Surrogate Preferences on the Physician Orders for Life-Sustaining Treatment Form

Surrogate Preferences on the Physician Orders for Life-Sustaining Treatment Form Abstract Background and Objectives The purpose of this study is to compare treatment preferences of patients to those of surrogates on the Physician Orders for Life-Sustaining Treatment (POLST) forms. Research Design and Methods Data were collected from a sequential selection of 606 Massachusetts POLST (MOLST) forms at 3 hospitals, and corresponding electronic patient health records. Selections on the MOLST forms were categorized into All versus Limit Life-Sustaining Treatment. Multivariable mixed effects (grouped by clinician) logistic regression models estimated the impact of using a surrogate decision maker on choosing All Treatment, controlling for patient characteristics (age, severity of illness, sex, race/ethnicity), clinician (physician vs non-physician), and hospital (site). Results Surrogates signed 253 of the MOLSTs (43%). A multivariable logistic regression model taking into consideration patient, clinician, and site variables showed that surrogate decision makers were 60% less likely to choose All Treatment than patients who made their own decisions (odds ratio = 0.39 [95% confidence interval = 0.24–0.65]; p < .001). This model explained 44% of the variation in the dependent variable (Pseudo-R2 = 0.442; p < .001); mixed effects logistic regression grouped by clinician showed no difference between the models (LR test = 4.0e-13; p = 1.00). Discussion and Implications Our study took into consideration variation at the patient, clinician, and site level, and showed that surrogates had a propensity to limit life-sustaining treatment. Surrogate decision makers are frequently needed for hospitalized patients, and nearly all states have adopted the POLST. Researchers may want study decision-making processes for patients versus surrogates when the POLST paradigm is employed. End-of-life care, Palliative care, Advance care planning The Physician Orders for Life-Sustaining Treatment (POLST) paradigm offers a structured approach for physicians, nurse practitioners, and physician assistants to discuss and document patient preferences for life-sustaining treatments. The POLST is not a substitute for advance directives, which are meant to name a surrogate decision maker and describe care preferences for future, unknown medical situations (National POLST Paradigm, 2017b). The Institute of Medicine recommends the POLST when disease advances and when patients are facing their final year of life (Institute of Medicine, 2014). If a patient needs a surrogate decision maker, such as in situations when a patient is deemed by a physician to be temporarily or permanently “not competent” to make medical decisions (Drane, 1984), the POLST reflects preferences resulting from discussions between clinicians and patients’ surrogates. The POLST Form is a medical order and is transferrable across care settings; this is particularly helpful given that individuals in the last 6 months of life are often under the care of more than 10 different physicians across multiple settings (Dartmouth Institute for Health Policy and Clinical Practice, 2014). Program efficacy studies indicate that the POLST improved documentation in medical records about end-of-life care, and that terminal interventions were largely consistent with documented preferences (Hickman et al., 2009; Hickman et al., 2011; Richardson, Fromme, Zive, Fu, & Newgard, 2014; Schmidt, Hickman, Tolle, & Brooks, 2004; Tolle, Tilden, Nelson, & Dunn, 1998). The POLST paradigm has been, or is in the process of being implemented in 48 states (National POLST Paradigm, 2017a), and its use will likely grow in the coming years along with the numbers of individuals aging with chronic illness. Previous research using POLST data showed that a majority of completed POLST records indicate a preference to limit life-sustaining treatments (Fritz & Barclay, 2014; Fromme, Zive, Schmidt, Cook, & Tolle, 2014; Fromme, Zive, Schmidt, Olszewski, & Tolle, 2012; Hammes, Rooney, Gundrum, Hickman, & Hager, 2012; Hickman, Keevern, & Hammes, 2014; Kim, Ersek, Bradway, & Hickman, 2015; Schmidt, Zive, Fromme, Cook, & Tolle, 2014; Tarzian & Cheevers, 2017; Tolle et al., 1998) even though the POLST allows patients or surrogates to indicate a preference to apply all treatment. One reason for the large proportions preferring to limit treatment may be due to clinicians’ interpretation of the need for a POLST form; that is, administering full treatment to sustain life is the default situation, and POLST forms are thus only needed if individuals prefer treatment limitations. Another reason may be attributed to study populations, largely comprised of non-Hispanic White individuals, who are more likely to prefer treatment limitations than those from minority populations (Rahman, Bressette, Gassoumis, & Enguidanos, 2016). Studies have not examined treatment preferences when surrogates or health care proxies make decisions documented on POLST forms. Almost one-half of hospitalized older adults use a surrogate decision maker; and three out of four physicians in a hospital sample reported that they have made a major decision with a surrogate in the past 30 days (Torke et al., 2014; Torke, Siegler, Abalos, Moloney, & Alexander, 2009). Guidance from the American Medical Association (2016) recommends that physicians guide surrogates to employ the ethical principle of “substituted judgment,” that is, to choose the treatment path the patient would have chosen if he or she were able. However, researchers have questioned the accuracy of substituted judgment for more than 25 years, and have shown in experimental vignettes that surrogates are largely unable to predict patient decisions about life-sustaining treatments (Barrio-Cantalejo et al., 2009; Hinderer, Friedmann, & Fins, 2015; Seckler, Meier, Mulvihill, & Cammer Paris, 1991; Shalowitz, Garrett-Mayer, & Wendler, 2006; Suhl, Simons, Reedy, & Garrick, 1994; Torke, Alexander, & Lantos, 2008). When there is no evidence of what a patient would have wanted, clinicians may apply the “best interest” standard, which is based on the “the pain and suffering associated with the intervention,” “the degree of and potential for benefit,” and “impairments that may result from the intervention” (American Medical Association, 2016). There is evidence of strong correlation between proxy and self-reports for observed behaviors, such as walking limitations after a stroke (Powell, Johnston, & Johnston, 2007) and quality of life measures associated with physical functions, such as breathing, hearing, sleeping (Elliott, Lazarus, & Leeder, 2006). Socio-demographic characteristics and the type of chronic conditions can confound results, but controlling for these confounders has confirmed that proxy reports are accurate for observable behaviors, such as the ability to walk a certain distance, but potentially inaccurate for subjective behaviors, such as the ability to manage money or unobserved behaviors, such as difficulty in using the toilet (Li, Harris, & Lu, 2015). Specific to patients nearing the end of life, family members can be reliable proxy reporters of objective quality of life measures, such as walking, nausea, and vomiting, but with mixed results for subjective measures, such as fatigue or pain. Further, when there is a discrepancy, proxies tend to report worse subjective quality of life than patients (Ferri & Pruchno, 2009; Kirou-Mauro, Harris, Sinclair, Selby, & Chow, 2006; Kutner, Bryant, Beaty, & Fairclough, 2006; Novella et al., 2001). In a sample of patients in an inpatient palliative care unit, proxy family or physician reports of patient symptoms were poorly correlated with patient reports during early hospitalization (Day 3), improved over time (Day 6), but were still low (<35% agreement) on measures of physical symptoms and psychological well-being (Jones et al., 2011). Family caregivers tended to overestimate the intensity or frequency of lack of energy, anxiety, sadness, and pain distress compared with patients. On the other hand, physicians generally underestimated symptom burden compared with patients (Oechsle, 2013). Concept Model This study is designed to estimate the odds that surrogate decision makers’ would choose aggressive life-sustaining treatment, taking into account patient, clinician, and upstream factors that might have channeled patients to sites of care. We incorporated patient variables to take into account factors that may influence whether a patient needed a surrogate, such as age and severity of illness, which may also affect preferences (Figure 1). On the other hand, clinician type, patient race/ethnicity, and sex were thought to affect only preferences. We acknowledge that prior decisions might have influenced a patient’s site of care, such as insurance limitations, patient or clinician preferences, and geography. These factors, in addition to site-associated practice norms and prior advance care plans, could affect patient or surrogate preferences for life-sustaining treatments in ways that were not measured. Figure 1. View largeDownload slide Concept model—factors affecting life-sustaining treatment preferences for hospitalized adults with limited life expectancy. Figure 1. View largeDownload slide Concept model—factors affecting life-sustaining treatment preferences for hospitalized adults with limited life expectancy. Methods The Institutional Review Board at Partners HealthCare approved this study, which uses 606 patient records from three of its hospitals. Research Electronic Data Capture (REDCap), an electronic data capture tool hosted by the hospital system’s research management office, was used to collect and manage study data (Harris, Taylor, Payne, Gonzalez, & Conde, 2009). De-identified data were exported for analysis using STATA 13.1 (StataCorp. 2013, 2013). Setting Characteristics The three hospital sites were part of the same integrated care network, and were among the first hospitals to implement Massachusetts POLST (MOLST) in Eastern Massachusetts. As the first hospitals to implement MOLST, our data represent de novo exposure to MOLST, and nearly eliminates the possibility that patients had arrived with a form completed at another hospital or in an outpatient setting. Each site employed a slightly different staffing model, and differed as to how broadly the MOLST was deployed. Site 1, a long-term acute care hospital (LTACH), was staffed largely by employed physicians and augmented by a few nurse practitioners; none were trained in palliative care. MOLST was implemented throughout the entire 180-bed hospital, where all patients were admitted directly from an acute care hospital stay. LTACHs serve patients with severe illness, who need hospital-level care for a long stay, which may last one or more months (American Hospital Association, 2017; Center for Medicare and Medicaid Services, October 2016). Thirty-one different clinicians signed MOLSTs at Site 1. Site 2 is a community hospital, where MOLSTs were largely administered by a consulting palliative care practice comprised of five palliative care physicians and nurse practitioners, who signed nearly all (96%) of the forms; another five clinicians outside of this palliative care practice signed the remaining 4% of the forms at this location. Site 3 is an inpatient intensive palliative care unit with 12 beds in a quaternary care academic medical center with more than 800 beds. This inpatient palliative care unit was staffed by employed physician assistants under the direction of a small number of palliative care physicians. Eleven different clinicians completed MOLSTs at Site 3. Patient selection also differed from site to site, but all patients had advanced illness and were nearing the end of life. At Site 1 (LTACH), clinicians used the “surprise” question, which has been validated in a number of settings as a simple prognostic tool for patients with limited life expectancy (Da Silva Gane, Braun, Stott, Wellsted, & Farrington, 2013; Javier et al., 2017; Moss et al., 2008; Vick, Pertsch, Hutchings, Neville, & Bernacki, 2016). That is, clinicians asked themselves “Would I be surprised if this patient died in the next 12 months?” A “no” triggered a “goals of care” conversation, followed by documentation on a MOLST form. Monthly tracking data over the first year of MOLST implementation at Site 1 showed that 15%–40% of discharged patients completed MOLSTs. The palliative care consulting practice (Site 2) did not use the “surprise” question for patients, but nearly all patients would have qualified had it been employed. Site 3, the inpatient palliative care unit, did not always use the “surprise” question because nearly all patients were expected to have less than 1 year of remaining life. Clinicians completed MOLST forms only for patients who were discharged to another care setting or to home, which accounted for approximately 50% of patients at this site. Despite differing patient selection criteria, patients at all sites with a completed MOLST were expected to have limited life expectancy. Sample The sample comprised a sequential selection of patients who had completed the Massachusetts Medical Orders for Life-Sustaining Treatment (MOLST) forms while inpatient at three hospitals between July 9, 2012 and January 17, 2014. The final analytic data set contained 593 MOLST forms after discarding 2% (N = 13 from 606) with missing clinician signatures, making the order invalid. Two-hundred and eighty-eight (N = 288; 49%) of the analytic data set came from the LTACH (Site 1); 204 (34%) from the palliative care consulting practice at a community hospital (Site 2); and 101 (17%) from the inpatient intensive palliative care unit at the academic medical center (Site 3). The sample comprised patients in both intensive care (ICU) and non-intensive care units, and residing in both long-term care settings and in the community. A small number of patients completed multiple MOLST forms during 1 hospital stay, or completed another MOLST form during a second or third hospitalization over the course of the study period. In these circumstances, data were collected only from the first MOLST, reserving the issue of changes to MOLST forms for another study. Dependent Variable The dependent variable (All vs Limit Treatment) was constructed from responses on Page 1 of the MOLST form, which provided a sequence of dichotomous “do not attempt” or “attempt” choices beginning with resuscitation, followed by ventilation, then transfer to a hospital (Massachusetts Medical Orders for Life Sustaining Treatment, 2014). Page 2 of the form seeks preferences about the duration of treatments on Page 1, and offers an opportunity to specify decisions regarding other life-sustaining treatments, such as kidney dialysis and artificial nutrition or hydration. Page 2 is not required in order for the MOLST to be valid, and too few were completed for meaningful analysis. The pattern of responses generally followed logically from the first choice, resuscitation (Figure 2). Ventilation options were divided into two categories: invasive (intubation) and noninvasive, such as continuous positive airway pressure (CPAP); information about CPAP was not used due to the large number (N = 119; 20%) of forms missing these data. Patients were grouped into those who preferred All Treatment (“Yes” to resuscitation, intubation, AND transfer) or Limit Treatment (“No” to resuscitation, intubation, OR transfer). A separate analysis confirmed that there was no significant difference between the group missing CPAP data and the group not missing CPAP data for the primary variable of interest (surrogate vs patient decisions; Pearson χ2 = 0.21; p = .644; Supplementary Table 1). Figure 2. View largeDownload slide Pattern of responses on MOLST. Figure 2. View largeDownload slide Pattern of responses on MOLST. Main Effect Data for the main variable of interest (patient or surrogate signature) were collected from the MOLST form, which requests that signatories identify their relationship to the patient. If the signatory did not respond, we compared the patient name with the signature and associated printed name. Forms with different signatory and patient names were categorized into the surrogate group. All printed names were legible. Control Variables Patient Demographics Demographic data were obtained from patient electronic health records, which included age at the time the MOLST was signed and patient race/ethnicity. Logistic regression models employed a categorical age variable approximating the bottom quartile (<60), interquartile range (60–73), and top quartile (80 and over) at Site 1, where there was the most variation between preferences for All versus Limit Treatment. Severity of Illness We used ICD-9 codes associated with the hospitalization during which the MOLST form was signed to compute the Charlson Score as an indicator of illness severity (Al Feghali et al., 2016; Chang et al., 2016; Tremblay, Arnsten, & Southern, 2016). The Charlson Weighted Index of Comorbidity was developed as a prognostic indicator for individuals with chronic illnesses. Each chronic illness on the Index has a score associated with mortality risk (1, 2, 3, or 6; Charlson, Pompei, Ales, & MacKenzie, 1987). For example, uncomplicated diabetes is 1, diabetes with end organ damage is 2, moderate or severe liver disease is 3, and metastatic solid tumor is 6. Scores for comorbid conditions are summed. In the original Charlson validation cohort, 1-year mortality was 100% for patients who were hospitalized, survived to discharge, and had a score greater than 5 (Charlson et al., 1987). Survival has improved since the original Charlson validation cohort for many illnesses captured by its Index, but the tool remains useful in research as a measure for burden of illness (Crooks, West, & Card, 2015; D’Hoore, Sicotte, & Tilquin, 1993). Some studies have incorporated a decade-based age adjustment specified by Charlson to reflect additional mortality risk associated with age beginning at age 50 (Charlson et al., 1987; Dias-Santos, Ferrone, Zheng, Lillemoe, & Fernandez-del Castillo, 2015; Kaesmann, Janssen, Schild, & Rades, 2016; Lorenzon et al., 2017; St-Louis et al., 2015). Conceptually, this adjustment was meant to reflect that the same illness in a 50-year-old and an 80-year-old represent different risks for 1-year mortality. We did not include this adjustment, which is based on age decade, to allow for greater precision using our age variable. Clinician Type Clinician certification (physician vs nurse practitioner or physician assistant) was collected from signatures on the MOLST forms. If certification data were missing, then a research assistant matched the signature against internal staffing records. Site The variables Site 1, Site 2, and Site 3 identified each hospital, and served as a means to hold constant site-associated practice variation, such as differing staffing models and the use of palliative care specialists. The Site variable also took into account unmeasured factors that may have channeled patients to each site, such as insurance coverage and patient or clinician preferences farther upstream in the care continuum. For example, patients at the LTACH may not have been offered palliative care, or were offered the service and refused; as a result, these patients were transferred to the LTACH for high intensity treatment. Patients under the care of clinicians in the palliative care practices were presented with an option for palliative care and accepted, which may be associated with a willingness to accept limitations to life-sustaining treatment. Statistical Analysis We examined sample characteristics by site and by decision maker using the Pearson χ2 test. Single-variable logistic regression analyses were first performed to estimate the odds that each independent variable would predict a preference for All (vs Limit) Treatment. Multivariable logistic regression models estimated the odds for All Treatment first for Surrogate (vs Patient) Decision Makers, followed by patient characteristics, then concluding with the impact of Site. Mixed effects logistic regression was performed using STATA’s xtmelogit command to account for covariance due to clustering of patients within clinicians. Results Baseline Patient Characteristics Nearly all patient characteristics differed between sites (Table 1). Surrogates signed approximately 35% of the MOLST forms at Sites 1 (LTACH without palliative care) and 3 (palliative care at academic medical center) compared with 60% at Site 2 (palliative care at community hospital), which also had the oldest patients. Only one in four surrogates chose All Treatment compared with nearly one in two patients doing the same. Individuals at the academic medical center had higher burden of illness than the other two sites as measured by the Charlson Score (median for Sites 1 and 2 = 5; Site 3 = 7). Patient primary diagnoses were not different between surrogate and patient-signed forms with the exception of cancers, where a larger proportion of patient signers had cancers (26%) compared with patients who had surrogates (11%; p < .001). Table 1. Sample Characteristics by Site and by Treatment Preference Group N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  Notes: MOLST, Massachusetts Physician Orders for Life-Sustaining Treatment. aSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed academic medical center. bMassachusetts Medical Orders for Life-Sustaining Treatment Form. cScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. p-Value using the two-sample test of proportion comparing Surrogate and Patient groups: *≤0.05; **≤0.01; ***≤0.001. View Large Table 1. Sample Characteristics by Site and by Treatment Preference Group N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  Notes: MOLST, Massachusetts Physician Orders for Life-Sustaining Treatment. aSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed academic medical center. bMassachusetts Medical Orders for Life-Sustaining Treatment Form. cScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. p-Value using the two-sample test of proportion comparing Surrogate and Patient groups: *≤0.05; **≤0.01; ***≤0.001. View Large Multivariable Logistic Regression Surrogate decision makers were much less likely to choose All Treatment than patients who made their own decisions (OR = 0.34–0.46; p ≤ .001). This effect remained unchanged as we added patient, clinician, and site characteristics (Table 2; Models 1–5). Patients in the highest age group (≥80) were consistently less likely to choose All Treatment than those in the lowest age group (<60; OR = 0.14–0.22; p ≤ .001; Models 2–5). Individuals with high Charlson Scores (>5) were less likely to choose All Treatment than those with lower Charlson Scores (<5; OR = 0.26; p ≤ .001; Model 3 and 4), but site-associated effects reduced the magnitude of this effect (OR = 0.59; p ≤ .05; Model 5). Overall, patients and surrogates at the palliative care sites had >90% lower odds for choosing All Treatment than those at the LTACH (Site 2: odds ratio, OR = 0.02 [95% confidence interval, CI, 0.01–0.05]; Site 3: OR = 0.06 [95% CI 0.02–0.10]; p ≤ .001). Table 2. Odds Ratios (95% Confidence Interval) for Logistic Regression Models Estimating the Odds for All Treatment versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  Notes: aScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. bSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed at an academic medical center. cPseudo-R2 is an indicator of model fit and can be interpreted as the amount of variation in All versus Limit Treatment attributed to each variable. +p-Value of odds ratio ≤.10; *≤.05; **≤.01; ***≤.001. ++++p-Value of the model ≤.001. View Large Table 2. Odds Ratios (95% Confidence Interval) for Logistic Regression Models Estimating the Odds for All Treatment versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  Notes: aScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. bSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed at an academic medical center. cPseudo-R2 is an indicator of model fit and can be interpreted as the amount of variation in All versus Limit Treatment attributed to each variable. +p-Value of odds ratio ≤.10; *≤.05; **≤.01; ***≤.001. ++++p-Value of the model ≤.001. View Large A multivariable mixed effects logistic regression model, grouped by clinicians, using all observations and the full complement of independent variables was no different from a model that did not cluster patients by clinician (Standard Deviation of the Intercept = 2.58e-10 [95% CI 0.00–0.00]; Likelihood Ratio Test = 4.0e-13; p = 1.00). Post Hoc Logistic Regression Using Only Data from the LTACH We tested our model post hoc using only data from the LTACH (Site 1) to eliminate the potential for confounding due to clinicians channeling patients into palliative care settings due to prior patient or surrogate preferences to limit life-sustaining treatments (Table 3). Using only LTACH data, surrogate decision makers experienced 52% lower odds for choosing All Treatment compared with patient decision makers (OR = 0.48 [95% CI 0.28–0.84]; p = .01). These results were substantially the same as those using data pooled for all three sites. Mixed effects multivariable logistic regression, grouped by 31 clinicians again showed no difference between this model and a comparable model that did not cluster patients within clinicians (Standard Deviation of Intercept: 1.58e-09 [95% CI 0.00–0.00]; p = 1.00). Table 3. Post Hoc Analysis – Odds Ratios and (95% Confidence Interval) for a Multivariable Logistic Regression Model Estimating the Odds for All Treatmenta versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form Using Only Site 1b Data N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  Notes: aAll Treatment = “Yes” to All (resuscitate, intubate, transfer) versus Limit Treatment = “No” to At Least One (resuscitate, intubate, transfer). bSite 1 is a 180-bed long-term acute care hospital with no palliative care specialist. cScore derived from the Charlson weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. View Large Table 3. Post Hoc Analysis – Odds Ratios and (95% Confidence Interval) for a Multivariable Logistic Regression Model Estimating the Odds for All Treatmenta versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form Using Only Site 1b Data N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  Notes: aAll Treatment = “Yes” to All (resuscitate, intubate, transfer) versus Limit Treatment = “No” to At Least One (resuscitate, intubate, transfer). bSite 1 is a 180-bed long-term acute care hospital with no palliative care specialist. cScore derived from the Charlson weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. View Large Discussion The goal of this study was to estimate the odds that surrogate decision makers would choose aggressive life-sustaining treatment decisions by taking into consideration patient, clinician, and site variables, which accounted for unmeasured factors that channeled patients to each location, such as patient insurance and patient or clinician preferences farther upstream in the care continuum. We collected data from the Massachusetts version of the POLST form and corresponding patient medical records at three hospitals in the greater Boston area. An important strength of this study is that data represent decisions made in a health care setting in the context of patients and families experiencing an event associated with serious, life-limiting illness. Surrogates signed 43% of the MOLST forms in our study, which is consistent with prior results (47%; Torke et al., 2014). Multivariable logistic regression results showed that surrogates were 60% less likely to choose All Treatment than patients who did not use a surrogate. Because decisions in palliative care settings may have confounded the results, we performed a post hoc analysis using only data from the LTACH, which did not have a palliative care practice. This post hoc analysis revealed essentially the same result, that surrogate decision makers were 52% less likely to choose All Treatment than patients not using a surrogate. Our results are consistent with findings using POLST data from a nursing home population (Rahman et al., 2016), which also used data gathered from a clinical setting. Other studies that examined both patient and surrogate decisions are not directly comparable because study methods were intended to test the accuracy of surrogate decisions, and used hypothetical scenarios (Barrio-Cantalejo et al., 2009; Shalowitz et al., 2006), which may not reflect contextual cues that could influence patient or surrogate decisions when life-sustaining treatment preferences are sought in clinical settings (Reamy, Kim, Zarit, & Whitlatch, 2011). Surrogate decision makers face intense emotional distress as they balance forces favoring more intervention against those favoring less. Prior research suggest that negative cues are most salient when surrogates need to make decisions, such as a patient’s unconscious state or perceived pain, and a focus on what patients do not want instead of what they do want (Dionne-Odom, 2015). The process of discussing end-of-life treatments with surrogates is a delicate balance between prior patient preferences and the best interests of the patient going forward (Torke, Moloney, Siegler, Abalos, & Alexander, 2010). We should not assume that our results reflect prior discussions between surrogate and patient pairs, as these pairs have shown low agreement (62%) when asked whether discussions about life-sustaining treatments had taken place (Fried, Redding, Robbins, O’Leary, & Iannone, 2011), and prior discussions also showed no impact on the accuracy of substituted judgment (Pruchno, Lemay Jr., Field, & Levinsky, 2005). Rather, our results seem to indicate that surrogate decision makers are systematically different from patient decision makers, which would be consistent with prior research showing that proxy respondents represented individuals in worse physical health, older, and were less educated than those not using a proxy (Elliott, Beckett, Chong, Hambarsoomians, & Hays, 2008). Prior research about the accuracy of proxy reporting suggests that surrogates tend to overestimate the burden of illness, especially on subjective quality of life measures (Ferri & Pruchno, 2009; Kirou-Mauro et al., 2006; Kutner et al., 2006; Novella et al., 2001). This tendency can be problematic whether clinicians guide surrogates to employ the principle of “substituted judgment” or “best interest” when making life-sustaining treatment decisions. Clinicians may want to take into consideration that in some situations, surrogates may overestimate burden, such as when life-sustaining treatment decisions are based on unobservable or subjective quality of life outcomes for the patient. Because the POLST is recommended for patients facing their last year of life, clinicians may want to attend to completing a POLST when patients are still able to make their own decisions and a surrogate is not yet needed. Limitations Our study used a sample that completed MOLSTs at three hospitals; thus, selection bias cannot be ignored. In addition to upstream forces that channeled individuals to these hospitals, clinicians at each location selected patients for MOLST, and patients or surrogates agreed to complete a MOLST form. We acknowledge that our data failed to capture patients and surrogates who were not asked to complete a MOLST form. Some clinicians may have been uncomfortable discussing end-of-life care goals with patients or surrogates, whereas others may have thought their patients would survive beyond 1 year, and still others may have decided not to complete MOLST forms because patients and surrogates wanted All Treatment. These biases in the data set would likely result in odds ratios being too low for surrogates. By the same token, patients and surrogates who chose not to complete a MOLST form were choosing All Treatment by default, and because we did not track refusals, our estimates regarding a preference to Limit Treatment may be high. Furthermore, we lacked background data about patients, which would have predisposed them to treatment versus treatment limitations, such as religiosity or prior declarations in advance care directive. While we have no evidence pointing to underlying systematic selection bias in our sample, we also have no evidence to support a lack of systematic selection bias. It is reasonable to assume that clinicians and patients in the palliative care settings were already biased against choosing All Treatment. Patients or surrogates agreed to treatment in these settings, and the definition of palliative care is to treat symptoms, not disease. The post hoc analysis using data from the LTACH, which did not have a palliative care practice, allowed us to eliminate the potential for this bias by examining results at one location where there was substantial variation in the dependent variable; that is, 70% chose All Treatment. The result for the main effect using single-site data was substantially the same as that using the full data set, which suggests that whatever forces led patients to the palliative care practices as opposed to the LTACH did not change our conclusion. If anything, one would expect that clinicians in an LTACH might approach goals of care discussions in a manner that could bias patient or surrogate decisions toward aggressive treatment. Again, our single-site results do not support this direction of thinking; rather, they confirm that even in an aggressive treatment pathway, surrogate decision makers are less likely to choose all life-sustaining treatments than patients. Using only one scale, the Charlson Score, to measure severity of illness is also an important limitation in this study. Our regression models showed that the surrogate variable acted independently from the Charlson Score, which suggests that the Charlson Index did not fully measure conditions that would lead to needing a surrogate decision maker, such as delirium, high fever, or nausea that often lead to hospitalization for individuals with advanced illness. Further, dementia is assigned only a score of 1 on the Charlson Index, with no further adjustment for higher severity, even though the Index adjusts for higher severity of cancers, diabetes, and heart disease. This limitation in the Charlson Score, coupled with other research showing insufficient reporting or documentation of underlying dementia in older populations (Crowther, Bennett, & Holmes, 2017; van den Dungen et al., 2012; Ostbye, Taylor, Clipp, Van Scoyoc, & Plassman, 2008), and a propensity for surrogates to favor limitations when dementia is present (Feltz & Samayoa, 2012), indicate that future studies should incorporate multiple measures to reflect conditions faced in the last days or weeks of life. Finally, our mixed effects models grouped by clinicians did not show variation between clinicians even though prior research point to an association between patient preferences and the clinicians’ end-of-life care preferences for themselves (Wilkinson & Truog, 2013). A larger sample may have yielded different results. A review of the numbers of MOLST forms completed by each of the 31 clinicians at Site 1 showed that close to 50% completed just one or two forms while 30% completed 11–36 forms. A larger sample, collected over a longer period of time, would have provided more MOLST forms for the large proportion of clinicians who had only completed one or two forms, and may have yielded different results. Conclusion The results in this study show that surrogate decision makers for a hospitalized population with serious and advancing illness were more likely to indicate a preference to limit life-sustaining treatments than patient decision makers. In multivariable logistic regression models, this effect operated independently from age, practice patterns associated with site, and the Charlson Score. Prior research demonstrating that proxies tend to overestimate symptom burden compared with patients presents a troubling ethical problem in our results, which is whether decisions to limit treatment by surrogates can be construed as an accurate reflection of what the patient would have wanted. Given the frequency with which surrogates are needed for older, hospitalized patients, provider organizations may want to incorporate new strategies to reach surrogates as an additional patient decision partner. Providers may also want to consider discussing with patients and documenting their preferences for temporary versus permanent incapacity. Finally, when communicating with surrogate decision makers, providers may want to keep in mind a surrogate’s tendency to overestimate symptom burden.. Supplementary Data Supplementary data are available at The Gerontologist online. Acknowledgments Elizabeth E. Chen had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. A portion of the results in this article was presented at the Gerontological Society of America Research Meeting in New Orleans, November 2016, at the International Association of Gerontology and Geriatrics World Congress in San Francisco, July 2017, and via webinar hosted by the National POLST on January 25, 2018. We gratefully acknowledge the contributions of Mary Fairbanks, BS, who generously volunteered hundreds of hours during the data collection phase of this study; Edward Alan Miller, PhD, MPA, who provided important critique for earlier versions of this research; and Susan T. Moore, RN, MPH for facilitating the data collection process and providing important comments to the final draft. References Al Feghali, K. A., Robbins, J. R., Mahan, M., Burmeister, C., Khan, N. T., Rasool, N.,… Elshaikh, M. A. ( 2016). Predictive capacity of 3 comorbidity indices in estimating survival endpoints in women with early-stage endometrial carcinoma. International Journal of Gynecological Cancer: Official Journal of The International Gynecological Cancer Society , 26, 1455– 1460. doi: 10.1016/j.ygyno.2016.04.205. Google Scholar CrossRef Search ADS   American Hospital Association. ( 2017). Long-term care hospitals (LTCHs). Retrieved October 22, 2017, from http://www.aha.org/advocacy-issues/postacute/ltach/index.shtml American Medical Association. ( 2016). Code of Medical Ethics Opinions on Consent, Communications and Decision Making Section 2.2.1. Retrieved March 3, 2018, from https://www.ama-assn.org/sites/default/files/media-browser/code-of-medical-ethics-chapter-2.pdf Barrio-Cantalejo, I. M., Molina-Ruiz, A., Simón-Lorda, P., Cámara-Medina, C., Toral López, I., del Mar Rodríguez del Aguila, M., & Bailón-Gómez, R. M. ( 2009). Advance directives and proxies’ predictions about patients’ treatment preferences. Nursing Ethics , 16, 93– 109. doi: 10.1177/0969733008097995 Google Scholar CrossRef Search ADS   Center for Medicare and Medicaid Services. ( October 2016). Long term care hospital prospective payment system. Retrieved October 22, 2017, from https://www.cms.gov/Outreach-and-Education/Medicare-Learning-Network-MLN/MLNProducts/Downloads/Long-Term-Care-Hospital-PPS-Fact-Sheet-ICN006956.pdf Chang, C., Yin, W., Wei, C., Wu, C., Su, Y., Yu, C., Lee, C.C., et al.   ( 2016). Adjusted age-adjusted Charlson Comorbidity Index score as a risk measure of perioperative mortality before cancer surgery. PLoS One , 11, e0148076. doi: 10.1371/journal.pone.0148076 Google Scholar CrossRef Search ADS   Charlson, M. E., Pompei, P., Ales, K. L., & MacKenzie, C. R. ( 1987). A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. Journal of Chronic Diseases , 40, 373– 383. doi: 10.1016/0021-9681(87)90171-8 Google Scholar CrossRef Search ADS   Crooks, C. J., West, J., & Card, T. R. ( 2015). A comparison of the recording of comorbidity in primary and secondary care by using the Charlson Index to predict short-term and long-term survival in a routine linked data cohort. BMJ Open , 5, e007974. doi: 10.1136/bmjopen-2015-007974 Google Scholar CrossRef Search ADS   Crowther, G. J., Bennett, M. I., & Holmes, J. D. ( 2017). How well are the diagnosis and symptoms of dementia recorded in older patients admitted to hospital? Age and Ageing , 46, 112– 118. doi: 10.1093/ageing/afw169 Da Silva Gane, M., Braun, A., Stott, D., Wellsted, D., & Farrington, K. ( 2013). How robust is the ‘surprise question’ in predicting short-term mortality risk in haemodialysis patients? Nephron. Clinical Practice , 123, 185– 193. doi: 10.1159/000353735 Google Scholar CrossRef Search ADS   Dartmouth Institute for Health Policy and Clinical Practice. ( 2014). Care of chronic illness in last two years of life: Number of different physicians seen per decedent in the last 6 months of life. Retrieved October 22, 2017, from http://www.dartmouthatlas.org/data/table.aspx?ind=12&loct=2&loc=2%2c3%2c4%2c5%2c6%2c7%2c8%2c9%2c10%2c11%2c12%2c13%2c14%2c15%2c16%2c17%2c18%2c19%2c20%2c21%2c22%2c23%2c24%2c25%2c26%2c27%2c28%2c29%2c30%2c31%2c32%2c33%2c34%2c35%2c36%2c37%2c38%2c39%2(TRUNCATED) D’Hoore, W., Sicotte, C., & Tilquin, C. ( 1993). Risk adjustment in outcome assessment: The Charlson comorbidity index. Methods of Information in Medicine , 32, 382– 387. doi: 10.1055/s-0038-1634956 Google Scholar CrossRef Search ADS   Dias-Santos, D., Ferrone, C. R., Zheng, H., Lillemoe, K. D., & Fernández-Del Castillo, C. ( 2015). The Charlson age comorbidity index predicts early mortality after surgery for pancreatic cancer. Surgery , 157, 881– 887. doi: 10.1016/j.surg.2014. 12.006 Google Scholar CrossRef Search ADS   Dionne-Odom, J. N., Willis, D. G., Bakitas, M., Crandall, B., & Grace, P. J. ( 2015). Conceptualizing surrogate decision making at end of life in the intensive care unit using cognitive task analysis. Nursing Outlook , 63, 331– 340. doi: 10.1016/j.outlook.2014.10.004 Google Scholar CrossRef Search ADS   Drane, J. F. ( 1984). Competency to give an informed consent. A model for making clinical assessments. JAMA , 252, 925– 927. doi: 10.1001/jama.1984.03350070043021 Google Scholar CrossRef Search ADS   van den Dungen, P., van Marwijk, H. W., van der Horst, H. E., Moll van Charante, E. P., Macneil Vroomen, J., van de Ven, P. M., & van Hout, H. P. ( 2012). The accuracy of family physicians’ dementia diagnoses at different stages of dementia: A systematic review. International Journal of Geriatric Psychiatry , 27, 342– 354. doi: 10.1002/gps.2726 Elliott, D., Lazarus, R., Leeder, S. R. ( 2006). Proxy respondents reliably assessed the quality of life of elective cardiac surgery patients. Journal of Clinical Epidemiology , 59, 153– 159. doi: 10.1097/DCC.0000000000000097 Google Scholar CrossRef Search ADS   Elliott, M. N., Beckett, M. K., Chong, K., Hambarsoomians, K., & Hays, R. D. ( 2008). How do proxy responses and proxy-assisted responses differ from what Medicare beneficiaries might have reported about their health care? Health Services Research , 43, 833– 848. doi: 10.1111/j.1475-6773.2007.00820.x Google Scholar CrossRef Search ADS   Feltz, A., & Samayoa, S. ( 2012). Heuristics and life-sustaining treatments. Journal of Bioethical Inquiry , 9, 443– 455. doi: 10.1007/s11673-012-9396-5 Google Scholar CrossRef Search ADS   Ferri, C. V., & Pruchno, R. A. ( 2009). Quality of life in end-stage renal disease patients: Differences in patient and spouse perceptions. Aging & Mental Health , 13, 706– 714. doi: 10.1080/13607860902845558 Google Scholar CrossRef Search ADS   Fried, T. R., Redding, C. A., Robbins, M. L., O’Leary, J. R., & Iannone, L. ( 2011). Agreement between older persons and their surrogate decision-makers regarding participation in advance care planning. Journal of the American Geriatrics Society , 59, 1105– 1109. doi: 10.1111/j.1532-5415.2011.03412.x Google Scholar CrossRef Search ADS   Fritz, Z. B., & Barclay, S. I. ( 2014). Patients’ resuscitation preferences in context: Lessons from POLST. Resuscitation , 85, 444– 445. doi: 10.1016/j.resuscitation.2014.01.016 Google Scholar CrossRef Search ADS   Fromme, E. K., Zive, D., Schmidt, T. A., Cook, J. N., & Tolle, S. W. ( 2014). Association between physician orders for life-sustaining treatment for scope of treatment and in-hospital death in Oregon. Journal of the American Geriatrics Society , 62, 1246– 1251. doi: 10.1111/jgs.12889 Google Scholar CrossRef Search ADS   Fromme, E. K., Zive, D., Schmidt, T. A., Olszewski, E., & Tolle, S. W. ( 2012). POLST Registry do-not-resuscitate orders and other patient treatment preferences. JAMA , 307, 34– 35. doi: 10.1001/jama.2011.1956 Google Scholar CrossRef Search ADS   Hammes, B. J., Rooney, B. L., Gundrum, J. D., Hickman, S. E., & Hager, N. ( 2012). The POLST program: A retrospective review of the demographics of use and outcomes in one community where advance directives are prevalent. Journal of Palliative Medicine , 15, 77– 85. doi: 10.1089/jpm.2011.0178 Google Scholar CrossRef Search ADS   Harris, P. A., Taylor, R., Thielke, R., Payne, J., Gonzalez, N., & Conde, J. G. ( 2009). Research electronic data capture (REDCap)–a metadata-driven methodology and workflow process for providing translational research informatics support. Journal of Biomedical Informatics , 42, 377– 381. doi: 10.1016/j.jbi.2008.08.010 Google Scholar CrossRef Search ADS   Hickman, S., Keevern, E., & Hammes, B. J. ( 2014). The effect of the POLST (Physician Orders for Life-Sustaining Treatment) program on clinical care: A systematic review. Journal of Palliative Care , 30, 226– 227. doi: 10.1089/jpm.2008.0196 Hickman, S. E., Nelson, C. A., Moss, A. H., Hammes, B. J., Terwilliger, A., Jackson, A., & Tolle, S. W. ( 2009). Use of the Physician Orders for Life-Sustaining Treatment (POLST) paradigm program in the hospice setting. Journal of Palliative Medicine , 12, 133– 141. doi: 10.1089/jpm.2008.0196 Google Scholar CrossRef Search ADS   Hickman, S. E., Nelson, C. A., Moss, A. H., Tolle, S. W., Perrin, N. A., & Hammes, B. J. ( 2011). The consistency between treatments provided to nursing facility residents and orders on the physician orders for life-sustaining treatment form. Journal of the American Geriatrics Society , 59, 2091– 2099. doi: 10.1111/j.1532-5415.2011.03656.x Google Scholar CrossRef Search ADS   Hinderer, K. A., Friedmann, E., Fins, J. J. ( 2015). Withdrawal of life-sustaining treatment: patient and proxy agreement. Dimensions of Critical Care Nursing , 34, 91– 99. doi: 10.1097/DCC.0000000000000097 Google Scholar CrossRef Search ADS   Institute of Medicine. ( 2014). Dying in America: Improving quality and honoring individual preferences near the end of life . Washington, DC: National Academies Press. doi: 10.17226/18748 Javier, A. D., Figueroa, R., Siew, E. D., Salat, H., Morse, J., Stewart, T. G.,… Abdel-Kader, K. ( 2017). Reliability and utility of the surprise question in CKD stages 4 to 5. American Journal of Kidney Diseases: The Official Journal of the National Kidney Foundation , 70, 93– 101. doi: 10.1053/j.ajkd.2016.11.025 Google Scholar CrossRef Search ADS   Jones, J. M., McPherson, C. J., Zimmermann, C., Rodin, G., Le, L. W., & Cohen, S. R. ( 2011). Assessing agreement between terminally ill cancer patients’ reports of their quality of life and family caregiver and palliative care physician proxy ratings. Journal of Pain and Symptom Management , 42, 354– 365. doi: 10.1016/j.jpainsymman.2010.11.018 Google Scholar CrossRef Search ADS   Kaesmann, L., Janssen, S., Schild, S. E., & Rades, D. ( 2016). Value of comorbidity scales for predicting survival after radiochemotherapy of small cell lung cancer. Lung , 194, 295– 298. doi: 10.1007/s00408-016-9857-4 Google Scholar CrossRef Search ADS   Kim, H., Ersek, M., Bradway, C., & Hickman, S. E. ( 2015). Physician orders for life-sustaining treatment for nursing home residents with dementia. Journal of the American Association of Nurse Practitioners , 27, 606– 614. doi: 10.1002/2327-6924.12258 Google Scholar CrossRef Search ADS   Kirou-Mauro, A., Harris, K., Sinclair, E., Selby, D., & Chow, E. ( 2006). Are family proxies a valid source of information about cancer patients’ quality of life at the end-of-life? A literature review. Journal of Cancer Pain & Symptom Palliation , 2, 23– 33. doi: 10.1300/J427v02n02_04 Google Scholar CrossRef Search ADS   Kutner, J. S., Bryant, L. L., Beaty, B. L., & Fairclough, D. L. ( 2006). Symptom distress and quality-of-life assessment at the end of life: The role of proxy response. Journal of Pain and Symptom Management , 32, 300– 310. DOI: 10.1016/j.jpainsymman.2006.05.009 Google Scholar CrossRef Search ADS   Li, M., Harris, I., and Lu, Z. K. ( 2015). Differences in proxy-reported and patient-reported outcomes: Assessing health and functional status among medicare beneficiaries. BMC Medical Research Methodology , 15, 62. doi: 10.1186/s12874-015-0053-7 Google Scholar CrossRef Search ADS   Lorenzon, L., Costa, G., Massa, G., Frezza, B., Stella, F., & Balducci, G. ( 2017). The impact of frailty syndrome and risk scores on emergency cholecystectomy patients. Surgery Today , 47, 74– 83. doi: 10.1007/s00595-016-1361-1 Google Scholar CrossRef Search ADS   Massachusetts Medical Orders for Life Sustaining Treatment. ( 2014). About MOLST. Accessed October 22, 2017http://molst-ma.org/about Moss, A. H., Ganjoo, J., Sharma, S., Gansor, J., Senft, S., Weaner, B.,… Schmidt, R. ( 2008). Utility of the “surprise” question to identify dialysis patients with high mortality. Clinical Journal of the American Society of Nephrology: CJASN , 3, 1379– 1384. doi: 10.2215/CJN.00940208 Google Scholar CrossRef Search ADS   National POLST Paradigm. ( 2017a). Programs in your state. Accessed October 22, 2017http://polst.org/programs-in-your-state/ National POLST Paradigm. ( 2017b). POLST & advance directives. Accessed December 17, 2017http://polst.org/advance-care-planning/polst-and-advance-directives/ Novella, J. L., Jochum, C., Jolly, D., Morrone, I., Ankri, J., Bureau, F., & Blanchard, F. ( 2001). Agreement between patients’ and proxies’ reports of quality of life in Alzheimer’s disease. Quality of Life Research , 10, 443– 452. doi: 10.1023/A:1012522013817 Google Scholar CrossRef Search ADS   Oechsle, K., Goerth, K., Bokemeyer, C., & Mehnert, A. ( 2013). Symptom burden in palliative care patients: Perspectives of patients, their family caregivers, and their attending physicians. Supportive Care in Cancer , 21, 1955– 1962. doi: 10.1007/s00520-013-1747-1 Google Scholar CrossRef Search ADS   Ostbye, T., Taylor, D. H. Jr, Clipp, E. C., Scoyoc, L. V., & Plassman, B. L. ( 2008). Identification of dementia: Agreement among national survey data, medicare claims, and death certificates. Health Services Research , 43( 1 Pt 1), 313– 326. doi: 10.1111/j.1475-6773.2007.00748.x Powell, R., Johnston, M., & Johnston, D. W. ( 2007). Assessing walking limitations in stroke survivors: Are self-reports and proxy-reports interchangeable? Rehabilitation Psychology , 52, 177– 183. doi: 10.1037/0090-5550.52.2.177 Google Scholar CrossRef Search ADS   Pruchno, R. A., Lemay, E. P. Jr, Feild, L., & Levinsky, N. G. ( 2005). Spouse as health care proxy for dialysis patients: Whose preferences matter? The Gerontologist , 45, 812– 819. doi: 10.1093/geront/45.6.812 Google Scholar CrossRef Search ADS   Rahman, A. N., Bressette, M., Gassoumis, Z. D., & Enguidanos, S. ( 2016). Nursing home residents’ preferences on physician orders for life sustaining treatment. The Gerontologist , 56, 714– 722. doi: 10.1093/geront/gnv019 Google Scholar CrossRef Search ADS   Reamy, A. M., Kim, K., Zarit, S. H., & Whitlatch, C. J. ( 2011). Understanding discrepancy in perceptions of values: Individuals with mild to moderate dementia and their family caregivers. The Gerontologist , 51, 473– 483. doi: 10.1093/geront/gnr010 Google Scholar CrossRef Search ADS   Richardson, D. K., Fromme, E., Zive, D., Fu, R., & Newgard, C. D. ( 2014). Concordance of out-of-hospital and emergency department cardiac arrest resuscitation with documented end-of-life choices in Oregon. Annals of Emergency Medicine , 63, 375– 383. doi: 10.1016/j.annemergmed.2013.09.004 Google Scholar CrossRef Search ADS   Schmidt, T. A., Hickman, S. E., Tolle, S. W., & Brooks, H. S. ( 2004). The physician orders for life-sustaining treatment program: Oregon emergency medical technicians’ practical experiences and attitudes. Journal of the American Geriatrics Society , 52, 1430– 1434. doi: 10.1111/j.1532-5415.2004.52403.x Google Scholar CrossRef Search ADS   Schmidt, T. A., Zive, D., Fromme, E. K., Cook, J. N., & Tolle, S. W. ( 2014). Physician orders for life-sustaining treatment (POLST): Lessons learned from analysis of the Oregon POLST Registry. Resuscitation , 85, 480– 485. doi: 10.1016/j.resuscitation.2013.11.027 Google Scholar CrossRef Search ADS   Seckler, A. B., Meier, D. E., Mulvihill, M., & Paris, B. E. ( 1991). Substituted judgment: How accurate are proxy predictions? Annals of Internal Medicine , 115, 92– 98. doi: 10.7326/0003-4819-115-2-92 Google Scholar CrossRef Search ADS   Shalowitz, D. I., Garrett-Mayer, E., & Wendler, D. ( 2006). The accuracy of surrogate decision makers: A systematic review. Archives of Internal Medicine , 166, 493– 497. doi: 10.1001/archinte.166.5.493 Google Scholar CrossRef Search ADS   StataCorp. 2013. ( 2013). Stata statistical software release 13.1 . College Station, TX: StataCorp LP. St-Louis, E., Iqbal, S., Feldman, L. S., Sudarshan, M., Deckelbaum, D. L., Razek, T. S., & Khwaja, K. ( 2015). Using the age-adjusted Charlson comorbidity index to predict outcomes in emergency general surgery. The Journal of Trauma and Acute Care Surgery , 78, 318– 323. doi: 10.1097/TA.0000000000000457 Google Scholar CrossRef Search ADS   Suhl, J., Simons, P., Reedy, T., & Garrick, T. ( 1994). Myth of substituted judgment. Surrogate decision making regarding life support is unreliable. Archives of Internal Medicine , 154, 90– 96. doi: 10.1001/archinte.154.1.90 Google Scholar CrossRef Search ADS   Tarzian, A. J., & Cheevers, N. B. ( 2017). Maryland’s medical orders for life-sustaining treatment form use: Reports of a statewide survey. Journal of Palliative Medicine , 20, 939– 945. doi: 10.1089/jpm.2016.0440 Google Scholar CrossRef Search ADS   Tolle, S. W., Tilden, V. P., Nelson, C. A., & Dunn, P. M. ( 1998). A prospective study of the efficacy of the physician order form for life-sustaining treatment. Journal of the American Geriatrics Society , 46, 1097– 1102. doi: 10.1111/j.1532-5415.1998.tb06647.x Google Scholar CrossRef Search ADS   Torke, A. M., Alexander, G. C., & Lantos, J. ( 2008). Substituted judgment: The limitations of autonomy in surrogate decision making. Journal of General Internal Medicine , 23, 1514– 1517. doi: 10.1007/s11606-008-0688-8 Google Scholar CrossRef Search ADS   Torke, A. M., Moloney, R., Siegler, M., Abalos, A., & Alexander, G. C. ( 2010). Physicians’ views on the importance of patient preferences in surrogate decision-making. Journal of the American Geriatrics Society , 58, 533– 538. doi: 10.1111/j.1532-5415.2010.02720.x Google Scholar CrossRef Search ADS   Torke, A. M., Sachs, G. A., Helft, P. R., Montz, K., Hui, S. L., Slaven, J. E., & Callahan, C. M. ( 2014). Scope and outcomes of surrogate decision making among hospitalized older adults. JAMA Internal Medicine , 174, 370– 377. doi: 10.1001/jamainternmed.2013.13315 Google Scholar CrossRef Search ADS   Torke, A. M., Siegler, M., Abalos, A., Moloney, R. M., & Alexander, G. C. ( 2009). Physicians’ experience with surrogate decision making for hospitalized adults. Journal of General Internal Medicine , 24, 1023– 1028. doi: 10.1007/s11606-009-1065-y Google Scholar CrossRef Search ADS   Tremblay, D., Arnsten, J. H., & Southern, W. N. ( 2016). A simple and powerful risk-adjustment tool for 30-day mortality among inpatients. Quality Management in Health Care , 25, 123– 128. doi: 10.1097/QMH.0000000000000096 Google Scholar CrossRef Search ADS   Vick, J., Pertsch, N., Hutchings, M., Neville, B., & Bernacki, R. ( 2016). The utility of the surprise question in identifying patients most at risk of death [Abstract]. Journal of Pain and Symptom Management , 51, 342. doi: 10.1016/j.jpainsymman.2015.12.177 Google Scholar CrossRef Search ADS   Wilkinson, D. J., & Truog, R. D. ( 2013). The luck of the draw: Physician-related variability in end-of-life decision-making in intensive care. Intensive Care Medicine , 39, 1128– 1132. doi: 10.1007/s00134-013-2871-6 Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Gerontologist Oxford University Press

Surrogate Preferences on the Physician Orders for Life-Sustaining Treatment Form

Loading next page...
 
/lp/ou_press/surrogate-preferences-on-the-physician-orders-for-life-sustaining-KrPQcsLxNI
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
0016-9013
eISSN
1758-5341
D.O.I.
10.1093/geront/gny042
Publisher site
See Article on Publisher Site

Abstract

Abstract Background and Objectives The purpose of this study is to compare treatment preferences of patients to those of surrogates on the Physician Orders for Life-Sustaining Treatment (POLST) forms. Research Design and Methods Data were collected from a sequential selection of 606 Massachusetts POLST (MOLST) forms at 3 hospitals, and corresponding electronic patient health records. Selections on the MOLST forms were categorized into All versus Limit Life-Sustaining Treatment. Multivariable mixed effects (grouped by clinician) logistic regression models estimated the impact of using a surrogate decision maker on choosing All Treatment, controlling for patient characteristics (age, severity of illness, sex, race/ethnicity), clinician (physician vs non-physician), and hospital (site). Results Surrogates signed 253 of the MOLSTs (43%). A multivariable logistic regression model taking into consideration patient, clinician, and site variables showed that surrogate decision makers were 60% less likely to choose All Treatment than patients who made their own decisions (odds ratio = 0.39 [95% confidence interval = 0.24–0.65]; p < .001). This model explained 44% of the variation in the dependent variable (Pseudo-R2 = 0.442; p < .001); mixed effects logistic regression grouped by clinician showed no difference between the models (LR test = 4.0e-13; p = 1.00). Discussion and Implications Our study took into consideration variation at the patient, clinician, and site level, and showed that surrogates had a propensity to limit life-sustaining treatment. Surrogate decision makers are frequently needed for hospitalized patients, and nearly all states have adopted the POLST. Researchers may want study decision-making processes for patients versus surrogates when the POLST paradigm is employed. End-of-life care, Palliative care, Advance care planning The Physician Orders for Life-Sustaining Treatment (POLST) paradigm offers a structured approach for physicians, nurse practitioners, and physician assistants to discuss and document patient preferences for life-sustaining treatments. The POLST is not a substitute for advance directives, which are meant to name a surrogate decision maker and describe care preferences for future, unknown medical situations (National POLST Paradigm, 2017b). The Institute of Medicine recommends the POLST when disease advances and when patients are facing their final year of life (Institute of Medicine, 2014). If a patient needs a surrogate decision maker, such as in situations when a patient is deemed by a physician to be temporarily or permanently “not competent” to make medical decisions (Drane, 1984), the POLST reflects preferences resulting from discussions between clinicians and patients’ surrogates. The POLST Form is a medical order and is transferrable across care settings; this is particularly helpful given that individuals in the last 6 months of life are often under the care of more than 10 different physicians across multiple settings (Dartmouth Institute for Health Policy and Clinical Practice, 2014). Program efficacy studies indicate that the POLST improved documentation in medical records about end-of-life care, and that terminal interventions were largely consistent with documented preferences (Hickman et al., 2009; Hickman et al., 2011; Richardson, Fromme, Zive, Fu, & Newgard, 2014; Schmidt, Hickman, Tolle, & Brooks, 2004; Tolle, Tilden, Nelson, & Dunn, 1998). The POLST paradigm has been, or is in the process of being implemented in 48 states (National POLST Paradigm, 2017a), and its use will likely grow in the coming years along with the numbers of individuals aging with chronic illness. Previous research using POLST data showed that a majority of completed POLST records indicate a preference to limit life-sustaining treatments (Fritz & Barclay, 2014; Fromme, Zive, Schmidt, Cook, & Tolle, 2014; Fromme, Zive, Schmidt, Olszewski, & Tolle, 2012; Hammes, Rooney, Gundrum, Hickman, & Hager, 2012; Hickman, Keevern, & Hammes, 2014; Kim, Ersek, Bradway, & Hickman, 2015; Schmidt, Zive, Fromme, Cook, & Tolle, 2014; Tarzian & Cheevers, 2017; Tolle et al., 1998) even though the POLST allows patients or surrogates to indicate a preference to apply all treatment. One reason for the large proportions preferring to limit treatment may be due to clinicians’ interpretation of the need for a POLST form; that is, administering full treatment to sustain life is the default situation, and POLST forms are thus only needed if individuals prefer treatment limitations. Another reason may be attributed to study populations, largely comprised of non-Hispanic White individuals, who are more likely to prefer treatment limitations than those from minority populations (Rahman, Bressette, Gassoumis, & Enguidanos, 2016). Studies have not examined treatment preferences when surrogates or health care proxies make decisions documented on POLST forms. Almost one-half of hospitalized older adults use a surrogate decision maker; and three out of four physicians in a hospital sample reported that they have made a major decision with a surrogate in the past 30 days (Torke et al., 2014; Torke, Siegler, Abalos, Moloney, & Alexander, 2009). Guidance from the American Medical Association (2016) recommends that physicians guide surrogates to employ the ethical principle of “substituted judgment,” that is, to choose the treatment path the patient would have chosen if he or she were able. However, researchers have questioned the accuracy of substituted judgment for more than 25 years, and have shown in experimental vignettes that surrogates are largely unable to predict patient decisions about life-sustaining treatments (Barrio-Cantalejo et al., 2009; Hinderer, Friedmann, & Fins, 2015; Seckler, Meier, Mulvihill, & Cammer Paris, 1991; Shalowitz, Garrett-Mayer, & Wendler, 2006; Suhl, Simons, Reedy, & Garrick, 1994; Torke, Alexander, & Lantos, 2008). When there is no evidence of what a patient would have wanted, clinicians may apply the “best interest” standard, which is based on the “the pain and suffering associated with the intervention,” “the degree of and potential for benefit,” and “impairments that may result from the intervention” (American Medical Association, 2016). There is evidence of strong correlation between proxy and self-reports for observed behaviors, such as walking limitations after a stroke (Powell, Johnston, & Johnston, 2007) and quality of life measures associated with physical functions, such as breathing, hearing, sleeping (Elliott, Lazarus, & Leeder, 2006). Socio-demographic characteristics and the type of chronic conditions can confound results, but controlling for these confounders has confirmed that proxy reports are accurate for observable behaviors, such as the ability to walk a certain distance, but potentially inaccurate for subjective behaviors, such as the ability to manage money or unobserved behaviors, such as difficulty in using the toilet (Li, Harris, & Lu, 2015). Specific to patients nearing the end of life, family members can be reliable proxy reporters of objective quality of life measures, such as walking, nausea, and vomiting, but with mixed results for subjective measures, such as fatigue or pain. Further, when there is a discrepancy, proxies tend to report worse subjective quality of life than patients (Ferri & Pruchno, 2009; Kirou-Mauro, Harris, Sinclair, Selby, & Chow, 2006; Kutner, Bryant, Beaty, & Fairclough, 2006; Novella et al., 2001). In a sample of patients in an inpatient palliative care unit, proxy family or physician reports of patient symptoms were poorly correlated with patient reports during early hospitalization (Day 3), improved over time (Day 6), but were still low (<35% agreement) on measures of physical symptoms and psychological well-being (Jones et al., 2011). Family caregivers tended to overestimate the intensity or frequency of lack of energy, anxiety, sadness, and pain distress compared with patients. On the other hand, physicians generally underestimated symptom burden compared with patients (Oechsle, 2013). Concept Model This study is designed to estimate the odds that surrogate decision makers’ would choose aggressive life-sustaining treatment, taking into account patient, clinician, and upstream factors that might have channeled patients to sites of care. We incorporated patient variables to take into account factors that may influence whether a patient needed a surrogate, such as age and severity of illness, which may also affect preferences (Figure 1). On the other hand, clinician type, patient race/ethnicity, and sex were thought to affect only preferences. We acknowledge that prior decisions might have influenced a patient’s site of care, such as insurance limitations, patient or clinician preferences, and geography. These factors, in addition to site-associated practice norms and prior advance care plans, could affect patient or surrogate preferences for life-sustaining treatments in ways that were not measured. Figure 1. View largeDownload slide Concept model—factors affecting life-sustaining treatment preferences for hospitalized adults with limited life expectancy. Figure 1. View largeDownload slide Concept model—factors affecting life-sustaining treatment preferences for hospitalized adults with limited life expectancy. Methods The Institutional Review Board at Partners HealthCare approved this study, which uses 606 patient records from three of its hospitals. Research Electronic Data Capture (REDCap), an electronic data capture tool hosted by the hospital system’s research management office, was used to collect and manage study data (Harris, Taylor, Payne, Gonzalez, & Conde, 2009). De-identified data were exported for analysis using STATA 13.1 (StataCorp. 2013, 2013). Setting Characteristics The three hospital sites were part of the same integrated care network, and were among the first hospitals to implement Massachusetts POLST (MOLST) in Eastern Massachusetts. As the first hospitals to implement MOLST, our data represent de novo exposure to MOLST, and nearly eliminates the possibility that patients had arrived with a form completed at another hospital or in an outpatient setting. Each site employed a slightly different staffing model, and differed as to how broadly the MOLST was deployed. Site 1, a long-term acute care hospital (LTACH), was staffed largely by employed physicians and augmented by a few nurse practitioners; none were trained in palliative care. MOLST was implemented throughout the entire 180-bed hospital, where all patients were admitted directly from an acute care hospital stay. LTACHs serve patients with severe illness, who need hospital-level care for a long stay, which may last one or more months (American Hospital Association, 2017; Center for Medicare and Medicaid Services, October 2016). Thirty-one different clinicians signed MOLSTs at Site 1. Site 2 is a community hospital, where MOLSTs were largely administered by a consulting palliative care practice comprised of five palliative care physicians and nurse practitioners, who signed nearly all (96%) of the forms; another five clinicians outside of this palliative care practice signed the remaining 4% of the forms at this location. Site 3 is an inpatient intensive palliative care unit with 12 beds in a quaternary care academic medical center with more than 800 beds. This inpatient palliative care unit was staffed by employed physician assistants under the direction of a small number of palliative care physicians. Eleven different clinicians completed MOLSTs at Site 3. Patient selection also differed from site to site, but all patients had advanced illness and were nearing the end of life. At Site 1 (LTACH), clinicians used the “surprise” question, which has been validated in a number of settings as a simple prognostic tool for patients with limited life expectancy (Da Silva Gane, Braun, Stott, Wellsted, & Farrington, 2013; Javier et al., 2017; Moss et al., 2008; Vick, Pertsch, Hutchings, Neville, & Bernacki, 2016). That is, clinicians asked themselves “Would I be surprised if this patient died in the next 12 months?” A “no” triggered a “goals of care” conversation, followed by documentation on a MOLST form. Monthly tracking data over the first year of MOLST implementation at Site 1 showed that 15%–40% of discharged patients completed MOLSTs. The palliative care consulting practice (Site 2) did not use the “surprise” question for patients, but nearly all patients would have qualified had it been employed. Site 3, the inpatient palliative care unit, did not always use the “surprise” question because nearly all patients were expected to have less than 1 year of remaining life. Clinicians completed MOLST forms only for patients who were discharged to another care setting or to home, which accounted for approximately 50% of patients at this site. Despite differing patient selection criteria, patients at all sites with a completed MOLST were expected to have limited life expectancy. Sample The sample comprised a sequential selection of patients who had completed the Massachusetts Medical Orders for Life-Sustaining Treatment (MOLST) forms while inpatient at three hospitals between July 9, 2012 and January 17, 2014. The final analytic data set contained 593 MOLST forms after discarding 2% (N = 13 from 606) with missing clinician signatures, making the order invalid. Two-hundred and eighty-eight (N = 288; 49%) of the analytic data set came from the LTACH (Site 1); 204 (34%) from the palliative care consulting practice at a community hospital (Site 2); and 101 (17%) from the inpatient intensive palliative care unit at the academic medical center (Site 3). The sample comprised patients in both intensive care (ICU) and non-intensive care units, and residing in both long-term care settings and in the community. A small number of patients completed multiple MOLST forms during 1 hospital stay, or completed another MOLST form during a second or third hospitalization over the course of the study period. In these circumstances, data were collected only from the first MOLST, reserving the issue of changes to MOLST forms for another study. Dependent Variable The dependent variable (All vs Limit Treatment) was constructed from responses on Page 1 of the MOLST form, which provided a sequence of dichotomous “do not attempt” or “attempt” choices beginning with resuscitation, followed by ventilation, then transfer to a hospital (Massachusetts Medical Orders for Life Sustaining Treatment, 2014). Page 2 of the form seeks preferences about the duration of treatments on Page 1, and offers an opportunity to specify decisions regarding other life-sustaining treatments, such as kidney dialysis and artificial nutrition or hydration. Page 2 is not required in order for the MOLST to be valid, and too few were completed for meaningful analysis. The pattern of responses generally followed logically from the first choice, resuscitation (Figure 2). Ventilation options were divided into two categories: invasive (intubation) and noninvasive, such as continuous positive airway pressure (CPAP); information about CPAP was not used due to the large number (N = 119; 20%) of forms missing these data. Patients were grouped into those who preferred All Treatment (“Yes” to resuscitation, intubation, AND transfer) or Limit Treatment (“No” to resuscitation, intubation, OR transfer). A separate analysis confirmed that there was no significant difference between the group missing CPAP data and the group not missing CPAP data for the primary variable of interest (surrogate vs patient decisions; Pearson χ2 = 0.21; p = .644; Supplementary Table 1). Figure 2. View largeDownload slide Pattern of responses on MOLST. Figure 2. View largeDownload slide Pattern of responses on MOLST. Main Effect Data for the main variable of interest (patient or surrogate signature) were collected from the MOLST form, which requests that signatories identify their relationship to the patient. If the signatory did not respond, we compared the patient name with the signature and associated printed name. Forms with different signatory and patient names were categorized into the surrogate group. All printed names were legible. Control Variables Patient Demographics Demographic data were obtained from patient electronic health records, which included age at the time the MOLST was signed and patient race/ethnicity. Logistic regression models employed a categorical age variable approximating the bottom quartile (<60), interquartile range (60–73), and top quartile (80 and over) at Site 1, where there was the most variation between preferences for All versus Limit Treatment. Severity of Illness We used ICD-9 codes associated with the hospitalization during which the MOLST form was signed to compute the Charlson Score as an indicator of illness severity (Al Feghali et al., 2016; Chang et al., 2016; Tremblay, Arnsten, & Southern, 2016). The Charlson Weighted Index of Comorbidity was developed as a prognostic indicator for individuals with chronic illnesses. Each chronic illness on the Index has a score associated with mortality risk (1, 2, 3, or 6; Charlson, Pompei, Ales, & MacKenzie, 1987). For example, uncomplicated diabetes is 1, diabetes with end organ damage is 2, moderate or severe liver disease is 3, and metastatic solid tumor is 6. Scores for comorbid conditions are summed. In the original Charlson validation cohort, 1-year mortality was 100% for patients who were hospitalized, survived to discharge, and had a score greater than 5 (Charlson et al., 1987). Survival has improved since the original Charlson validation cohort for many illnesses captured by its Index, but the tool remains useful in research as a measure for burden of illness (Crooks, West, & Card, 2015; D’Hoore, Sicotte, & Tilquin, 1993). Some studies have incorporated a decade-based age adjustment specified by Charlson to reflect additional mortality risk associated with age beginning at age 50 (Charlson et al., 1987; Dias-Santos, Ferrone, Zheng, Lillemoe, & Fernandez-del Castillo, 2015; Kaesmann, Janssen, Schild, & Rades, 2016; Lorenzon et al., 2017; St-Louis et al., 2015). Conceptually, this adjustment was meant to reflect that the same illness in a 50-year-old and an 80-year-old represent different risks for 1-year mortality. We did not include this adjustment, which is based on age decade, to allow for greater precision using our age variable. Clinician Type Clinician certification (physician vs nurse practitioner or physician assistant) was collected from signatures on the MOLST forms. If certification data were missing, then a research assistant matched the signature against internal staffing records. Site The variables Site 1, Site 2, and Site 3 identified each hospital, and served as a means to hold constant site-associated practice variation, such as differing staffing models and the use of palliative care specialists. The Site variable also took into account unmeasured factors that may have channeled patients to each site, such as insurance coverage and patient or clinician preferences farther upstream in the care continuum. For example, patients at the LTACH may not have been offered palliative care, or were offered the service and refused; as a result, these patients were transferred to the LTACH for high intensity treatment. Patients under the care of clinicians in the palliative care practices were presented with an option for palliative care and accepted, which may be associated with a willingness to accept limitations to life-sustaining treatment. Statistical Analysis We examined sample characteristics by site and by decision maker using the Pearson χ2 test. Single-variable logistic regression analyses were first performed to estimate the odds that each independent variable would predict a preference for All (vs Limit) Treatment. Multivariable logistic regression models estimated the odds for All Treatment first for Surrogate (vs Patient) Decision Makers, followed by patient characteristics, then concluding with the impact of Site. Mixed effects logistic regression was performed using STATA’s xtmelogit command to account for covariance due to clustering of patients within clinicians. Results Baseline Patient Characteristics Nearly all patient characteristics differed between sites (Table 1). Surrogates signed approximately 35% of the MOLST forms at Sites 1 (LTACH without palliative care) and 3 (palliative care at academic medical center) compared with 60% at Site 2 (palliative care at community hospital), which also had the oldest patients. Only one in four surrogates chose All Treatment compared with nearly one in two patients doing the same. Individuals at the academic medical center had higher burden of illness than the other two sites as measured by the Charlson Score (median for Sites 1 and 2 = 5; Site 3 = 7). Patient primary diagnoses were not different between surrogate and patient-signed forms with the exception of cancers, where a larger proportion of patient signers had cancers (26%) compared with patients who had surrogates (11%; p < .001). Table 1. Sample Characteristics by Site and by Treatment Preference Group N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  Notes: MOLST, Massachusetts Physician Orders for Life-Sustaining Treatment. aSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed academic medical center. bMassachusetts Medical Orders for Life-Sustaining Treatment Form. cScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. p-Value using the two-sample test of proportion comparing Surrogate and Patient groups: *≤0.05; **≤0.01; ***≤0.001. View Large Table 1. Sample Characteristics by Site and by Treatment Preference Group N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  N = 593  Site 1a (N = 288)  Site 2a (N = 204)  Site 3a (N = 101)  MOLSTb signature  Surrogate (N = 253)  Patient (N = 340)  All (vs Limit) life-sustaining treatment  199 (69%)  5 (3%)  12 (12%)  58 (23%)  158 (46%)***  Surrogate decision maker  100 (35%)  119 (58%)  34 (34%)  253 (100%)  —  Patient age: median (range)  68 (20–93)  83 (42–102)  60 (24–87)  80 (24–102)  68 (20–100)***  Patient age group             <60  69 (24%)  12 (6%)  50 (50%)  43 (17%)  88 (26%)**   60–79   152 (53%)  52 (25%)  41 (41%)  76 (30%)  169 (50%)***   80 and over  67 (23%)  140 (69%)  10 (10%)  134 (53%)  83 (24%)***  Charlson Scorec: median (range)  5 (0–12)  5 (0–15)  7 (2–10)  5 (0–12)  6 (0–15)*  Male (vs female) patient  162 (56%)  83 (41%)  48 (48%)  124 (49%)  169 (50%)  Race/ethnicity of patient             Non-Hispanic White  214 (74%)  191 (94%)  87 (86%)  282 (83%)  210 (83%)   Non-Hispanic Black  17 (6%)  1 (<1%)  5 (5%)  9 (4%)  14 (4%)   Non-Hispanic other and Hispanic  57 (20%)  12 (6%)  9 (9%)  34 (13%)  44 (13%)  Physician (vs NP or PA)  179 (62%)  115 (56%)  13 (4%)  135 (53%)  172 (51%)  Primary diagnoses (top 5)       Cancers  29 (11%)  87 (26%)***   Cardiovascular diseases  44 (17%)  45 (13%)   Respiratory disease/dysfunction  40 (17%)  52 (15%)   Infection/non-specific fever  34 (13%)  40 (12%)   Gastrointestinal disease/dysfunction  18 (7%)  18 (5%)  Notes: MOLST, Massachusetts Physician Orders for Life-Sustaining Treatment. aSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed academic medical center. bMassachusetts Medical Orders for Life-Sustaining Treatment Form. cScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. p-Value using the two-sample test of proportion comparing Surrogate and Patient groups: *≤0.05; **≤0.01; ***≤0.001. View Large Multivariable Logistic Regression Surrogate decision makers were much less likely to choose All Treatment than patients who made their own decisions (OR = 0.34–0.46; p ≤ .001). This effect remained unchanged as we added patient, clinician, and site characteristics (Table 2; Models 1–5). Patients in the highest age group (≥80) were consistently less likely to choose All Treatment than those in the lowest age group (<60; OR = 0.14–0.22; p ≤ .001; Models 2–5). Individuals with high Charlson Scores (>5) were less likely to choose All Treatment than those with lower Charlson Scores (<5; OR = 0.26; p ≤ .001; Model 3 and 4), but site-associated effects reduced the magnitude of this effect (OR = 0.59; p ≤ .05; Model 5). Overall, patients and surrogates at the palliative care sites had >90% lower odds for choosing All Treatment than those at the LTACH (Site 2: odds ratio, OR = 0.02 [95% confidence interval, CI, 0.01–0.05]; Site 3: OR = 0.06 [95% CI 0.02–0.10]; p ≤ .001). Table 2. Odds Ratios (95% Confidence Interval) for Logistic Regression Models Estimating the Odds for All Treatment versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  Notes: aScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. bSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed at an academic medical center. cPseudo-R2 is an indicator of model fit and can be interpreted as the amount of variation in All versus Limit Treatment attributed to each variable. +p-Value of odds ratio ≤.10; *≤.05; **≤.01; ***≤.001. ++++p-Value of the model ≤.001. View Large Table 2. Odds Ratios (95% Confidence Interval) for Logistic Regression Models Estimating the Odds for All Treatment versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  N = 593  Model 1  Model 2  Model 3  Model 4  Model 5  Surrogate (vs patient signature)  0.34*** (0.24–0.49)  0.46*** (0.31–0.67)  0.39*** (0.26–0.58)  0.36*** (0.24–0.55)  0.39*** (0.24–0.65)  Age group: (reference <60)             60–79    0.76 (0.49–1.18)   0.74 (0.47–1.16)   0.65+ (0.41–1.05)  0.41** (0.21–0.78)   80 and over    0.22*** (0.13–0.37)  0.15*** (0.09–0.26)  0.14*** (0.08–0.24)  0.19*** (0.09–0.41)  Charlson Scorea >5 (vs Charlson≤5)      0.26*** (0.17–0.39)  0.27*** (0.18–0.41)  0.59* (0.36–0.99)  Male (vs female patient)        1.71** (1.16–2.53)  1.20 (0.75–1.94)  Race/ethnicity of patient (reference: non-Hispanic White)             Non-Hispanic Black         2.59* (1.01–6.65)  1.37 (0.49–3.81)   Non-Hispanic other and Hispanic        2.10** (1.19–3.69)  1.12 (0.59–2.12)  Physician (vs NP or PA)        1.73** (1.17–2.57)  0.98 (0.59–1.63)  Site (reference site 1)             Site 2b          0.02*** (0.01–0.05)   Site 3b          0.04*** (0.02–0.10)  Pseudo-R2c  0.046++++  0.103++++  0.163++++  0.195++++  0.442++++  Notes: aScore derived from the Charlson Weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. bSite 1: 180-bed long-term acute care hospital with no palliative care specialist; Site 2: Palliative care consulting practice in 300-bed community acute care teaching hospital; Site 3: Inpatient palliative care unit in 800+ bed at an academic medical center. cPseudo-R2 is an indicator of model fit and can be interpreted as the amount of variation in All versus Limit Treatment attributed to each variable. +p-Value of odds ratio ≤.10; *≤.05; **≤.01; ***≤.001. ++++p-Value of the model ≤.001. View Large A multivariable mixed effects logistic regression model, grouped by clinicians, using all observations and the full complement of independent variables was no different from a model that did not cluster patients by clinician (Standard Deviation of the Intercept = 2.58e-10 [95% CI 0.00–0.00]; Likelihood Ratio Test = 4.0e-13; p = 1.00). Post Hoc Logistic Regression Using Only Data from the LTACH We tested our model post hoc using only data from the LTACH (Site 1) to eliminate the potential for confounding due to clinicians channeling patients into palliative care settings due to prior patient or surrogate preferences to limit life-sustaining treatments (Table 3). Using only LTACH data, surrogate decision makers experienced 52% lower odds for choosing All Treatment compared with patient decision makers (OR = 0.48 [95% CI 0.28–0.84]; p = .01). These results were substantially the same as those using data pooled for all three sites. Mixed effects multivariable logistic regression, grouped by 31 clinicians again showed no difference between this model and a comparable model that did not cluster patients within clinicians (Standard Deviation of Intercept: 1.58e-09 [95% CI 0.00–0.00]; p = 1.00). Table 3. Post Hoc Analysis – Odds Ratios and (95% Confidence Interval) for a Multivariable Logistic Regression Model Estimating the Odds for All Treatmenta versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form Using Only Site 1b Data N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  Notes: aAll Treatment = “Yes” to All (resuscitate, intubate, transfer) versus Limit Treatment = “No” to At Least One (resuscitate, intubate, transfer). bSite 1 is a 180-bed long-term acute care hospital with no palliative care specialist. cScore derived from the Charlson weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. View Large Table 3. Post Hoc Analysis – Odds Ratios and (95% Confidence Interval) for a Multivariable Logistic Regression Model Estimating the Odds for All Treatmenta versus Limit Treatment on the Massachusetts Medical Orders for Life-Sustaining Treatment Form Using Only Site 1b Data N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  N = 288  Odds ratio (95% CI)  p value  Surrogate (vs patient signature)  0.48 (0.28–0.84)  .010  Charlson Score >5c  0.56 (0.32–0.97)  .038  Age: (reference ≤60)       60–79  0.33 (0.15–0.73)  .006   80 and over  0.16 (0.07–0.40)  ≤.001  Male  1.14 (0.66–1.96)  .638  Race/ethnicity (reference = non-Hispanic White)       Non-Hispanic Black  1.70 (0.54–5.33)  .364   Non-Hispanic other and Hispanic  1.05 (0.53–2.07)  .889  Physician (vs NP/PA signature)  1.06 (0.62–1.85)  .815  Notes: aAll Treatment = “Yes” to All (resuscitate, intubate, transfer) versus Limit Treatment = “No” to At Least One (resuscitate, intubate, transfer). bSite 1 is a 180-bed long-term acute care hospital with no palliative care specialist. cScore derived from the Charlson weighted Index of Comorbidities using discharge diagnoses (ICD-9) and does not incorporate the decade-based age adjustment described in the original Charlson (1987) paper. View Large Discussion The goal of this study was to estimate the odds that surrogate decision makers would choose aggressive life-sustaining treatment decisions by taking into consideration patient, clinician, and site variables, which accounted for unmeasured factors that channeled patients to each location, such as patient insurance and patient or clinician preferences farther upstream in the care continuum. We collected data from the Massachusetts version of the POLST form and corresponding patient medical records at three hospitals in the greater Boston area. An important strength of this study is that data represent decisions made in a health care setting in the context of patients and families experiencing an event associated with serious, life-limiting illness. Surrogates signed 43% of the MOLST forms in our study, which is consistent with prior results (47%; Torke et al., 2014). Multivariable logistic regression results showed that surrogates were 60% less likely to choose All Treatment than patients who did not use a surrogate. Because decisions in palliative care settings may have confounded the results, we performed a post hoc analysis using only data from the LTACH, which did not have a palliative care practice. This post hoc analysis revealed essentially the same result, that surrogate decision makers were 52% less likely to choose All Treatment than patients not using a surrogate. Our results are consistent with findings using POLST data from a nursing home population (Rahman et al., 2016), which also used data gathered from a clinical setting. Other studies that examined both patient and surrogate decisions are not directly comparable because study methods were intended to test the accuracy of surrogate decisions, and used hypothetical scenarios (Barrio-Cantalejo et al., 2009; Shalowitz et al., 2006), which may not reflect contextual cues that could influence patient or surrogate decisions when life-sustaining treatment preferences are sought in clinical settings (Reamy, Kim, Zarit, & Whitlatch, 2011). Surrogate decision makers face intense emotional distress as they balance forces favoring more intervention against those favoring less. Prior research suggest that negative cues are most salient when surrogates need to make decisions, such as a patient’s unconscious state or perceived pain, and a focus on what patients do not want instead of what they do want (Dionne-Odom, 2015). The process of discussing end-of-life treatments with surrogates is a delicate balance between prior patient preferences and the best interests of the patient going forward (Torke, Moloney, Siegler, Abalos, & Alexander, 2010). We should not assume that our results reflect prior discussions between surrogate and patient pairs, as these pairs have shown low agreement (62%) when asked whether discussions about life-sustaining treatments had taken place (Fried, Redding, Robbins, O’Leary, & Iannone, 2011), and prior discussions also showed no impact on the accuracy of substituted judgment (Pruchno, Lemay Jr., Field, & Levinsky, 2005). Rather, our results seem to indicate that surrogate decision makers are systematically different from patient decision makers, which would be consistent with prior research showing that proxy respondents represented individuals in worse physical health, older, and were less educated than those not using a proxy (Elliott, Beckett, Chong, Hambarsoomians, & Hays, 2008). Prior research about the accuracy of proxy reporting suggests that surrogates tend to overestimate the burden of illness, especially on subjective quality of life measures (Ferri & Pruchno, 2009; Kirou-Mauro et al., 2006; Kutner et al., 2006; Novella et al., 2001). This tendency can be problematic whether clinicians guide surrogates to employ the principle of “substituted judgment” or “best interest” when making life-sustaining treatment decisions. Clinicians may want to take into consideration that in some situations, surrogates may overestimate burden, such as when life-sustaining treatment decisions are based on unobservable or subjective quality of life outcomes for the patient. Because the POLST is recommended for patients facing their last year of life, clinicians may want to attend to completing a POLST when patients are still able to make their own decisions and a surrogate is not yet needed. Limitations Our study used a sample that completed MOLSTs at three hospitals; thus, selection bias cannot be ignored. In addition to upstream forces that channeled individuals to these hospitals, clinicians at each location selected patients for MOLST, and patients or surrogates agreed to complete a MOLST form. We acknowledge that our data failed to capture patients and surrogates who were not asked to complete a MOLST form. Some clinicians may have been uncomfortable discussing end-of-life care goals with patients or surrogates, whereas others may have thought their patients would survive beyond 1 year, and still others may have decided not to complete MOLST forms because patients and surrogates wanted All Treatment. These biases in the data set would likely result in odds ratios being too low for surrogates. By the same token, patients and surrogates who chose not to complete a MOLST form were choosing All Treatment by default, and because we did not track refusals, our estimates regarding a preference to Limit Treatment may be high. Furthermore, we lacked background data about patients, which would have predisposed them to treatment versus treatment limitations, such as religiosity or prior declarations in advance care directive. While we have no evidence pointing to underlying systematic selection bias in our sample, we also have no evidence to support a lack of systematic selection bias. It is reasonable to assume that clinicians and patients in the palliative care settings were already biased against choosing All Treatment. Patients or surrogates agreed to treatment in these settings, and the definition of palliative care is to treat symptoms, not disease. The post hoc analysis using data from the LTACH, which did not have a palliative care practice, allowed us to eliminate the potential for this bias by examining results at one location where there was substantial variation in the dependent variable; that is, 70% chose All Treatment. The result for the main effect using single-site data was substantially the same as that using the full data set, which suggests that whatever forces led patients to the palliative care practices as opposed to the LTACH did not change our conclusion. If anything, one would expect that clinicians in an LTACH might approach goals of care discussions in a manner that could bias patient or surrogate decisions toward aggressive treatment. Again, our single-site results do not support this direction of thinking; rather, they confirm that even in an aggressive treatment pathway, surrogate decision makers are less likely to choose all life-sustaining treatments than patients. Using only one scale, the Charlson Score, to measure severity of illness is also an important limitation in this study. Our regression models showed that the surrogate variable acted independently from the Charlson Score, which suggests that the Charlson Index did not fully measure conditions that would lead to needing a surrogate decision maker, such as delirium, high fever, or nausea that often lead to hospitalization for individuals with advanced illness. Further, dementia is assigned only a score of 1 on the Charlson Index, with no further adjustment for higher severity, even though the Index adjusts for higher severity of cancers, diabetes, and heart disease. This limitation in the Charlson Score, coupled with other research showing insufficient reporting or documentation of underlying dementia in older populations (Crowther, Bennett, & Holmes, 2017; van den Dungen et al., 2012; Ostbye, Taylor, Clipp, Van Scoyoc, & Plassman, 2008), and a propensity for surrogates to favor limitations when dementia is present (Feltz & Samayoa, 2012), indicate that future studies should incorporate multiple measures to reflect conditions faced in the last days or weeks of life. Finally, our mixed effects models grouped by clinicians did not show variation between clinicians even though prior research point to an association between patient preferences and the clinicians’ end-of-life care preferences for themselves (Wilkinson & Truog, 2013). A larger sample may have yielded different results. A review of the numbers of MOLST forms completed by each of the 31 clinicians at Site 1 showed that close to 50% completed just one or two forms while 30% completed 11–36 forms. A larger sample, collected over a longer period of time, would have provided more MOLST forms for the large proportion of clinicians who had only completed one or two forms, and may have yielded different results. Conclusion The results in this study show that surrogate decision makers for a hospitalized population with serious and advancing illness were more likely to indicate a preference to limit life-sustaining treatments than patient decision makers. In multivariable logistic regression models, this effect operated independently from age, practice patterns associated with site, and the Charlson Score. Prior research demonstrating that proxies tend to overestimate symptom burden compared with patients presents a troubling ethical problem in our results, which is whether decisions to limit treatment by surrogates can be construed as an accurate reflection of what the patient would have wanted. Given the frequency with which surrogates are needed for older, hospitalized patients, provider organizations may want to incorporate new strategies to reach surrogates as an additional patient decision partner. Providers may also want to consider discussing with patients and documenting their preferences for temporary versus permanent incapacity. Finally, when communicating with surrogate decision makers, providers may want to keep in mind a surrogate’s tendency to overestimate symptom burden.. Supplementary Data Supplementary data are available at The Gerontologist online. Acknowledgments Elizabeth E. Chen had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. A portion of the results in this article was presented at the Gerontological Society of America Research Meeting in New Orleans, November 2016, at the International Association of Gerontology and Geriatrics World Congress in San Francisco, July 2017, and via webinar hosted by the National POLST on January 25, 2018. We gratefully acknowledge the contributions of Mary Fairbanks, BS, who generously volunteered hundreds of hours during the data collection phase of this study; Edward Alan Miller, PhD, MPA, who provided important critique for earlier versions of this research; and Susan T. Moore, RN, MPH for facilitating the data collection process and providing important comments to the final draft. References Al Feghali, K. A., Robbins, J. R., Mahan, M., Burmeister, C., Khan, N. T., Rasool, N.,… Elshaikh, M. A. ( 2016). Predictive capacity of 3 comorbidity indices in estimating survival endpoints in women with early-stage endometrial carcinoma. International Journal of Gynecological Cancer: Official Journal of The International Gynecological Cancer Society , 26, 1455– 1460. doi: 10.1016/j.ygyno.2016.04.205. Google Scholar CrossRef Search ADS   American Hospital Association. ( 2017). Long-term care hospitals (LTCHs). Retrieved October 22, 2017, from http://www.aha.org/advocacy-issues/postacute/ltach/index.shtml American Medical Association. ( 2016). Code of Medical Ethics Opinions on Consent, Communications and Decision Making Section 2.2.1. Retrieved March 3, 2018, from https://www.ama-assn.org/sites/default/files/media-browser/code-of-medical-ethics-chapter-2.pdf Barrio-Cantalejo, I. M., Molina-Ruiz, A., Simón-Lorda, P., Cámara-Medina, C., Toral López, I., del Mar Rodríguez del Aguila, M., & Bailón-Gómez, R. M. ( 2009). Advance directives and proxies’ predictions about patients’ treatment preferences. Nursing Ethics , 16, 93– 109. doi: 10.1177/0969733008097995 Google Scholar CrossRef Search ADS   Center for Medicare and Medicaid Services. ( October 2016). Long term care hospital prospective payment system. Retrieved October 22, 2017, from https://www.cms.gov/Outreach-and-Education/Medicare-Learning-Network-MLN/MLNProducts/Downloads/Long-Term-Care-Hospital-PPS-Fact-Sheet-ICN006956.pdf Chang, C., Yin, W., Wei, C., Wu, C., Su, Y., Yu, C., Lee, C.C., et al.   ( 2016). Adjusted age-adjusted Charlson Comorbidity Index score as a risk measure of perioperative mortality before cancer surgery. PLoS One , 11, e0148076. doi: 10.1371/journal.pone.0148076 Google Scholar CrossRef Search ADS   Charlson, M. E., Pompei, P., Ales, K. L., & MacKenzie, C. R. ( 1987). A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. Journal of Chronic Diseases , 40, 373– 383. doi: 10.1016/0021-9681(87)90171-8 Google Scholar CrossRef Search ADS   Crooks, C. J., West, J., & Card, T. R. ( 2015). A comparison of the recording of comorbidity in primary and secondary care by using the Charlson Index to predict short-term and long-term survival in a routine linked data cohort. BMJ Open , 5, e007974. doi: 10.1136/bmjopen-2015-007974 Google Scholar CrossRef Search ADS   Crowther, G. J., Bennett, M. I., & Holmes, J. D. ( 2017). How well are the diagnosis and symptoms of dementia recorded in older patients admitted to hospital? Age and Ageing , 46, 112– 118. doi: 10.1093/ageing/afw169 Da Silva Gane, M., Braun, A., Stott, D., Wellsted, D., & Farrington, K. ( 2013). How robust is the ‘surprise question’ in predicting short-term mortality risk in haemodialysis patients? Nephron. Clinical Practice , 123, 185– 193. doi: 10.1159/000353735 Google Scholar CrossRef Search ADS   Dartmouth Institute for Health Policy and Clinical Practice. ( 2014). Care of chronic illness in last two years of life: Number of different physicians seen per decedent in the last 6 months of life. Retrieved October 22, 2017, from http://www.dartmouthatlas.org/data/table.aspx?ind=12&loct=2&loc=2%2c3%2c4%2c5%2c6%2c7%2c8%2c9%2c10%2c11%2c12%2c13%2c14%2c15%2c16%2c17%2c18%2c19%2c20%2c21%2c22%2c23%2c24%2c25%2c26%2c27%2c28%2c29%2c30%2c31%2c32%2c33%2c34%2c35%2c36%2c37%2c38%2c39%2(TRUNCATED) D’Hoore, W., Sicotte, C., & Tilquin, C. ( 1993). Risk adjustment in outcome assessment: The Charlson comorbidity index. Methods of Information in Medicine , 32, 382– 387. doi: 10.1055/s-0038-1634956 Google Scholar CrossRef Search ADS   Dias-Santos, D., Ferrone, C. R., Zheng, H., Lillemoe, K. D., & Fernández-Del Castillo, C. ( 2015). The Charlson age comorbidity index predicts early mortality after surgery for pancreatic cancer. Surgery , 157, 881– 887. doi: 10.1016/j.surg.2014. 12.006 Google Scholar CrossRef Search ADS   Dionne-Odom, J. N., Willis, D. G., Bakitas, M., Crandall, B., & Grace, P. J. ( 2015). Conceptualizing surrogate decision making at end of life in the intensive care unit using cognitive task analysis. Nursing Outlook , 63, 331– 340. doi: 10.1016/j.outlook.2014.10.004 Google Scholar CrossRef Search ADS   Drane, J. F. ( 1984). Competency to give an informed consent. A model for making clinical assessments. JAMA , 252, 925– 927. doi: 10.1001/jama.1984.03350070043021 Google Scholar CrossRef Search ADS   van den Dungen, P., van Marwijk, H. W., van der Horst, H. E., Moll van Charante, E. P., Macneil Vroomen, J., van de Ven, P. M., & van Hout, H. P. ( 2012). The accuracy of family physicians’ dementia diagnoses at different stages of dementia: A systematic review. International Journal of Geriatric Psychiatry , 27, 342– 354. doi: 10.1002/gps.2726 Elliott, D., Lazarus, R., Leeder, S. R. ( 2006). Proxy respondents reliably assessed the quality of life of elective cardiac surgery patients. Journal of Clinical Epidemiology , 59, 153– 159. doi: 10.1097/DCC.0000000000000097 Google Scholar CrossRef Search ADS   Elliott, M. N., Beckett, M. K., Chong, K., Hambarsoomians, K., & Hays, R. D. ( 2008). How do proxy responses and proxy-assisted responses differ from what Medicare beneficiaries might have reported about their health care? Health Services Research , 43, 833– 848. doi: 10.1111/j.1475-6773.2007.00820.x Google Scholar CrossRef Search ADS   Feltz, A., & Samayoa, S. ( 2012). Heuristics and life-sustaining treatments. Journal of Bioethical Inquiry , 9, 443– 455. doi: 10.1007/s11673-012-9396-5 Google Scholar CrossRef Search ADS   Ferri, C. V., & Pruchno, R. A. ( 2009). Quality of life in end-stage renal disease patients: Differences in patient and spouse perceptions. Aging & Mental Health , 13, 706– 714. doi: 10.1080/13607860902845558 Google Scholar CrossRef Search ADS   Fried, T. R., Redding, C. A., Robbins, M. L., O’Leary, J. R., & Iannone, L. ( 2011). Agreement between older persons and their surrogate decision-makers regarding participation in advance care planning. Journal of the American Geriatrics Society , 59, 1105– 1109. doi: 10.1111/j.1532-5415.2011.03412.x Google Scholar CrossRef Search ADS   Fritz, Z. B., & Barclay, S. I. ( 2014). Patients’ resuscitation preferences in context: Lessons from POLST. Resuscitation , 85, 444– 445. doi: 10.1016/j.resuscitation.2014.01.016 Google Scholar CrossRef Search ADS   Fromme, E. K., Zive, D., Schmidt, T. A., Cook, J. N., & Tolle, S. W. ( 2014). Association between physician orders for life-sustaining treatment for scope of treatment and in-hospital death in Oregon. Journal of the American Geriatrics Society , 62, 1246– 1251. doi: 10.1111/jgs.12889 Google Scholar CrossRef Search ADS   Fromme, E. K., Zive, D., Schmidt, T. A., Olszewski, E., & Tolle, S. W. ( 2012). POLST Registry do-not-resuscitate orders and other patient treatment preferences. JAMA , 307, 34– 35. doi: 10.1001/jama.2011.1956 Google Scholar CrossRef Search ADS   Hammes, B. J., Rooney, B. L., Gundrum, J. D., Hickman, S. E., & Hager, N. ( 2012). The POLST program: A retrospective review of the demographics of use and outcomes in one community where advance directives are prevalent. Journal of Palliative Medicine , 15, 77– 85. doi: 10.1089/jpm.2011.0178 Google Scholar CrossRef Search ADS   Harris, P. A., Taylor, R., Thielke, R., Payne, J., Gonzalez, N., & Conde, J. G. ( 2009). Research electronic data capture (REDCap)–a metadata-driven methodology and workflow process for providing translational research informatics support. Journal of Biomedical Informatics , 42, 377– 381. doi: 10.1016/j.jbi.2008.08.010 Google Scholar CrossRef Search ADS   Hickman, S., Keevern, E., & Hammes, B. J. ( 2014). The effect of the POLST (Physician Orders for Life-Sustaining Treatment) program on clinical care: A systematic review. Journal of Palliative Care , 30, 226– 227. doi: 10.1089/jpm.2008.0196 Hickman, S. E., Nelson, C. A., Moss, A. H., Hammes, B. J., Terwilliger, A., Jackson, A., & Tolle, S. W. ( 2009). Use of the Physician Orders for Life-Sustaining Treatment (POLST) paradigm program in the hospice setting. Journal of Palliative Medicine , 12, 133– 141. doi: 10.1089/jpm.2008.0196 Google Scholar CrossRef Search ADS   Hickman, S. E., Nelson, C. A., Moss, A. H., Tolle, S. W., Perrin, N. A., & Hammes, B. J. ( 2011). The consistency between treatments provided to nursing facility residents and orders on the physician orders for life-sustaining treatment form. Journal of the American Geriatrics Society , 59, 2091– 2099. doi: 10.1111/j.1532-5415.2011.03656.x Google Scholar CrossRef Search ADS   Hinderer, K. A., Friedmann, E., Fins, J. J. ( 2015). Withdrawal of life-sustaining treatment: patient and proxy agreement. Dimensions of Critical Care Nursing , 34, 91– 99. doi: 10.1097/DCC.0000000000000097 Google Scholar CrossRef Search ADS   Institute of Medicine. ( 2014). Dying in America: Improving quality and honoring individual preferences near the end of life . Washington, DC: National Academies Press. doi: 10.17226/18748 Javier, A. D., Figueroa, R., Siew, E. D., Salat, H., Morse, J., Stewart, T. G.,… Abdel-Kader, K. ( 2017). Reliability and utility of the surprise question in CKD stages 4 to 5. American Journal of Kidney Diseases: The Official Journal of the National Kidney Foundation , 70, 93– 101. doi: 10.1053/j.ajkd.2016.11.025 Google Scholar CrossRef Search ADS   Jones, J. M., McPherson, C. J., Zimmermann, C., Rodin, G., Le, L. W., & Cohen, S. R. ( 2011). Assessing agreement between terminally ill cancer patients’ reports of their quality of life and family caregiver and palliative care physician proxy ratings. Journal of Pain and Symptom Management , 42, 354– 365. doi: 10.1016/j.jpainsymman.2010.11.018 Google Scholar CrossRef Search ADS   Kaesmann, L., Janssen, S., Schild, S. E., & Rades, D. ( 2016). Value of comorbidity scales for predicting survival after radiochemotherapy of small cell lung cancer. Lung , 194, 295– 298. doi: 10.1007/s00408-016-9857-4 Google Scholar CrossRef Search ADS   Kim, H., Ersek, M., Bradway, C., & Hickman, S. E. ( 2015). Physician orders for life-sustaining treatment for nursing home residents with dementia. Journal of the American Association of Nurse Practitioners , 27, 606– 614. doi: 10.1002/2327-6924.12258 Google Scholar CrossRef Search ADS   Kirou-Mauro, A., Harris, K., Sinclair, E., Selby, D., & Chow, E. ( 2006). Are family proxies a valid source of information about cancer patients’ quality of life at the end-of-life? A literature review. Journal of Cancer Pain & Symptom Palliation , 2, 23– 33. doi: 10.1300/J427v02n02_04 Google Scholar CrossRef Search ADS   Kutner, J. S., Bryant, L. L., Beaty, B. L., & Fairclough, D. L. ( 2006). Symptom distress and quality-of-life assessment at the end of life: The role of proxy response. Journal of Pain and Symptom Management , 32, 300– 310. DOI: 10.1016/j.jpainsymman.2006.05.009 Google Scholar CrossRef Search ADS   Li, M., Harris, I., and Lu, Z. K. ( 2015). Differences in proxy-reported and patient-reported outcomes: Assessing health and functional status among medicare beneficiaries. BMC Medical Research Methodology , 15, 62. doi: 10.1186/s12874-015-0053-7 Google Scholar CrossRef Search ADS   Lorenzon, L., Costa, G., Massa, G., Frezza, B., Stella, F., & Balducci, G. ( 2017). The impact of frailty syndrome and risk scores on emergency cholecystectomy patients. Surgery Today , 47, 74– 83. doi: 10.1007/s00595-016-1361-1 Google Scholar CrossRef Search ADS   Massachusetts Medical Orders for Life Sustaining Treatment. ( 2014). About MOLST. Accessed October 22, 2017http://molst-ma.org/about Moss, A. H., Ganjoo, J., Sharma, S., Gansor, J., Senft, S., Weaner, B.,… Schmidt, R. ( 2008). Utility of the “surprise” question to identify dialysis patients with high mortality. Clinical Journal of the American Society of Nephrology: CJASN , 3, 1379– 1384. doi: 10.2215/CJN.00940208 Google Scholar CrossRef Search ADS   National POLST Paradigm. ( 2017a). Programs in your state. Accessed October 22, 2017http://polst.org/programs-in-your-state/ National POLST Paradigm. ( 2017b). POLST & advance directives. Accessed December 17, 2017http://polst.org/advance-care-planning/polst-and-advance-directives/ Novella, J. L., Jochum, C., Jolly, D., Morrone, I., Ankri, J., Bureau, F., & Blanchard, F. ( 2001). Agreement between patients’ and proxies’ reports of quality of life in Alzheimer’s disease. Quality of Life Research , 10, 443– 452. doi: 10.1023/A:1012522013817 Google Scholar CrossRef Search ADS   Oechsle, K., Goerth, K., Bokemeyer, C., & Mehnert, A. ( 2013). Symptom burden in palliative care patients: Perspectives of patients, their family caregivers, and their attending physicians. Supportive Care in Cancer , 21, 1955– 1962. doi: 10.1007/s00520-013-1747-1 Google Scholar CrossRef Search ADS   Ostbye, T., Taylor, D. H. Jr, Clipp, E. C., Scoyoc, L. V., & Plassman, B. L. ( 2008). Identification of dementia: Agreement among national survey data, medicare claims, and death certificates. Health Services Research , 43( 1 Pt 1), 313– 326. doi: 10.1111/j.1475-6773.2007.00748.x Powell, R., Johnston, M., & Johnston, D. W. ( 2007). Assessing walking limitations in stroke survivors: Are self-reports and proxy-reports interchangeable? Rehabilitation Psychology , 52, 177– 183. doi: 10.1037/0090-5550.52.2.177 Google Scholar CrossRef Search ADS   Pruchno, R. A., Lemay, E. P. Jr, Feild, L., & Levinsky, N. G. ( 2005). Spouse as health care proxy for dialysis patients: Whose preferences matter? The Gerontologist , 45, 812– 819. doi: 10.1093/geront/45.6.812 Google Scholar CrossRef Search ADS   Rahman, A. N., Bressette, M., Gassoumis, Z. D., & Enguidanos, S. ( 2016). Nursing home residents’ preferences on physician orders for life sustaining treatment. The Gerontologist , 56, 714– 722. doi: 10.1093/geront/gnv019 Google Scholar CrossRef Search ADS   Reamy, A. M., Kim, K., Zarit, S. H., & Whitlatch, C. J. ( 2011). Understanding discrepancy in perceptions of values: Individuals with mild to moderate dementia and their family caregivers. The Gerontologist , 51, 473– 483. doi: 10.1093/geront/gnr010 Google Scholar CrossRef Search ADS   Richardson, D. K., Fromme, E., Zive, D., Fu, R., & Newgard, C. D. ( 2014). Concordance of out-of-hospital and emergency department cardiac arrest resuscitation with documented end-of-life choices in Oregon. Annals of Emergency Medicine , 63, 375– 383. doi: 10.1016/j.annemergmed.2013.09.004 Google Scholar CrossRef Search ADS   Schmidt, T. A., Hickman, S. E., Tolle, S. W., & Brooks, H. S. ( 2004). The physician orders for life-sustaining treatment program: Oregon emergency medical technicians’ practical experiences and attitudes. Journal of the American Geriatrics Society , 52, 1430– 1434. doi: 10.1111/j.1532-5415.2004.52403.x Google Scholar CrossRef Search ADS   Schmidt, T. A., Zive, D., Fromme, E. K., Cook, J. N., & Tolle, S. W. ( 2014). Physician orders for life-sustaining treatment (POLST): Lessons learned from analysis of the Oregon POLST Registry. Resuscitation , 85, 480– 485. doi: 10.1016/j.resuscitation.2013.11.027 Google Scholar CrossRef Search ADS   Seckler, A. B., Meier, D. E., Mulvihill, M., & Paris, B. E. ( 1991). Substituted judgment: How accurate are proxy predictions? Annals of Internal Medicine , 115, 92– 98. doi: 10.7326/0003-4819-115-2-92 Google Scholar CrossRef Search ADS   Shalowitz, D. I., Garrett-Mayer, E., & Wendler, D. ( 2006). The accuracy of surrogate decision makers: A systematic review. Archives of Internal Medicine , 166, 493– 497. doi: 10.1001/archinte.166.5.493 Google Scholar CrossRef Search ADS   StataCorp. 2013. ( 2013). Stata statistical software release 13.1 . College Station, TX: StataCorp LP. St-Louis, E., Iqbal, S., Feldman, L. S., Sudarshan, M., Deckelbaum, D. L., Razek, T. S., & Khwaja, K. ( 2015). Using the age-adjusted Charlson comorbidity index to predict outcomes in emergency general surgery. The Journal of Trauma and Acute Care Surgery , 78, 318– 323. doi: 10.1097/TA.0000000000000457 Google Scholar CrossRef Search ADS   Suhl, J., Simons, P., Reedy, T., & Garrick, T. ( 1994). Myth of substituted judgment. Surrogate decision making regarding life support is unreliable. Archives of Internal Medicine , 154, 90– 96. doi: 10.1001/archinte.154.1.90 Google Scholar CrossRef Search ADS   Tarzian, A. J., & Cheevers, N. B. ( 2017). Maryland’s medical orders for life-sustaining treatment form use: Reports of a statewide survey. Journal of Palliative Medicine , 20, 939– 945. doi: 10.1089/jpm.2016.0440 Google Scholar CrossRef Search ADS   Tolle, S. W., Tilden, V. P., Nelson, C. A., & Dunn, P. M. ( 1998). A prospective study of the efficacy of the physician order form for life-sustaining treatment. Journal of the American Geriatrics Society , 46, 1097– 1102. doi: 10.1111/j.1532-5415.1998.tb06647.x Google Scholar CrossRef Search ADS   Torke, A. M., Alexander, G. C., & Lantos, J. ( 2008). Substituted judgment: The limitations of autonomy in surrogate decision making. Journal of General Internal Medicine , 23, 1514– 1517. doi: 10.1007/s11606-008-0688-8 Google Scholar CrossRef Search ADS   Torke, A. M., Moloney, R., Siegler, M., Abalos, A., & Alexander, G. C. ( 2010). Physicians’ views on the importance of patient preferences in surrogate decision-making. Journal of the American Geriatrics Society , 58, 533– 538. doi: 10.1111/j.1532-5415.2010.02720.x Google Scholar CrossRef Search ADS   Torke, A. M., Sachs, G. A., Helft, P. R., Montz, K., Hui, S. L., Slaven, J. E., & Callahan, C. M. ( 2014). Scope and outcomes of surrogate decision making among hospitalized older adults. JAMA Internal Medicine , 174, 370– 377. doi: 10.1001/jamainternmed.2013.13315 Google Scholar CrossRef Search ADS   Torke, A. M., Siegler, M., Abalos, A., Moloney, R. M., & Alexander, G. C. ( 2009). Physicians’ experience with surrogate decision making for hospitalized adults. Journal of General Internal Medicine , 24, 1023– 1028. doi: 10.1007/s11606-009-1065-y Google Scholar CrossRef Search ADS   Tremblay, D., Arnsten, J. H., & Southern, W. N. ( 2016). A simple and powerful risk-adjustment tool for 30-day mortality among inpatients. Quality Management in Health Care , 25, 123– 128. doi: 10.1097/QMH.0000000000000096 Google Scholar CrossRef Search ADS   Vick, J., Pertsch, N., Hutchings, M., Neville, B., & Bernacki, R. ( 2016). The utility of the surprise question in identifying patients most at risk of death [Abstract]. Journal of Pain and Symptom Management , 51, 342. doi: 10.1016/j.jpainsymman.2015.12.177 Google Scholar CrossRef Search ADS   Wilkinson, D. J., & Truog, R. D. ( 2013). The luck of the draw: Physician-related variability in end-of-life decision-making in intensive care. Intensive Care Medicine , 39, 1128– 1132. doi: 10.1007/s00134-013-2871-6 Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

The GerontologistOxford University Press

Published: May 17, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off