The Influence of Race/Ethnicity and Education on Family Ratings of the Quality of Dying in the ICU

Abstract

Context

Racial and ethnic differences in end-of-life care may be attributable to both patient preferences and healthcare disparities. Identifying factors that differentiate preferences from disparities may enhance end-of-life care for critically ill patients and their families.

Objectives

To understand the association of minority race/ethnicity and education with family ratings of the quality of dying and death, taking into consideration possible markers of patient and family preferences for end-of-life care as mediators of this association.

Methods

Data were obtained from 15 ICUs participating in a cluster-randomized trial of a palliative care intervention. Family members of decedents completed self-report surveys evaluating quality of dying. We used regression analyses to identify associations between race/ethnicity, education and quality of dying ratings. We then used path analyses to investigate whether advance directives and life-sustaining treatment acted as mediators between patient characteristics and ratings of quality of dying.

Results

Family members returned 1290 surveys for 2850 decedents. Patient and family minority race/ethnicity were both associated with lower ratings of quality of dying. Presence of a living will and dying in the setting of full support mediated the relationship between patient race and family ratings; patient race exerted an indirect, rather than direct, effect on quality of dying. Family minority race had a direct effect on lower ratings of quality of dying. Neither patient nor family education was associated with quality of dying.

Conclusion

Minority race/ethnicity was associated with lower family ratings of quality of dying. This association was mediated by factors that may be markers of patient and family preferences (living will, death in the setting of full support); family member minority race/ethnicity was directly associated with lower ratings of quality of dying. Our findings generate hypothesized pathways that require future evaluation.

INTRODUCTION

High-quality healthcare should be available to all patients, regardless of race or ethnicity. However, racial and ethnic disparities in healthcare persist. Compared to non-minority patients, individuals of minority race and ethnicity have less access to preventative services [1], receive fewer potentially life-saving procedures [2–4], and experience higher morbidity and mortality from both acute and chronic illnesses [2, 5, 6]. These differences in care extend across the spectrum of healthcare and include important variations in end-of-life care [7]. Minorities are less likely to have advance directives [8–10], are more likely to receive high intensity care at the end of life [11, 12], and are less likely to receive care consistent with stated preferences [13, 14]. In addition, rates of hospice utilization are lower and rates of hospice disenrollment higher for minorities [15–17]. Although the great majority of these studies have focused on African American and, to a lesser degree Hispanic minorities, limited data available from other racial/ethnic groups support similar findings [2, 7].

Although some racial and ethnic differences in end-of-life care may be attributed to patients’ informed preferences and decision-making, some differences may not reflect patient values and goals. Instead, they may be the product of poor quality communication about goals of care and limited access to hospice and palliative care [7, 18, 19]. Differences that result from informed decision-making should be respected, but differences in end-of-life care that result from inadequate communication or a lack of resources must be addressed and corrected. For example, compared to whites, family members of African American decedents are more likely to cite absent or problematic communication with physicians about end-of-life care [20], and African American patients report less satisfaction with quality of care received at the end of life [8]. It can be difficult to separate preference-driven differences from healthcare disparities. However, an understanding of these distinctions is an important step to improving quality of end-of-life care for all.

In order to understand differences in end-of-life care by race/ethnicity, other complex relationships with race/ethnicity need to be considered. These include factors such as patients’ and families’ spirituality, culture, health literacy, and socioeconomic status [2, 10, 21–24]. Educational attainment, notable for its role in health care disparities in general [25, 26], is an additional factor that may contribute to differences in end of life care preferences. Although few studies have examined these associations, findings suggest an independent role for educational attainment controlling for race/ethnicity, with a lower level of educational attainment associated with preferences for more aggressive end-of-life care among both African American and Latino subjects [22, 27]. If education is found to be associated with racial and ethnic differences in end-of-life care, interventions utilizing innovative methods may help overcome barriers associated with lower educational attainment [22, 27].

In this study, we assessed associations between race/ethnicity, education and quality of end-of-life care in the intensive care unit (ICU). In addition, we examined these associations as mediated by two processes of care measures that serve as markers of patient and family preferences for intensity of care at the end of life: the presence of a living will prior to ICU admission and death in the setting of full support. We examined the following questions: 1) are minority racial/ethnic status or education independently associated with family ratings of quality of dying?, and 2) do advance care planning and markers of the intensity of care at the end of life mediate the relationship between minority race/ethnicity or education and family ratings of quality of dying?

METHODS

Design Overview

This was an observational analysis evaluating associations between race/ethnicity, education level and family ratings of quality of dying in the ICU. Data were drawn from a multicenter, cluster-randomized trial of an interdisciplinary, multi-faceted intervention to improve ICU clinicians’ ability to provide end-of-life care to critically ill patients and their families [28]. The intervention was not associated with improvements in the quality of end-of-life care, so all patients were combined into one dataset for the purposes of this analysis. All study procedures were approved by institutional review boards at all participating institutions.

Population and Setting

Eligible patients were those who died in an ICU after a minimum stay of 6 hours or who died within 30 hours following transfer from the ICU to another hospital location. Patients were identified by daily examination of admission, discharge, and/or transfer records at 15 Seattle-Tacoma area hospitals, with study deaths extending from August 2003 to February 2008. Of the 15 hospitals, three were university-affiliated teaching hospitals, three were community-based teaching hospitals and nine were community-based non-teaching hospitals.

Data Collection and Variables

Methods for survey data collection have been previously described [28]. Briefly, study materials were sent to patients’ homes 4 to 6 weeks after death, addressed to the family of the patient. Responses were requested from the person most knowledgeable about the patient’s end-of-life experiences. Questionnaire materials included an incentive, postage-paid return envelope, and questionnaire booklet. Additional follow-up mailings included reminder/thank-you postcards sent 3 weeks after initial distribution and a second set of materials sent to non-respondents after 5 weeks.

In addition to questionnaire data, demographic and illness information was identified from medical records (i.e., patient age, patient sex, presence of a living will, dying in the setting of full support) and publicly available Washington State death certificates (i.e., patient race/ethnicity, marital status, education, cause of death). Trained chart abstractors reviewed patients’ medical records using a standardized protocol. Training included a minimum of 80 hours of practice abstraction with instruction on the protocol, guided practice charts, and independent chart review. Practice abstraction was followed by reconciliation with the research abstraction trainers, requiring 95% agreement with the trainer before being permitted to perform independent review. Five percent of all charts were co-reviewed to ensure greater than 95% agreement on all data elements [28].

Outcome

Our outcome measure, obtained from family surveys, was a single item providing an overall rating of quality of dying and death (QODD-1). This item is part of a larger 22-item questionnaire (QODD) that includes symptom control, preparation, connectedness, and transcendence domains [29]. The QODD-1 asks family members to rate the following on a scale from 0 “terrible” to 10, “almost perfect”: “Overall, how would you rate the quality of your loved one’s dying?” The QODD-1 score has been shown to be significantly associated with quality of care in the ICU. It has been shown to differentiate higher quality of dying and death for decedents receiving lower intensity of care at the end of life and for families participating in patient- and family-centered decision-making [30]. Significant differences of 1 point have been found between patients for whom comfort care orders were in place (or all orders were discontinued) as compared to patients without comfort care orders.

Predictors

Our primary predictors were race/ethnicity and education level, and because prior research has suggested that ratings of quality of care at the end of life may vary depending on both patient and family characteristics, we considered both [30, 31]. Patient race/ethnicity was determined primarily from death certificates as these were more complete than the medical record. We compared race/ethnicity from the medical record and the death certificate when both were available and found a high level of concordance between these two sources of data for the four largest race/ethnicity categories (κ, 0.85–0.95) [32]. Patients were divided into two groups: non-minority (non-Hispanic white) and minority (Hispanic and/or any other non-white race). Education level of the patient was also obtained from the death certificate and was grouped into 6 levels, ranging from ≤ 8^th grade level to post-college study. Race/ethnicity and education level of the family member were obtained from a questionnaire.

Mediators

Past studies have demonstrated that minority status is associated with a lower likelihood of having a living will [10, 20] and a higher likelihood of dying in the setting of full support [11, 12, 32]. In order to address the potential influence of these factors, we adjusted our model to include two mediating variables: 1) presence of a living will; and 2) dying in the setting of full support. A living will was defined as present if there was documented evidence in the medical record that a living will had been completed prior to the patient’s admission to the ICU. Dying in the setting of full support was defined as death in the presence of life-support measures (i.e., use of mechanical ventilation, vasopressors, IV hydration, nutrition, CPR); if a “do not attempt resuscitation” (DNAR) order was placed prior to death but the patient continued to receive life-support measures at the time of death, this was considered death in the setting of full support. This definition of full support has been used successfully in prior analyses to examine the quality of care [33].

Potential Confounders

For models including patient characteristics, we examined patient age, sex, cause of death (cancer, trauma, other) and assessment period (before or after the intervention) as possible confounders. For models including family characteristics, we examined family age, sex, being a spouse and assessment period (before or after the intervention). In all models, hospital site was included as a covariate because we have previously found important difference in outcomes by site [34].

Analysis

Differences between minority vs. non-minority racial/ethnic groups were identified using t-tests and Pearson chi square statistics as appropriate. We used robust multivariable linear regression models to assess associations between predictors and the outcome (QODD-1); models were run separately for patient and family characteristics and were initially run without the mediating variables (presence of a living will, dying in the setting of full support). We tested for confounding only if the association between a predictor of interest and the outcome in a “base model” (i.e., a model with hospital site as the only covariate) had a P-value ≤0.20. If addition of a potential confounder changed the coefficient of the predictor of interest by more than 10% as compared with the base model, the variable was defined as a confounder and used as a covariate in the final model. Significant findings were defined as P <0.05.

Finally, we developed and tested a theoretically sound path model in which the available patient characteristics (age, racial/ethnic minority status, and education) served as direct predictors of two mediating variables (presence of a living will, dying in the setting of full support) and the outcome (QODD-1). Presence of a living will directly predicted death in the setting of full support and ratings of quality of dying; receipt of life support directly predicted ratings of quality of dying; and available family characteristics (racial/ethnic minority status and education) served as direct predictors of ratings of quality of dying. From this model, we sequentially removed paths that were not statistically significant, in decreasing order of the P-values associated with their coefficient estimates, until all remaining paths had P-values <0.05. All associations in the model were adjusted for hospital site as a fixed effect, with all hospital associations retained, irrespective of their statistical significance. We evaluated the fit of the final model based on the following criteria: P-value for the chi-square test of fit >0.05; probability >0.50 that the root mean square error of approximation (RMSEA) was ≤0.05; value <0.06 for the top of the RMSEA’s 90% confidence interval; Comparative Fit Index (CFI) >0.95; and Tucker-Lewis Index (TLI) >0.95 [35].

RESULTS

Patient and Family Characteristics

Of the 2850 eligible decedents for whom an address was available, 1290 family member returned surveys (RR 45%:1290/2850). Eighty-six percent of patients with a returned survey were non-minority and 14% were minority. Decedents identified as minority included: Hispanic, n=31 (2.4%), African American, n=49 (3.8%), Asian, n=72 (5.6%), Native American or Alaskan Native, n=17 (1.3%), Pacific Islanders, n=7 (0.5%) and other races, n=6 (0.5%). Six decedents were identified by both ethnicity (Hispanic) and a minority race. Compared to non-minority patients, minority patients were significantly younger, had lower levels of educational attainment, were less likely to have a living will, and were more likely to die in the setting of full support (Table 1). Patients whose family members did not return a survey did not vary by age, gender, education or cause of death.

Table 1.

Patient characteristics (n=1290)

Characteristic	Non-minority (n=1114)	Minority (n=176)	P*
Age (mean± SD)	70.8 ±14.5	64.5 ±17.1	<0.001
Female % (n)	41.3% (460)	38.6% (68)	0.505
Education Level % (n)a			<0.001
</= 8th grade	5.6% (62)	15.3% (26)
Some high school	7.7% (85)	9.4% (16)
High school grad/GED	37.6% (416)	35.3% (60)
Some college	25.7% (285)	27.1% (46)
4 year college	16.6% (184)	10% (17)
Post college degree	6.8% (75)	2.9% (5)
Cause of death % (n)			0.345
Trauma	10.5% (117)	11.9% (21)
Cancer	15.4% (172)	11.4% (20)
Other cause	74.1% (825)	76.7% (135)
Documentation of having a living will % (n)b	45.9% (482)	25.5% (42)	<0.001
Dying in the setting of full support % (n)c	21.4% (224)	31.9% (52)	0.003

^*P values are based on t-tests for continuous variables and Pearson chi-square tests for categorical variables

^a. 13 missing information

^b. 74 missing information

^c. 81 missing information

Of the 1290 family surveys returned, 1249 provided responses regarding their race/ethnicity. The percent of non-minority and minority individuals in the family sample was the same as that of patients, with 86% identifying as non-minority and 14% identifying as minority, but the correlation between the race/ethnicity of patients and families was 0.72 suggesting that not all patients and families fell into the same categories. Families reported as minority included: Hispanic, n=31 (2.4%), African-American, n=50 (3.9%), Asian, n=68 (5.3%), Native American or Alaskan Native, n=34 (2.6%), Pacific Islanders, n=10 (0.8%) and other races, n=2 (0.2%). Two family members were identified by both ethnicity (Hispanic) and a minority race. Minority family members were significantly younger and less likely to be a spouse; they did not differ by sex or education level (Table 2). Family members of minority patients were significantly less likely to return a survey as compared with those who did (25.1% vs.13.6%, p<0.001).

Table 2.

Family characteristics and survey scores (n=1249)^a

	Non-minority (n=1068)	Minority (n=181)	Pb
Characteristic
Age (mean ± SD)c	59.3 ± 14.2	50.9 ± 14.7	<0.001
Female % (n)d	67.6% (719)	72.9% (132)	0.152
Spouse % (n)e	47.7% (508)	34.8% (63)	0.001
Education Level % (n)f			0.198
</= 8th grade	0.8% (9)	2.8% (5)
Some high school	3% (32)	2.2% (4)
High school grad/GED	17.6% (187)	19% (34)
Some college	42.7% (453)	40.2% (72)
4 year college	19.5% (207)	22.3% (40)
Post college degree	16.4% (174)	13.4% (24)
Family survey score (mean± SD)
QODD-1g	7.1 ± 3.0	6.1 ± 3.6	0.001

^a. 41 respondents did not provide answers to allow identification as non-minority/minority

^b. P values are based on t-tests for continuous variables and Pearson chi-square tests for categorical variables

^c. 4 missing information

^d. 4 missing information

^e. 2 missing information

^f. 8 missing information

^g. 76 missing information

Table 2 compares the average QODD-1 scores between non-minority and minority family members. In an unadjusted analysis, minority family members reported significantly lower QODD-1 (p< 0.001) scores compared to non-minority family members.

Race/Ethnicity, Education, and Ratings of Quality of Dying

In models adjusted for education level and confounders (hospital, age), patient and family minority race remained significantly associated with lower QODD-1 scores (Table 3). In models adjusted for minority race and confounders (hospital, age), education was not significantly associated with ratings on the QODD-1 for either patient or family models (Table 3).

Table 3.

Adjusted associations between race/ethnicity, education^a and family-reported QODD-1

	N	b	95% CI	P
Patient characteristicsb	1199
Minority		−0.784	−1.335, −0.234	0.005
Education		0.055	−.0928, 0.203	0.464
Family characteristicsc	1165
Minority		−0.660	−1.224, −0.097	0.022
Education		0.0418	−0.140, 0.223	0.651

^a. Education used as an ordinal variable, with the following 6 levels: <=8^th grade, some high school, high school graduate/GED, some college, 4 year college degree, post-college study

^b. Adjusted for hospital and patient age

^c. Adjusted for hospital and family age

Race/Ethnicity, Mediating Variables and Ratings of Quality of Dying

Path analysis tested both direct and indirect associations with the family ratings of the patient’s quality of dying. The final model, which showed excellent fit to the data, included three exogenous predictors (patient age and race/ethnicity and family race/ethnicity), two mediating variables (presence of a living will, death in the setting of full support), and one outcome (family ratings of quality of dying). (Figure 1) Older patients were more likely to have a living will, which was in turn associated with a lower likelihood of death in the setting of full support. Death in the absence of full support was associated with higher ratings of quality of dying. Older patients had higher ratings of quality of dying irrespective of their living will and life support status. Unlike patient age, patient racial/ethnic minority status exerted its full effect on downstream events solely through its association with living wills. Racial/ethnic minority patients were significantly less likely than non-minority patients to have living wills; this difference indirectly affected their receipt of life support at the time of death, with death in the setting of full support being associated with lower family ratings of quality of dying. In contrast, family race/ethnicity had a direct effect on ratings of quality of dying that was not mediated by the other included variables, with family respondents of minority race/ethnicity providing lower ratings of quality of dying. An initial model that included both patient and family member education levels revealed no associations between these variables and either the mediators or the outcome.

Figure 1. Path analysis of factors influencing family ratings of the quality of dying and death.

*** = P < 0.001; ** = P < 0.01; * = P < 0.05

Model fit:

Chi-square₅ = 5.222, P = 0.5156

RMSEA = 0.000; 90% C.I. = 0.000, 0.034; probability of RMSEA≤.05 = 0.998

CFI = 1.000; TLI = 1.011

DISCUSSION

We found that patient and family minority race/ethnicity were associated with significantly lower family ratings of quality of dying in the ICU. These findings support previous studies reporting lower quality of end-of-life care as experienced by minority patients whether measured by ratings provided by bereaved family member ratings in a variety of settings [20], with processes of care measures such as concordance between preferences and care received [13, 14], or the decedent’s receipt of palliative care measures [32]. Given the already high but increasing proportion of deaths that occur in or shortly after a stay in the ICU [36–39], this additional documentation of racial/ethnic variability in ratings of quality of dying and death in the ICU reinforces the need to identify and remedy differences that may be attributable to disparities rather than preferences.

In contrast to our finding of differences by race/ethnicity, we did not find an association between educational attainment and family members’ quality of end-of-life ratings, controlling for race/ethnicity. This finding differs from some prior studies in which education has been found to play a role in end-of-life care among minority populations [40, 41] but may be more in line with studies suggesting that health literacy, rather than educational attainment, may account for differences in care experienced by racial/ethnic minorities [22, 27, 42, 43]. It is also possible that our setting (the ICU) and outcome (quality of dying and death rating) may have altered or influenced the expression of differences attributable to education. For instance, previous study of ratings of satisfaction with physician communication as a measure of quality of end-of-life care found racial/ethnic differences mediated by socioeconomic status [44], suggesting that the influence of race/ethnicity and education may vary depending on the specific assessment of quality of care used. Importantly, because many recommendations to eliminate racial and ethnic disparities in health care in general [45], and end of life care specifically [18, 46], have focused on improving patient education and knowledge, it is important to continue to explore the role educational attainment, along with health literacy, may play in assessments of quality of end-oflife care. This exploration would benefit from the addition of other measures assessing quality of end-of-life care (e.g. symptom assessment, treatment; concordance between care desired and received; patient- and family-centered communication), in combination with family ratings.

We also performed a path analysis that included two variables measuring processes of care at the end of life: the presence of a living will and death in the setting of full support. Multiple studies have demonstrated that minorities are less likely to have advance directives [10, 20] and are more likely to die in the setting of full support [11, 12, 32]. We used these variables as markers of patient and family preferences for intensity of care at the end of life, reasoning that if the association between race/ethnicity and ratings of quality of dying were due exclusively to differences in preferences for intensity of care at the end of life, these differences may be more likely to represent differences in preferences rather than healthcare disparities. The presence of an advance directive indicates pre-hospital decision-making by the patient, which generally prescribes a preference for limitations on intensity of care at the end of life. Similarly, death in the setting of full support likely represents, in part, a combination of patient and family preferences for intensity of care at the end of life. Our analyses suggested that these factors did in fact influence ratings of quality of dying provided by minority family members. For patients, race/ethnicity had an indirect effect on family ratings of quality of dying through the presence or absence of a living will, which in turn influenced the likelihood of death in the setting of full support. For family members, race/ethnicity directly influenced ratings of quality of dying. These associations inform, but do not resolve, the question of whether preferences, disparities or a combination of the two account for differences in the quality of end-of-life care as experienced by patients and families from racial and ethnic minorities. Were these differences in care determined by actual preferences for more life extending care? Or rather, did these differences in care arise from inadequate communication about advance care planning - planning that may have reduced the receipt of aggressive measures at the end of life? Our findings suggest a hypothesized path for these relationships, but further exploration of measures that may reflect informed preferences for care at the end of life will be important in order to fully elucidate these relationships.

This study had several important limitations. First, we focused on the role of education, as defined by educational attainment, as it relates to race/ethnicity and quality of end-of-life care; we did not include health literacy, an associated factor that has been found to play an important role distinct from educational attainment in patient and family experiences with end-of-life care [22, 27, 42, 43]. Second, some of our variables, such as documentation of the presence of an advance directive in the medical record and identification of race from the death certificate, may be incompletely documented or inaccurately recorded [47]. Third, we were unable to examine the experiences of specific minority groups due to their relatively small number. Thus, the generalizability of our findings to particular minority experiences is limited. Fourth, we examined quality of dying and death with a single measure, family ratings on the QODD-1, and may have missed associations that would have been supported with other validated measures of quality of end-of-life care, such as symptom management, communication, and care concordance. Further, although scores on the QODD-1 varied significantly between minority and non-minority groups, the minimally clinically significant difference on the QODD-1 has not been defined and, therefore, the meaningfulness of these differences should be cautiously interpreted. Last, although our path model displayed adequate fit and was designed to test hypotheses that were informed by existing evidence, it was only one of several models that we could have proposed; findings from this model should be tested for reproducibility in other data.

We found that patient or family minority race/ethnicity was associated with lower ratings of quality of dying, independent of education. We also found that the presence of a living will and death in the setting of full support mediated the relationship between race/ethnicity and ratings of quality of dying, but only when examining patient race/ethnicity. When examining family experiences of quality of end-of-life care, family race/ethnicity was not mediated by the presence of a living will or death in the setting of full support, but was directly associated with ratings of quality of dying. To the extent that evidence of a living will or death in the setting of full support are effective markers of informed preferences for intensity of care, our findings suggest that lower ratings of quality of dying may be a downstream result of patient and family preferences for more aggressive care at the end of life. However, if lack of a living will and death in the setting of full support reflect preferences that arose in the absence of high-quality communication from providers, then these differences in ratings of quality of dying may be more representative of healthcare disparities. Further research that aids in our understanding of these associations will be essential in order to differentiate end-of-life care directed by informed preferences from end-of-life care chosen in the presence of deficient communication.