Advertisement

Terminally Ill Cancer Patients' Concordance Between Preferred Life-Sustaining Treatment States in Their Last Six Months of Life and Received Life-Sustaining Treatment States in Their Last Month: An Observational Study

Open ArchivePublished:July 16, 2018DOI:https://doi.org/10.1016/j.jpainsymman.2018.07.003

      Abstract

      Context/Objective

      The extent to which patients' preferences for end-of-life (EOL) care are honored may be distorted if preferences are measured long before death, a common approach of existing research. We examined the concordance between cancer patients' states of life-sustaining treatments (LSTs) received in their last month and LST preference states assessed longitudinally over their last six months.

      Methods

      We examined states of preferred and received LSTs (cardiopulmonary resuscitation, intensive care unit care, chest compression, intubation with mechanical ventilation, intravenous nutrition, and nasogastric tube feeding) in 271 cancer patients' last six months by a transition model with hidden Markov modeling (HMM). The extent of concordance was measured by a percentage and a kappa value.

      Results

      HMM identified four LST preference states: life-sustaining preferring, comfort preferring, uncertain, and nutrition preferring. HMM identified four LST states received in patients' last month: generally received LSTs, LSTs uniformly withheld, selectively received LSTs, and received intravenous nutrition only. LSTs received concurred poorly with patients' preferences estimated right before death (39.5% and kappa value: 0.06 [95% CI: −0.02, 0.13]). Patients in the life-sustaining–preferring, uncertain, and nutrition-preferring states primarily received no LSTs, and patients in three of four states received intravenous nutrition against their preferences. Concordance was strongest for comfort-preferring patients.

      Conclusions

      Concordance was poor between patients' preferred and received LST states. Interventions are needed to clarify patients' EOL care goals and to facilitate their understanding about LST's ineffectiveness in prolonging life at EOL. Such interventions might increase patients' comfort preference and ensure concordance between their preferred and received EOL care.

      Key Words

      Introduction

      Concordance between terminally ill cancer patients' preferred and received care is valued as an essential element of high-quality patient-centered end-of-life (EOL) care.
      Institute of Medicine
      Dying in America: improving quality and honoring individual preferences near the end of life.
      • Sudore R.L.
      • Heyland D.K.
      • Lum H.D.
      • et al.
      Outcomes that define successful advance care planning: a Delphi panel consensus.
      Respecting patients' preferences for avoiding life-sustaining treatments (LSTs)
      • Mack J.W.
      • Weeks J.C.
      • Wright A.A.
      • et al.
      EOL discussions, goal attainment, and distress at the end of life: predictors and outcomes of receipt of care consistent with preferences.
      • Silveira M.J.
      • Kim S.Y.
      • Langa K.M.
      Advance directives and outcomes of surrogate decision making before death.
      • Tang S.T.
      • Liu T.W.
      • Wen F.H.
      • et al.
      A decade of changes in terminally ill cancer patients' preferences for life-sustaining treatments.
      and clarifying their misunderstandings about LST efficacy at EOL
      • Heyland D.K.
      • Frank C.
      • Groll D.
      • et al.
      Understanding cardiopulmonary resuscitation decision making: perspectives of seriously ill hospitalized patients and family members.
      • Cox C.E.
      • Martinu T.
      • Sathy S.J.
      • et al.
      Expectations and outcomes of prolonged mechanical ventilation.
      not only “individualizes” EOL cancer care
      • Peppercorn J.M.
      • Smith T.J.
      • Helft P.R.
      • et al.
      American society of clinical oncology statement: toward individualized care for patients with advanced cancer.
      but may also counteract the international trend toward increasingly aggressive and costly EOL cancer care,
      • Falchook A.D.
      • Dusetzina S.B.
      • Tian F.
      • et al.
      Aggressive EOL care for metastatic cancer patients younger than age 65 years.
      • Bekelman J.E.
      • Halpern S.D.
      • Blankart C.R.
      • et al.
      Comparison of site of death, health care utilization, and hospital expenditures for patients dying with cancer in 7 developed countries.
      • Hung Y.N.
      • Liu T.W.
      • Wen F.H.
      • Chou W.C.
      • Tang S.T.
      Escalating healthcare expenditures in cancer decedents’ last year of life: a decade of evidence from a retrospective population-based cohort study in Taiwan.
      which endangers the viability of health care systems.
      • Luengo-Fernandez R.
      • Leal J.
      • Gray A.
      • et al.
      Economic burden of cancer across the European Union: a population-based cost analysis.
      • Sullivan R.
      • Peppercorn J.
      • Sikora K.
      • et al.
      Delivering affordable cancer care in high-income countries.
      However, among the few studies
      • Mack J.W.
      • Weeks J.C.
      • Wright A.A.
      • et al.
      EOL discussions, goal attainment, and distress at the end of life: predictors and outcomes of receipt of care consistent with preferences.
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Detering K.M.
      • Hancock A.D.
      • Reade M.C.
      • et al.
      The impact of advance care planning on end of life care in elderly patients: randomised controlled trial.
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      • Song M.K.
      • Ward S.E.
      • Hanson L.C.
      • et al.
      Determining consistency of surrogate decisions and EOL care received with patient goals-of-care preferences.
      that prospectively investigated concordance between patients' preferred and received EOL care (including two randomized, advanced care planning clinical trials),
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Detering K.M.
      • Hancock A.D.
      • Reade M.C.
      • et al.
      The impact of advance care planning on end of life care in elderly patients: randomised controlled trial.
      preferences were measured only once (at baseline),
      • Mack J.W.
      • Weeks J.C.
      • Wright A.A.
      • et al.
      EOL discussions, goal attainment, and distress at the end of life: predictors and outcomes of receipt of care consistent with preferences.
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Detering K.M.
      • Hancock A.D.
      • Reade M.C.
      • et al.
      The impact of advance care planning on end of life care in elderly patients: randomised controlled trial.
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      • Song M.K.
      • Ward S.E.
      • Hanson L.C.
      • et al.
      Determining consistency of surrogate decisions and EOL care received with patient goals-of-care preferences.
      with a median time before death of 76–146 days
      • Mack J.W.
      • Weeks J.C.
      • Wright A.A.
      • et al.
      EOL discussions, goal attainment, and distress at the end of life: predictors and outcomes of receipt of care consistent with preferences.
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      • Song M.K.
      • Ward S.E.
      • Hanson L.C.
      • et al.
      Determining consistency of surrogate decisions and EOL care received with patient goals-of-care preferences.
      or mean (SD) of 362.2–388.8 (255.7–288.4)
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
       days. EOL care preferences are stable for most patients (50.7%–88.9%),
      • Auriemma C.L.
      • Nguyen C.A.
      • Bronheim R.
      • et al.
      Stability of EOL preferences: a systematic review of the evidence.
      with a substantial minority (11.1%–49.3%) changing their preferences as death approached. These patients' preferences cannot be captured if assessed at baseline only, thus distorting the extent to which patients' EOL care preferences are honored. Furthermore, applications of research findings in busy clinical settings may be complicated and impractical when assessing preferences for multiple LSTs.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      • Auriemma C.L.
      • Nguyen C.A.
      • Bronheim R.
      • et al.
      Stability of EOL preferences: a systematic review of the evidence.
      Rather than focusing on individual LSTs, clinicians can parsimoniously identify LST preference patterns/states,
      • Tang S.T.
      • Wen F.H.
      • Chang W.C.
      • et al.
      Preferences for life-sustaining treatments examined by hidden Markov modeling are mostly stable in terminally ill cancer patients’ last 6 months of life.
      thus minimizing the time spent assessing LST preferences and avoiding burdening terminally ill patients with choosing among multiple individual LSTs when they are physically and psychologically frail. Moreover, none of the previous prospective studies
      • Mack J.W.
      • Weeks J.C.
      • Wright A.A.
      • et al.
      EOL discussions, goal attainment, and distress at the end of life: predictors and outcomes of receipt of care consistent with preferences.
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Detering K.M.
      • Hancock A.D.
      • Reade M.C.
      • et al.
      The impact of advance care planning on end of life care in elderly patients: randomised controlled trial.
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      • Song M.K.
      • Ward S.E.
      • Hanson L.C.
      • et al.
      Determining consistency of surrogate decisions and EOL care received with patient goals-of-care preferences.
      was from Asian countries such as Taiwan where patient autonomy is commonly subordinate to families' power for EOL care under the Confucian doctrine of filial piety.
      • Lee L.
      Filial duty as the moral foundation of caring for the elderly: its possibility and limitations.
      Therefore, this study was conducted to examine the concordance between terminally ill Taiwanese cancer patients' LST states received in the last month and LST preference states estimated longitudinally during their last six months.

      Methods

      Design and Sample

      Data for this study were from a longitudinal study on the quality of death and dying. The stability of patients' state-specific changes in preferences for cardiopulmonary resuscitation (CPR), intensive care unit (ICU) care, chest compression, intubation with mechanical ventilation, and nutrition support in their last six months has been reported.
      • Tang S.T.
      • Wen F.H.
      • Chang W.C.
      • et al.
      Preferences for life-sustaining treatments examined by hidden Markov modeling are mostly stable in terminally ill cancer patients’ last 6 months of life.
      Methodological details have been reported.
      • Tang S.T.
      • Wen F.H.
      • Chang W.C.
      • et al.
      Preferences for life-sustaining treatments examined by hidden Markov modeling are mostly stable in terminally ill cancer patients’ last 6 months of life.
      Briefly, adult cancer patients were recruited by convenience in 2009–2012 and followed until December 2015. Patients were referred by their oncologist who declared them terminally ill when their disease continued to progress and was unresponsive to curative treatments. Cognitively competent participants were interviewed by trained, experienced oncology nurses approximately once every two weeks during hospitalization or at outpatient clinics, until they declined to participate or died. The study site's institutional review board approved the study (98-0476B). All participants signed a written informed consent.

      Preferences for Life-Sustaining Treatments

      To assess participants' current preferences for CPR, ICU care, chest compression, intubation with mechanical ventilation, intravenous nutrition, and nasogastric tube feeding when they would someday need each treatment, we used an adapted interview protocol (Table 1),
      • Stapleton R.D.
      • Nielsen E.L.
      • Engelberg R.A.
      • et al.
      Association of depression and life-sustaining treatment preferences in patients with COPD.
      without informing participants of the likelihood of benefit or risk from each treatment. For each LST, patients were asked whether they 1) wanted, 2) did not want, or 3) were undecided about the treatment. LST preferences were assessed without informing participants of their prognosis (disease curability or estimated survival) because prognostic awareness was under investigation in the original study and published in another article.
      • Chen C.H.
      • Wen F.H.
      • Hou M.M.
      • et al.
      Transitions in prognostic awareness among terminally ill cancer patients in their last 6 months of life examined by multi-state Markov modeling.
      Table 1Interview Questions Regarding Preferences for Life-Sustaining Treatments
      • 1.
        At the first interview, before participants offered their preferences regarding cardiopulmonary resuscitation (CPR), they were told, “If your heart were to stop beating and your life were in danger, your health care professionals might provide CPR. CPR comprises a combination of electric shocks to the heart, pumping the chest to stimulate the heart, placing a tube through the mouth or nose into the lungs and attaching this tube to a breathing machine to help with breathing, and heart medications given through the veins.” Participants were then asked, “If your life were in danger, would you want to receive CPR?”
      • 2.
        For life-sustaining treatments, participants were asked, “If you were dying and
        • (1)
          your heart stopped beating, would you want your chest to be pumped to stimulate the heart to beat?
        • (2)
          If you were unable to breathe on your own, would you want to be intubated and on a breathing machine? In this situation, a tube would be placed through your mouth or nose into your lungs. This tube would be attached to a breathing machine. During that time, you would have to be continuously on the breathing machine and would be unable to talk and might be sedated.
        • (3)
          If you need intensive care, would you like to stay in an intensive care unit (ICU)? An ICU is an isolated care unit that heavily uses health technology to provide intensive care with more nursing staff to closely monitor you. If you receive care in an ICU, you could only have contact with your family at specific visiting times.
        • (4)
          If you cannot eat by yourself, would you be willing to be fed by artificial means, such as feeding through a nasogastric tube or receiving nutritional support by injection?

      Life-Sustaining Treatments Received

      Data on receipt of the aforesaid six LSTs in patients' last month were retrieved from medical records. For patient participants who had not been hospitalized at the study hospital or did not return there for clinical visits in their last month, LST receipt was obtained from family caregivers during bereavement follow-ups.

      Analysis

      Baseline characteristics and LST preferences of patients comprising the final sample and those excluded from analysis were compared by chi-square tests. Identification of and changes in distinct LST preference states between consecutive time points were examined by a transition model with hidden Markov modeling (HMM).
      • Vermunt J.K.
      • Tran B.
      • Magidson J.
      Latent class models in longitudinal research.
      In HMM, LST preferences were treated as patterns or sets of treatment preferences (“latent states”) rather than individual treatment preferences. Latent states were estimated by observed response variables (i.e., preferences for the six selected LSTs) to reflect the unobserved (latent) concepts (i.e., LST-preference states). HMM simultaneously examined LST preferences as latent treatment preference states, described dynamic changes in patients' LST preferences in their last six months, and estimated LST transition probabilities.
      • Eddy S.R.
      Profile hidden Markov models.
      Data were analyzed using Latent GOLD 5.0 (Belmont, MA).
      The first part of HMM modeling assigned patients to a limited number of mutually exclusive LST preference states based on common characteristics that discriminate patients in each state. Choosing the optimal number of states in the model was determined by three factors: 1) examining fit indices with information criteria (ICs), for example, the Akaike information criterion (AIC),
      • Akaike H.
      Factor analysis and AIC.
      the Bayesian information criterion (BIC),
      • Schwarz G.
      Estimating the dimension of a model.
      the consistent AIC (CAIC),
      • Bozdogan H.
      Model selection and Akaike's information criterion (AIC). The general theory and its analytical extensions.
      and log-likelihood (LL)
      • Nylund K.L.
      • Asparouhov T.
      • Muthén B.O.
      Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study.
      ; 2) parsimony with adequate sample size for each state
      • Vermunt J.K.
      • Tran B.
      • Magidson J.
      Latent class models in longitudinal research.
      • Nylund K.L.
      • Asparouhov T.
      • Muthén B.O.
      Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study.
      • Muthén B.
      Latent variable analysis: growth mixture modeling and related techniques for longitudinal data.
      ; and 3) clinical meaningfulness of latent-state results. Lower AIC, BIC, and CAIC but higher LL values indicate a better model fit. However, in plots of IC values versus state number, the flattening of IC values between consecutive numbers of states suggests that higher state numbers are not statistically meaningful.
      • Muthén B.
      Latent variable analysis: growth mixture modeling and related techniques for longitudinal data.
      These criteria (i.e., generally lower IC and higher LL values, with more weight on flattening IC values between consecutive numbers of states) were used to determine the optimal number of states. The second part of HMM estimated state-transition probabilities.
      • Eddy S.R.
      Profile hidden Markov models.
      Transition probability represented the likelihood that a patient would prefer a specific set of LSTs at time t, given his/her preference for a specific LST set at time t − 1.
      States of LSTs received in patients' last month were identified using the same procedures as described previously, except those for estimating state-transition probabilities. Concordance was evaluated by cross-tabulating LST preference states estimated just before death (based on each state's initial and transition probabilities) and LST states received in the last month. Concordance is expressed both as a percentage (received/preferred LSTs) and by a kappa coefficient for chance-corrected agreement.
      • Landis J.R.
      • Koch G.G.
      The measurement of observer agreement for categorical data.
      Concordance (kappa value) was determined as poor (≤0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), or almost perfect (0.81–1.00).
      • Landis J.R.
      • Koch G.G.
      The measurement of observer agreement for categorical data.

      Results

      Sample Characteristics

      Of 433 eligible terminally ill cancer patients, 380 were enrolled (participation rate = 87.8%) (Supplemental Figure 1). Among the 380 patients enrolled, 317 died by the end of follow-ups and information of LST preferences was available for 303 patients. The final sample included 271 patients with complete information on preferred and received LSTs. These patients' baseline characteristics and LST preferences did not differ significantly from those excluded from analysis except for a few diagnoses (data not shown; available from the corresponding author).
      The majority of participants were male (55.0%) (Table 2), with a mean (SD) age of 57.9 (12.7) years, and married (82.6%). The most common cancer sites were pancreas (17.3%), stomach (16.2%), liver (13.7%), lung (11.8%), and head and neck (9.2%). At enrollment, participants had been diagnosed on average 17.2 (SD = 32.3) months ago. After enrollment, participants survived 170.2 days (SD = 214.3; median = 89.0; range = 3–1506) and were assessed on average 5.3 times (SD = 3.5; median = 5.0, range = 1–18; 73.8% had more than two assessments) in their last six months. The following results are based on 1372 assessments, separated by 18.7 days on average (SD = 7.7; median = 16.0; range = 10–67). The last assessment was on average 25.3 days (SD = 25.8; median = 17.0; range = 1–166) before death. For details on initial and final preferences for each LST and LSTs received in the last month, see Table 2.
      Table 2Participants' Demographic and Clinical Characteristics, Preferred Life-Sustaining Treatments, and Received Life-Sustaining Treatments (N = 271)
      Characteristicn%Characteristicn%
      GenderEducational level
       Male14955.0 ≤Elementary school10639.7
       Female12245.0 Junior high school5119.1
      Age (yrs) Senior high school8130.3
       Mean ± SD57.92 ± 12.68 >Senior high school2910.9
      Marital statusCancer site
       Single124.4 Pancreas4717.3
       Married22382.6 Stomach4416.2
       Divorced/separated186.7 Liver3713.7
       Widowed176.3 Lung3211.8
      With chronic disease Head and neck259.2
       Yes16862.0 Esophagus217.8
       No10338.0 Colon-rectum176.3
      Metastasis Breast124.4
       Yes20776.7 Other3613.3
       No6323.3Postdiagnosis survival (months)
      Mean ± SD17.16 ± 32.26
      Characteristicn%Characteristicn%
      Initial preferences for life-sustaining treatments
      CPRICU care
       Yes2810.3 Yes4014.8
       No16761.6 No15055.6
       Unsure7628.1 Unsure8029.6
      Chest compressionIntubation
       Yes3312.2 Yes248.9
       No16962.4 No18066.4
       Unsure6925.5 Unsure6724.7
      Tube feedingIntravenous nutrition
       Yes6624.4 Yes12646.5
       No13048.0 No8732.1
       Unsure7527.7 Unsure5821.4
      Final preferences for life-sustaining treatments
      CPR ICU care
       Yes259.3 Yes3613.4
       No18066.7 No16260.2
       Unsure6524.1 Unsure7126.4
      Chest compressionIntubation
       Yes269.6 Yes217.8
       No18468.1 No19070.4
       Unsure6022.2 Unsure5921.9
      Tube feedingIntravenous nutrition
       Yes7327.0 Yes13550.0
       No13349.3 No8832.6
       Unsure6423.7 Unsure4717.4
      Life-sustaining treatments received in the last month of life (N = 271)
       CPR114.1 ICU care197.0
       Chest compression93.3 Intubation207.4
       Tube feeding8029.5 Intravenous nutrition9033.2
      Hospice care (N = 270)18267.4Place of death
       Hospital14553.5
       Home12646.5
      CPR = cardiopulmonary resuscitation; ICU = intensive care unit.

      Hidden Markov Modeling of Preferred and Received LST States and Transitions Between LST Preference States in Participants' Last Six Months

      Evaluation of model fit indexes (Supplemental Table 1), the AIC, BIC, and CAIC plots (Supplemental Figure 2), and clinical meaningfulness supports selection of a four-state solution for LST preferences as optimal and parsimonious. These four states' emission probabilities and sizes (state probabilities) are given in Table 3. The states were labeled life-sustaining preferring, comfort preferring, uncertain, and nutrition preferring (principally by intravenous nutrition support). When LST preferences were initially assessed, the most prevalent state was comfort preferring (40.9%) (Table 3), followed by uncertain (26.6%) and nutrition preferring (22.6%) states. About one-tenth of participants preferred all LSTs (9.9%).
      Table 3Emission Probabilities and Initial Sizes of the Four Life-Sustaining Treatment-Preference States (N = 303)
      Preferences for Life-Sustaining TreatmentsStateLife-Sustaining PreferringComfort PreferringUncertainNutrition Preferring
      Initial Size (%)9.940.926.622.6
      Want treatment
       Cardiopulmonary resuscitation0.9440.0000.0100.011
       Intensive care unit care0.9780.0020.0110.137
       Chest compression0.9350.0030.0130.034
       Intubation0.8600.0000.0000.009
       Nasogastric tube feeding0.9630.0160.0840.425
       Intravenous nutrition support1.0000.2740.2690.766
      Don't want treatment
       Cardiopulmonary resuscitation0.0560.9980.0550.852
       Intensive care unit care0.0070.9980.0000.685
       Chest compression0.0580.9980.0280.938
       Intubation0.1331.0000.0680.976
       Nasogastric tube feeding0.0300.9830.0370.318
       Intravenous nutrition support0.0000.7200.0020.099
      Undecided
       Cardiopulmonary resuscitation0.0000.0020.9350.138
       Intensive care unit care0.0150.0000.9890.179
       Chest compression0.0070.0000.9590.029
       Intubation0.0080.0000.9320.016
       Nasogastric tube feeding0.0070.0010.8800.257
       Intravenous nutrition support0.0000.0060.7290.136
      Emission probability represents the observed probability that each patient would want, not want, or be undecided about each LST in each identified state. Bold-italicized values indicate that participants in the life-sustaining preference, comfort preference, and uncertain states had uniformly high probabilities of wanting, rejecting, and being undecided about, respectively, each treatment. Bold values indicate the emission probabilities for participants in each state wanting, rejecting, and being undecided about each treatment.
      LST preferences were highly stable over patients' last six months as evident by remaining in their original LST preference state (the diagonal in Table 4, 92.8%–97.7%) rather than changing to other preference states. Therefore, the prevalence (state probability) fluctuated within a narrow range over the last six months (Fig. 1). As time passed and death approached, the proportion of patients in the comfort- and nutrition-preferring states increased, whereas the proportion of those in the uncertain state tended to decrease (the only state to do so) from about 27% to 21%. Over time, the nutrition-preferring state became the second-most prevalent LST preference state, only surpassed by the comfort-preferring state.
      Table 4Transition Probabilities of Life-Sustaining Treatment States From Time [t − 1] to Time [t] (N = 303)
      Time [t]Time t [t − 1]
      StateLife-Sustaining PreferringComfort PreferringUncertainNutrition Preferring
      Life-sustaining preferring0.9490.0000.0090.010
      Comfort preferring0.0000.9770.0220.023
      Uncertain0.0280.0110.9280.029
      Nutrition preferring0.0230.0120.0420.938
      Bold indicates the highest transition probability between different times.
      Figure thumbnail gr1
      Fig. 1Probability (size) for each LST preference state estimated at different times after enrollment in patients' last six months; life-sustaining treatment (LST) preferences: State 1: LST preference; State 2: comfort preference; State 3: uncertain; State 4: nutrition preference.
      HMM identified four states of LSTs received by cancer patients in their last month (Supplemental Table 2 and Supplemental Figure 2). These four states were identified as generally received LSTs (State 1), LSTs uniformly withheld (State 2), selectively received LSTs (State 3), and received intravenous nutrition only (State 4) (Table 5). Patients in State 1 predominantly received aggressive LSTs, for example, CPR, chest compression, and intubation with mechanical ventilation support (probabilities: 75.6%–97.4%), and had an approximately equal chance of receiving or not receiving nutrition support by intravenous and nasogastric tube feeding. By contrast, State 3 patients selectively received ICU care, intubation with mechanical ventilation support, and nasogastric tube feeding, whereas other LSTs were withheld. The most prevalent received LST state was receiving intravenous nutrition only (State 4, size: 50.5%), followed by LSTs uniformly withheld (State 2, size: 41.7%). Less than 5% of participants received mostly or selected aggressive LSTs before they died.
      Table 5Probabilities and Sizes of the Four Received Life-Sustaining Treatment States (N = 274)
      Life-Sustaining Treatments ReceivedStateGenerally Received LSTsLSTs Uniformly WithheldSelectively Received LSTsReceived Only IV Nutrition
      Size (%)3.441.74.450.5
      Treatment received
       Cardiopulmonary resuscitation0.9740.0020.0010.012
       Intensive care unit care0.3250.0040.9320.031
       Chest compression0.9700.0000.0010.000
       Intubation0.7560.0220.8620.000
       Nasogastric tube feeding0.4390.2820.7800.255
       Intravenous nutrition support0.4410.0320.2950.576
      Treatment not received
       Cardiopulmonary resuscitation0.0270.9980.9990.988
       Intensive care unit care0.6750.9960.0680.969
       Chest compression0.0311.0000.9991.000
       Intubation0.2440.9780.1381.000
       Nasogastric tube feeding0.5610.7180.2200.745
       Intravenous nutrition support0.5590.9680.7050.424
      LST = life-sustaining treatment.
      Probability represents the observed probability that each patient did or did not receive each life-sustaining treatment in each identified state. Bold indicates the probabilities for participants in each state receiving or not receiving each treatment.

      Concordance Between Preferred and Received LST States at EOL

      The concordance between patients' estimated preferred and received LST states was 39.5% (Table 6, bolded values). For patients in the uncertain state, preferred and received LST states were evaluated as concordant when they received some LSTs in the last month because they neither accepted nor rejected all LSTs estimated in the last assessment. LST preferences were least honored for patients in the life-sustaining–preferring state (0%; 0 of 24), followed by the uncertain state (6.9%; four of 58), nutrition-preferring state (38.4%; 28 of 73), and comfort-preferring state (64.7%; 75 of 116). Indeed, among patients estimated to be in the life-sustaining-preferring, uncertain, or nutrition-preferring states right before death, approximately one-half to two-thirds received no LSTs in their last month (14, 36, and 40 of 24, 58, and 73 patients, respectively, Table 6). Receiving intravenous nutrition only was the second-most likely LST state received for patients across all LST preference states. Concordance between preferred and received LST states was poor as shown by kappa = 0.06 (95% CI: −0.02, 0.13).
      Table 6Concordance Between States of Preferred and Received Life-Sustaining Treatments (N = 271)
      Life-Sustaining Treatment Preference StateReceived Life-Sustaining Treatment State
      Life-sustaining treatments received: State 1: generally received LSTs; State 2: LSTs uniformly withheld; State 3: selectively received LSTs (i.e., received ICU care, intubation with mechanical ventilation, and nasogastric tube feeding with other treatments withheld); State 4: received intravenous nutrition support only.
      Generally Received LSTsLSTs Uniformly WithheldSelectively Received LSTsReceived Only Intravenous NutritionTotal
      Life-sustaining preferring0141924
      Comfort preferring375335116
      Uncertain4364
      Preferred and received LST states were evaluated as concordant when some LSTs were received in the last month of life based on patients' failure to accept or reject all LSTs examined.
      1458
      Nutrition preferring24032873
      Total91651186271
      LST = life-sustaining treatment.
      Bold indicates the number of patients whose preferred and received LST states were concordant.
      a Life-sustaining treatments received: State 1: generally received LSTs; State 2: LSTs uniformly withheld; State 3: selectively received LSTs (i.e., received ICU care, intubation with mechanical ventilation, and nasogastric tube feeding with other treatments withheld); State 4: received intravenous nutrition support only.
      b Preferred and received LST states were evaluated as concordant when some LSTs were received in the last month of life based on patients' failure to accept or reject all LSTs examined.

      Discussion

      Terminally ill Taiwanese cancer patients' preferred and received LST states did not agree beyond chance (only 39.5%), with poor concordance (kappa value: 0.06 [95% CI: −0.02, 0.13]). Discordance between preferred and received LST states was most likely when no LST was provided to patients who uniformly preferred/were unsure about LSTs or preferred nutrition support but rejected other LSTs in their last month. The second major discordance between preferred and received LST states came from participants in the life-sustaining-preferring, comfort-preferring, and uncertain states receiving intravenous nutrition at EOL.
      Our findings concur with reports that concordance was infrequent between terminally or seriously ill patients' preferences for EOL care and EOL care received, whether measured by individual LSTs
      • Detering K.M.
      • Hancock A.D.
      • Reade M.C.
      • et al.
      The impact of advance care planning on end of life care in elderly patients: randomised controlled trial.
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
      or goal of EOL care.
      • Mack J.W.
      • Weeks J.C.
      • Wright A.A.
      • et al.
      EOL discussions, goal attainment, and distress at the end of life: predictors and outcomes of receipt of care consistent with preferences.
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      • Song M.K.
      • Ward S.E.
      • Hanson L.C.
      • et al.
      Determining consistency of surrogate decisions and EOL care received with patient goals-of-care preferences.
      Preference for CPR was honored for only 40%
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
      and 58.9%
      • Detering K.M.
      • Hancock A.D.
      • Reade M.C.
      • et al.
      The impact of advance care planning on end of life care in elderly patients: randomised controlled trial.
      of decedents with congestive heart failure/end-stage renal disease and elderly decedents, respectively. EOL care goals were successfully attained for 54.2% of elderly patients,
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      39.6%–68.0% of patients with cancer,
      • Mack J.W.
      • Weeks J.C.
      • Wright A.A.
      • et al.
      EOL discussions, goal attainment, and distress at the end of life: predictors and outcomes of receipt of care consistent with preferences.
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      68.6% of patients under dialysis,
      • Song M.K.
      • Ward S.E.
      • Hanson L.C.
      • et al.
      Determining consistency of surrogate decisions and EOL care received with patient goals-of-care preferences.
      and 69.1% of patients with congestive heart failure and end-stage renal disease.
      • Kirchhoff K.T.
      • Hammes B.J.
      • Kehl K.A.
      • et al.
      Effect of a disease-specific advance care planning intervention on EOL care.
      Our observed concordance between preferred and received LST states is at the lowest end of evidence, likely stemming from cultural differences in respecting patient autonomy.
      • Lee L.
      Filial duty as the moral foundation of caring for the elderly: its possibility and limitations.
      In western cultures, individual autonomy is highly valued, including at EOL.
      • Blackhall L.J.
      • Murphy S.T.
      • Frank G.
      • et al.
      Ethnicity and attitudes toward patient autonomy.
      By contrast, in a Confucian doctrine–influenced society such as Taiwan, where family power is strongly exercised in medical care decision making, including EOL care,
      • Yang Y.I.
      A family-oriented Confucian approach to advance directives in EOL decision making for incompetent elderly patients.
      Taiwanese families have the authority to make medical decisions on behalf of terminally ill relatives even when they are physically capable or consciously competent.
      • Tang S.T.
      • Liu T.W.
      • Lai M.S.
      • et al.
      Congruence of knowledge, experiences and preferences for disclosure of diagnosis and prognosis between terminally-ill cancer patients and their family caregivers in Taiwan.
      Respecting patients' autonomy in Asian countries is further impeded by physicians' infrequently disclosing prognosis to cancer patients
      • Tang S.T.
      • Liu T.W.
      • Lai M.S.
      • et al.
      Congruence of knowledge, experiences and preferences for disclosure of diagnosis and prognosis between terminally-ill cancer patients and their family caregivers in Taiwan.
      • Yun Y.H.
      • Kwon Y.C.
      • Lee M.K.
      • et al.
      Experiences and attitudes of patients with terminal cancer and their family caregivers toward the disclosure of terminal illness.
      and EOL care seldom being discussed between patients and their families.
      • Yun Y.H.
      • Lee M.K.
      • Chang Y.J.
      • et al.
      The life-sustaining treatments among cancer patients at end of life and the caregiver’s experience and perspectives.
      Discussing issues about death and dying is taboo in Asia,
      • Kim S.
      • Hahm K.H.
      • Park H.W.
      • et al.
      A Korean perspective on developing a global policy for advance directives.
      resulting in families inaccurately understanding patients' preferences
      • Kim S.
      • Hahm K.H.
      • Park H.W.
      • et al.
      A Korean perspective on developing a global policy for advance directives.
      • Tang S.T.
      • Wen F.H.
      • Liu L.N.
      • et al.
      A decade of changes in family caregivers' preferences for life-sustaining treatments for terminally ill cancer patients at end of life in the context of a family-oriented society.
      and projecting their own preferences to shape patients' EOL care (by overtreating or undertreating, as discussed in the following sections).
      Our findings of discordance between preferred and received LST states concur with reports that such discrepancies often tend toward patients receiving less aggressive care than they prefer.
      • Cosgriff J.A.
      • Pisani M.
      • Bradley E.H.
      • et al.
      The association between treatment preferences and trajectories of care at the end-of-life.
      • Parr J.D.
      • Zhang B.
      • Nilsson M.E.
      • et al.
      The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
      • Gao X.
      • Prigerson H.G.
      • Diamond E.L.
      • et al.
      Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
      The large proportion of participants preferring or being undecided about all/some LSTs (59.1%; Table 3) may have resulted from lack of accurate prognostic awareness,
      • Tang S.T.
      • Liu T.W.
      • Chow J.M.
      • et al.
      Associations between accurate prognostic understanding and EOLcare preferences and its correlates among Taiwanese terminally ill cancer patients surveyed in 2011-2012.
      coupled with misunderstanding/overexpectations of LSTs' effectiveness in prolonging life at EOL.
      • Heyland D.K.
      • Frank C.
      • Groll D.
      • et al.
      Understanding cardiopulmonary resuscitation decision making: perspectives of seriously ill hospitalized patients and family members.
      • Cox C.E.
      • Martinu T.
      • Sathy S.J.
      • et al.
      Expectations and outcomes of prolonged mechanical ventilation.
      However, no patients in the life-sustaining-preferring state received their preferred treatments, and one-half to two-thirds of patients in the life-sustaining-preferring, uncertain, and nutrition-preferring states received no LST before death. These observations, that is, our participants received less aggressive care than they were willing to undergo, probably reflect Taiwan's success in diffusing hospice philosophy to limit futile LSTs. The number of Taiwanese hospice programs increased substantially over the past decade.

      Health Promotion Administration. 2015 Health Promotion Administration Annual Report. pp. 109. Health Promotion Administration, Ministry of Health and Welfare, R.O.C. Taipei, Taiwan. Accessed on July 28, 2017.

      Clinicians became more familiar with hospice philosophy and better appreciated the futility of providing LSTs to reverse the natural course of the dying process. Taiwanese clinicians usually endorse the family's power in decision making, communicating the costs and benefits of LSTs to them, and obtaining their consent to limit LSTs, not only when a patient's death is imminent, and he/she cannot make EOL care decisions, but also when patients prefer or are uncertain about receiving LSTs. Besides, the government-run National Health Insurance, the only payer of health care services in Taiwan, advocates avoiding futile LSTs to counteract increasingly aggressive EOL care in Taiwan.
      • Bekelman J.E.
      • Halpern S.D.
      • Blankart C.R.
      • et al.
      Comparison of site of death, health care utilization, and hospital expenditures for patients dying with cancer in 7 developed countries.
      Therefore, terminally ill cancer patients' preferences for aggressive LSTs may not be honored to benefit both patients and society at large.
      Preferences of patients in the comfort-preferring state were most likely to be violated by receiving intravenous nutrition before they died. This treatment was also the LST most likely to be provided to participants in the life-sustaining-preferring and uncertain states right before death. Given the Confucian doctrine of filial duty
      • Lee L.
      Filial duty as the moral foundation of caring for the elderly: its possibility and limitations.
      • Blackhall L.J.
      • Murphy S.T.
      • Frank G.
      • et al.
      Ethnicity and attitudes toward patient autonomy.
      and the family's authority in decision making,
      • Lee L.
      Filial duty as the moral foundation of caring for the elderly: its possibility and limitations.
      • Blackhall L.J.
      • Murphy S.T.
      • Frank G.
      • et al.
      Ethnicity and attitudes toward patient autonomy.
      Taiwanese families feel obliged to provide food and nutrition to keep a parent (patient) “alive,” not only to stave off their loved one's physical deterioration but also to provide humanistic EOL care by not abandoning terminally ill patients to die miserably. In Taiwanese culture, a person who dies hungry is believed to become a “starving soul” or “hungry ghost/spirit” in hell. Therefore, even when patients clearly and uniformly rejected LSTs and families recognized that LSTs could cause unbearable suffering for patients in the life-sustaining-preferring and uncertain states, Taiwanese families might elect to forgo LSTs to avoid needlessly protracting the dying process but still insist on providing intravenous nutrition till the patient's death.
      • Tang S.T.
      • Wen F.H.
      • Liu L.N.
      • et al.
      A decade of changes in family caregivers' preferences for life-sustaining treatments for terminally ill cancer patients at end of life in the context of a family-oriented society.

      Study Strengths and Limitations

      The strengths of our study include evaluating the concordance between preferred and received LST states by longitudinally assessing LST preferences over each patient's last six months and using advanced statistics to explore a parsimonious number of preferred and received LST states rather than examining multiple individual LSTs. Health care professionals can efficiently differentiate among patients with different LST preferences by using a maximum of two sequential questions (Supplemental Table 3) to routinely assess their LST preferences, thus facilitating earlier, timelier, and individualized discussions about LST preferences at EOL.
      • Peppercorn J.M.
      • Smith T.J.
      • Helft P.R.
      • et al.
      American society of clinical oncology statement: toward individualized care for patients with advanced cancer.
      However, our sample's representation of the target population and the generalizability of our findings may have been compromised by convenience sampling from a single medical center. Generalization of our findings may also have been limited by a remarkable proportion of patients withdrawing or being excluded from analysis. Our findings from Taiwan need to be replicated for terminally ill cancer patients in other countries where cultural, societal, and health care characteristics may substantially differ. Our investigation into preferred and received LST states was limited to the six LSTs assessed, and our participants were not given an in-depth risk-benefit analysis for each LST. Furthermore, “undecided” was treated as a valid response with the “wanted” and “not wanted” responses. Indecision indicates patients' ambivalence about LSTs in decision making and categorizing the large proportion of “undecided” responses as “wanting” the treatment inflates the preference rate.
      • Kim S.Y.
      Improving medical decisions for incapacitated persons: does focusing on "accurate predictions" lead to an inaccurate picture?.
      On the other hand, excluding “undecided” responses would prevent understanding the needs of patients in the uncertain state and the EOL care they received, limiting opportunities to improve their EOL care quality. However, the appropriateness of our approach to determining preferred-received LST state concordance for terminally ill cancer patients who are uncertain about their LST preferences warrants further validation. Furthermore, we could not capture any changes in LST preferences between the last assessment and the patient's death to evaluate the concordance between preferred and received LST states. We also recognize the limitation inherent in not exploring terminally ill cancer patients' concerns about their LST preferences and barriers to not prominently shifting toward preferring less aggressive LSTs or reducing uncertainty in LST preferences even when death approached. Qualitative research is suggested to understand these issues in depth. Finally, we did not explore factors predisposing patients to have their LST preference state honored, including family caregivers' attitudes and preferences toward LSTs for their relative, warranting further investigation.

      Conclusion and Clinical Implications

      In conclusion, LSTs received by terminally ill Taiwanese cancer patients in their last month concurred poorly with their preferences, resulting in receiving less aggressive care than they preferred and receiving intravenous nutrition than they did not prefer. Honoring patients' unrealistic expectations
      • Heyland D.K.
      • Frank C.
      • Groll D.
      • et al.
      Understanding cardiopulmonary resuscitation decision making: perspectives of seriously ill hospitalized patients and family members.
      • Cox C.E.
      • Martinu T.
      • Sathy S.J.
      • et al.
      Expectations and outcomes of prolonged mechanical ventilation.
      of using LSTs to combat forthcoming death may cause more harm than benefits. Patients in the life-sustaining-preferring or uncertain states may need interventions to facilitate understanding of prognosis, clarify their EOL care goals, and realize the ineffectiveness of LSTs in prolonging life at EOL. Exploring and understanding the physical and psychological burden of receiving intravenous nutrition against terminally ill cancer patients' preferences is highly desirable to facilitate patient-family discussions about EOL care and to deliver personalized EOL care.
      • Peppercorn J.M.
      • Smith T.J.
      • Helft P.R.
      • et al.
      American society of clinical oncology statement: toward individualized care for patients with advanced cancer.
      With appropriate interventions tailored to the unique needs of terminally ill cancer patients at each mismatched preferred and received LST state identified in this study, value-based EOL care may be provided to achieve a good death consistent with patients' wishes
      Institute of Medicine
      Dying in America: improving quality and honoring individual preferences near the end of life.
      while avoiding potentially futile aggressive EOL care.

      Disclosures and Acknowledgments

      This study was funded by 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).
      All authors declare no financial or other conflict of interest.
      No funding sources had any role in designing and conducting the study; collecting, managing, analyzing, and interpreting the data; or preparing, reviewing, or approving the article.
      The corresponding author has full access to all study data, analyzed the data with Dr. Fur-Hsing Wen, and takes responsibility for the integrity of the data and accuracy of the data analysis.

      Appendix

      Figure thumbnail fx1
      Supplemental Figure 1Participant flowchart.
      Figure thumbnail fx2
      Supplemental Figure 2(A) Plots of AIC, BIC, and CAIC values for life-sustaining treatment preferences, information criterion value. (B) Plots of AIC, BIC, and CAIC values for life-sustaining treatments received, information criterion value. AIC = Akaike information criterion; BIC = Bayesian information criterion; CAIC = consistent AIC.
      Supplemental Table 1Model Fit Indexes for One- to Six-State Solutions of Preferences for Life-Sustaining Treatments
      State NumberLog Likelihood (LL)AIC (LL)BIC (LL)CAIC (LL)Number of ParametersDegrees of Freedom
      1−8169.6216,363.2316,407.7916,419.7912291
      2−5086.6410,227.2710,327.5410,354.5427276
      3−3923.597935.188098.588142.5844259
      4−3427.766981.517215.487278.4863240
      5−3241.656651.316963.267047.2684219
      6−3186.846587.686985.057092.05107196
      Supplemental Table 2Model Fit Indexes for One- to Five-State Solutions of Life-Sustaining Treatments Received
      State NumberLog Likelihood (LL)AIC (LL)BIC (LL)CAIC (LL)Number of ParametersDegrees of Freedom
      1−566.891145.781167.451173.456184.97
      2−502.751031.501078.481091.481356.70
      3−483.841007.691079.951099.952018.88
      4−482.101018.191115.751142.752715.39
      5−478.461024.931147.771181.77348.12
      AIC = Akaike information criterion; BIC = Bayesian information criterion; CAIC = consistent AIC.
      Bold indicates best-fitting model by each model fit index.
      Supplemental Table 3Differentiation Among LST Preference States
      TreatmentState
      1234
      CPR
      CPR can be replaced by ICU care, chest compression, intubation with mechanical ventilation.
      Yes
      Yes: prefer treatment.
      No
      No: reject treatment.
      UnsureNo
      No: reject treatment.
      Intravenous nutritional supportYes
      Yes: prefer treatment.
      No
      No: reject treatment.
      UnsureYes
      Yes: prefer treatment.
      LST = life-sustaining treatment; CPR = cardiopulmonary resuscitation; ICU = intensive care unit.
      a CPR can be replaced by ICU care, chest compression, intubation with mechanical ventilation.
      b Yes: prefer treatment.
      c No: reject treatment.

      References

        • Institute of Medicine
        Dying in America: improving quality and honoring individual preferences near the end of life.
        The National Academies Press, Washington, DC2014
        • Sudore R.L.
        • Heyland D.K.
        • Lum H.D.
        • et al.
        Outcomes that define successful advance care planning: a Delphi panel consensus.
        J Pain Symptom Manage. 2018; 55: 245-255
        • Mack J.W.
        • Weeks J.C.
        • Wright A.A.
        • et al.
        EOL 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
        • Silveira M.J.
        • Kim S.Y.
        • Langa K.M.
        Advance directives and outcomes of surrogate decision making before death.
        N Engl J Med. 2010; 362: 1211-1218
        • Tang S.T.
        • Liu T.W.
        • Wen F.H.
        • et al.
        A decade of changes in terminally ill cancer patients' preferences for life-sustaining treatments.
        J Natl Compr Canc Netw. 2015; 13: 1510-1518
        • Heyland D.K.
        • Frank C.
        • Groll D.
        • et al.
        Understanding cardiopulmonary resuscitation decision making: perspectives of seriously ill hospitalized patients and family members.
        Chest. 2006; 130: 419-428
        • Cox C.E.
        • Martinu T.
        • Sathy S.J.
        • et al.
        Expectations and outcomes of prolonged mechanical ventilation.
        Crit Care Med. 2009; 37: 2888-2894
        • Peppercorn J.M.
        • Smith T.J.
        • Helft P.R.
        • et al.
        American society of clinical oncology statement: toward individualized care for patients with advanced cancer.
        J Clin Oncol. 2011; 29: 755-760
        • Falchook A.D.
        • Dusetzina S.B.
        • Tian F.
        • et al.
        Aggressive EOL care for metastatic cancer patients younger than age 65 years.
        J Natl Cancer Inst. 2017; 109: djx028
        • Bekelman J.E.
        • Halpern S.D.
        • Blankart C.R.
        • et al.
        Comparison of site of death, health care utilization, and hospital expenditures for patients dying with cancer in 7 developed countries.
        JAMA. 2016; 315: 272-283
        • Hung Y.N.
        • Liu T.W.
        • Wen F.H.
        • Chou W.C.
        • Tang S.T.
        Escalating healthcare expenditures in cancer decedents’ last year of life: a decade of evidence from a retrospective population-based cohort study in Taiwan.
        Oncologist. 2017; 22: 460-469
        • Luengo-Fernandez R.
        • Leal J.
        • Gray A.
        • et al.
        Economic burden of cancer across the European Union: a population-based cost analysis.
        Lancet Oncol. 2013; 14: 1165-1174
        • Sullivan R.
        • Peppercorn J.
        • Sikora K.
        • et al.
        Delivering affordable cancer care in high-income countries.
        Lancet Oncol. 2011; 12: 933-980
        • Cosgriff J.A.
        • Pisani M.
        • Bradley E.H.
        • et al.
        The association between treatment preferences and trajectories of care at the end-of-life.
        J Gen Intern Med. 2007; 22: 1566-1571
        • Detering K.M.
        • Hancock A.D.
        • Reade M.C.
        • et al.
        The impact of advance care planning on end of life care in elderly patients: randomised controlled trial.
        BMJ. 2010; 340: c1345
        • Kirchhoff K.T.
        • Hammes B.J.
        • Kehl K.A.
        • et al.
        Effect of a disease-specific advance care planning intervention on EOL care.
        J Am Geriatr Soc. 2012; 60: 946-950
        • Parr J.D.
        • Zhang B.
        • Nilsson M.E.
        • et al.
        The influence of age on the likelihood of receiving EOL care consistent with patient treatment preferences.
        J Palliat Med. 2010; 13: 719-726
        • Gao X.
        • Prigerson H.G.
        • Diamond E.L.
        • et al.
        Minor cognitive impairments in cancer patients magnify the effect of caregiver preferences on EOL care.
        J Pain Symptom Manage. 2013; 45: 650-659
        • Song M.K.
        • Ward S.E.
        • Hanson L.C.
        • et al.
        Determining consistency of surrogate decisions and EOL care received with patient goals-of-care preferences.
        J Palliat Med. 2016; 19: 610-616
        • Auriemma C.L.
        • Nguyen C.A.
        • Bronheim R.
        • et al.
        Stability of EOL preferences: a systematic review of the evidence.
        JAMA Intern Med. 2014; 174: 1085-1092
        • Tang S.T.
        • Wen F.H.
        • Chang W.C.
        • et al.
        Preferences for life-sustaining treatments examined by hidden Markov modeling are mostly stable in terminally ill cancer patients’ last 6 months of life.
        J Pain Symptom Manage. 2017; 54: 628-636
        • Lee L.
        Filial duty as the moral foundation of caring for the elderly: its possibility and limitations.
        in: Fun R.P. Family-Oriented Informed Consent: East Asian and American Perspectives. Springer International Publishing, Cham, Switzerland2015: 137-147
        • Stapleton R.D.
        • Nielsen E.L.
        • Engelberg R.A.
        • et al.
        Association of depression and life-sustaining treatment preferences in patients with COPD.
        Chest. 2005; 127: 328-334
        • Chen C.H.
        • Wen F.H.
        • Hou M.M.
        • et al.
        Transitions in prognostic awareness among terminally ill cancer patients in their last 6 months of life examined by multi-state Markov modeling.
        Oncologist. 2017; 22: 1135-1142
        • Vermunt J.K.
        • Tran B.
        • Magidson J.
        Latent class models in longitudinal research.
        in: Menard S. Handbook of Longitudinal Research: Design, Measurement, and Analysis. Elsevier, Burlington, MA2008: 373-385
        • Eddy S.R.
        Profile hidden Markov models.
        Bioinformatics. 1998; 14: 755-763
        • Akaike H.
        Factor analysis and AIC.
        Psychometrika. 1987; 52: 317-332
        • Schwarz G.
        Estimating the dimension of a model.
        Ann Stat. 1978; 6: 461-464
        • Bozdogan H.
        Model selection and Akaike's information criterion (AIC). The general theory and its analytical extensions.
        Psychometrika. 1987; 52: 345-370
        • Nylund K.L.
        • Asparouhov T.
        • Muthén B.O.
        Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study.
        Struct Equ Model. 2007; 14: 535-569
        • Muthén B.
        Latent variable analysis: growth mixture modeling and related techniques for longitudinal data.
        in: Kaplan D. The Sage handbook of quantitative methodology for the social sciences. Sage, Thousand Oaks, CA2004: 345-369
        • Landis J.R.
        • Koch G.G.
        The measurement of observer agreement for categorical data.
        Biometrics. 1977; 33: 159-174
        • Blackhall L.J.
        • Murphy S.T.
        • Frank G.
        • et al.
        Ethnicity and attitudes toward patient autonomy.
        JAMA. 1995; 274: 820-825
        • Yang Y.I.
        A family-oriented Confucian approach to advance directives in EOL decision making for incompetent elderly patients.
        in: Fun R.P. Family-Oriented Informed Consent: East Asian and American Perspectives. Springer International Publishing, Basel, Switzerland2015: 257-270
        • Tang S.T.
        • Liu T.W.
        • Lai M.S.
        • et al.
        Congruence of knowledge, experiences and preferences for disclosure of diagnosis and prognosis between terminally-ill cancer patients and their family caregivers in Taiwan.
        Cancer Invest. 2006; 24: 360-366
        • Yun Y.H.
        • Kwon Y.C.
        • Lee M.K.
        • et al.
        Experiences and attitudes of patients with terminal cancer and their family caregivers toward the disclosure of terminal illness.
        J Clin Oncol. 2010; 28: 1950-1957
        • Yun Y.H.
        • Lee M.K.
        • Chang Y.J.
        • 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
        • Kim S.
        • Hahm K.H.
        • Park H.W.
        • et al.
        A Korean perspective on developing a global policy for advance directives.
        Bioethics. 2010; 24: 113-117
        • Tang S.T.
        • Wen F.H.
        • Liu L.N.
        • et al.
        A decade of changes in family caregivers' preferences for life-sustaining treatments for terminally ill cancer patients at end of life in the context of a family-oriented society.
        J Pain Symptom Manage. 2016; 51: 907-915
        • Tang S.T.
        • Liu T.W.
        • Chow J.M.
        • et al.
        Associations between accurate prognostic understanding and EOLcare preferences and its correlates among Taiwanese terminally ill cancer patients surveyed in 2011-2012.
        Psycho-Oncol. 2014; 23: 780-787
      1. Health Promotion Administration. 2015 Health Promotion Administration Annual Report. pp. 109. Health Promotion Administration, Ministry of Health and Welfare, R.O.C. Taipei, Taiwan. Accessed on July 28, 2017.

        • Kim S.Y.
        Improving medical decisions for incapacitated persons: does focusing on "accurate predictions" lead to an inaccurate picture?.
        J Med Philos. 2014; 39: 187-195