Hospital Variation in Do-Not-Resuscitate Orders and End-of-Life Healthcare Use in the United States

To the Editor:

Use of health care resources by individuals at the end of life (EOL) varies widely across the United States (1). Whereas individual patients who participate in advance care planning tend to have lower rates of in-hospital death and higher rates of hospice use (2), scant information in national databases regarding patient advance directives has led to knowledge gaps regarding the associations between hospital use of advance directives and EOL healthcare use. We leveraged new International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9) (3) identifiers of patient “do-not-resuscitate” (DNR) status to characterize national variation in acute care hospital DNR orders—a common method of documenting directives for treatment limitation—across the United States, and correlated hospital DNR rates with measures of EOL healthcare use.

Methods

We used the Hospital Cost and Utilization Project, National Inpatient Sample (NIS), a representative sample of hospitalized patients in the United States during 2011 and 2012 (4). During 2011, the NIS was a 20% stratified probability sample of nonfederal acute-care hospitals, with hospital identifiers allowing linkage to the Dartmouth Atlas of Healthcare measures of hospital EOL healthcare use (http://www.dartmouthatlas.org/tools/downloads.aspx?tab=40). During 2012, the NIS eliminated hospital identifiers, but used a 20% sample of patients from nonfederal acute-care hospitals across the United States, allowing improved characterization of variation across U.S. hospitals. We used ICD-9-Clinical Modification code V49.86 (introduced October 1, 2010) to identify patient DNR status, and conducted survey-weighted analyses to identify population estimates of hospital DNR rates. Analyses included cases 65 years old or older at each U.S. hospital, excluding patients admitted to hospitals with 0 DNR orders (5%). We calculated risk-standardized hospital DNR rates for 2011 and 2012 from multivariable hierarchical logistic regression models adjusted for patient demographics, 235 Clinical Classification Codes characterizing principal reason for hospitalization, comorbidities (5), acute organ failures (6), and hospital characteristics. We summarized hospital variation in DNR orders using the median odds ratio (7), a measure of the median odds of DNR status for similar patients selected from among all possible pairs of hospitals. We abstracted a priori 12 measures of hospital EOL healthcare utilization from the Dartmouth Atlas that correspond to proposed measures of quality care at EOL (8), and used linear regression to evaluate associations between risk-standardized DNR rates and EOL health care utilization. Study procedures were deemed nonhuman subjects research by the Baystate Medical Center Institutional Review Board. Statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC; α = 0.05).

Results

We analyzed hospital variation in DNR orders among 12,057,620 (±11,3314) (survey weighted) patients aged 65 years or older discharged from 3,210 U.S. hospitals during 2012. Patients were median 77 (interquartile range = 71–84) years old, 56% were female, and 74% white. The median risk-standardized hospital rate of DNR orders was 9.2% (interquartile range = 5.2–14.8; Figure 1). The median odds ratio for DNR status between hospitals was 2.65 (95% confidence interval = 2.58–2.72). By comparison, a diagnosis of metastatic cancer was associated with 2.54-fold increased odds of patient DNR status, suggesting that the hospital to which a patient was admitted had as strong an influence on the possibility of a patient receiving a DNR order as the presence of advanced cancer.

Figure 1.

Caterpillar plot showing variation in risk-standardized do-not-resuscitate (DNR) rate (%, with 95% confidence intervals) (y axis: % of patients aged 65 yr or older with a DNR order) across hospitals in the United States. Hospitals are ranked along the x axis in order of increasing DNR rates.

We identified 1,714,289 (unweighted) patients among 330 hospitals in the 2011 NIS that were able to be linked to Dartmouth Health Atlas measures of EOL care. Higher risk-standardized hospital DNR rates were associated with lower EOL healthcare use across multiple measures, including days in the hospital at EOL (e.g., 0.20±0.04 [SE] fewer hospitals days per 1% increase in hospital DNR rate), and deaths that included an intensive care unit admission (e.g., 0.18±0.04% [SE] fewer deaths that included intensive care unit admission per 1% increase in DNR rate) (Figure 2).

Figure 2.

Association between hospital risk-adjusted do-not-resuscitate (DNR) rate and measures of end-of-life healthcare utilization; β estimate per 1% increase in hospital DNR rate (SE); P value. (A) Hospital care intensity index; β = −0.015 (0.002); P < 0.001. (B) Total Medicare part B spending ($) per patient in last 2 years of life; β = −144 (32); P < 0.001. (C) Intensive care beds per 1,000 decedents during last 2 years of life; β = −0.41 (0.09); P < 0.001. (D) Hospital reimbursement ($) per decedent in last 2 years of life; β = −142.1 (83.6); P < 0.001. (E) Hospital days per decedent; β = −0.20 (0.04); P < 0.001. (F) Intensive care days per decedent; β = −0.09 (0.02); P < 0.001. (G) Percent of deaths occurring in the hospital; β = −0.10 (0.05); P = 0.06. (H) Percent of deaths that included an intensive care unit (ICU) admission; β = −0.18 (0.04); P < 0.001. (I) Percent of patients seeing 10 or more different physicians; β = −0.38 (0.09); P < 0.001. (J) Number of different physicians seen per decedent; β = −0.09 (0.02); P < 0.001. (K) Percent of decedents enrolled in hospice; β = −0.10 (0.05); P = 0.26. (L) Hospice days per decedent; β = −0.05 (0.05); P = 0.29.

Discussion

We observed wide variation between hospitals in rates of DNR orders among older adults in the United States. The odds of patients with similar clinical and demographic characteristics having DNR status varied approximately 2.5-fold based only upon the hospital to which they were admitted. Our findings suggest that variation in hospital DNR rates has important ramifications—hospitals with higher DNR rates demonstrated a pattern of lower-intensity healthcare use at EOL and greater use of measures proposed as representative of quality EOL care (8).

Due to the lack of identifiers for orders limiting life-sustaining treatments in most clinical databases, few studies have investigated hospital variation in DNR orders and EOL treatment intensity. We were able to leverage new ICD-9 codes identifying a representative sample of patients with DNR orders across the United States. Prior studies from the state of California—which routinely collects information on patient DNR status at hospital admission—have shown similarly high levels of hospital variation in DNR orders (9) and inverse associations between hospital DNR rates and use of life-support interventions (but not measures of hospital EOL treatment intensity) among patients with DNR orders (10).

Our findings from a nationally representative sample of hospitalized patients complement results of prior survey-based studies of EOL treatment preferences. When presented with hypothetical clinical scenarios, clinicians vary widely in interventions they would provide to patients nearing EOL (11–14). Notably, clinician beliefs—rather than patient preferences—most strongly correlate with regional EOL medical care utilization (12, 15). We observed similarly large variation in use of DNR orders between hospitals, regardless of patient clinical and demographic characteristics, supporting the hypothesis that local hospital norms represent strong drivers of EOL care practices (11, 13, 15, 16).

Our findings should be considered in the context of study limitations. ICD-9 identifiers of patient DNR status have not been well validated against chart review, and hospital variation in coding (rather than DNR ordering) may account for some of the observed variation in DNR rates. However, variation in coding alone would not explain associations between DNR rates and EOL healthcare use. Given the cross-sectional study design, we were unable to identify etiological relationships between hospital DNR rates and EOL medical resource use.

In conclusion, we identified large variation in use of DNR orders among U.S. hospitals, with correlations between higher rates of DNR orders and proposed measures of EOL quality of care. Our findings support studies of programs focused on improving hospital norms and procedures for eliciting and documenting patient preferences as potentially effective means to promote quality, patient-centered EOL care.