sepsis	−0.070	0.151	0.641	0.016	0.120	0.891
Neurologic disorder	−0.370	0.193	0.056	−0.254	0.154	0.099
Respiratory disorder	0.282	0.282	0.199	0.615	0.148	<0.0001
Other	−0.130	0.186	0.484	−0.288	0.153	0.059
APR-DRG risk of mortality
Minor	−2.129	0.239	<0.0001	−2.371	0.176	<0.0001
Moderate	−0.602	0.141	<0.0001	−0.423	0.094	<0.0001
Major	0.988	0.111	<0.0001	1.013	0.080	<0.0001
Extreme	0	—	—	0	—	—
ICU during index hospitalization
Yes	0.138	0.066	0.036	0.149	0.043	0.0005
No	0	—		0	—
ICU >6 days during index hospitalization
Yes	0.232	0.090	0.01	0.265	0.056	<0.0001
No	0	—		0	—
Seen by oncology in index hospitalization
Yes	0.284	0.088	0.001	0.266	0.051	<0.0001
No	0	—		0	—
Admitted to hospital 30 days earlier
Yes	1.435	0.097	<0.0001	1.589	0.064	<0.0001
No	0	—		0	—
Direct cost of index hospitalization (impact per $1,000)a	2.92 × 10−3	1.52 × 10−3	0.0535	1.72 × 10−3	9.55 × 10−4	0.072

^aFor every increase of $1,000 in direct costs, the estimate shows the expected increase in getting a palliative care consult.

Bold indicates a significant p value with significance taken at the 0.05 level.

SE, standard error.

Individual-level propensity scores were created for each patient in the sample by using output from the logistic regression models, by racial group. These individual propensity scores were then ranked and stratified into propensity tiers, from highest to lowest, to identify which patients were most likely to have received PCC (Table 3). For each racial group's model, the patients in the PCC and Non-PCC groups were similar in terms of their propensity scores, which enabled a direct comparison of PCC patients to Non-PCC patients within tiers, within each racial group (Tables 4 and 5). To understand outcomes among patients with the highest likelihood of receiving PCC per racial group, subsequent analysis focused on the two highest tiers of propensity scores (blended) for each model. Significance tests for continuous variables were analyzed by using a pooled t test, and percentage variables were analyzed by using a Cochran–Mantel–Haenszel test. Statistical significance was taken at the 0.05 level.

Table 3.

Propensity Scores Ranked from Highest to Lowest Tiers, by Race

^aPalliative care consult to discuss goals-of-care.

Shading indicates tiers that were selected for subsequent analysis to understand outcomes among patients with the highest likelihood of receiving PCC.

Table 4.

Future Costs and Acute Care Utilization After Propensity Score Matching, African American Patients with High Propensity for PCC

^aSample sizes are consistent within each column (across all outcome variables).

^bSignificance tests for the percentage variables involved a Cochran–Mantel–Haenszel test; significance tests for the continuous variables involved a pooled t test.

Bold indicates a significant p value with significance taken at the 0.05 level.

Table 5.

Future Costs and Acute Care Utilization After Propensity Score Matching, White Patients with High Propensity for PCC

^aSample sizes are consistent within each column (across all outcome variables).

^bSignificance tests for the percentage variables involved a Cochran–Mantel–Haenszel test; significance tests for the continuous variables involved a pooled t test.

Bold indicates a significant p value with significance taken at the 0.05 level.

Results

The sample for this study included 35,154 patients who were discharged from an inpatient hospitalization at an academic medical center in the Northeast region (Table 1). The sample included 1197 patients who received PCC before discharge and 33,957 patients who did not. Specifically, the sample was composed of 11,160 African Americans (PCC = 383, Non-PCC = 10,777) and 23,994 Whites (PCC = 814, Non-PCC = 23,180). At baseline, before propensity score matching, African Americans with PCC significantly differed from African Americans in the Non-PCC group (Table 1). Whites with PCC also significantly differed from Whites in the Non-PCC group before propensity score matching (Table 1). In both racial groups, PCC patients were more likely to be older; have major or extreme severity of illness; have major or extreme risk of mortality; have been hospitalized in the prior 30 days; have received ICU care (including for longer than six days); and have been seen by oncology services during index hospitalization. These differences are reflected in PCC patients' higher index hospitalization utilization and costs, suggesting that PCC patients are sicker than the general hospitalized population (Table 1). After propensity score matching, patients in each racial group did not differ in their propensity for having had PCC (Table 2). African Americans with and without PCC became similar, and Whites with and without PCC became similar, enabling us to identify differences in outcomes associated with PCC by race.

Cost outcomes

Any future acute care cost

Compared with African Americans with PCC, African Americans without PCC were more likely to incur “any future acute care cost” (blended top two tiers, 31.9% vs. 37.7%, p = 0.047) (Table 4). White PCC patients were also significantly less likely than Whites without PCC to incur “any future acute care cost” (blended top two tiers, 25.9% vs. 35.5%, p < 0.0001) (Table 5).

Total future acute care costs

Although average accumulated future acute care costs were lower for African Americans with PCC compared with those without PCC (blended top two tiers, $11,651 vs. $15,050, p = 0.09), the differences were not statistically significant (Table 4). Data support a trend toward an average difference in future acute care costs of $4,415 between PCC and Non-PCC African Americans in the highest propensity tier, but the difference was not statistically significant (p = 0.055). However, average future acute care costs were significantly lower among Whites with PCC (blended top two tiers, $8,095 vs. $16,799, p < 0.001) (Table 5). In effect, PCC in Whites is associated with an average difference in future acute care costs of $6,693 (highest tier) to $10,745 (second highest tier) per person ($8,704 difference in blended top tiers). A large effect-size difference is represented among Whites with PCC.

Health care utilization outcomes

Thirty-day readmissions

Significant differences among African Americans with and without PCC were not evident in 30-day readmission rates (blended top two tiers, 16.6% vs. 15.4%, p = 0.58), but they were evident among Whites with PCC compared with Whites without PCC (blended top two tiers, 10.2% vs. 16.7%, p < 0.0001).

Future days in hospital post-discharge

Significant differences in the number of future hospitalized days were not found among African Americans with and without PCC (blended top two tiers, 5.5 vs. 6.4 days, p = 0.34), but they were found among Whites with PCC compared with Whites without PCC (blended top two tiers, 3.7 vs. 6.3 days, p < 0.0001).

Future admission to ICU

African Americans with PCC were no less likely than African Americans without PCC to be admitted to the ICU after index hospitalization discharge (blended top two tiers, 16.6% vs. 14.2%, p = 0.25). Results were similar for Whites with PCC vs. those without PCC (blended top two tiers, 12.5% vs. 14.4%, p = 0.20).

Future days in ICU post-discharge

Significant differences in the number of future ICU days were not found among African Americans with PCC compared with those without PCC (blended top two tiers, 1.3 vs. 0.99 days, p = 0.3). Among Whites with and without PCC, significant differences in the number of future ICU days were evident in the highest propensity group but not in the top blended tiers (0.6 vs. 0.9 days, p = 0.042; blended top two tiers, 0.8 vs. 1.0 days, p = 0.15).

Discussion

This propensity-matched study found that PCC is associated with significant differences in 30-day readmissions, number of future days spent hospitalized, and accumulated mean future acute care costs (resulting in an average reduction of $8,704 per patient) in Whites, but no statistically significant differences in costs or acute care utilization in African Americans.

Differences in outcomes between African American and White PCC groups may be explained by variables not measured in our study. Socioeconomic variables such as income and education may have contributed.⁴⁴ Compared with Whites, African Americans endure systematic disadvantages, resulting in poor economic resources, higher unemployment, and lower incomes, which impact health, health literacy, and access to quality care.^44,45 Patients with less education have been found to have lower levels of health literacy,^44,46 which can influence how a patient understands, recalls, and applies information exchanged during PCC or other health care encounters.^46,47 Patients with lower health literacy have an increased risk of hospital admission,^41,42 use an inefficient mix of health care services, and have higher health care costs.^43,48

Socioeconomic disadvantages may have contributed to avoidable use of the ED.⁴⁹ African Americans also more frequently use EDs for care that does not result in hospitalization,⁵⁰ but our study excluded such visits because the variable was not available in the dataset. Cost differences among PCC and Non-PCC patients, especially African Americans, may have been greater had all ED visits been included. Research is needed to understand how PCC and Non-PCC patients, by race, use the ED and how ED use influences costs. In addition, religiosity, which is higher for African Americans⁵¹ and associated with preferences for more treatment and life prolongation near EOL,⁵² was not available as a confounder.

Unmeasured disparities in concordance between care preferences expressed during PCC and actual care received during subsequent hospitalizations¹⁹ and the time from admission to PCC^53,54 may also partly explain results. Research is needed to understand how PCC differs across racial groups and whether unmeasured factors influence findings. Finally, although health systems benefit from reducing costs, saving money should never be the primary reason for engaging patients in PCC or making clinical decisions.⁵⁵

Limitations

This study has several limitations. First, our analysis was limited to variables available in the parent dataset. As is the case with all propensity-score studies, our models cannot account for unmeasured or unknown confounding variables.^4,56 Unmeasured variables may be distributed differently in patients who received PCC and those who did not.⁴ Human error in variables involving clinician input of data, such as APR-DRGs, may also introduce bias due to misclassification errors or other inaccuracies.

Second, our analysis only examined future acute care direct costs and utilization incurred within the health system after index discharge.⁴ It did not include costs incurred between hospitalizations, outside the health system, or ED visits that did not result in hospitalization.⁴ The health system is the largest acute care provider in the Northeast, so a few patients were likely admitted to hospitals outside the system. In addition, the readmission variable applies only to patients alive 30 days after discharge from index hospitalization, which is not the entire sample population, as deaths occurred during the study period.

Finally, our study examined a single academic medical center with high acuity and a well-established palliative care team and may not represent all hospitals. Despite these limitations, study results show possible cost and utilization outcomes associated with PCC beyond the initial hospitalization among African Americans and Whites in a sample matched, within races, on known covariates.^3,4,6

Conclusion

In propensity-matched cohorts of African Americans and Whites based on which patients did or did not receive PCC, we found cost-savings and reductions in acute care use among Whites with PCC but no significant differences in African Americans with PCC. Further research is needed to explain why acute care utilization and cost disparities persist among African Americans despite PCC, and to identify workable interventions to overcome these disparities.

Funding Information

The authors acknowledge the following financial support for the research, authorship, and/or publication of this article: Dr. Starr received support from the Ruth L. Kirschstein National Research Service Award training program in Individualized Care for At Risk Older Adults at the University of Pennsylvania, National Institute of Nursing Research T32NR009356; the Rita and Alex Hillman Foundation's Hillman Scholars Program; and Jonas Philanthropies' Jonas Nurse Leaders Scholars program. Dr. Ulrich is currently supported by a National Cancer Institute/National Institutes of Health Award (R01CA196131). Dr. Meghani is currently supported by a National Institute of Health/National Institute of Nursing Research Award (R01NR017853).

Author Disclosure Statement

No competing financial interests exist.