Evolution and Predictors of Patient–Caregiver Concordance on States of Life-Sustaining Treatment Preferences over Terminally Ill Cancer Patients' Last Six Months of Life

Abstract

Background:

Patient–caregiver concordance on end-of-life (EOL) care preferences is poor, but changes in this concordance have not been longitudinally explored as patient death approaches, potentially distorting the extent of concordance. Cross-sectional studies cannot disentangle whether the extent of concordance is facilitated or hindered by dyads' specific preferences, prognostic awareness, physical and psychological well-being, and quality of life, or whether these variables were enhanced or worsened by patient–caregiver concordance on EOL care preferences.

Objective:

To examine the evolution of and factors facilitating or hindering patient–caregiver concordance on life-sustaining treatment (LST) preferences over cancer patients' last six months.

Design:

Longitudinal study design.

Methods/Subjects:

Patient–caregiver concordance on LST preference states (patterns) was examined among 215 cancer patient–caregiver dyads in patients' last six months by hidden Markov modeling. Concordance on LST preference states was determined by percent agreement and kappa coefficients. Predictors of concordance on LST preference states were tested by hierarchical generalized linear modeling with logistic regression, with concordance and time-varying, modifiable independent variables arranged in a distinct time sequence.

Results:

Patient–caregiver concordance on LST preference states was poor and improved only slightly over cancer patients' last six months. Concordance on LST preference states was significantly more likely in patients with greater physical symptom distress. Caregivers were more likely to concur with their relative's LST preference states if caregivers uniformly rejected all LSTs or accepted nutritional support while rejecting other aggressive LSTs for their relative.

Discussion/Conclusion:

Patient symptom distress and caregiver rejection of aggressive LSTs predicted greater patient–caregiver concordance on LST preference states in patients' last six months. To encourage patients and caregivers to discuss LST preferences, clinicians should facilitate caregivers' understanding of patients' LST preferences and LST efficacy at EOL and adjustment to their beloved's inevitable death when his/her physical symptoms still wax and wane, thus providing personalized and value-concordant EOL care for dying cancer patients.

Introduction

Providing value-concordant end-of-life (EOL) care requires eliciting patients' EOL care preferences, especially on life-sustaining treatments (LSTs), and tailoring treatments to these preferences whenever medically appropriate.¹ However, dying patients' LST preferences are often unidentified due to their physical deterioration and cognitive impairment,^2–4 necessitating proxy decision making to honor incapacitated patients' EOL care values/goals. Nevertheless, EOL care preferences are rarely discussed,^5,6 especially in Asian countries.^7,8 Thus, family caregivers may inaccurately understand patients' preferences^5,6,9 and project their own preferences to shape patients' EOL care,¹⁰ especially in Confucian-influenced societies (e.g., Taiwan) where the cultural values of filial piety and families' relative power are strongly exercised in EOL care decision making.¹¹ Subsequently, patients may receive more aggressive care than they prefer⁴ and suffer depression,¹² whereas family members endure psychological burden (e.g., decisional conflict,^3,5,6 anxiety,³ depression,^3,12,13 and post-traumatic stress disorder¹⁴) both before and after patient death.³

Poor patient–caregiver concordance on EOL care preferences has been documented for dyads of caregivers and patients with terminal chronic diseases⁹ and terminal/advanced cancer.^14–19 However, almost all reviewed studies were cross-sectional,^9,14–19 that is, they did not explore changes in patient–caregiver concordance on EOL care preferences as death approached. The extent of caregiver concurrence with patients' preferences may be distorted if assessments are measured long before death, when both parties may be less realistic about LST efficacy, and caregivers may be less knowledgeable about patient's preferences than closer to death.

Given caregivers' poor agreement with patients' EOL care wishes,^9,14–19 identifying the factors associated with patient–caregiver concordance on EOL care preferences can play an important role in improving the likelihood of honoring dying patients' wishes. Nevertheless, few studies have examined the factors predicting patient–caregiver concordance on EOL care preferences from the perspectives of family caregivers and care receivers as a unit. Furthermore, cross-sectional studies cannot disentangle the direction of the relationship between dyads' concordance and prognostic awareness,^20,21 LST preferences,²² physical²³ and psychological^8,24,25 well-being, and quality of life (QOL).²⁴ In other words, does patient–caregiver concordance on EOL care preferences enhance or worsen their prognostic awareness, physical and psychological well-being, and QOL or do these factors facilitate or hinder concordance? Therefore, the purposes of this study were to examine the evolution of and to identify factors predicting patient–caregiver concordance on LST preferences over cancer patients' last six months.

Methods

Design and sample

We examined longitudinal data on the quality of death and dying from a convenience sample of dyads of caregivers for terminally ill Taiwanese cancer patients recruited in 2009 to 2012 and followed through 2015. Methodological details have been reported.²⁶ Briefly, adult cancer patients were referred by their oncologist who had recognized their metastatic disease as progressive and unresponsive to curative treatments. Adult family caregivers were included if they were identified by patients as the family member most involved in their daily care, were willing to participate, and could communicate with data collectors. If any party refused participation, the patient–caregiver dyad was excluded. Dyad participants were interviewed separately by experienced oncology nurses approximately every two to four weeks when the patient was hospitalized or returned for clinical care until they withdrew or the patient died. The study site's ethical committee approved the study, and each participant individually signed informed consent.

Preferences for LSTs

Dyads were assessed using an adapted interview protocol (Appendices 1 and 2; Appendices 1 and 2 are available online at www. liebertpub.com/jpm) on their (or their beloved's) LST preferences for cardiopulmonary resuscitation (CPR), intensive care unit care, chest compression, intubation with mechanical ventilation, intravenous nutrition support, and nasogastric tube feeding. For each LST, participants were asked whether they (or for their beloved): (1) wanted, (2) did not want, or (3) were undecided about the treatment.

Predictors of patient–caregiver concordance on LST preferences

Potential predictors^8,20–25 included both time-invariant patient (gender and age) and caregiver (relationship with the patient) characteristics and modifiable time-varying variables (accurate prognostic awareness,^20,21 physical²³ and psychological well-being,^8,24,25 QOL,²³ LST preferences,²² and caregiving burden).

Accurate prognostic awareness

Prognostic awareness was evaluated by asking participants if they knew their (relative's) prognosis, and if so, whether the disease (1) was curable; (2) might recur in the future, but their (relative's) life was not currently in danger; and (3) could not be cured, and they (their relative) would probably die soon. Participants were recognized as accurately understanding their (relative's) prognosis only if they chose option 3.

Patient physical well-being

Patient physical symptom distress was measured by the 13-item Symptom Distress Scale (SDS),²⁷ which assesses cancer patients' common symptoms. SDS scores range from 13 to 65; higher scores indicate greater distress.

Patient functional dependence was measured by the 10-item Enforced Social Dependency Scale (ESDS).²⁸ ESDS scores range from 10 to 51; higher scores reflect greater dependence on help for personal and social functioning.

Psychological well-being

Patient anxiety and depressive symptoms were measured by the Hospital Anxiety and Depression Scale (HADS) anxiety (HADS-A) and depression (HADS-D) subscales, respectively.²⁹ HADS-A and HADS-D subscales' scores range from 0 to 21; higher scores indicate more anxiety and depressive symptoms.

Caregivers' depressive symptoms were assessed by the 20-item Center for Epidemiological Studies Depression Scale (CES-D).³⁰ CES-D scores range from 0 to 60; higher scores indicate more depressive symptoms.

Quality of life

Patient QOL was measured by a modified 13-item McGill Quality of Life Questionnaire (MQOL).³¹ The original 16-item MQOL emphasizes psychological, social, and existential well-being. The MQOL was modified by omitting three items for the most distressing symptoms to avoid overlap with symptom distress (SDS scores). Modified MQOL scores range from 0 to 130; higher scores indicate better QOL.

Caregiver QOL was measured with the 35-item Caregiver Quality of Life Index-Cancer (CQOLC).³² The CQOLC measures the effect of cancer patients' illness and caregiving on family caregivers' physical, emotional, social, and family functioning. CQOLC scores range from 0 to 140; higher scores indicate better QOL.

Caregivers' subjective caregiving burden was measured by the Caregiver Reaction Assessment.³³ Total scores range from 24 to 120; higher scores represent stronger negative caregiving impact on schedule, health, finances, caregiver support, and self-esteem.

Analysis

Participants' baseline characteristics were analyzed by descriptive statistics. Distinct LST preference states and changes in these states between consecutive assessments were identified and examined using a transition model with hidden Markov modeling (HMM).³⁴ In HMM, LST preferences were treated as treatment preference patterns/sets (“latent states”) rather than individual treatment preferences. This approach allows clinicians to parsimoniously identify patients' LST preference patterns/states rather than preferences for individual LSTs, thus minimizing clinicians' time spent in assessing LST preferences and avoiding burdening terminally ill patients with choosing among multiple LSTs when they are physically and psychologically frail. Distinct LST preference states were identified across patient and family caregiver cohorts in a single model using Latent GOLD 5.0.

HMM simultaneously examined LST preferences as treatment preference states (“latent states”), described dynamic changes in participants' LST preference states at EOL, and estimated between-state transition probabilities. The first part of HMM assigned patients and caregivers to a finite number of mutually exclusive probabilistic LST preference states based on shared characteristics of members in each state. Emission probability for each identified state reflects the observed probability that each participant will want, not want, or be undecided about each LST, conditional on his/her state membership.³⁵ Model choice, regarding the optimal number of states, was determined by examining fit indices with the Akaike information criterion (AIC),³⁶ Bayesian information criterion (BIC),³⁷ and consistent AIC (CAIC).³⁸ Lower AIC, BIC, and CAIC scores indicate a better model fit. These criteria and clinical meaningfulness of latent-state results were used to determine the optimal state number. The second part of HMM estimated between-state transition probabilities,³⁴ which represent the likelihood that a participant will prefer a specific LST set at time t, given his/her preference for a specific LST set at time t − 1.

Patient–caregiver concordance on LST preference states was analyzed by percent agreement and Cohen's kappa coefficient (from chi-square tests) to correct for the concordance expected by chance alone.³⁹ The strength of concordance measured by kappa was determined as follows: poor, ≤0.20; fair, 0.21 to 0.40; moderate, 0.41 to 0.60; substantial, 0.61 to 0.80; and almost perfect, 0.81 to 1.00.³⁹

Predictors of patient–caregiver concordance on LST preference states were identified using hierarchical generalized linear modeling (HGLM⁴⁰) with logistic regression. HGLM uses random intercepts to account for within-subject correlations of repeated observations from each dyad.⁴⁰ HGLM also allows different waves of data across participants to accommodate variable numbers of follow-ups, inconsistent time intervals between data collections, and missing data for dependent variables, thus eliminating the need to delete observations in analyses.⁴⁰ Our HGLM analysis modeled the outcome variable as a function of proposed lagged time-varying predictors in the previous assessment wave. The lagged measure was added to identify variables predicting patient–caregiver concordance by arranging proposed time-varying modifiable predictors and the outcome variable in a distinct time sequence to ensure a clear time sequence. The regression parameter for each independent variable was exponentiated to transform into adjusted odds ratio (AOR) with 95% confidence interval (CI).

Results

Sample characteristics

Our final sample comprised terminally ill cancer patient–family caregiver dyads (N = 215) who participated until the patient's death and were assessed at least twice. This approach allowed adequate analysis of transitions in LST preference states between consecutive times and factors predicting patient–caregiver concordance on LST preference states over patients' dying process. Detailed participant characteristics and potential predictors of patient–caregiver concordance on LST preference states are shown for patients and caregivers assessed once (n = 51) and repeatedly (Table 1). Characteristics did not differ significantly between groups, but patients assessed only once had greater symptom distress, functional dependence, and depressive symptoms, with lower QOL. On average, patients survived 190.99 days (standard deviation [SD] = 213.77; range: 17–1444) after enrollment. Only data from patients' last six months were included. Participants completed on average 5.32 (SD = 2.48; range: 2–12) assessments in patients' last six months, with the last assessment completed 27.90 (SD = 26.20; range 1–137) days before death.

Table 1.

Demographic and Disease Characteristics, with Selected Baseline Variables of Patient and Caregiver Participants in the Final Sample (Repeatedly Assessed) and Those Assessed Once

	Patients
	Repeated (N = 215)		Once (N = 61)
Characteristic	n	%	n	%	χ ² /t-test	p
Gender					0.68	0.410
Male	124	57.7	22	36.1
Female	91	42.3	39	63.9
Age (years)
Mean (SD)	58.68 (12.42)		60.44 (13.83)		−0.84	0.402
Range	25–89		35–95
Marital status			(n = 60)		0.65	0.723
Single	12	5.6	2	3.3
Married/cohabiting	176	81.9	50	83.3
Separated/divorced/widowed	27	12.6	8	13.3
Educational level	(n = 212)		(n = 60)		2.03	0.154
>Junior high school	39	18.4	44	73.3
≤Junior high school	173	81.6	16	26.1
Cancer site					3.27	0.774
Liver–pancreas	70	32.6	24	39.3
Stomach	40	18.6	9	14.8
Lung	23	10.7	8	13.1
Head–neck	21	9.8	6	9.8
Esophageal	13	6.0	2	3.3
Colon–rectum	12	5.6	4	6.6
Others	36	16.8	8	13.1
Comorbidity					3.77	0.052
Yes	140	65.1	31	50.8
No	75	34.9	30	49.2
Symptom distress (SDS) (Cronbach's α = 0.80)	26.71 (7.65)		30.62 (8.26)		−4.11	<0.001
Functional dependence (ESDS) (Cronbach's α = 0.90)	24.43 (8.94)		32.14 (10.33)		−5.91	<0.001
Anxiety symptoms (HADS-A) (Cronbach's α = 0.87)	7.09 (4.82)		8.06 (5.02)		−1.65	0.100
Depressive symptoms (HADS-D) (Cronbach's α = 0.90)	11.10 (5.72)		13.56 (5.68)		−3.36	0.001
Quality of life (MQOL) (Cronbach's α = 0.90)	86.25 (25.06)		75.95 (25.51)		3.18	0.002

	Family caregivers
	Repeated (N = 215)		Once (N = 61)
Characteristic	n	%	n	%	χ²/t-test	p
Gender					0.99	0.318
Male	86	40.0	28	45.9
Female	129	60.0	33	54.1
Age (years)					1.64	0.102
Mean (SD)	48.89 (11.98)		45.71 (12.77)
Range	20–76		23–78
Marital status	(n = 210)				1.13	0.568
Single	32	14.9	12	19.7
Married/cohabiting	165	76.8	44	72.1
Separated/divorced/widowed	13	6.2	5	8.2
Relationship	(n = 210)				2.61	0.217
Spouse	120	57.1	27	44.3
Child	56	26.7	22	36.1
Other	34	16.2	12	19.7
Living with the patient	(n = 209)				0.31	0.581
Yes	170	81.3	47	77
No	39	18.7	14	23
Primary decision maker	(n = 209)				0.06	0.803
Yes	109	52.2	33	54.1
No	100	47.8	28	45.9
	Mean (SD)		Mean (SD)		t-test	p
Depressive symptoms (CES-D) (Cronbach's α = 0.93)	17.73 (11.17)		18.98 (10.70)		−0.41	0.681
Quality of life (CQOLC) (Cronbach's α = 0.92)	85.96 (21.50		80.27 (20.01)		1.07	0.283
Caregivers' subjective caregiving burden (CRA) (Cronbach's α = 0.88)	63.00 (11.69)		64.51 (12.60)		−0.87	0.383

CES-D, Center for Epidemiological Studies Depression Scale; CQOLC, Caregiver Quality of Life Index-Cancer; CRA, Caregiver Reaction Assessment; ESDS, Enforced Social Dependency Scale; HADS-A, Hospital Anxiety and Depression Scale anxiety subscale; HADS-D, Hospital Anxiety and Depression Scale depressive subscale; MQOL, McGill Quality of Life Questionnaire; SDS, Symptom Distress Scale.

HMM of patients' and family caregivers' LST preference states and transitions of LST preference states in patients' last six months

Model fit indexes are shown in Appendix Figure A1 (Appendix Figure A1 is available online at www.liebertpub.com/jpm). A four-state solution for LST preferences, with dyad-state covariance and measurement invariance, was selected as optimal based on the AIC, BIC, and CAIC plots (Appendix Fig. A1) and clinical meaningfulness. Therefore, the emission probabilities for LST preference states were identical across patient and caregiver groups. These four states' emission probabilities and sizes (state probabilities) are presented in Table 2.

Table 2.

Life-Sustaining Treatment Preference–Emission Probabilities for Classes of Life-Sustaining Treatment Preferences, Separately by Patient and Family Caregiver Groups for Initial Class Sizes

		State
	Group	1	2	3	4
	Patients	9.48	42.66	22.91	24.95
Class size at baseline (%)	Caregivers	13.05	22.07	23.26	41.62
Want treatment
CPR		0.974	0.001	0.011	0.006
ICU care		0.985	0.003	0.034	0.139
Chest compression		0.975	0.002	0.006	0.044
Intubation		0.930	0.000	0.004	0.014
Tube feeding		0.976	0.018	0.149	0.587
Intravenous nutrition support		1.000	0.315	0.339	0.865
Do not want treatment
CPR		0.004	0.991	0.052	0.878
ICU care		0.004	0.991	0.003	0.677
Chest compression		0.020	0.998	0.014	0.941
Intubation		0.056	1.000	0.048	0.975
Tube feeding		0.017	0.976	0.038	0.212
Intravenous nutrition support		0.000	0.671	0.002	0.036
Undecided
CPR		0.023	0.008	0.937	0.116
ICU care		0.011	0.006	0.963	0.184
Chest compression		0.005	0.000	0.980	0.015
Intubation		0.014	0.000	0.948	0.012
Tube feeding		0.008	0.006	0.813	0.201
Intravenous nutrition support		0.000	0.014	0.659	0.100

State 1: uniformly preferring, State 2: uniformly rejecting, State 3: uniformly uncertain about LSTs, and State 4: favoring nutrition support but rejecting other LSTs. Shading indicates the probability of LST preference response, conditional on class membership, which characterizes participants in each class and distinguishes them from participants in other classes. This characteristic LST preference response probability was used to name each state. Emission probabilities for classes 1–4 were assumed to be similar across patient and family caregiver cohorts by setting the measurement as invariant in our model building.

CPR, cardiopulmonary resuscitation; ICU, intensive care unit; LST, life-sustaining treatment.

These states were labeled as uniformly preferring, uniformly rejecting, and uniformly uncertain about LSTs (states 1–3), and favoring nutrition support by intravenous/nasogastric tube feeding but rejecting other LSTs (state 4). When LST preferences were initially assessed, patients' most prevalent state was uniformly rejecting LSTs (state 2, 42.66%; Table 2), followed by favoring nutrition support but rejecting other LSTs (state 4, 24.95%), uniformly uncertain about LSTs (state 3, 22.91%), and uniformly preferring LSTs (state 1, 9.48%). Caregivers' LST preference states were substantially different from patients', in descending prevalence as follows: favoring nutrition support but rejecting other LSTs (state 4, 41.62%), uniformly uncertain about LSTs (state 3, 23.26%), uniformly rejecting LSTs (state 2, 22.07%), and uniformly preferring LSTs (state 1, 13.05%). Therefore, caregivers generally preferred LSTs more than patients as shown by caregivers' higher state-membership probabilities for state 1 (uniformly preferring LSTs) and state 4 (favoring nutritional support but rejecting other treatments), and lower membership in state 2 (uniformly rejecting LSTs).

The transition probabilities for patients and caregivers in a specific state moving between states from time t − 1 to time t are presented in Table 3. LST preferences were highly stable over patients' last six months as evident by both groups remaining in their original LST preference state (diagonal in Table 3, 93.4% to 97.4% and 88.9% to 95.7% for patients and caregivers, respectively) rather than shifting to other preference states.

Table 3.

Transition Probabilities of States of Life-Sustaining Treatment Preferences from Time Point (t − 1) to Time Point (t)

Time point (t)	Time point (t − 1)
Group	Patients
State	1	2	3	4
1	0.941	0.001	0.009	0.008
2	0.006	0.974	0.025	0.026
3	0.039	0.012	0.934	0.025
4	0.014	0.013	0.032	0.940
	Family caregivers
1	0.905	0.004	0.036	0.006
2	0.008	0.957	0.021	0.015
3	0.055	0.010	0.889	0.036
4	0.032	0.029	0.054	0.944

Bold indicates the highest transition probability between different time points. State 1: uniformly preferring, State 2: uniformly rejecting, State 3: uniformly uncertain about LSTs, and State 4: favoring nutrition support but rejecting other LSTs.

Concordance on patients' and family caregivers' LST preference states

Patient–caregiver concordance on LST preference states increased over patients' last six months from 24.48% (95% CI: 18.34–30.62) to 36.86% (95% CI: 30.90–42.82; Table 4). Patient–caregiver concordance on LST preference states was not significant in patients' last four to six months (κ range = −0.013 [95% CI: −0.089 to 0.063] to 0.045 [95% CI: −0.045 to 0.135]). In contrast, concordance on LST preference states became significant in patients' last three months (κ range = 0.077 [95% CI: 0.001–0.153] to 0.146 [95% CI: 0.072–0.221]). Despite this improvement in patient–caregiver concordance on LST preference states, it was at best poor.

Table 4.

Changes in Patient–Family Caregiver Concordance on States of Life-Sustaining Treatment Preferences in the Last Six Months of Life (N = 215)

Time before death, months	6	5	4	3	2	1
Assessments, n	192	143	178	218	255	233
Patient–caregiver dyads in agreement, n	47	45	55	69	94	85
Patient–caregiver agreement, % (95% CI)	24.48 (18.34–30.62)	31.47 (23.76–39.17)	30.90 (24.04–37.75)	31.65 (25.43–37.87)	36.86 (30.90–42.82)	36.48 (30.25–42.71)
Patient–caregiver agreement, kappa (95% CI)	−0.013 (−0.089 to 0.063)	0.061 (−0.039 to 0.161)	0.045 (−0.045 to 0.135)	0.077 (0.001–0.153)	0.146 (0.072–0.221)	0.124 (0.042–0.206)
p	0.745	0.207	0.297	0.039	< 0.001	0.001

Participants may be assessed multiple times during each time period.

Predictors of patient–caregiver concordance on LST preference states

Our logistic HGLM showed that patient–caregiver concordance on LST preference states was more likely in cancer patients with greater physical symptom distress (AOR [95% CI]: 1.063 [1.011–1.118] with each unit increase in SDS score, p = 0.017) in the previous wave of assessment (Table 5). Caregivers who uniformly rejected all LSTs (AOR [95% CI]: 5.300 [1.339–20.974], p = 0.018) or accepted nutrition support for their relative while rejecting other aggressive LSTs (AOR [95% CI]: 3.334 [1.267–8.775], p = 0.015) were more likely to concur with their relative's LST preference states in the subsequent assessment wave. Other potential predictors of patient–caregiver concordance on LST preferences, especially patients' and caregivers' accurate prognostic awareness, were not significant in our HGLM model.

Table 5.

Predictors of Patient–Family Caregiver Concordance on States of Life-Sustaining Treatment Preferences in the Last Six Months of Life

		95% CI
Predictor	AOR	Upper	Lower	t-ratio	p
Patient variables^a
Gender
Male	0.746	0.309	1.803	−0.654	0.514
Female	Ref
Age	1.001	0.968	1.048	0.344	0.731
Accurate prognostic awareness
Yes	1.411	0.684	2.909	0.934	0.350
No	Ref
Symptom distress	1.063	1.011	1.118	2.400	0.017
Functional dependence	1.012	0.967	1.059	0.499	0.618
Quality of life	1.001	0.989	1.029	0.878	0.380
Depressive symptoms	0.920	0.842	1.005	−1.863	0.063
Anxiety symptoms	1.055	0.966	1.152	1.191	0.234
States of LST preferences
Uniformly rejecting LSTs	0.514	0.131	2.013	−0.957	0.339
Uniformly uncertain about LSTs	1.069	0.286	4.002	0.099	0.921
Favoring nutritional support while rejecting other treatments	0.861	0.246	3.013	−0.234	0.815
Uniformly preferring LSTs	Ref
Family caregiver variables^a
Relationship with the patient
Spouse	1.590	0.455	5.555	0.731	0.466
Adult child	0.793	0.181	3.479	−0.310	0.757
Other	Ref
Accurate prognostic awareness
Yes	0.741	0.350	1.571	−0.783	0.434
No	Ref
Subjective caregiving burden	1.020	0.983	1.058	1.066	0.287
Depressive symptoms	0.956	0.894	1.023	−1.306	0.192
Quality of life	1.001	0.967	1.036	0.027	0.978
States of LST preferences
Uniformly rejecting LSTs	5.300	1.339	20.974	2.381	0.018
Uniformly uncertain about LSTs	1.748	0.601	5.087	1.027	0.305
Favoring nutritional support while rejecting other treatments	3.334	1.267	8.775	2.444	0.015
Uniformly preferring LSTs	Ref

Bold indicates statistically significant.

Except for patient gender and age as well as caregiver relationship with the patient, all potential predictors were lagged time-varying variables and measured in the previous wave of assessment.

AOR, adjusted odds ratio; Ref, reference.

Discussion

We longitudinally confirmed previous cross-sectional evidence that LST preferences of terminally ill cancer patients and their family caregivers differed substantially or had poor concordance.^14–19 Furthermore, such discordance on LST preference states only improved slightly as death approached, contrary to our expectation but in line with a report that concordance of family member–lung cancer patient dyads' perceived/actual symptom trajectories remained poor until patient death.⁴¹

Caregivers were more likely to concur with terminally ill patients' LST preference state in the subsequent assessment wave if patients reported greater physical symptom distress, as reported.⁴² Patient-proxy LST preferences were reported to be least concordant when patients' condition waxed and waned.⁴² Likewise, proxy knowledge of patients' EOL care preferences was significantly associated with elderly decedents' consistently higher symptom burden,²³ and family caregivers' willingness to forgo CPR, which is generally declined by patients, was associated with patients' uncontrolled pain.²⁴ Family caregivers may be readier to forgo LSTs to honor terminally ill cancer patients' wishes when patients have harsher symptom distress because caregivers have already considered that further LSTs would less effectively prolong their relative's life without increasing suffering.

Taiwanese family caregivers were more likely to concur with their dying relatives' LST preference state when caregivers uniformly rejected or selectively rejected aggressive LSTs while accepting nutrition support for the patient in the previous assessment, as reported.²² Taiwanese family caregivers preferred LSTs more than cancer patients wanted for themselves (Table 2), similar to family caregivers worldwide.^17–19,24 This higher Taiwanese caregiver preference for LSTs for their dying relative than his/her preference may be primarily due to “filial duty”—the cultural backbone of Confucianism.^11,43 Fulfilling one's filial duty is a step toward personal integrity and self-perfection, whereas ignoring one's filial duty is condemned as a serious sin.⁴³ Therefore, Taiwanese family caregivers have a strong moral obligation to try every effort to keep a relative alive or to continuously provide nutrition support to dying patients to avoid their becoming a “starving soul” or “hungry ghost/spirit” in hell, even when caregivers acknowledge death as inevitable and are willing to forgo aggressive LSTs for their ill relative. However, such philosophical/cultural practices may differ from patients' preferences when their physical condition continually deteriorates. Patients may choose to forgo LSTs to avoid protracting the dying process and let nature take its course. Only when family caregivers reconcile with the letting go process to successfully resolve anticipatory grief⁴⁴ and forgo LSTs, can they respect and value patients' EOL care wishes.

Our finding that patient–caregiver concordance on LST preference states was not associated with patient and caregiver psychological distress (anxiety and depressive symptoms, and subjective caregiving burden), contrasts with the literature,^8,24,25 especially the negative association between family caregivers' emotional state and patient–caregiver concordance.^8,25 However, our finding not only underscores the distinction between understanding a dying relative's EOL care preferences and the amount of relationship strain (either depression or caregiving burden) but also suggests that such strain does not indicate a caregiver's incapability to serve as a substitute decision maker, as suggested.⁹

Our most unexpected finding is that patient–caregiver concordance on LST preference states was not associated with either patients' or caregivers' accurate prognostic awareness. Despite inconclusive findings from two studies,^20,21 accurate prognostic awareness has been identified as a cornerstone of terminal cancer patients' value-concordant EOL care decision making⁴⁵ and was associated with patient-proxy concordance on LST decisions at EOL.⁴² However, we did not observe such a relationship, which we speculate reflects the lack of patient–caregiver communication on EOL care preferences even when both parties have accurate prognostic awareness. EOL care discussions between Asian patients and their families remain rare^7,8 and culturally taboo. Without adequate EOL care discussions, caregivers may not be aware of similarities and differences between the patient's values and their own, leading to discordance on LST preference states.

The strengths of our study include longitudinally assessing patient–caregiver concordance on LST preferences till the patient's death and identifying factors predicting such concordance from a dyad perspective by arranging time-varying, modifiable independent variables in a distinct time sequence. However, patient participants who had family caregivers and survived long enough to be repeatedly assessed were recruited by convenience from a single Taiwanese medical center, possibly compromising our findings' generalizability and their representation of national and international target populations. We assessed six LST preferences and six predictors of patient–caregiver concordance on LST preferences, possibly creating a burden for terminal patients. This burden was likely minimal since only 21 of 344 enrolled patients withdrew.²⁶ We can never exclude the possible association of patient–caregiver concordance on LST preferences with unmeasured residuals in our observational study, for example, caregiver attitudes toward patient autonomy and involvement in EOL care decisions as well as the extent and quality of caregiver–patient communication/relationship. These issues warrant further investigation to guide interventions that promote patient–caregiver concordance on LST preference states to honor terminally ill cancer patients' EOL care preferences.^1,2

Conclusion

Terminally ill Taiwanese patient–caregiver concordance on LST preference states was poor and only improved slightly over the cancer patient's last six months. Greater patient symptom distress and family caregivers' rejection of aggressive LSTs predicted greater patient–caregiver concordance on LST preference states at EOL. To encourage an open dialogue between terminal cancer patients and their family caregivers about patients' desired LSTs, healthcare providers should inform caregivers about patients' LST preferences, facilitate caregivers' realistic understanding of the efficacy of aggressive LSTs at EOL,⁴⁶ and facilitate their adjustment to the patient's inevitable death when his/her physical symptoms are waxing and waning. This approach may lead to personalized and value-concordant EOL care for dying cancer patients to honor their LST preferences at EOL.

Footnotes

Acknowledgments

The study was supported by the National Health Research Institutes (NHRI-EX107–10704PI) and Ministry of Science and Technology (MOST 104-2314-B-182-027-MY3) and Chang Gung Memorial Hospital (BMRP888). The corresponding author has full access to all study data, analyzed the data with F.-H.W., and takes responsibility for the integrity of the data and accuracy of the data analysis.

Author Disclosure Statement

No competing financial interests exist.

References

Peppercorn

, Smith

, Helft

, et al. American Society of Clinical Oncology statement: Toward individualized care for patients with advanced cancer. J Clin Oncol, 2011; 29:755–760.

Heyland

, Barwich

, Pichora

, et al.: Failure to engage hospitalized elderly patients and their families in advance care planning. JAMA Intern Med, 2013; 173:778–787.

Wendler

, Rid

: Systematic review: The effect on surrogates of making treatment decisions for others. Ann Intern Med, 2011; 154:336–346.

Silveira

, Kim

, Langa

: Advance directives and outcomes of surrogate decision making before death. N Engl J Med, 2010; 362:1211–1218.

Majesko

, Hong

, Weissfeld

, White

: Identifying family members who may struggle in the role of surrogate decision maker. Crit Care Med, 2012; 40:2281–2286.

Chiarchiaro

, Buddadhumaruk

, Arnold

, White

: Prior advance care planning is associated with less decisional conflict among surrogates for critically ill patients. Ann Am Thorac Soc, 2015; 12:1528–1533.

Yun

, Lee

, Chang

, et al.: The life-sustaining treatments among cancer patients at end of life and the caregiver's experience and perspectives. Support Care Cancer, 2010; 18:189–196.

Tsai

, Lee

, et al.: Depression of family caregivers is associated with disagreements on life-sustaining preferences for treating patients with dementia. PLoS One, 2015; 10:e0133711.

Shalowitz

, Garrett-Mayer

, Wendler

: The accuracy of surrogate decision makers. Arch Intern Med, 2006; 166:493–497.

10.

Moorman

, Inoue

: Predicting a partner's end-of-life preferences, or substituting one's own?. J Marriage Family, 2013; 75:734–745.

11.

Chen

, Fan

: The family and harmonious medical decision making: Cherishing an appropriate Confucian moral balance. J Med Philos, 2010; 35:573–586.

12.

Zhang

, Zyzanski

, Siminoff

: Differential patient-caregiver opinions of treatment and care for advanced lung cancer patients. Soc Sci Med, 2010; 70:1155–1158.

13.

Gries

, Engelberg

, Kross

, et al.: Predictors of symptoms of posttraumatic stress and depression in family members after patient death in the ICU. Chest, 2010; 137:280–287.

14.

Phipps

, True

, Horris

, et al.: Approaching the end of life: Attitudes, preferences, and behaviors of African-American and white patients and their family caregivers. J Clin Oncol, 2003; 21:549–554.

15.

Engelberg

, Patrick

, Curtis

: Correspondence between patients' preferences and surrogates' understandings for dying and death. J Pain Symptom Manage, 2005; 30:498–509.

16.

Yun

, You

, Lee

, et al.: Understanding disparities in aggressive care preferences between patients with terminal illness and their family members. J Pain Symptom Manage, 2006; 31:513–521.

17.

Malhotra

, Farooqui

, Kanesvaran

, et al.: Comparison of preferences for end-of-life care among patients with advanced cancer and their caregivers: A discrete choice experiment. Palliat Med, 2015; 29:842–850.

18.

Shin

, Cho

, Kim

, et al.: Discordance among patient preferences, caregiver preferences, and caregiver predictions of patient preferences regarding disclosure of terminal status and end-of-life choices. Psychooncology, 2015; 24:212–219.

19.

Ackroyd

, Russon

, Newell

: Views of oncology patients, their relatives and oncologists on cardiopulmonary resuscitation (CPR): Questionnaire-based study. Palliat Med, 2007; 21:139–144.

20.

Ahn

, Shin

, Choi

, et al.: The impact of awareness of terminal illness on quality of death and care decision making: A prospective nationwide survey of bereaved family members of advanced cancer patients. PsychoOncol, 2013; 22:2771–2778.

21.

Lee

, Yun

, An

, et al.: The understanding of terminal cancer and its relationship with attitudes toward end-of-Life care issues. Med Decis Making, 2014; 34:720–734.

22.

Hauke

, Reiter-Theil

, Hoster

, et al.: The role of relatives in decisions concerning life-prolonging treatment in patients with end-stage malignant disorders: Informants, advocates or surrogate decision-makers?. Ann Oncol, 2011; 22:2667–2674.

23.

Klinkenberg

, Willems

, Onwuteaka-Philipsen

, et al.: Preferences in end-of-life care of older persons: After-death interviews with proxy respondents. Soc Sci Med, 2004; 59:2467–2477.

24.

Hwang

, Keam

, Kim

, Yun

: Factors related to the differential preference for cardiopulmonary resuscitation between patients with terminal cancer and that of their respective family caregivers. Am J Hosp Palliat Care, 2016; 33:20–26.

25.

Turan

, Goldstein

, Garber

, Carstensen

: Knowing loved ones' end-of-life health care wishes: Attachment security predicts caregivers' accuracy. Health Psychol, 2011; 30:814–818.

26.

, Chou

, Hou

, et al.: Changes in and modifiable patient- and family caregiver-related factors associated with cancer patients' high self-perceived burden to others at the end of life: A longitudinal study. Eur J Cancer Care. (In press.)

27.

McCorkle

, Young

: Development of a symptom distress scale. Cancer Nurs, 1978; 1:373–378.

28.

Benoliel

, McCorkle

, Young

: Development of a social dependency scale. Res Nurs Health, 1980; 3:3–10.

29.

Zigmond

, Snaith

: The Hospital Anxiety and Depression Scale. Acta Psychiatr Scan, 1983; 67:361–370.

30.

Radloff

: The CES-D: Scale a self-report depression scale for researching the general population. Appl Psychol Measures, 1977; 1; 385–401.

31.

Cohen

, Mount

, Tomas

JJN

, et al.: Existential well-being is an important determinant of quality of life. Evidence from the McGill Quality of Life Questionnaire. Cancer, 1996; 77:576–586.

32.

Weitzner

, Jacobsen

, Wagner

Jr , et al.: The Caregiver Quality of Life Index-Cancer (CQOLC) scale: Development and validation of an instrument to measure quality of life of the family caregiver of patients with cancer. Qual Life Res, 1999; 8:55–63.

33.

Given

, Given

, Stommel

, et al.: The Caregiver Reaction Assessment (CRA) for caregivers to persons with chronic physical and mental impairments. Res Nurs Health, 1992; 15:271–283.

34.

Zucchini

, MacDonald

, Langrock

: Hidden Markov Models for Time Series, An Introduction Using R, 2nd ed. CRC Press, Boca Raton, FL, 2016.

35.

Eddy

: Profile hidden Markov models. Bioinformatics, 1998; 14:755–763.

36.

Akaike

: Factor analysis and AIC. Psychometrika, 1987; 52:317–332.

37.

Schwarz

: Estimating the dimension of a model. Ann Stat, 1978; 6:461–464.

38.

Bozdogan

: Model selection and Akaike's information criterion (AIC): The general theory and its analytical extensions. Psychometrika, 1987; 52:345–370.

39.

Viera

, Garrett

: Understanding interobserver agreement: The kappa statistic. Fam Med, 2005; 37:360–363.

40.

Mowen

, Culhane

: Modeling recidivism within the study of reentry: Hierarchical generalized linear models and lagged dependent variable models. Crim Justice Behav, 2017; 44:85–102.

41.

Lyons

, Lee

, Bennett

, et al.: Symptom incongruence trajectories in lung cancer dyads. J Pain Symptom Manage, 2014; 48:1031–1040.

42.

Hinderer

, Friedmann

, Fins

: Withdrawal of life-sustaining treatment: Patient and proxy agreement: A secondary analysis of “contracts, covenants, and advance care planning.”. Dimens Crit Care Nurs, 2015; 34:91–99.

43.

Lee

: Filial duty as the moral foundation of caring for the elderly: Its possibility and limitations. In: Fun

(eds): Family-Oriented Informed Consent: East Asian and American Perspectives. Springer International Publishing, Switzerland, 2015, pp.137–147.

44.

Coelho

, Barbosa

: Family anticipatory grief: An integrative literature review. Am J Hosp Palliat Care, 2017; 34:774–785.

45.

Mack

, Weeks

, Wright

, et al.: End-of-life discussions, goal attainment, and distress at the end of life: Predictors and outcomes of receipt of care consistent with preferences. J Clin Oncol, 2010; 28:1203–1208.

46.

Shif

, Doshi

, Almoosa

: What CPR means to surrogate decision makers of ICU patients. Resuscitation, 2015; 90:73–78.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.05 MB