Characteristics of the Sample
As shown in , patients ranged in age from 59 to 92 years, in education from 6 to 22 years, and in education-adjusted MMSE score from 22 to 30. Proxies ranged in age from 21 to 84 years, in education from 5 to 24 years, and in education-adjusted MMSE score from 24 to 30. A chi-square analysis indicated no significant differences between African Americans and Caucasians in regard to family status of the proxy, χ2(4, N = 64) = 2.393, p = .664. Seventy percent of African American and 74% of Caucasian proxies were adult children; 20% of African American and 24% of Caucasian proxies, spouses; and 10% of African American and 3% of Caucasian proxies, other family members.
As shown in , patient and proxy income differed significantly between race/ethnicity (Mann–Whitney U = 341, p = .018, and Mann–Whitney U = 281, p = .002, respectively). One Caucasian proxy declined to provide income information. Caucasian patients and proxies reported higher income than African American patients and proxies. Caucasians scored slightly higher than African Americans on the adjusted MMSE; however, this difference was not clinically significant.
Illness Experience and Advance Care Planning.—
As shown in , African Americans reported worse health than Caucasians (patients: Mann–Whitney U = 334, p = .009; proxies: Mann–Whitney U = 308, p = .002). Most patients and proxies reported having some experience with life-threatening illness either as a patient or caring for an ill loved one. There were no racial/ethnic differences in levels of experience with life-threatening illness for patients or proxies. A minority of participants possessed a living will (35%) or durable power of attorney for health care (31%). Compared with Caucasian patients, significantly fewer African American patients reported having a living will, χ2(1, N = 64) = 5.869, p = .015, or durable power of attorney for health care, χ2(1, N = 64) = 5.173, p = .023. Among proxies, however, no significant differences were found by race/ethnicity.
Illness Experience and Advance Care Planning (N =128 consisting of 64 dyads, with 34 Caucasian dyads and 30 African American dyads)
Most patients and proxies (62.5%) reported that they had talked with at least one other family member about their end-of-life medical wishes (see ). For patients, there were no significant differences between African Americans and Caucasians in their reports of talking to family members about end-of-life wishes, χ2(1, N = 64) = 1.318, p = .251. However, among proxies, a greater number of Caucasians than African Americans reported engaging in such discussions with one or more family members, χ2(1, N = 64) = 11.628, p = .009.
Measure of Understanding.—
To ensure that patients and proxies understood the questions on the LSPQ, they were asked to define each of the illnesses (AD, emphysema, coma, and cancer) and each of the treatments (CPR, artificial feeding and fluids, and palliative care) in their own words. Thirteen (10%; 6 African American, 7 Caucasian) participants included definition by example as an adjunct to at least one of their answers (e.g., the participant informed the interviewer that he or she was a caregiver for a relative living with AD). For each item missed, the corresponding items in that participant’s LSPQ were removed from the analysis. Two Caucasian patients, three African American patients, and one African American proxy missed one definition each. Three missed the definition of emphysema and three missed a definition of one of the treatment options (palliative care, CPR, and artificial feeding and fluids, respectively).
LSPQ Means by Family Status and Race/Ethnicity.—
Overall desire for treatment among patients was significantly higher for African Americans, t(57) = −4.539, p < .001. Overall desire for treatment for patients as reported by proxies was also higher among African Americans compared with Caucasians, t(52.488) = −2.176, p = .034.
Outcome Variable: Patient–Proxy Agreement/Raw and Absolute Difference Scores
We conducted a 5 (illness) × 3 (treatment) × 2 (race) mixed ANOVA predicting the raw difference scores between patient and proxy, with illness and treatment analyzed as within-subjects factors and race analyzed as a between-subjects factor. The mean raw difference scores within this design are presented in . Within-subjects tests were analyzed using Wilk’s lambda, a commonly used multivariate test that, unlike repeated measures analysis, does not make the assumption of sphericity. Post hoc analyses were only carried out on significant effects making the use of Fisher’s least significant difference (LSD) measure appropriate. The only significant result was a main effect of race, F(1, 57) = 5.304, p = .03. Specifically, Caucasian proxies indicated a tendency to overtreat in comparison with patients’ wishes (M = 0.54), whereas African American proxies tended to undertreat (M = −0.17) their older family members. As hypothesized, this mean reflects better accuracy in predicting patients’ end-of-life treatment preferences on the part of African Americans (perfect prediction would be represented by M = 0) when overtreatment and undertreatment errors are combined.
Raw Differences on Life-Support Preferences/Predictions Questionnaire by Illness, Treatment, and Race
We next used the same 5 (illness) × 3 (treatment) × 2 (race) design to predict the absolute value of the differences between patient and proxy. The mean absolute values of the difference scores within this design are presented in . We observed a significant main effect of illness, Wilk’s lambda = .703, F(4, 54) = 5.714, p = .001. LSD post hoc analyses indicate that the absolute value differences were significantly smaller for coma (M = 0.957) than for AD (M = 1.46), emphysema (M = 1.53), cancer without pain (M = 1.48), and cancer with pain (M = 1.12). None of the other mean comparisons was significant. We also observed a significant interaction between race and illness, Wilk’s lambda = .838, F(4, 54) = 2.61, p = .05. LSD post hoc analyses showed that, for emphysema, African Americans had smaller absolute value differences than Caucasians (Ms = 1.25 and 1.81, respectively) but that there were no significant racial differences for the other illnesses.
Absolute Differences on Life-Support Preferences/Predictions Questionnaire by Illness, Treatment, and Race
Mediators and Moderators of Race Effects
We investigated whether religiousness/spirituality, decision-making style, trust in the health care system, and prior advance care planning could possibly explain the observed relation between race/ethnicity and the patient–proxy difference scores. Using the guidelines proposed by Baron and Kenny (1986)
, we first investigated whether these variables had significant relations with patient–proxy difference scores. None of the relations with the difference scores was significant, so we concluded that none of these variables could be used as mediators to explain the effects of race.
Next, moderation analyses were conducted to explore the possible influences of religiousness/spirituality, decision-making style, trust in the health care system, and prior advance care planning on the relation between race/ethnicity and patient–proxy treatment preference agreement. The general linear model was used to predict raw difference scores (combining across illness and treatment) from race, the main effects of the proposed moderators, and the interaction between each moderator and race/ethnicity. The only significant interaction effect was found between race/ethnicity and advance care planning, F(1, 47) = 4.15, p = .047, accounting for 8% of patient–proxy agreement. Specifically, when advance care planning is high, African Americans and Caucasians display relatively high patient–proxy agreement on end-of-life decisions. When advance care planning is low, African American proxies tend to make more undertreatment errors and Caucasian proxies tend to make more overtreatment errors ().
Overall difference score between patients and proxies for end-of-life treatment preferences as a function of the interaction of race/ethnicity and advance care planning.
We conducted additional analyses to determine if the effect of race on the overall absolute value difference scores (again combining across illness and treatment) was moderated by religiousness/spirituality, decision-making style, trust in the health care system, or prior advance care planning. There were no significant effects in this model (all ps > .14).