[Skip to Navigation]
Sign In
Figure. Geographic Variation in Advance Directives
Figure. Geographic Variation in Advance Directives

The unadjusted proportion of decedents with advance directives across hospital referral regions in the United States. Of the 3302 Health and Retirement Study participants who died between 1998 and 2007, 454 lived in low-spending regions (median, $8787; range, $7252-$9707), 1847 in medium-spending regions (median, $10 848; range $10 242-$12 404), 1001 in high-spending regions (median, $15 744; range, $12 446-$29 797). Error bars indicate 95% CIs. Any advance directive includes decedents with a written advance directive or durable power of attorney. Treatment-limiting advance directives are written documents specifying limitations in the types of treatment provided.

Table 1. Characteristics of Decedents by Treatment-Limiting Advance Directive Statusa
Table 1. Characteristics of Decedents by Treatment-Limiting Advance Directive Statusa
Table 2. Decedent Characteristics by Level of Regional End-of-Life Spendinga
Table 2. Decedent Characteristics by Level of Regional End-of-Life Spendinga
Table 3. Regression-Adjusted Total Medicare Spending in Last 6 Months of Life: Adjusted Association With Regional Spending Levels and Advance Directive Usea
Table 3. Regression-Adjusted Total Medicare Spending in Last 6 Months of Life: Adjusted Association With Regional Spending Levels and Advance Directive Usea
Table 4. Predicted Probability of Treatments in the Last 6 Months of Life: Adjusted Association With Regional Spending Levels and Advance Directive Usea
Table 4. Predicted Probability of Treatments in the Last 6 Months of Life: Adjusted Association With Regional Spending Levels and Advance Directive Usea
1.
Riley GF, Lubitz JD. Long-term trends in Medicare payments in the last year of life.  Health Serv Res. 2010;45(2):565-57620148984PubMedGoogle ScholarCrossref
2.
Wennberg JE, Cooper M. The Dartmouth Atlas of Health Care. http://www.dartmouthatlas.org. Accessed January 3, 2011
3.
Barnato AEH, Herndon MB, Anthony DL,  et al.  Are regional variations in end-of-life care intensity explained by patient preferences?: a study of the US Medicare population.  Med Care. 2007;45(5):386-39317446824PubMedGoogle ScholarCrossref
4.
Silveira MJ, Kim SYH, Langa KM. Advance directives and outcomes of surrogate decision making before death.  N Engl J Med. 2010;362(13):1211-121820357283PubMedGoogle ScholarCrossref
5.
Pritchard RS, Fisher ES, Teno JM,  et al; SUPPORT Investigators.  Influence of patient preferences and local health system characteristics on the place of death.  J Am Geriatr Soc. 1998;46(10):1242-12509777906PubMedGoogle Scholar
6.
Barnato AE, Anthony DL, Skinner J, Gallagher PM, Fisher ES. Racial and ethnic differences in preferences for end-of-life treatment.  J Gen Intern Med. 2009;24(6):695-70119387750PubMedGoogle ScholarCrossref
7.
Teno JM, Lynn J, Wenger NS,  et al; SUPPORT Investigators.  Advance directives for seriously ill hospitalized patients: effectiveness with the patient self-determination act and the SUPPORT intervention.  J Am Geriatr Soc. 1997;45(4):500-5079100721PubMedGoogle Scholar
8.
Taylor JS, Heyland DK, Taylor SJ. How advance directives affect hospital resource use: systematic review of the literature.  Can Fam Physician. 1999;45:2408-241310540700PubMedGoogle Scholar
9.
Fagerlin A, Schneider CE. Enough: the failure of the living will.  Hastings Cent Rep. 2004;34(2):30-4215156835PubMedGoogle ScholarCrossref
10.
Kelley AS, Ettner SL, Morrison RS, Du Q, Wenger NS, Sarkisian CA. Determinants of medical expenditures in the last 6 months of life.  Ann Intern Med. 2011;154(4):35-24221320939PubMedGoogle ScholarCrossref
11.
Juster FT, Suzman R. An overview of the Health and Retirement Study.  J Hum Resour. 1995;30(5):S7-S56Google ScholarCrossref
12.
Barnato AE, Farrell MH, Chang CC, Lave JR, Roberts MS, Angus DC. Development and validation of hospital “end-of-life” treatment intensity measures.  Med Care. 2009;47(10):1098-110519820614PubMedGoogle ScholarCrossref
13.
Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data.  Med Care. 1998;36(1):8-279431328PubMedGoogle ScholarCrossref
14.
The Dartmouth Atlas of Health Care.  Data by region. http://www.dartmouthatlas.org/data/region/. Accessed 20 September 2010
15.
Manning WG, Basu A, Mullahy J. Generalized modeling approaches to risk adjustment of skewed outcomes data.  J Health Econ. 2005;24(3):465-48815811539PubMedGoogle ScholarCrossref
16.
Greene WH. Econometric Analysis. 6th ed. Upper Saddle River, NJ: Prentice–Hall; 2008
17.
Zhang B, Wright AA, Huskamp HA,  et al.  Health care costs in the last week of life: associations with end-of-life conversations.  Arch Intern Med. 2009;169(5):480-48819273778PubMedGoogle ScholarCrossref
18.
Wright AA, Zhang B, Ray A,  et al.  Associations between end-of-life discussions, patient mental health, medical care near death, and caregiver bereavement adjustment.  JAMA. 2008;300(14):1665-167318840840PubMedGoogle ScholarCrossref
19.
Teno JM, Lynn J, Connors AF Jr,  et al; SUPPORT Investigators.  The illusion of end-of-life resource savings with advance directives.  J Am Geriatr Soc. 1997;45(4):513-5189100723PubMedGoogle Scholar
20.
Curtis JR, Patrick DL, Engelberg RA, Norris K, Asp C, Byock I. A measure of the quality of dying and death: initial validation using after-death interviews with family members.  J Pain Symptom Manage. 2002;24(1):17-3112183092PubMedGoogle ScholarCrossref
21.
Teno JM, Gruneir A, Schwartz Z, Nanda A, Wetle T. Association between advance directives and quality of end-of-life care: a national study.  J Am Geriatr Soc. 2007;55(2):189-19417302654PubMedGoogle ScholarCrossref
22.
Christakis NA, Iwashyna TJ. The health impact of health care on families: a matched cohort study of hospice use by decedents and mortality outcomes in surviving, widowed spouses.  Soc Sci Med. 2003;57(3):465-47512791489PubMedGoogle ScholarCrossref
23.
Sudore RL, Fried TR. Redefining the “planning” in advance care planning: preparing for end-of-life decision making.  Ann Intern Med. 2010;153(4):256-26120713793PubMedGoogle ScholarCrossref
24.
Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending, I: the content, quality, and accessibility of care.  Ann Intern Med. 2003;138(4):273-28712585825PubMedGoogle ScholarCrossref
25.
Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending, II: health outcomes and satisfaction with care.  Ann Intern Med. 2003;138(4):288-29812585826PubMedGoogle ScholarCrossref
26.
Xu JQ, Kochanek KD, Murphy SL, Tejada-Vera B. Deaths: final data for 2007.  Natl Vital Stat Rep. 2010;58(19):1-135Google Scholar
27.
Bach PB, Schrag D, Begg CB. Resurrecting treatment histories of dead patients: a study design that should be laid to rest.  JAMA. 2004;292(22):2765-277015585737PubMedGoogle ScholarCrossref
28.
Fried TR, Redding CA, Robbins ML, O’Leary JR, Iannone L. Agreement between older persons and their surrogate decision-makers regarding participation in advance care planning.  J Am Geriatr Soc. 2011;59(6):1105-110921649619PubMedGoogle ScholarCrossref
29.
McPherson CJ, Addington-Hall JM. Judging the quality of care at the end of life: can proxies provide reliable information?  Soc Sci Med. 2003;56(1):95-10912435554PubMedGoogle ScholarCrossref
Original Contribution
October 5, 2011

Regional Variation in the Association Between Advance Directives and End-of-Life Medicare Expenditures

Author Affiliations

Author Affiliations: Institute for Social Research (Drs Nicholas, Langa, Iwashyna, and Weir), University of Michigan; Department of Medicine, University of Michigan Medical School, and Veterans Affairs Health Services Research and Development Center of Excellence (Drs Langa and Iwashyna), Ann Arbor.

JAMA. 2011;306(13):1447-1453. doi:10.1001/jama.2011.1410
Abstract

Context It is unclear if advance directives (living wills) are associated with end-of-life expenditures and treatments.

Objective To examine regional variation in the associations between treatment-limiting advance directive use, end-of-life Medicare expenditures, and use of palliative and intensive treatments.

Design, Setting, and Patients Prospectively collected survey data from the Health and Retirement Study for 3302 Medicare beneficiaries who died between 1998 and 2007 linked to Medicare claims and the National Death Index. Multivariable regression models examined associations between advance directives, end-of-life Medicare expenditures, and treatments by level of Medicare spending in the decedent's hospital referral region.

Main Outcome Measures Medicare expenditures, life-sustaining treatments, hospice care, and in-hospital death over the last 6 months of life.

Results Advance directives specifying limits in care were associated with lower spending in hospital referral regions with high average levels of end-of-life expenditures (−$5585 per decedent; 95% CI, −$10 903 to −$267), but there was no difference in spending in hospital referral regions with low or medium levels of end-of-life expenditures. Directives were associated with lower adjusted probabilities of in-hospital death in high- and medium-spending regions (−9.8%; 95% CI, −16% to −3% in high-spending regions; −5.3%; 95% CI, −10% to −0.4% in medium-spending regions). Advance directives were associated with higher adjusted probabilities of hospice use in high- and medium-spending regions (17%; 95% CI, 11% to 23% in high-spending regions, 11%; 95% CI, 6% to 16% in medium-spending regions), but not in low-spending regions.

Conclusion Advance directives specifying limitations in end-of-life care were associated with significantly lower levels of Medicare spending, lower likelihood of in-hospital death, and higher use of hospice care in regions characterized by higher levels of end-of-life spending.

End-of-life health care is a frequent target for efforts to control Medicare spending. In 2006, treatment for patients in their last year of life accounted for more than one-quarter of Medicare spending.1 Marked geographic variation in Medicare end-of-life spending is well documented,2 and this variation is believed to be driven by physician practice style rather than by differences in patients' preferences for aggressiveness of treatment at the end of life.3 There is concern that this expensive care may have limited clinical effectiveness and may be contrary to what patients want. Surveys report that many patients do not wish to receive aggressive treatment at the end of their lives; however, these preferences are often undocumented.4,5 A national study by Barnato and colleagues6 found that 42% of white Medicare beneficiaries worried about receiving too much care at the end-of-life, whereas an equal proportion worried about receiving too little.

Patients can use advance directives to document their preferences for the use or avoidance of life-sustaining treatments (living wills) or to appoint a surrogate to make treatment decisions should they become no longer competent to make those decisions (durable power of attorney for health care). Although advance directives have become more common in the past few decades, evidence is mixed on whether they change the course of treatment provided near the end-of-life.7-10

It is sometimes overlooked that an advance directive can only influence treatment when the patient's wishes are inconsistent with what would be provided absent an advance directive. The wide variation in end-of-life Medicare expenditures across geographic regions suggests that default treatment levels also vary regionally. Advance directives specifying limits at the end-of-life may have their greatest impact in regions where the norms are to provide very high-intensity end-of-life treatment. Given that there exist regions with consistently high levels of spending at the end of life that may be of limited benefit and contrary to patient preferences, we used nationally representative survey data from the Health and Retirement Study (HRS) linked to respondents' Medicare claims to examine the relationship of advance directives with the cost and aggressiveness of end-of-life treatment in geographic regions across the United States characterized by high, medium, and low average end-of-life expenditures.

Methods
Study Population

We analyzed survey and Medicare claims data for HRS respondents who died between 1998 and 2007 at age 65 years or older or after qualifying for Medicare through disability or end-stage renal disease. The HRS is a large, nationally representative panel survey that conducts biennial interviews of older US residents.11 The HRS also conducts an interview with a proxy informant, typically next-of-kin, after the respondent's death. During the postmortem interview, informants are asked about the decedent's experience at the end of life, including the nature and type of their advance directives. Oral informed consent was obtained from participants and proxies in the original study. The institutional review board of the University of Michigan approved the study protocol.

Because our primary interest was in end-of-life Medicare spending and use, we limited our analytic sample to patients continuously enrolled in fee-for-service Medicare during the last 6 months of life. We used data on deaths through 2007, which were the most recently available data linked to Medicare claims.

Advance Directives

We measured the presence and type of advance directive from interviews with respondents' next-of-kin conducted after death. For living wills, informants were asked: “Did [first name] provide written instructions about the treatment or care [she/he] wanted to receive during the final days of [her/his] life?” If respondents had a written advance directive in place, additional questions were asked about the type of written instructions. Our analysis included those who indicated living wills limiting the type of care provided and those requesting all care. These informant reports have been used to document advance directive status previously.4 For durable power of attorney, informants were asked: “Did [first name] make any legal arrangements for a specific person or persons to make decisions about [his/her] care or medical treatment if [he/she] could not make those decisions [himself/herself]?”

Medicare Claims

During biennial interviews, HRS respondents were asked to provide their Medicare number and consent to the release of their claims for research purposes. For each decedent, we calculated Medicare spending in the last 6 months of life across all care settings (inpatient, outpatient, carrier, durable medical, hospice, home health, and skilled nursing). All spending measures were adjusted to 2007 US dollars using the medical consumer price index.

End-of-life hospital treatment is a major driver of end-of-life expense, and a setting in which many aggressive procedures to sustain life are performed. We used MedPAR and hospice files to identify all hospitalizations and hospice use in the last 6 months of life. We used International Classification of Diseases, Ninth Revision, Clinical Modification procedure codes to identify a set of life-sustaining interventions that have been previously used as measures of end-of-life treatment intensity (intubation and mechanical ventilation, tracheostomy, gastrostomy tube placement, hemodialysis, enteral and parenteral nutrition, and cardiopulmonary resuscitation).12 We examined any hospice use at the end-of-life as a measure of palliative care. We assessed comorbidities in the year preceding the last 6 months of life using the Elixhauser method for inpatient data.13

Regional Intensity of End-of-Life Care

We used hospital referral region measures of average Medicare spending per decedent in the last 6 months of life reported by the Dartmouth Atlas of Health Care averaged across the years 1999 through 2005.14 We characterized regions by quartiles of end-of-life spending averaged across the 7-year period. Decedent's region intensity was determined by zip code of residence.

Statistical Analysis

We used generalized linear models with a log link function and gamma distribution to fit skewed Medicare end-of-life expenditure data.15 Multivariable regression models assessed the relationships between hospital referral region end-of-life treatment intensity, advance directives specifying limits in treatment, and Medicare expenditures in the last 6 months of life. We first estimated the regressions typically reported in the literature, which assume a constant relationship between advance directives and individual end-of-life resource use. We next allowed the relationship between treatment-limiting advance directives and spending to vary for patients in high-, medium-, and low-spending regions by including interactions between treatment-limiting advance directives and region spending type. Regressions also controlled for sex, self-reported nonwhite race (due to the sample size, we collapsed respondents reporting black and other race relative to the reference group of white decedents), 5-year age categories, whether the respondent was in the highest tertile of household wealth at the last interview, indicators for having less than a high school education and a high school degree (relative to at least some college), being widowed or single or separated or divorced (vs married), self-reported chronic conditions in the interview prior to the last 6 months of life, Elixhauser comorbidities calculated from hospitalizations in the 12-month period before the last 6 months of life, and a linear time trend.

We tested for differential effects of treatment-limiting advance directives across regions by calculating average marginal effects of advance directives in low-, medium-, and high-spending regions and by conducting nonlinear Wald tests16 of the equality of spending on decedents at the end of life with and without treatment-limiting advance directives within and across levels of region spending. Two-sided significance testing was used throughout, and a P value of .05 was considered statistically significant.

We estimated logistic regressions of life-sustaining treatments, in-hospital death, and hospice use among decedents hospitalized during the last 6 months of life controlling for patient characteristics described above to test their relationship between treatment-limiting advance directives and regional spending.

Our sensitivity analyses included examining components of Medicare spending, estimating models for decedents with and without at least 1 end-of-life hospitalization separately, and analyses stratifying by cause of death using 15-cause recodes from the National Death Index. We used SAS version 9.2 (SAS Institute Inc, Cary, North Carolina) for data manipulation and Stata 11 MP (StataCorp, College Station, Texas) for analyses.

Results

A total of 4761 HRS participants who died between 1998 and 2007 (82% of 5810 decedents with postmortem interviews) also had linked Medicare data. To accurately account for end-of-life health care use, we excluded 1445 decedents with any managed care enrollment during the last 6 months of life and excluded 14 decedents who reported military coverage who might have received care through Veterans Affairs facilities.

Our cohort included 3302 decedents. Their mean age at death was 82.8 years. Fifty-six percent were women; 70% were hospitalized at least once in the last 6 months of life; 41% died in a hospital; 61% had either a living will or written durable power of attorney; and 39% completed a written, treatment-limiting advance directive. Table 1 presents the characteristics of individuals with and without a treatment-limiting advance directive. Median Medicare spending in the last 6 months of life did not vary by treatment-limiting advance directive status. In unadjusted comparisons, those with treatment-limiting advance directives had lower rates of life-sustaining treatment (34% vs 39%, P = .002), lower rates of in-hospital death (37% vs 43%, P < .001), and higher rates of hospice use (40% vs 26%, P < .001). Those with advance directives were more likely to be white, affluent, and highly educated.

We compared end-of-life care of decedents living in hospital referral regions in the lowest quartile of Medicare spending (median unadjusted spending, $8787; range, $7252 to $9707), middle half (median, $10 848; range, $10 242 to $12 404), and highest quartile (median, $15 744; range, $12 446 to $29 797; see Figure). There was substantial geographic diversity in the rates of advance directives among regions. Decedents residing in low-spending regions were more likely to have a treatment-limiting advance directive than decedents in high-spending regions (42% vs 36%; P < .001; Figure). After adjusting for demographic and socioeconomic characteristics, decedents in high-spending regions continued to face lower odds of having a treatment-limiting advance directive (odds ratio [OR], 0.69; 95% CI, 0.54 to 0.88). Although there was considerable variation in advance directive use (Figure) and median end-of-life spending (Table 2) across regions, relatively little difference existed in cause of death or comorbidity prior to the end of life (Table 2). Regression-adjusted end-of-life spending was significantly lower for decedents in low-spending regions (predicted spending, $21 741 [95% CI, $19 668 to $23 816]; difference, $4400 [95% CI, $2083 to $6717]) and higher for those in high-spending regions (predicted spending, $37 841 [95% CI, $34 855 to $40 107]; difference, $11 340 [95% CI, $8479 to $14 200]) relative to those in medium-spending regions. Spending was considerably higher for nonwhite decedents (difference, $6561; 95% CI, $3293 to $9829) and lower for decedents aged 90 years or older than younger decedents (difference, −$7871; 95% CI, −$11 212 to −$4,530). After adjusting for patient characteristics and hospital referral region-spending intensity, there was no difference in Medicare spending in the last 6 months of life for those with ($28 348; 95% CI, $26 698 to $29 999) and without advance directives ($29 352 [95% CI, $27 885 to $30 819]; difference, −$1004 [95% CI, −$3366 to $1359]) when using regressions that forced the association with advance directives to be the same for all regions.

However, there was important geographic heterogeneity in the relationship between advance directives and end-of-life spending (Table 3). In high-spending regions, adjusted spending on patients with a treatment-limiting advance directive was $33 933 (95% CI, $30 233 to $37 681), whereas adjusted spending for patients without an advance directive was $39 518 (95% CI, $35 871 to $43 167; difference, −$5585; 95% CI, −$10 903 to −$267). Having a treatment-limiting advance directive was not associated with differences in aggregate end-of-life spending for decedents in low- and medium-spending regions.

Treatment-limiting advance directives were associated with site of death and palliative care for decedents in medium- and high-spending regions (Table 4). Directives were associated with lower probabilities of in-hospital death in high- and medium-spending regions but not in low-spending regions. Thus, in high-spending regions, patients without an advance directive had a 47% adjusted probability of in-hospital death (95% CI, 44% to 51%), whereas those with an advance directive had a 38% probability of in-hospital death (95% CI, 32% to 43%; difference, −9.8%; 95% CI, −16% to −3%). The equivalent results for in-hospital death for those in medium-spending regions were 42% without an advance directive (95% CI, 39% to 45%) and 37% with an advance directive (95% CI, 33% to 41%; difference, −5.3%; 95% CI, −10% to −0.4%). In high-spending regions, patients without a limiting advance directive had a 24% adjusted probability of hospice use (95% CI, 21% to 28%), whereas those with a directive had an adjusted probability of hospice use of 41% (95% CI, 36% to 46%; difference 17%, 95% CI; 11% to 23%). Similar differences in hospice use occurred in medium-spending regions. There was no statistically significant relationship between treatment-limiting advance directive use and receipt of life-sustaining treatments during hospitalizations in the last 6 months of life, although the point estimates were in the same direction as for total end-of-life expenditures in high-spending regions, ie, lower likelihood of use of life-sustaining treatments among those with treatment-limiting advance directives (P = .11).

The differences in Medicare spending observed among those with advance directives in high-spending regions appear to be driven mainly by lower inpatient spending ($7509, 95% CI, $3404-$11 614) slightly offset by higher hospice spending ($976; 95% CI, $294-$1658; eTable 1). The differences in end-of-life spending across regions and advance directive status are concentrated among the 2384 decedents (72%) experiencing at least 1 hospitalization in the last 6 months of life (eTable 2). There was no evidence of heterogeneity of the advance directive effect in high-spending region by cause of death (P = .44 for joint test of significance; eFigure).

Nearly 4% of those with advance directives (1.5% of decedents overall) requested all care possible in their advance directive. There were too few such decedents in our study to reliably estimate the effects of such advance directives; these decedents are included in the no-limiting directive group. On average, these decedents used $8060 more care at the end-of-life (P = .02) than decedents with treatment-limiting directives.

Our results were consistent across numerous alternative specifications including an ordinary least squares regression, excluding all veterans because Veterans Affairs care is unobserved in the Medicare claims, excluding disabled decedents who are younger than 65 years but observed in the Medicare claims, excluding decedents with cancer as the cause of death, and excluding those who wrote advance directives in the last 6 months of life (eTable 3-6; eFigure).

Comment

Using linked personal interviews and Medicare claims for decedents in a large, nationally representative study, we found that advance directives specifying limits in treatment were more common in areas with lower levels of end-of-life spending. When patients in high-spending areas had advance directives limiting treatments, they averaged significantly lower end-of-life Medicare spending, were less likely to have an in-hospital death, and had significantly greater odds of hospice use than decedents without advance directives in these regions.

Our findings may reconcile prior, seemingly conflicting evidence that advance directives both reduce17,18 and do not affect10,19 end-of-life health care expenditures and use of life-sustaining treatments. We replicate the absence of a global relationship between advance directives and resource use but extend the analysis to show that treatment-limiting advance directives are associated with lower Medicare expenditures for beneficiaries living in geographic regions characterized by aggressive end-of-life care. Most studies—including the 1980s landmark Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatment (SUPPORT)7—relied on samples from a small number of hospitals, precluding such geographic comparisons.17

These results suggest both the power and limitations of advance directives. One interpretation of these data is that advance directives are most effective when one prefers treatment that is different from the local norms. Thus, in high-intensity regions, more limited treatment requires an explicit statement. Treatment-limiting advance directives were associated with increased likelihood of palliative hospice care and lower rates of in-hospital deaths in high- and medium-spending regions, where patients were most likely to receive aggressive end-of-life treatment. This has important implications for patient quality of death and well-being of family and friends. Previous research has found that next-of-kin report that the quality of death is higher for decedents who die at home20 or in hospice care21 relative to hospital and nursing home settings, that caregivers report worse overall physical and mental health following deaths characterized by use of aggressive end-of-life treatment,18 and surviving spouses of hospice decedents are less likely to die within 18 months of widowhood relative to those whose spouses did not use hospice.22

Interestingly, we found that while treatment-limiting advance directives were associated with significantly lower total end-of-life Medicare expenditures in high-spending hospital referral regions, the relationship between treatment-limiting advance directives and the receipt of aggressive life-sustaining treatments (eg, intubation and mechanical ventilation, hemodialysis, and enteral and parenteral nutrition) was less strong. This may suggest that treatment-limiting advance directives are associated with a quicker withdrawal of these aggressive and expensive interventions, even if the likelihood of initiating these treatments is less strongly affected. Given the significant consequences of these treatments for both patient and family well-being and health care costs, future research should be aimed at better understanding how advance directives, durable power of attorney provisions, and newer initiatives to improve in-the-moment end-of-life decision making23 affect the initiation and withdrawal of life-sustaining treatments.

Our results indicate a statistically and economically significant relationship between advance directives and regional practice patterns. The regional variations literature has asserted that significant savings to the Medicare population could be achieved if high-spending regions practiced more like low-spending regions.24,25 However, we note that patient preferences may also contribute to observed differences, possibly in a mutually reinforcing pattern. If an additional 6% of decedents in high-intensity regions had had treatment-limiting advance directives in place (matching preferences in low-spending regions), our estimates suggest that Medicare spending on the 790 061 beneficiaries dying in high-spending hospital referral regions in 2006 would have been $265 million lower. We urgently need studies to examine the extent to which greater advance directive use in high-intensity regions would result in treatment that is more concordant with patient preferences and to understand the patient, physician, and health system characteristics that lead to higher rates of use in low-spending regions.

Our findings should be interpreted in light of several limitations. Because we used Medicare claims data to measure use, we were unable to include decedents not in fee-for-service Medicare. The associations and impact of advance directives may be different in other populations. However, Medicare-eligible adults account for more than 70% of deaths in the United States.26 We also lacked information about decedents who had not consented to the HRS-Medicare data linkage, even though demographics of decedents in the linkage were similar to the full sample (eTable 8).

Because our study was observational, we were unable to assess causal effects of advance directive use. We focused on the last 6 months of life so were unable to address the question of how advance directives alter spending patterns over an individual's life course nor to address the extent to which they influence the length of life. Although limitations of the look-back approach to studying end-of-life intensity are well-known, this approach addresses the policy-relevant question of health care use just before death.27 Our sensitivity analyses suggest that our results are not driven by heterogeneous patient characteristics. Furthermore, we did not have patients' own reports of preferences or copies of their advance directives. However, the postmortem interviews are conducted with proxy informants, often widows or adult children, who are likely to know of advance directives if they were used; 78% were involved in decedents' end-of-life decisions. Recent studies of older adults and their surrogate decision makers support the reliability of this approach.28,29

Advance directives are associated with important differences in treatment during the last 6 months of life for patients who live in areas of high medical expenditures but not in other regions. This suggests that the clinical effect of advance directives is critically dependent on the context in which a patient receives care. Advance directives may be especially important for ensuring treatment consistent with patients' preferences for those who prefer less aggressive treatment at the end of life but are patients in systems characterized by high intensity of treatment.

Back to top
Article Information

Corresponding Author: Lauren Hersch Nicholas, PhD, MPP, 3005 Institute for Social Research, 426 Thompson St, Ann Arbor, MI 48104 (lnichola@umich.edu).

Author Contributions: Dr Nicholas had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Nicholas, Langa, Iwashyna, and Weir.

Acquisition of data: Nicholas, Langa, and Iwashyna.

Analysis and interpretation of data: Nicholas, Langa, Iwashyna, and Weir.

Drafting of the manuscript: Nicholas, Langa, Iwashyna, and Weir.

Critical revision of the manuscript for important intellectual content: Nicholas, Langa, Iwashyna, and Weir.

Statistical analysis: Nicholas.

Administrative, technical, or material support: Nicholas, Langa, Iwashyna, and Weir.

Study supervision: Nicholas.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.

Funding/Support: This work was supported by grants U01 AG09740, R01 AG030155, K08 HL091249 from the National Institutes of Health and Clinical and Translational Science Award grant UL1RR024986 from the Michigan Institute for Clinical and Health Research.

Role of the Sponsor: The National Institutes of Health had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript. The Health and Retirement Study is conducted by the University of Michigan with funding from the National Institute on Aging.

Disclaimer: The views expressed are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the US government.

Online-Only Material: The Author Video Interview is available here.

Additional Contributions: We thank Tish Shapiro, MA, for programming assistance and Morris Hamilton, BA, for research assistance. Both are employees of the University of Michigan and did not receive additional compensation for their involvement.

References
1.
Riley GF, Lubitz JD. Long-term trends in Medicare payments in the last year of life.  Health Serv Res. 2010;45(2):565-57620148984PubMedGoogle ScholarCrossref
2.
Wennberg JE, Cooper M. The Dartmouth Atlas of Health Care. http://www.dartmouthatlas.org. Accessed January 3, 2011
3.
Barnato AEH, Herndon MB, Anthony DL,  et al.  Are regional variations in end-of-life care intensity explained by patient preferences?: a study of the US Medicare population.  Med Care. 2007;45(5):386-39317446824PubMedGoogle ScholarCrossref
4.
Silveira MJ, Kim SYH, Langa KM. Advance directives and outcomes of surrogate decision making before death.  N Engl J Med. 2010;362(13):1211-121820357283PubMedGoogle ScholarCrossref
5.
Pritchard RS, Fisher ES, Teno JM,  et al; SUPPORT Investigators.  Influence of patient preferences and local health system characteristics on the place of death.  J Am Geriatr Soc. 1998;46(10):1242-12509777906PubMedGoogle Scholar
6.
Barnato AE, Anthony DL, Skinner J, Gallagher PM, Fisher ES. Racial and ethnic differences in preferences for end-of-life treatment.  J Gen Intern Med. 2009;24(6):695-70119387750PubMedGoogle ScholarCrossref
7.
Teno JM, Lynn J, Wenger NS,  et al; SUPPORT Investigators.  Advance directives for seriously ill hospitalized patients: effectiveness with the patient self-determination act and the SUPPORT intervention.  J Am Geriatr Soc. 1997;45(4):500-5079100721PubMedGoogle Scholar
8.
Taylor JS, Heyland DK, Taylor SJ. How advance directives affect hospital resource use: systematic review of the literature.  Can Fam Physician. 1999;45:2408-241310540700PubMedGoogle Scholar
9.
Fagerlin A, Schneider CE. Enough: the failure of the living will.  Hastings Cent Rep. 2004;34(2):30-4215156835PubMedGoogle ScholarCrossref
10.
Kelley AS, Ettner SL, Morrison RS, Du Q, Wenger NS, Sarkisian CA. Determinants of medical expenditures in the last 6 months of life.  Ann Intern Med. 2011;154(4):35-24221320939PubMedGoogle ScholarCrossref
11.
Juster FT, Suzman R. An overview of the Health and Retirement Study.  J Hum Resour. 1995;30(5):S7-S56Google ScholarCrossref
12.
Barnato AE, Farrell MH, Chang CC, Lave JR, Roberts MS, Angus DC. Development and validation of hospital “end-of-life” treatment intensity measures.  Med Care. 2009;47(10):1098-110519820614PubMedGoogle ScholarCrossref
13.
Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data.  Med Care. 1998;36(1):8-279431328PubMedGoogle ScholarCrossref
14.
The Dartmouth Atlas of Health Care.  Data by region. http://www.dartmouthatlas.org/data/region/. Accessed 20 September 2010
15.
Manning WG, Basu A, Mullahy J. Generalized modeling approaches to risk adjustment of skewed outcomes data.  J Health Econ. 2005;24(3):465-48815811539PubMedGoogle ScholarCrossref
16.
Greene WH. Econometric Analysis. 6th ed. Upper Saddle River, NJ: Prentice–Hall; 2008
17.
Zhang B, Wright AA, Huskamp HA,  et al.  Health care costs in the last week of life: associations with end-of-life conversations.  Arch Intern Med. 2009;169(5):480-48819273778PubMedGoogle ScholarCrossref
18.
Wright AA, Zhang B, Ray A,  et al.  Associations between end-of-life discussions, patient mental health, medical care near death, and caregiver bereavement adjustment.  JAMA. 2008;300(14):1665-167318840840PubMedGoogle ScholarCrossref
19.
Teno JM, Lynn J, Connors AF Jr,  et al; SUPPORT Investigators.  The illusion of end-of-life resource savings with advance directives.  J Am Geriatr Soc. 1997;45(4):513-5189100723PubMedGoogle Scholar
20.
Curtis JR, Patrick DL, Engelberg RA, Norris K, Asp C, Byock I. A measure of the quality of dying and death: initial validation using after-death interviews with family members.  J Pain Symptom Manage. 2002;24(1):17-3112183092PubMedGoogle ScholarCrossref
21.
Teno JM, Gruneir A, Schwartz Z, Nanda A, Wetle T. Association between advance directives and quality of end-of-life care: a national study.  J Am Geriatr Soc. 2007;55(2):189-19417302654PubMedGoogle ScholarCrossref
22.
Christakis NA, Iwashyna TJ. The health impact of health care on families: a matched cohort study of hospice use by decedents and mortality outcomes in surviving, widowed spouses.  Soc Sci Med. 2003;57(3):465-47512791489PubMedGoogle ScholarCrossref
23.
Sudore RL, Fried TR. Redefining the “planning” in advance care planning: preparing for end-of-life decision making.  Ann Intern Med. 2010;153(4):256-26120713793PubMedGoogle ScholarCrossref
24.
Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending, I: the content, quality, and accessibility of care.  Ann Intern Med. 2003;138(4):273-28712585825PubMedGoogle ScholarCrossref
25.
Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending, II: health outcomes and satisfaction with care.  Ann Intern Med. 2003;138(4):288-29812585826PubMedGoogle ScholarCrossref
26.
Xu JQ, Kochanek KD, Murphy SL, Tejada-Vera B. Deaths: final data for 2007.  Natl Vital Stat Rep. 2010;58(19):1-135Google Scholar
27.
Bach PB, Schrag D, Begg CB. Resurrecting treatment histories of dead patients: a study design that should be laid to rest.  JAMA. 2004;292(22):2765-277015585737PubMedGoogle ScholarCrossref
28.
Fried TR, Redding CA, Robbins ML, O’Leary JR, Iannone L. Agreement between older persons and their surrogate decision-makers regarding participation in advance care planning.  J Am Geriatr Soc. 2011;59(6):1105-110921649619PubMedGoogle ScholarCrossref
29.
McPherson CJ, Addington-Hall JM. Judging the quality of care at the end of life: can proxies provide reliable information?  Soc Sci Med. 2003;56(1):95-10912435554PubMedGoogle ScholarCrossref
×