Are Advance Directives Associated with Better Hospice Care? Kevin Ache, DO,* Joan Harrold, MD,† Pamela Harris, MD,‡ Meredith Dougherty, MS,§ and David Casarett, MD, MA§
Key words: advance directive; hospice OBJECTIVES: To describe individuals with advance directives at the time of hospice enrollment and to determine whether they have patterns of care and outcomes that are different from those of individuals without advance directives. DESIGN: Electronic health record–based retrospective cohort study with propensity score–adjusted analysis. SETTING: Three hospice programs in the United States. PARTICIPANTS: Individuals admitted to hospice between January 1, 2008, and May 15, 2012 (N = 49,370). MEASUREMENTS: Timing of hospice enrollment before death, rates of voluntary withdrawal from hospice, and site of death. RESULTS: Most participants (35,968, 73%) had advance directives at the time of hospice enrollment. These participants were enrolled in hospice longer (median 29 vs 15 days) and had longer survival times before death (adjusted hazard ratio = 0.62; 95% confidence interval (CI) 0.58–0.66; P < .001). They were less likely to die within the first week after hospice enrollment (24.3% vs 33.2%; adjusted odds ratio (aOR) = 0.83, 95% CI = 0.78–0.88; P < .001). Participants with advance directives were less likely to leave hospice voluntarily (2.2% vs 3.4%; aOR = 0.82, 95% CI = 0.74–0.90; P = .003) and more likely to die at home or in a nursing home than in an inpatient unit (15.3% vs 25.8%; aOR = 0.82, 95% CI = 0.77–0.87; P < .001). CONCLUSION: Participants with advance directives were enrolled in hospice for a longer period of time before death than those without and were more likely to die in the setting of their choice. J Am Geriatr Soc 62:1091– 1096, 2014.
From the *Physician Services, Suncoast Hospice, Clearwater, Florida; † Hospice and Community Care, Lancaster, Pennsylvania; ‡Kansas City Hospice and Palliative Care, Kansas City, Missouri; and §Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pennsylvania. Address correspondence to Dr. David Casarett, University of Pennsylvania, 3615 Chestnut Street, Philadelphia, PA 19104. E-mail: [email protected]
JAGS 62:1091–1096, 2014 © 2014, Copyright the Authors Journal compilation © 2014, The American Geriatrics Society
dvance directives are legal documents that allow people to communicate their preferences about medical care to family, friends, and healthcare professionals in the event that they are unable to make those decisions themselves. The Patient Self-Determination Act of 1990 required entities that participate in Medicare or Medicaid to advise individuals of their rights to make healthcare decisions using advance directives.1–6 Since then, advance directives have been advocated widely as a way to enable individuals to shape the course of their medical care.7–11 There is growing evidence that advance directives are associated with less-aggressive treatment and better outcomes in a variety of healthcare settings. For example, individuals with heart failure who have advance directives may receive less-aggressive treatment,12 and families of individuals with dementia may be more satisfied with care when they spend more time engaged in advance care planning discussions.13 Advance care planning has also been shown to be beneficial for children with life-limiting conditions, helping families to contemplate and articulate their goals,14 but it is not known whether advance directives are associated with better hospice care. For instance, it is not known whether individuals with advance directives enroll in hospice earlier. Nor is it known whether individuals who come to hospice with an advance directive have different trajectories of care or different outcomes. These questions are important because more than 1.65 million individuals use hospice every year.15 Even small differences in care for individuals with advance directives could affect large numbers of individuals. Therefore, the goals of this study were to describe the population of individuals who have an advance directive at the time of hospice enrollment and to determine whether these individuals have patterns of care and outcomes that are different from those of individuals without advance directives.
METHODS Participant data were extracted from the electronic medical records (EMRs) of three hospices in the Coalition of
ACHE ET AL.
Hospices Organized to Investigate Comparative Effectiveness (CHOICE) network.16 CHOICE is a research-focused collaborative of hospices that all use common EMR software and have agreed to share their data for research purposes. A steering committee comprising leaders from all hospices in the network defines and approves CHOICE projects. Hospices participating in this study range in size from 400 to 2,700 individuals per day and are located in Florida, California, and Texas. All are nonprofit. CHOICE extracts data from a warehouse that participating hospices use for tracking, quality measurement, and benchmarking. Warehouse data reside on a secure server. Extracted data were stripped of identifiers to create a limited dataset that was compliant with the Health Insurance Portability and Accountability Act and was transferred as an encrypted file to the University of Pennsylvania for analysis. Individuals were included if they were admitted to a participating hospice between January 1, 2008, and May 15, 2012. Advance directives documented in participants’ electronic medical record at the time of hospice enrollment were identified. Living wills and durable powers of attorney for health care were included. Advance directives that were described as being “in process” or “unsigned” were not counted. Do-not-resuscitate (DNR) orders were counted as being distinct from advance directives. Although both reflect individual preferences, the former are legally binding orders that a physician has written, whereas the latter are individual-generated guides for care. Next, information on basic demographic variables (age, sex, race) and diagnoses (admitting diagnosis and up to three additional diagnoses) was extracted. Coding also included site of care at time of enrollment (e.g., home, long-term care facility, hospital, hospice inpatient unit). Clinical data elements that were markers of the severity of the illness and complexity of care (e.g., presence of pain, use of oxygen, and presence of intravenous access) were extracted. Finally, Palliative Performance Scale (PPS) scores were extracted for each participant. The PPS is an 11-point scale (scored 0–100 in 10-point increments) on which higher numbers indicate better function.17 It assesses five domains: ambulation (bedbound to full); activity (unable to work to normal); self-care (completely dependent to completely independent); intake (mouth care only to full diet); level of consciousness (drowsy or coma to fully alert). Scoring proceeds in this order so that the first categories (ambulation, activity) are given the greatest weight. For ease of interpretation in calculating predicted survivals, PPS scores were grouped into three categories (0–20, 30–40, 50–100) based on previous studies of prognosis in individuals in hospice.18,19 Next, a propensity score was calculated to account for nonrandom assignment between groups by building a logistic regression model in which the outcome was the presence of an advance directive at the time of hospice enrollment. A limited model was first built that included only participant characteristics from Table 1 that were independently associated with group assignment. A forward stepwise approach was used, and all variables that had an association that reached at least a moderate level of significant (P < .25) were considered.20
JUNE 2014–VOL. 62, NO. 6
The final model was used to predict the presence of an advance directive. The balance between the two groups was checked after the propensity score was adjusted for. Next, characteristics we added to the model, which was rechecked until adjustment by the propensity score balanced all characteristics in Table 1 (P > .25). This approach was used rather than propensity score matching because the majority of participants had advance directives. Therefore, matching would have decreased the sample size if acceptable matches could be found only for some individuals with advance directives. Using the propensity score to adjust for confounders allows the use of all observations in the data set, which is valuable when the sample size is small or group sizes are uneven. This approach is similar to that of multivariable regression analysis but offers several advantages.21 First, unlike multivariable adjustment, propensity score adjustment allows for a check of balancing. Second, as in a randomized controlled trial, propensity score adjustment allows the separation of a study’s design (e.g., balancing of groups) from its analysis. Third, propensity score analysis makes it possible to determine when little or no overlap exists between two groups, which in turn suggests that any comparison is likely to be problematic. Finally, propensity score adjustment offers greater efficiency, because several variables can be combined into a single score, which is also beneficial in small data sets. In models that included the propensity score, three outcomes of interest were evaluated, using robust jackknife standard errors, clustered according to hospice. First, the timing of enrollment in hospice relative to death was examined. Kaplan–Meier survival curves with the log rank test were used to evaluate unadjusted differences in survival, and a Cox proportional hazards model was used to examine propensity score–adjusted differences in survival between participants with and without advance directives. The proportion of participants with short (