Understanding End-of-Life Preferences: Predicting Life-Prolonging Treatment Preferences Among Community-Dwelling Older Americans

Abstract

Objectives.

To determine how demographic, socioeconomic, health, and psychosocial factors predict preferences to accept life-prolonging treatments (LPTs) at the end of life (EOL).

Methods.

This is a retrospective cohort study of a nationally representative sample of community-dwelling older Americans (N = 1648). Acceptance of LPT was defined as wanting to receive all LPTs in the hypothetical event of severe disability or severe chronic pain at the EOL. Participants with a durable power of attorney, living will, or who discussed EOL with family were determined to have expressed their EOL preferences. The primary analysis used survey-weighted logistic regression to measure the association between older adult characteristics and acceptance of LPT. Secondarily, the associations between LPT preferences and health outcomes were measured using regression models.

Results.

Approximately 31% of older adults would accept LPT. Nonwhite race/ethnicity (odds ratio [OR] 0.54; 95% CI 0.41, 0.70; white vs. nonwhite), self-realization (OR 1.34; 95% CI 1.01, 1.79), attendance of religious services (OR 1.44; 95% CI 1.07, 1.94), and expression of preferences (OR 0.54; 95% CI 0.40, 0.72) were associated with acceptance of LPT. LPT preferences were not independently associated with mortality or disability.

Conclusions.

Approximately one-third of older Americans would accept LPT in the setting of severe disability or severe chronic pain at the EOL. Adults who discussed their EOL preferences were more likely to reject LPT. Conversely, minorities were more likely to accept LPT. Sociodemographics, physical capacity, and health status were poor predictors of acceptance of LPT. A better understanding of the complexities of LPT preferences is important to ensuring patient-centered care.

Background

Receipt of life-prolonging treatment (LPT) at the end of life (EOL) varies widely.^1-4 One source of variation could be patient preferences for LPT. A better understanding of the variation in older adult LPT preferences could help guide patient-clinician communication, clinical decision making, and has societal implications in planning for the aging population.

Prior work elucidating EOL preferences has typically focused on specific EOL treatments and populations,^5-7 elucidated preferences from family members of decedents retrospectively,⁸ and most use data from regional or racially homogeneous populations.^9,10 Understanding racial variation in EOL preferences is also important as racial variation in EOL treatment intensity has been consistently documented, with black adults more likely to receive high-intensity care compared with white adults.^1,2,7,11 We will build on this prior work by exploring a nationally representative diverse population that oversamples blacks with LPT preferences ascertained before EOL.

In this context, we describe LPT preferences and examine predictors of LPT preferences among community-dwelling older adults in a nationally representative sample.

Methods

Data

We performed a retrospective cohort study of a nationally representative sample of community-dwelling older adults to estimate the association between acceptance of LPT and individual-level factors. Secondarily, we explored the association between acceptance of LPT and mortality and disability in a retrospective cohort study, using prospectively collected preference data. Data for both analyses are from the National Health and Aging Trends Study (NHATS), a longitudinal study of health, aging, and disability of 8245 Medicare recipients aged 65 and older in the U.S.¹² NHATS used age-stratified random sampling to obtain a representative sample of Medicare recipients,¹³ and black and older participants were oversampled. Preference questions were administered as part of the EOL module, which was received by one-third of the NHATS sample (N = 1999) in 2012. Annual follow-up interviews were conducted from 2013 to 2016, in which mortality and measures of function were collected (2012–2016, N = 1648).

Measures

EOL Plans and Care

Participants of the EOL plan and care module were asked questions about LPT in two hypothetical situations. Specifically, participants were asked to suppose that they were at the end of their life and had a serious illness, and whether they would want to receive or stop/reject all LPTs in the event: 1) they could speak, walk, and recognize others but were in constant and severe physical pain; and they were not in pain but could not speak, walk, or recognize others. A binary measure of acceptance of LPT was created by classifying respondents by whether they wanted to receive (accept) LPT in either scenario or whether they would want to stop (reject) LPT in both scenarios.

Expression of EOL preferences was affirmed by a positive response to any of the three options: 1) durable power of attorney for health care; 2) living will; and 3) spoken to anyone about the types of medical treatment they want/do not want if they became seriously ill in the future.

Predictors of Preferences

Martial status, living arrangement, education, income, hospital stays, religious services attendance, and assets were measured using NHATS survey questions. Race was self-reported. Health status was measured using a five-point scale (excellent-poor), combining excellent and good. Participants self-reported on ability to carry out a series of tasks, and a physical capacity score was created.¹⁴ A count of 10 self-reported comorbidities was included as a measure of health. An immediate word recall task was used as a measure of cognition.¹⁴ Depressive symptoms were assessed using the PHQ-2.¹⁵ Self-efficacy, resilience, and self-realization were assessed using items adapted from Midlife in the U.S study.¹²

Health Outcomes

Mortality was ascertained by NHATS field staff and the month and year at which that status was determined. Four measures of disability were created. 1) count of number of self-care/mobility tasks with any difficulty; 2) a dichotomous indicator of any difficulty on any self-care/mobility task; 3) count of number of self-care or mobility task with a high level of disability (high difficulty, needing assistance, or the inability to do the activities); and 4) a dichotomous indicator of high difficulty with any task.¹⁶ Self-care tasks included eating, dressing, toileting, and bathing/washing. Mobility tasks included getting out of bed, getting around inside, and getting around outside.

Statistical Analysis

We examined variation in LPT preferences and described predictive factors and expression of EOL preferences based on their LPT preference. We then built logistic regression models to estimate the association of participant characteristics and acceptance of LPT. First, we modeled acceptance of LPT adjusting for demographic and socioeconomic factors, as we expected race and socioeconomic status to be associated with LPT (Model 1). Next, we added health and medical factors (Model 2), which could confound or mediate the relationship between sociodemographic factors and LPT. Finally, we included psychosocial predictors (Model 3).

Then, in a secondary hypothesis-generating analysis, we modeled the association between LPT preferences and expression of preferences, with mortality and disability outcomes longitudinally. The pathways linking preferences and expression to outcomes are likely complex and variable. Moreover, within the context of our limited prior data and theory, it is possible to imagine different pathways linking preferences and outcomes resulting in the same, or different, associations. For example, preferences for LPT may result in increased survival, but with increased disability, they suffer a severe stroke (e.g., mediated via a decision to opt for feeding tube placement). But preferences against LPT may reflect increased reduced baseline functioning and therein serve as markers for reduced survival and increased disability. In both cases, the same disability patterns would emerge, but divergent mortality would be observed. As such, the purpose of our analysis was not to test a specific hypothesis but rather to inform future research by determining if a clear and marked pattern emerged relating preferences, expression, and outcomes in this data set, which is uniquely well positioned to explore this question.

All outcome models included LPT preferences, EOL preference expression, and the interaction between preferences and preference expression because we hypothesized that expressed preferences would have a stronger association with outcomes than unexpressed preferences. Cox proportional hazards regression models were then used to examine the independent association of EOL preferences and preference expression on mortality (2012–2016) after accounting for demographics, comorbidities, and cognition. For disability outcomes, multilevel poisson regression was used to estimate the association between the count of self-care and mobility difficulties and LPT and expression of EOL preferences after adjusting for the same covariates across NHATS waves with a random patient-level intercept. Binary disability models used the sample approach but with using multilevel logistic regression. Finally, to address possible reverse causation, we modeled associations between preferences and incident disability in participants with no self-care or mobility limitations at baseline, using Cox proportional hazard models. The complex survey design was accounted for in all analyses by applying survey weights using established methods,¹⁷ and analyses were run in Stata (Version 15; College Station, TX).¹⁸

Results

Description of the Sample

A total of 1999 participants were included, and of those, 57% were women, 68% were white, and 23% were 80 and older.

Description of NHATS Participants by LPT Preferences

Just more than 30% of the older adults would accept LPT, and 73% expressed their EOL preferences. The most prominent bivariate differences in EOL preferences for LPT were by race/ethnicity, living situation, religious service attendance, and expression of EOL preferences (Table 1). Minorities were more likely to accept LPT than non-Hispanic whites (72.6% vs. 27.6%, P < 0.01). Similarly, participants who attended religious services were more likely to accept LPT than their counterparts (63.6% vs. 54.5%). Older adults who expressed their EOL preferences were more likely to reject LPT at the EOL (72%) than those who had not expressed their preferences (60%) (Table 2).

Table 1.

Survey-Weighted Column Means (SE) and Proportions (%) of Model Covariates for NHATS Participants by LPT Preferences (Round 2, 2012)^a

	Reject LPT	Accept LPT
Sex (%)
Male	44.6	45.0
Female	55.4	55.0
Race (%)
Non-Hispanic white	73.3	85.1b
Minority	26.7	14.9
Age (%)
65–74	48.9	51.8b
75–84	38.6	34.9
85+	12.5	13.3
Married (%)	60.7	54.6
Income (natural log)	10.36 (0.05)	10.40 (0.04)
Education (%)
Less than high school	23.2	18.5b
High school	51.5	56.5
College and higher	25.3	25.0
Live alone (%)	24.7	32.4b
Parent (%)	89.4	90.5
Mother living (%)	4.7	5.6
Father living (%)	0.3	1.2b
Assets (%)
Lowest tertile	32.5	30.6b
Middle tertile	42.6	47.0
Highest tertile	24.9	22.4
Self-rated health (%)
Good	79.1	78.6b
Fair	15.1	16.3
Poor	5.8	5.1
Physical capacity (%)
Low	2.4	1.4
Moderate	33.3	34.1
High	65.2	64.6
Comorbidities (0–10	2.27 (0.07)	2.18 (0.05)
ADL limitations (1–35)	11.03 (0.22)	10.70 (0.17)
Cognition (0–10)	3.73 (0.10)	3.75 (0.07)
Multiple hospital stays (%)	6.8	6.1
Depressive symptoms (% more than half of the time)	7.2	6.7
Moderate/high self-efficacy (%)	65.3	60.1
High self-realization (%)	75.3	71.3b
Attends religious services (%)	63.6	54.5b
Expressed preferences (%)	67.0	78.2b

NHATS = National Health and Aging Trends Study; LPT = life-prolonging treatment; ADL = activities of daily living.

^aMeans and proportions are survey weighted to account for multistage survey design of NHATS.

^bSignificant between-group differences based on F test for means and at P < 0.005.

Table 2.

Survey-Weighted Proportions (%) of Expression of EOL Preferences Among NHATS Participants by LPT Preferences (Round 2, 2012)

	Accept LPT	Reject LPT
Overall
Preferences not expressed	9.6	14.5
Preferences expressed	21.3	54.6
Whites
Preferences not expressed	9.0	20.1
Preferences expressed	18.1	52.8
Blacks
Preferences not expressed	34.2	27.8
Preferences expressed	14.4	23.6

EOL = end of life; NHATS = National Health and Aging Trends Study; LPT = life-prolonging treatment.

Multivariable models indicated similar patterns of associations between individual-level factors and LPT preferences (Table 3). Minority status was the only demographic or socioeconomic predictor that was associated with preferences for LPT at the EOL. White adults had lower odds (odds ratio [OR] 0.44; 95% CI 0.34, 0.57) of accepting LPT in either EOL scenario compared with minority adults (Table 3; Model 1). None of the health variables considered in Model 2 were associated with preferences for LPT. The psychosocial predictors were associated with acceptance of LPT in this sample (Model 3). Older adults who had high levels of self-realization and who attended religious services had higher odds of accepting LPT (OR 1.34; 95% CI 1.01, 1.79 and OR 1.44; 95% CI 1.07, 1.94, respectively). As in the bivariate analysis, older adults who had expressed their EOL preferences had significantly lower odds of accepting LPT at the EOL (OR 0.54; 95% CI 0.40, 0.72). The model predictiveness for LPT based on area under the curve was better than chance but low (receiver operating characteristic area = 0.65).

Table 3.

ORs (95% CI) for Reporting an Acceptance of LPT, Among NHATS Participants (Round 2)

	Demographics Adjusted Model (N = 1859)	Health Factors Adjusted Model (N = 1806)	Fully Adjusted Model (N = 1803)
Male	0.84 (0.67, 1.05)	0.86 (0.70, 1.07)	0.87 (0.69, 1.10)
White	0.45 (0.35, 0.58)	0.44 (0.34, 0.56)	0.54 (0.41, 0.70)
Age (vs. 65-74)
75–84	1.24 (0.99, 1.57)	1.28 (1.00, 1.65)	1.31 (1.01, 1.70)
85+	1.15 (0.84, 1.58)	1.18 (0.88, 1.58)	1.26 (0.91, 1.75)
Married	1.40 (0.93, 2.12)	1.45 (0.96, 2.19)	1.38 (0.90, 2.13)
Income	0.92 (0.77, 1.08)	0.90 (0.77, 1.06)	0.93 (0.79, 1.11)
Education (vs. <high school)
High school	0.86 (0.62, 1.19)	0.86 (0.64, 1.16)	0.89 (0.65, 1.22)
College or higher	0.96 (0.63, 1.45)	0.95 (0.64, 1.40)	1.01 (0.67, 1.52)
Live alone	0.82 (0.57, 1.19)	0.83 (0.57, 1.23)	0.84 (0.56, 1.27)
Parent	0.87 (0.57, 1.31)	0.84 (0.54, 1.29)	0.85 (0.56, 1.29)
Mother living	0.85 (0.42, 1.71)	0.83 (0.41, 1.70)	0.88 (0.43, 1.80)
Father living	0.41 (0.07, 2.17)	0.50 (0.09, 2.77)	0.55 (0.10, 3.14)
Assets (vs. lowest tertile)
Middle tertile	0.87 (0.66, 1.15)	0.89 (0.67, 1.19)	0.92 (0.68, 1.25)
High tertile	1.26 (0.88, 1.80)	1.27 (0.88, 1.83)	1.33 (0.91, 1.93)
Self-rated health (vs. good)
Fair		0.68 (0.44, 1.05)	0.71 (0.45, 1.12)
Poor		1.04 (0.62, 1.75)	1.26 (0.74, 2.14)
Physical capacity (vs. low)
Moderate		0.72 (0.35, 1.48)	0.66 (0.31, 1.40)
High		0.78 (0.37, 1.61)	0.70 (0.33, 1.49)
Comorbidities		1.06 (0.95, 1.18)	1.08 (0.96, 1.21)
Cognition		1.08 (0.80, 1.45)	1.07 (0.79, 1.45)
Multiple past year hospital stays		1.06 (0.66, 1.69)	1.03 (0.63, 1.70)
Depressive symptoms more than half of the time			1.15 (0.65, 2.02)
Self-efficacy (moderate/high)			1.16 (0.87, 1.56)
Self-realization (high)			1.34 (1.01, 1.79)
Attend religious services			1.44 (1.07, 1.94)
Expressed EOL preferences			0.54 (0.40, 0.72)

ORs = odds ratios; LPT = life-prolonging treatment; NHATS = National Health and Aging Trends Study; EOL = end of life.

Association Between Preferences and Mortality

We secondarily examined the association of LPT preferences and the expression of EOL preferences on mortality and found that older adults who expressed their preferences had 1.64 times the hazard of mortality (95% CI 1.02, 2.64) compared with older adults with unexpressed preferences, after full adjustment (Appendix Table 1). The relationship between expression of preferences and mortality, however, may depend on preferences to accept/reject LPT, as the interaction between LPT preferences and expression of EOL preferences was of borderline significance (OR 0.50; 95% CI 0.25, 1.00; Appendix Table 1). Fig. 1 displays the Kaplan-Meier survival curves for combinations of EOL preferences indicated by the interaction term. Adults who reject LPT and expressed their EOL preferences had the lowest survival, whereas adults who accepted LPT and expressed their EOL preferences had the highest survival.

Fig. 1.

Kaplan-Meier curve summarizing mortality by preference expression for LPTs. EOL = end of life; LPT = life-prolonging treatment.

Association Between Preferences and Disability

There were no independent associations between preferences for LPT or expression of EOL preferences and disability (Appendix Tables 2 and 3). We found limited evidence that the association between preferences for LPT and the number of high difficulties with self-care and mobility activities was dependent on expression of preferences (Fig. 2). For older adults who expressed their EOL preferences, there was no significant difference in reporting high levels of difficulty with self-care and mobility across LPT preferences. However, for individuals who did not express preferences, rejecting LPT was associated with a greater disability. Cox proportional hazards models indicated similar results (Appendix Table 3).

Fig. 2.

Graph of the predicted mean count of high difficulty is self-care and mobility tasks based on expression (Exp) of EOL preferences and preferences for LPT, controlling for all model covariates. LPT = life-prolonging treatment.

Discussion

In a national sample of community-living older Americans, we found that one-third would accept LPT at the EOL despite constant pain or severe disability. Surprisingly, other than race/ethnicity, sociodemographic factors and health status were relatively poor predictors of LPT preferences. A better understanding of factors that drive LPT preferences is important and will likely require substantial theoretical and empirical work. In the absence of such an understanding, clinicians partaking in advance care planning should be mindful to avoid making assumptions about patient preferences, particularly based on health status.

Although our models explained only a limited proportion of the variance in LPT preferences, several important observations emerged. Older adults who expressed their EOL preferences were more likely to reject LPT at the EOL, which is consistent with previous findings in smaller hospital-based or regional samples.^9,19 Individuals who reject LPT may be more likely to prioritize documenting their EOL preferences or discussing them with family. It is also possible that adults who have expressed their preferences have done so in response to personal experiences; individuals who witnessed family members suffer before their EOL may be more likely to reject LPT for themselves.⁹

Despite prior evidence that health status and physical symptoms influence EOL preferences,^1,10,20 we did not find strong evidence supporting these relationships. This may reflect differences in sampling. Our study included a broad sample of community-dwelling older adults, whereas prior work has focused on samples of subjects in very old age⁶ or on samples of physicians.^6,9,10,19 Conversely, we found that psychological factors were strong predictors of LPT preferences. Older adults who felt more strongly that their life had meaning and purpose, and who felt confident about themselves and liked their living situation (indicators of self-realization), were more likely to accept LPT at the EOL. Similarly, religiosity or spirituality was associated with LPT at the EOL, consistent with previous research.^1,10 Our results, then, suggest that, first and foremost, clinicians should be cautious about making assumptions regarding patient preferences based on a patient’s level of baseline disability. Even to the extent that health status predicts preferences, these effects are modest, and many patients with substantial disability accept LPT, whereas many others with limited or no disability would reject it. Moreover, providers should be particularly cautious because provider’s real-world do-not-resuscitate recommendations appear to be strongly influenced by health status,²¹ which may have a lesser impact on patient preferences. Second, exploring non-health-related factors may be a key step in optimizing LPT conversations. Providers and family members who speak to older adults about personal values, religious beliefs, and past experiences might gain a better understanding of the patient’s EOL preferences. Although such conversations are often uncomfortable for providers,²² our results suggest that they may lead to more clarity regarding LPT preferences than a focus of health status.

Results from this study also may contribute to our understanding of late-life racial disparities in disability. Freedman and Spillman²³ found that older whites gained years of active life expectancy (delayed mortality and fewer years of disability), whereas older blacks did not experience the same compression of morbidity, resulting in more years lived but with significant disability. We found that racial/ethnic minorities had lower rates of preference expression and greater acceptance of LPT, both of which are consistent with prior observations.^1,2,24,25 One interpretation of our findings is that lack of preference expression may lead to increased survival (higher preference expression associated with lower mortality) among minorities, although may result in increased disability. If this is the case, racial differences in mortality may also reflect these differences in preference expression and suggests that the magnitude of preference-independent late-life mortality difference may be underestimated. This possibility merits further research.

In this relatively small sample, we found limited evidence that LPT preferences influence mortality and disability. Our findings do not lead to a clear interpretation about the relationship between LPT preference, EOL preference expression, and disability, as we did not identify consistent or large effects. There are multiple reasons why our data may not lead to clear inferences. First, we have limited statistical power for outcome analyses. Thus, subsequent work with superior power may lead to more reliable inferences. Second, it is likely that given the complexity underlying these pathways that our simple models may be misspecified. For example, it may be the case that the interaction between preferences and preference expression differs by race, yet we could not include a three-way interaction term because of limited power. Third, it may be the case that the true relationships are conflicting. LPT preferences may not only serve simultaneously as a causal factor influencing subsequent treatment decisions but also be a marker for underlying disease severity—resulting in divergent outcomes. Subsequent work in larger data sets with more granular measures and theoretical complexity is likely needed to explore these possibilities.

Limitations

This work has several limitations. Little is known about how EOL preferences evolve across complex and changing clinical situations. As this study uses the answers to hypothetical scenarios, it is possible that the measures of preferences used here incompletely or inaccurately capture the true spectrum of older adult preferences. This said, however, most advanced care planning conversations hinge on the ability of clinicians to speculate on prognosis and ask patients to think through hypothetical scenarios before they become ill in real life.^26,27 Another limitation is that the measure of preferences used in this study is a measure at a single time point, whereas it is likely that repeated measures would be both more reliable and sensitive to the inevitable fluctuations in preferences that occur over time. Our findings regarding LPT preferences and disability should be interpreted cautiously given that we performed multiple comparisons—defining disability several ways and considering prevalent and incident definitions of disability.

Conclusions

This study is the first to examine a range of individual-level predictors of preferences for LPT in a nationally representative sample of community-living older adults. We found that older Americans have varied preferences for LPT at the EOL. Surprisingly, little of the variance in LPT preferences was explained by the numerous factors considered. This suggests that there is much to learn about what predicts an individual’s EOL preferences, and that preferences are likely complex and influenced by multiple factors. Our findings underscore the need for additional research on preferences for LPT and health outcomes in large and racially diverse populations.

Appendix

Appendix Table 1.

Hazard Ratio for Mortality By Preferences for LST at the EOL and Expression of EOL Preferences^a

	Model 1 (Unadjusted)	Model 2 (Unadjusted)	Model 3 (Adjusted)	Model 4 (Adjusted)
Acceptance of LST (vs. rejection of LST)	0.81 (0.60, 1.08)	1.29 (0.71, 2.37)	0.81 (0.61, 1.09)	1.38 (0.75, 2.51)
Expressed preferences (vs. nonexpressed)	1.36 (0.93, 1.91)	1.70 (1.08, 2.68)	1.30 (0.91, 1.86)	1.64 (1.02, 2.64)
Acceptance of LST × expressed preferences (interaction term)		0.54 (0.27, 1.07)		0.50 (0.25, 1.00)b

LST = life-sustaining treatment; EOL = end of life

^aAdjusted for demographics, comorbidities, and cognition.

^bP = 0.05.

Appendix Table 2.

ORs and Incidence Ratios (95% CIs) for Adjusted Associations of Preference for LST and Expression Preferences at the EOL, With Late-Life Self-Care and Mobility Difficulty Among NHATS Participants^a

	Any Self-Care Difficulty	Any Self-Care or Mobility Difficulty	High Self-Care Difficulty	High Self-Care or Mobility Difficulty	Need Help/Cannot Do Self-Care	Need Help/Cannot Do Self-Care or Mobility
Odds of having self-care and mobility difficulty
Acceptance of LST (vs. rejection of LST)	0.70 (0.40, 1.22)	0.92 (0.49, 1.71)	0.58 (0.33, 1.01)b	0.67 (0.38, 1.16)	0.66 (0.27, 1.63)	0.70 (0.33, 1.45)
Expressed preferences (vs. nonexpressed)	0.78 (0.50, 1.22)	0.87 (0.51, 1.49)	0.83 (0.52, 1.34)	0.79 (0.49, 1.28)	1.06 (0.54, 2.08)	1.05 (0.58, 1.92)
Expression × acceptance LST (interaction term)	1.55 (0.82, 2.91)	1.22 (0.56, 2.68)	1.81 (0.94, 3.46)	1.87 (0.91, 3.82)	1.42 (0.56, 3.61)	1.71 (0.77, 3.83)
Incidence ratio for number of self-care and mobility limitations
Acceptance of LST (vs. rejection of LST)	0.81 (0.59, 1.10)	0.92 (0.69, 1.22)	0.76 (0.55, 1.04)	0.76 (0.54, 1.04)
Expressed preferences (vs. nonexpressed)	0.92 (0.71, 1.18)	0.87 (0.75, 1.25)	0.96 (0.72, 1.27)	0.96 (0.72, 1.27)
Expression × acceptance LST (interaction term)	1.34 (0.93, 1.94)	1.21 (0.84, 1.76)	1.53 (1.00, 2.33)	1.53 (1.00, 2.34)

ORs = odds ratios; LST = life-sustaining treatment; EOL = end of life; NHATS = National Health and Aging Trends Study.

^aAdjusted for age, sex, race, comorbidities, and cognition.

^bP = 0.055.

Appendix Table 3.

Incidence Ratios for Adjusted Associations of EOL Preferences Expression and Preference for LSTs With Late-Life Self-Care and Mobility Difficulty Limitation Counts Among NHATS Participants^a

	Any Difficulty in Self-Care Limitations	Any Difficulty in Self-Care and Mobility Limitations	High Difficulty in Self-Care Limitations/Needs Help/ Cannot Do	High Difficulty in Self-Care and Mobility Limitations/Needs Help/Cannot Do
Acceptance of LST (vs. rejection of LST)	0.81 (0.59, 1.10)	0.92 (0.69, 1.22)	0.76 (0.55, 1.04)	0.76 (0.54, 1.04)
Expressed preferences (vs. nonexpressed)	0.92 (0.71, 1.18)	0.87 (0.75, 1.25)	0.96 (0.72, 1.27)	0.96 (0.72, 1.27)
Expression × acceptance LST (interaction term)	1.34 (0.93, 1.94)	1.21 (0.84, 1.76)	1.53 (1.00, 2.33)	1.53 (1.00, 2.34)

EOL = end of life; LSTs = life-sustaining treatments; NHATS = National Health and Aging Trends Study.

^bP = 0.055.

^aAdjusted for age, sex, race, comorbidities, and cognition.