Skip to main content
NCBI home page
Health Services Research logoLink to Health Services Research
. 2017 Jan 18;53(3):1644–1661. doi: 10.1111/1475-6773.12651

Hospital Variation in Utilization of Life‐Sustaining Treatments among Patients with Do Not Resuscitate Orders

Allan J Walkey 1,✉, Janice Weinberg 2, Renda Soylemez Wiener 3,4, Colin R Cooke 5, Peter K Lindenauer 6
PMCID: PMC5980340  PMID: 28097649

Abstract

Objective

To determine between‐hospital variation in interventions provided to patients with do not resuscitate (DNR) orders.

Data Sources/Setting

United States Agency of Healthcare Research and Quality, Healthcare Cost and Utilization Project, California State Inpatient Database.

Study Design

Retrospective cohort study including hospitalized patients aged 40 and older with potential indications for invasive treatments: in‐hospital cardiac arrest (indication for CPR), acute respiratory failure (mechanical ventilation), acute renal failure (hemodialysis), septic shock (central venous catheterization), and palliative care. Hierarchical logistic regression to determine associations of hospital “early” DNR rates (DNR order placed within 24 hours of admission) with utilization of invasive interventions.

Data Collection/Extraction Methods

California State Inpatient Database, year 2011.

Principal Findings

Patients with DNR orders at high‐DNR‐rate hospitals were less likely to receive invasive mechanical ventilation for acute respiratory failure or hemodialysis for acute renal failure, but more likely to receive palliative care than DNR patients at low‐DNR‐rate hospitals. Patients without DNR orders experienced similar rates of invasive interventions regardless of hospital DNR rates.

Conclusions

Hospitals vary widely in the scope of invasive or organ‐supporting treatments provided to patients with DNR orders.

Keywords: Administrative data, end‐of‐life care, hierarchical regression models, hospice and palliative medicine, patient preference, quality assessment, risk adjustment, utilization, variation

Advance directives are meant to foster patient autonomy by documenting wishes regarding life‐sustaining treatments prior to loss of active decision‐making capacity. While efforts to increase the specificity of advance directives have gained traction (e.g., Patient Orders for Life‐Sustaining Treatment forms; National POLST), do not resuscitate (DNR) orders remain a common method to document wishes to forgo certain life‐sustaining treatments, particularly among patients requiring hospitalization. In the strictest interpretation, DNR orders are meant to convey wishes of patients not to receive cardiopulmonary resuscitation (CPR) during cardiac arrest. In reality, survey studies suggest DNR orders may be broadly interpreted by both patients and physicians (La Puma et al. 1988; Beach and Morrison 2002) to suggest limitation of a wide range of health care interventions (e.g., mechanical ventilation, hemodialysis, invasive procedures).

Between‐hospital differences in the rate of DNR orders placed at hospital admission (“early DNR orders”) and the procedures or therapies provided to patients with DNR orders may substantially impact patient experiences and outcomes, and may confound evaluations of treatment variation (Bradford et al. 2014) and quality (Tabak et al. 2005; Escobar et al. 2013; Kelly et al. 2014; Walkey et al. 2016). Somewhat paradoxically, prior studies demonstrated that patients with DNR orders tended to have higher mortality when admitted to hospitals with low DNR rates (Tabak et al. 2005; Zingmond and Wenger 2005; Escobar et al. 2013; Walkey et al. 2016). One potential explanation for this finding is that low‐DNR‐rate hospitals may apply DNR orders with more extensive scope of limitations on organ‐supportive therapies (e.g., a DNR order implies “no CPR, no mechanical ventilation, no dialysis, and no invasive procedures”) as compared with high‐DNR‐rate hospitals (where DNR may only imply “no CPR”). Although variation in the prevalence of DNR orders, mortality rates (Tabak et al. 2005; Zingmond and Wenger 2005; Escobar et al. 2013; Walkey et al. 2016), and hospital norms surrounding end‐of‐life care (Barnato et al. 2007, 2014) has been previously described, associations between hospital DNR rates and the scope of invasive or organ‐supportive therapies provided to patients admitted with DNR orders across hospitals are unclear.

To close knowledge gaps regarding variation in the implementation of DNR orders across hospitals, we examined associations of early DNR orders (placed within 24 hours of admission) with utilization of invasive procedures and organ‐supportive therapies (such as mechanical ventilation) among patients hospitalized with acute organ failures. Given prior reports of higher mortality rates for patients with DNR orders at low‐DNR‐rate hospitals (Tabak et al. 2005; Zingmond and Wenger 2005; Escobar et al. 2013; Walkey et al. 2016), we hypothesized that low‐DNR‐rate hospitals may apply DNR orders with more extensive scope of limitations on invasive procedures, with lower likelihood of utilizing invasive interventions among patients with DNR orders and indications for each invasive intervention (as compared to DNR patients in high‐DNR‐rate hospitals).

Methods

Cohort

We analyzed a population‐based cohort of hospitalized adults aged 40 and older abstracted from the 2011 Healthcare Cost and Utilization Project, California State Inpatient Database (CA SID; Healthcare Cost and Utilization Project, 2011), an administrative claims database containing all non‐federal acute care hospitalizations in California. A characteristic of the CA SID is a validated field that captures DNR orders written during the first 24 hours of hospitalization (“early DNR”) (Goldman et al. 2013). Using algorithms based upon International Classification of Diseases, 9th edition, Clinical Modification (ICD‐9‐CM) codes, we defined four nonexclusive cohorts of patients with potential indications for interventions or organ‐supportive therapies of interest: patients with any diagnosis of acute respiratory failure (to evaluate mechanical ventilation), acute renal failure (hemodialysis), septic shock (central venous catheterization), and cardiac arrest (CPR) (see Table S1). To avoid capturing out‐of‐hospital cardiac arrest, we excluded patients with cardiac arrest coded as present on admission. However, because mechanical ventilation, dialysis, or central venous catheterization is less likely initiated acutely out of hospital, we did not place restrictions on timing of acute respiratory failure, acute renal failure, or septic shock.

Early DNR Measures

Both patient‐level early DNR status and hospital‐level early DNR rates were identified. We defined hospital DNR rates as the percentage of patients with an early DNR order among all patients ages 40 and older at each hospital; we excluded severe outlier hospitals with DNR rates less than or greater than the 95th percentile (i.e., 0 percent or more than 25 percent). Hospital DNR rates among all patients correlated strongly with hospital DNR rates among patients with conditions of interest (r = 0.92).

Covariates

We developed a mortality risk index for comorbidities and acute organ failures among our cohort of hospitalized patients to improve statistical model performance. In separate logistic regression models including Elixhauser (Elixhauser et al. 1998) comorbidities and acute organ failures, we assigned integer values based upon effect estimates for each comorbidity (risk score calculations shown in Table S2) or acute organ failure (Table S3) as a predictor of mortality (Angus et al. 2001; Martin et al. 2003). Each patient was assigned a comorbidity score from the sum of the comorbidity values and an acute organ failure score from the sum of the organ failure values. We then risk‐adjusted models using patient demographics, hospital characteristics, the Elixhauser comorbidity risk index, and acute organ failures index (c‐statistic for full model to predict mortality among hospitalized cohort = 0.88).

Outcomes

We computed risk‐standardized hospital rates of invasive or organ‐supportive therapies among at‐risk patients, including (1) mechanical ventilation during acute respiratory failure, (2) hemodialysis during acute renal failure, (3) central venous catheterization during septic shock, and (4) CPR among patients with in‐hospital cardiac arrest. The four different procedures were selected to represent different levels of organ support, “invasiveness” and specificity in conventional DNR orders, ranging from CPR (multiorgan support, more invasive, and explicitly limited by a DNR order) to mechanical ventilation and hemodialysis (individual organ support, high‐to‐moderately invasive, inconsistently limited by DNR orders) to central venous catheters (partial organ support, likely considered less invasive and unlikely to be explicitly limited by a DNR order). We also examined associations of patient DNR status and hospital DNR rates with patient receipt of invasive or organ‐supporting interventions. As additional measures of resource utilization, we examined encounters for palliative care (ICD‐9‐CM V66.7, sensitivity 81 percent, specificity 97 percent) (Qureshi, Adil, and Suri 2013) among patients with DNR orders, as well as hospital length of stay (LOS) for patients included in any of the four organ failure cohorts.

Statistical Analysis

Summary data were examined across quartiles of hospital DNR rate. We also assessed the distribution of demographics, comorbid conditions, and acute organ failures stratified by individual patient DNR status. To assess potential differences in utilization rates by patient DNR status, we stratified analyses examining associations between hospital DNR rates and interventions by patient DNR status. We assessed effect modification by patient DNR status across hospital DNR rate using a hospital DNR rate by patient DNR status multiplicative interaction term. As previously described (Walkey et al. 2016), models assessing associations of hospital DNR rates with resource utilization were adjusted for fixed effects of demographics, hospital characteristics, Elixhauser comorbidity index, acute organ failures index, and patient‐level DNR status; models also accounted for hospital random intercepts as well as random DNR slope coefficients. Including DNR status as both a fixed‐effect and a random slope coefficient allowed the association between patient DNR status and resource utilization to vary for each hospital (Gould 1998; Agresti and Hartzel 2000; Localio et al. 2001; Finucane, Samet, and Horton 2007). We used hierarchical logistic regression to model utilization of interventions and organ‐supportive therapies with hospital random intercepts. Cox proportional hazard models censored on death and transfer, with robust variance estimators for hospital clustering, were used to model length of stay.

Hospital risk‐standardized intervention rates were calculated from the ratio of hospital risk‐adjusted rate to the average risk‐adjusted rate, multiplied by the average hospital rate in California (Bratzler et al. 2011). We evaluated the relative contribution of different covariate characteristics (patient DNR status, patient demographics/severity of illness, measured hospital characteristics, and hospital clustering effects) to model prediction for resource utilization by measuring change in Akaike's information criterion after exclusion of each characteristic of interest from a fully adjusted model (Harrell 2001; Gershengorn et al. 2014). We compared hospital variation in use of invasive and organ‐supportive therapies among decedents (to reduce confounding by indication)(Fisher et al. 2003; Wiener and Welch 2007) among patients with and without DNR orders using coefficients of variation, calculated as the standard deviation divided by mean hospital DNR rate (Verrill and Johnson 2007). We visualized variation in utilization among patients with DNR orders with “caterpillar” plots and evaluation of statistically significant hospital outliers of risk‐standardized utilization rates.

To further account for potential differences in case‐mix severity and ICD‐9‐CM coding differences between hospitals, we calculated the ratio of resource utilization for patients with and without DNR within each hospital. We correlated the within‐hospital “DNR: not DNR” utilization ratio with hospital DNR rates. Hospital‐level correlations were assessed quantitatively using Spearman rank correlation coefficients and visually using penalized b‐spline regression (Eilers and Marx 1996).

Sensitivity Analyses

Because eligibility to receive interventions may vary by unmeasured differences in severity of illness that confound relative estimates of resource utilization, we repeated analyses of associations between hospital DNR rates and patient risk for receiving interventions only among patients with DNR orders who did not survive the hospitalization (decedents). We reasoned that decedents in each respective cohort would be the sickest patients (100 percent mortality) and would thus be more likely to require the intervention of interest, reducing unmeasured confounding by indication (Fisher et al. 2003; Wiener and Welch 2007). Because age is strongly associated with DNR status, we also repeated analyses of associations between hospital DNR rates and patient risk for receiving interventions only among patients 80 years of age or older. We performed a third sensitivity analysis excluding patients whose outcomes may be biased by transfer in or out of the hospital, or who had a rehabilitation or “convalescence” ICD‐9‐CM code (V57.86,V66). We used SAS version 9.4 (Cary, NC, USA) and a two‐tailed alpha level of 0.05 for all analyses. All procedures were performed on de‐identified data and approved by Boston University Medical Center Institutional Review Board as exempt from review.

Results

Hospital and Patient Characteristics

Among 2.2 million adult admissions in 311 California hospitals reporting patient early DNR orders, we identified 376,793 patients with indications of interest. After excluding outlier hospitals beyond the 95th percentile, hospital early DNR rates averaged 7.6 percent with a range of 0.15 percent to 25.3 percent (Figure S1). As expected, patients with DNR orders were older, with more comorbid conditions and acute organ failures (Table S4) and were less likely to receive CPR, invasive mechanical ventilation, hemodialysis, and central venous catheters when compared to patients without DNR orders (Table  1).

Table 1.

Interventions for Patients with and without Early DNR Orders

Intervention % Given Intervention DNR vs. No DNR Adjusted Odds Ratio (95% CI)
Patients with DNR Order Patients with No DNR Order
CPR, among in‐hospital cardiac arrest, N = 8,581 32 54 0.39 (0.34–0.45)
Invasive mechanical ventilation, among acute respiratory failure, N = 162,723 31 46 0.56 (0.53–0.61)
Hemodialysis, among acute renal failure, N = 260,768 4.3 8.1 0.57 (0.52–0.62)
Central venous catheter, among septic shock, N = 43,927 40 49 0.65 (0.61–0.70)
Open in a new tab

Model adjusted for age, sex, race, payor, median income for residence zip code, comorbidity index, acute organ failure index, hospital urban/rural location, hospital control (e.g., not‐for‐profit, for profit), hospital teaching status, and number of licensed hospital beds.

Table  2 demonstrates patient and hospital characteristics for all 2.2 million adult admissions and for patients with indications of interest, according to hospital DNR rate quartile. Hospitals with the highest DNR rates were less likely to be urban or teaching hospitals, had fewer beds, and were more likely to be not‐for‐profit than hospitals with low DNR rates. Patients admitted to higher DNR rate hospitals were older, more likely to be white, more likely to have Medicare or private insurance, had higher median household incomes, and greater indices of disease severity as compared to patients admitted to hospitals with low DNR rates. However, among admissions with in‐hospital cardiac arrest, acute respiratory failure, acute renal failure, and septic shock, patients at high‐DNR‐rate hospitals tended to have lower comorbidity and acute organ failure indices.

Table 2.

Patient and Hospital Characteristics by Hospital Do Not Resuscitate Quartile

Quartile 1, <2.5% Quartile 2, 2.5–6.2% Quartile 3, 6.2–10.6% Quartile 4, >10.6%
All patients N = 488,964 N = 643,369 N = 596,881 N = 491,569
Age 64.4 ± 14.2 65.5 ± 14.5 67.5 ± 14.5 68.0 ± 14.5
Comorbidity index 16.6 ± 22.9 17.5 ± 23.0 18.1 ± 23.5 18.3 ± 23.5
Acute organ failure index 0.88 ± 2.22 0.94 ± 2.3 0.94 ± 2.3 0.96 ± 2.2
Sex, female 52% 53% 54% 54%
Race, white 43% 54% 62% 71%
Insurance
Medicare 50% 52% 58% 60%
Medicaid 20% 16% 10% 7%
Private 18% 22% 20% 28%
Highest income quartile 11% 23% 24% 29%
Hospital characteristics
Teaching hospital 27% 32% 4% 7%
Not‐for‐profit 53% 56% 74% 88%
Urban 95% 95% 92% 89%
Licensed beds 373 ± 219 361 ± 172 316 ± 135 274 ± 125
In‐hospital cardiac arrest N = 2,083 N = 2,743 N = 2,403 N = 1,662
Age 70.0 ± 13.4 70.0 ± 13.4 71.7 ± 13.2 71.7 ± 13.7
Comorbidity index 43.5 ± 29.8 42.2 ± 29.0 41.3 ± 29.3 39.2 ± 28.5
Acute organ failure index 3.9 ± 4.7 3.4 ± 4.5 3.2 ± 4.3 3.0 ± 4.1
Acute respiratory failure N = 31,714 N = 47,539 N = 45,794 N = 37,704
Age 69.2 ± 13.6 69.4 ± 13.9 71.1 ± 13.4 70.9 ± 13.4
Comorbidity index 38.3 ± 28.9 37.0 ± 27.7 36.4 ± 27.8 35.0 ± 27.7
Acute organ failure index 6.5 ± 4.2 6.5 ± 4.2 6.3 ± 3.9 6.3 ± 3.8
Acute renal failure N = 55,041 N = 79,611 N = 70,367 N = 55,786
Age 70.6 ± 13.9 72.5 ± 13.4 72.6 ± 13.4 73.0 ± 13.4
Comorbidity index 35.1 ± 26.5 36.4 ± 26.6 36.3 ± 26.7 35.7 ± 26.7
Acute organ failure index 4.0 ± 3.6 3.9 ± 3.5 3.9 ± 3.5 4.0 ± 3.4
Septic shock N = 9,435 N = 14,078 N = 11,435 N = 8,981
Age 69.5 ± 13.6 69.6 ± 13.5 71.0 ± 13.4 70.9 ± 13.5
Comorbidity index 49.3 ± 29.4 47.5 ± 28.3 46.9 ± 28.9 45.2 ± 28.8
Acute organ failure index 8.6 ± 5.2 8.7 ± 5.1 8.4 ± 5.0 8.3 ± 4.8
Open in a new tab

Variation in Treatment Utilization among Patients with Early DNR Orders

Among patients with organ failures, early DNR orders explained between 3 and 10 percent of model predictive ability for utilization of each intervention (Table S5). Hospital coefficients of variation for intervention rates were significantly greater among patients with DNR orders than patients without DNR orders for mechanical ventilation, hemodialysis, and central venous catheters, but not CPR (Table S6). Variation among hospitals in the proportion of patients with DNR orders who received mechanical ventilation for acute respiratory failure (Figure S2A, 32 outlier hospitals) and central venous catheter for septic shock (Figure S2B) was large when compared with CPR for cardiac arrest (Figure S2C, 2 outlier hospitals) and hemodialysis for acute renal failure (Figure S2D, 0 outlier hospitals).

Hospital DNR Rate and Treatment Variation

Patients without DNR orders admitted to high‐DNR‐rate hospitals did not have significantly different rates of CPR, mechanical ventilation, or hemodialysis than patients without DNR orders at low‐DNR hospitals; rates of central venous catheters were higher for patients without DNR orders at high‐DNR‐rate hospitals (Table  3). In contrast, patients with DNR orders admitted to high‐DNR‐rate hospitals were significantly less likely to receive invasive mechanical ventilation (multivariable‐adjusted odds ratio [aOR] DNR quartile 4 vs. quartile 1: 0.59, [95 percent CI 0.45–0.76]) or hemodialysis (aOR 0.58 [95 percent CI 0.41–0.84]) than patients with DNR orders at low‐DNR‐rate hospitals (Table  3). Associations between hospital DNR rates and use of mechanical ventilation, hemodialysis, and central venous catheters differed based upon patient DNR status (p interaction < 0.01, Table  3). The within‐hospital ratio of intervention rates for patients with DNR versus without DNR orders was inversely correlated with hospital DNR rates for mechanical ventilation (r = −0.19, p = .001, Figure 1a) and central venous catheters (r = −0.17, p = .004, Figure 1b), but not CPR (−0.04, p = .52) or hemodialysis (r = −0.07, p = .24).

Table 3.

Interventions for Patients Admitted to High‐ versus Low‐DNR‐Rate Hospitals According to Patient DNR Status

Intervention Adjusted Odds Ratio (95% CI) of Receiving Intervention, Hospital DNR Rate Quartile 4 (high) versus Quartile 1 (low) Comparison of Odds Ratios: Hospital DNR Rate and Intervention, DNR versus No DNR Patients
Patients with DNR Order Patients with No DNR Order p, Interaction
CPR, among in‐hospital cardiac arrest 1.02 (0.58–1.81), N = 1,374 1.34 (0.92–1.96), N = 7,207 .39
Invasive mechanical ventilation, among acute respiratory failure 0.59 (0.45–0.76), N = 24,609 0.95 (0.78–1.16), N = 130,134 <.001
Hemodialysis, among acute renal failure 0.58 (0.41–0.84), N = 33,100 0.86 (0.70–1.05), N = 218,964 .003
Central venous catheter, among septic shock 1.11 (0.78–1.58), N = 7,496 1.72 (1.27–2.32), N = 34,837 <.0001
Open in a new tab

Model adjusted for age, sex, race, payor, median income for residence zip code, comorbidity index, acute organ failure index, hospital urban/rural location, hospital control (e.g., not‐for‐profit, for profit), hospital teaching status, and number of licensed hospital beds.

Figure 1.

Figure 1

Open in a new tab

Association of Hospital Ratio of Utilization between Patients with and without DNR Orders to Hospital DNR Rate for (a) Mechanical Ventilation and (b) Central Venous Catheters

Among patients with DNR orders, 13,193 of 52,864 (25 percent) had a palliative care encounter. Hospitals with higher early DNR rates were more likely to utilize palliative care for patients with DNR orders (DNR rate quartile 4 vs. quartile 1 aOR 1.62, 95 percent CI 1.08–2.45). Compared with lowest quartile DNR rate hospitals, highest quartile DNR rate hospitals had shorter LOS (15 percent relative reduction in LOS [95 percent CI 7, 23 percent]).

Sensitivity Analyses

Sensitivity analyses evaluating decedents with DNR orders showed similar results as primary analyses, with lower utilization of mechanical ventilation (aOR 0.69, 95 percent CI 0.53–0.90) and hemodialysis (0.52, 95 percent CI 0.33–0.81), and similar rates of CPR (aOR 0.83, 95 percent CI 0.54–1.30) and CVL (aOR 1.00 95 percent CI 0.85–1.49) in highest DNR rate quartile hospitals. Exclusion of patients with rehabilitation diagnosis codes, transfers in or transfers out to other acute care hospitals (Table S7), or restriction to patients aged 80 years or older (Table S8) also did not substantively change results.

Discussion

We explored variation among hospitals in invasive and organ‐supportive interventions (e.g., mechanical ventilation during acute respiratory failure) provided to patients with early DNR orders. Rates of invasive and organ‐supportive therapies among patients with DNR orders varied greatly between hospitals. Hospitals with high DNR rates tended to use fewer organ support therapies and more palliative care for patients with DNR orders, but use of organ support interventions did not markedly differ by hospital DNR rates among patients without DNR orders. For example, a theoretical patient with acute respiratory failure without a DNR order would be equally likely to receive mechanical ventilation at a high‐ or low‐DNR‐rate hospital, but a patient with a DNR order would be nearly half as likely to receive mechanical ventilation depending on the DNR rate of the hospital to which he or she was admitted. Our findings suggest that patients with DNR orders may have considerably different experiences depending upon the hospital to which they are admitted, with ramifications for the reporting of hospital practices around wishes for life‐sustaining treatments, measurement of practice variation, and hospital quality.

Few prior studies have examined associations between DNR orders early in the course of hospitalization and resource utilization. Similar to our findings, Hart et al. identified wide variation in treatment limitations among 13,405 patients with pre‐existing DNR orders admitted to intensive care units included in the Project Impact database (Hart et al. 2015) and observed that a large proportion (23 percent) of patients with orders for treatment limitation also received CPR. Potential explanations of higher‐than‐expected CPR rates among patients with DNR orders include a clinician ignoring or patient/surrogate reversal of DNR orders, CPR performed prior to DNR decisions, or higher rates of misclassification for CPR than other ICD‐9‐CM procedure codes. Our findings among hospitalized patients with acute organ failures extend those of Hart et al., using patients without DNR orders as a control for between‐hospital variation in case mix, eliminating selection bias that may be due to hospital variation in intensive care unit admission among patients with DNR orders, and examining use of palliative care. Our findings differed from Hemphill et al. (2004), who showed that high hospital DNR rates among patients with intracerebral hemorrhage were associated with a general “nonaggressive approach,” regardless of individual patient DNR status. Rather than a general “nonaggressive approach” at higher DNR rate hospitals, measured indices of “aggressiveness” were similar (or greater) among patients without DNR orders at hospitals with high DNR rates.

Importantly, we found that hospitals with higher DNR rates tended to use a less invasive, more palliative approach among patients with DNR orders. The observation that utilization of organ‐supportive interventions was lower for patients with DNR orders at high‐DNR‐rate hospitals was contrary to our hypothesis that higher DNR rate hospitals would tend to use DNR orders with less extensive limitations on care. This makes it unlikely that previously described inverse associations between hospital DNR rates and hospital mortality among patients with DNR orders (Tabak et al. 2005) are explained by greater willingness to use invasive therapies to “rescue” patients with pre‐existing DNR orders (e.g., a patient with respiratory failure requiring mechanical ventilation) at higher DNR rate hospitals.

Based on our findings showing minimal variation between hospitals in use of CPR among patients with DNR orders, as well as similar odds of receiving CPR for cardiac arrest regardless of underlying hospital DNR rate, the strictest definition of DNR (“no CPR”) did not substantially vary between hospitals. However, organ‐replacement therapies that may not fall under “strict” definitions of DNR such as mechanical ventilation (i.e., “Do Not Intubate” orders) and dialysis were less likely to be used among DNR patients at hospitals with higher rates of DNR orders. Further, when compared to hospitals with low DNR rates, hospitals with high DNR rates used fewer central venous catheters and less mechanical ventilation among patients with DNR orders than patients without DNR orders. Our findings are in accordance with previous survey studies that demonstrated potentially wide physician‐level variation in application of DNR orders (La Puma et al. 1988; Beach and Morrison 2002; Garland and Connors 2007). However, our results extend prior physician survey results to real‐world practice and suggest that the scope of interventions provided to patients with DNR orders depends on the hospital to which they were admitted.

Fewer invasive interventions among patients with DNR orders at high‐DNR‐rate hospitals potentially signal different hospital practices and local cultural norms (Barnato et al. 2007, 2012; Halpern et al. 2013; Cutler et al. 2015; Dzeng et al. 2015) for discussing, eliciting, and documenting patient wishes regarding life‐sustaining treatments. Lower comorbidity and acute organ failure indices observed at high‐DNR‐rate hospitals also suggest that hospitals vary in the thresholds at which invasive interventions may be limited. Although we were unable to access details of physician–patient/surrogate discussions regarding decisions to limit life‐support interventions and were unable to address the extent to which limits were based upon patient‐driven, physician‐driven, or shared decisions, survey‐based studies show that health care utilization at the end of life may be more strongly associated with local physician practice style than patient beliefs (Barnato et al. 2007, 2012; Cutler et al. 2015). Our findings suggest that studies should continue to explore how interactions between patient beliefs and physician practice styles drive measured variation in hospital DNR rates and the scope of therapies associated with DNR orders. Further efforts to standardize documentation and increase the specificity of patient advance directives (National POST) may better align patient wishes with care received (Chen et al. 2014), potentially reducing the influence of individual physician beliefs or local hospital norms.

Because variation in health care utilization was partly explained by variation in preferences for life‐sustaining treatments, our results support identification of patient wishes to withhold life‐sustaining therapies in programs that seek to evaluate health care quality. For example, accurate measurement and description of the variation in DNR practices between hospitals would produce greater transparency in public reporting of hospital practices and potentially allow patients to choose hospitals with practice patterns that best align with their beliefs. Although wide variation in DNR rates and DNR scope between hospitals complicates evaluation of health care delivery (Tabak et al. 2005; Walkey et al. 2016), the lack of substantial differences in interventions used among patients without DNR orders supports strategies that assess robustness of quality rankings after adjusting for or excluding patients with early DNR orders (California Office of Statewide Health Planning and Development, Healthcare). However, given that many patients with DNR orders received organ‐supportive therapies—potentially indicating a commitment to full support short of CPR—methods that better account for variation in scope of DNR orders between hospitals should be further developed to compare patient outcomes (Walkey et al. 2016).

Our study has potential limitations. The early DNR variable of CA SID shows ~85 percent accuracy (Goldman et al. 2013); differential misclassification of DNR orders may affect our findings. DNR orders placed after the first day of hospitalization were unavailable in the CA SID dataset, but they are generally correlated with failure to respond to treatments, rather than pre‐existing wishes regarding life‐support and invasive treatments (Marrie et al. 2002). Because patients at high‐DNR‐rate hospitals had shorter LOS, further studies seeking to explore differences in hospital mortality among patients with DNR orders should explore 30‐day mortality rates or whether patients with DNR orders at high‐DNR‐rate hospitals are more likely to transfer to hospice care, information unavailable through the CA SID. In addition, unmeasured differences in severity of illness may potentially explain variation in invasive treatments based upon hospital DNR rates. However, several lines of evidence argue against strong unmeasured confounding by severity of illness. First, our covariate adjustment produced models with high resolution for predicting mortality outcomes, reducing likelihood of confounding by severity of illness (Sjoding et al. 2015). Second, our results were similar in analyses including only decedents, with likely severe illness (Fisher et al. 2003; Wiener and Welch 2007). Third, we did not find lower utilization of interventions according to hospital DNR rates among patients without DNR orders. Fourth, within‐hospital analyses that would better control for differences in case mix also showed greater reduction in utilization of some invasive procedures among patients with DNR orders at high‐DNR‐rate hospitals.

In conclusion, hospital variation in the scope of decisions to limit life‐sustaining treatments contributed to differences in health care utilization. Hospitals with higher DNR rates tended to have broader limits on life‐support interventions and greater use of palliative care among patients with DNR orders than hospitals with low DNR rates. Variation in the scope of DNR orders between hospitals has broad ramifications, from types of care delivered to patients, to the need for accurate reporting of health care delivery to patients and policymakers. Improved efforts to measure and report hospital practices regarding decisions to limit life‐sustaining treatments are warranted.

Supporting information

Appendix SA1: Author Matrix.

Click here for additional data file. (1.2MB, pdf)

Table S1: ICD‐9 Definitions.

Table S2: Effect Estimates for Associations between Comorbidities and Mortality with Resulting Risk Score Index Calculation.

Table S3: Effect Estimates for Associations between Acute Organ Failures and Mortality with Resulting Risk Score Index Calculation.

Table S4: Characteristics of Patients with Acute Organ Failures, According to Early Do Not Resuscitate (DNR) Order Status.

Table S5: Proportional Contribution to Model Prediction of Organ‐Supportive Interventions Explained for DNR Status, Patient and Hospital Characteristics, and Clustering Within Hospitals.

Table S6: Coefficients of Variation for Between‐Hospital Utilization Rates of Interventions among Decedents.

Table S7: Sensitivity Analysis: Interventions for Patients Admitted to High‐Versus Low‐DNR‐Rate Hospitals, Excluding Patients Transferred or with Rehabilitation Codes.

Table S8: Sensitivity Analysis: Interventions for Patients Admitted to High‐Versus Low‐DNR‐Rate Hospitals, Restricted to Age Greater Than or Equal to Eighty Years Old.

Figure S1: Distribution of Hospital DNR Rates.

Figure S2: (A–D) Risk‐Standardized Hospital Rates of Interventions Among Patients with DNR Orders.

Click here for additional data file. (73.9KB, docx)

Acknowledgments

Joint Acknowledgment/Disclosure Statement: The authors acknowledge the following sources of funding for conduct of research, not specifically allocated for the current manuscript: Walkey, NIH NHLBI K01 HL116768; Lindenauer, NIH NHLBI K24HL132008.

Disclosures: None.

Disclaimer: None.

References

  1. Agresti, A. , and Hartzel J.. 2000. “Strategies for Comparing Treatments on a Binary Response with Multi‐Centre Data.” Statistics in Medicine 19 (8): 1115–39. [DOI] [PubMed] [Google Scholar]
  2. Angus, D. C. , Linde‐Zwirble W. T., Lidicker J., Clermont G., Carcillo J., and Pinsky M. R.. 2001. “Epidemiology of Severe Sepsis in the United States: Analysis of Incidence, Outcome, and Associated Costs of Care.” Critical Care Medicine 29 (7): 1303–10. [DOI] [PubMed] [Google Scholar]
  3. Barnato, A. E. , Herndon M. B., Anthony D. L., Gallagher P. M., Skinner J. S., Bynum J. P., and Fisher E. S.. 2007. “Are Regional Variations in End‐of‐Life Care Intensity Explained by Patient Preferences? A Study of the US Medicare Population.” Medical Care 45 (5): 386–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Barnato, A. E. , Tate J. A., Rodriguez K. L., Zickmund S. L., and Arnold R. M.. 2012. “Norms of Decision Making in the ICU: A Case Study of Two Academic Medical Centers at the Extremes of End‐of‐Life Treatment Intensity.” Intensive Care Medicine 38 (11): 1886–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Barnato, A. E. , Mohan D., Lane R. K., Huang Y. M., Angus D. C., Farris C., and Arnold R. M.. 2014. “Advance Care Planning Norms May Contribute to Hospital Variation in End‐of‐Life ICU Use: A Simulation Study.” Medical Decision Making: An International Journal of the Society for Medical Decision Making 34 (4): 473–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Beach, M. C. , and Morrison R. S.. 2002. “The Effect of Do‐Not‐Resuscitate Orders on Physician Decision‐Making.” Journal of the American Geriatrics Society 50 (12): 2057–61. [DOI] [PubMed] [Google Scholar]
  7. Bradford, M. A. , Lindenauer P. K., Wiener R. S., and Walkey A. J.. 2014. “Do Not Resuscitate Status and Observational Comparative Effectiveness Research in Patients with Septic Shock.” Critical Care Medicine 42 (9): 2042–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bratzler, D. W. , Normand S. L., Wang Y., O'Donnell W. J., Metersky M., Han L. F., Rapp M. T., and Krumholz H. M.. 2011. “An Administrative Claims Model for Profiling Hospital 30‐Day Mortality Rates for Pneumonia Patients.” PLoS ONE 6 (4): e17401. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. California Office of Statewide Health Planning and Development, Healthcare . Community‐Acquired Pneumonia: Hospital Outcomes in California, 1999–2001. Sacramento, CA: California Office of Statewide Health Planning and Development, February 2004 [accessed on May 29, 2014 ]. Available at http://www.oshpd.ca.gov/HID/Products/PatDischargeData/ResearchReports/OutcomeRpts/CAP/Reports/1999_to_2001.html [Google Scholar]
  10. Chen, Y. Y. , Gordon N. H., Connors A. F., Garland A., Chang S. C., and Youngner S. J.. 2014. “Two Distinct Do‐Not‐Resuscitate Protocols Leaving Less to the Imagination: An Observational Study Using Propensity Score Matching.” BMC Medicine 12: 146. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Cutler, D. , Skinner J., Stern A. D., and Wennberg D. E.. 2015. Physician Beliefs and Patient Preferences: A new Look at Regional Variation in Healthcare Spending. Working Paper 15‐090. Available at http://www.hbs.edu/faculty/Publication%20Files/15-090_bfbc1026-8cd3-497a-845a-f57b382a348a.pdf: Harvard Business School.
  12. Dzeng, E. , Colaianni A., Roland M., Chander G., Smith T. J., Kelly M. P., Barclay S., and Levine D.. 2015. “Influence of Institutional Culture and Policies on Do‐Not‐Resuscitate Decision Making at the End of Life.” JAMA Internal Medicine 175 (5): 812–9. [DOI] [PubMed] [Google Scholar]
  13. Eilers, P. H. C. , and Marx B. D.. 1996. “Flexible Smoothing with B‐Splines and Penalties.” Statistical Science 11 (2): 89–121. [Google Scholar]
  14. Elixhauser, A. , Steiner C., Harris D. R., and Coffey R. M.. 1998. “Comorbidity Measures for Use with Administrative Data.” Medical Care 36 (1): 8–27. [DOI] [PubMed] [Google Scholar]
  15. Escobar, G. J. , Gardner M. N., Greene J. D., Draper D., and Kipnis P.. 2013. “Risk‐Adjusting Hospital Mortality Using a Comprehensive Electronic Record in an Integrated Health Care Delivery System.” Medical Care 51 (5): 446–53. [DOI] [PubMed] [Google Scholar]
  16. Finucane, M. M. , Samet J. H., and Horton N. J.. 2007. “Translational Methods in Biostatistics: Linear Mixed Effect Regression Models of Alcohol Consumption and HIV Disease Progression over Time.” Epidemiologic Perspectives & Innovations: EP+I 4: 8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Fisher, E. S. , Wennberg D. E., Stukel T. A., Gottlieb D. J., Lucas F. L., and Pinder E. L.. 2003. “The Implications of Regional Variations in Medicare Spending. Part 2: Health Outcomes and Satisfaction with Care.” Annals of Internal Medicine 138 (4): 288–98. [DOI] [PubMed] [Google Scholar]
  18. Garland, A. , and Connors A. F.. 2007. “Physicians' Influence over Decisions to Forego Life Support.” Journal of Palliative Medicine 10 (6): 1298–305. [DOI] [PubMed] [Google Scholar]
  19. Gershengorn, H. B. , Garland A., Kramer A., Scales D. C., Rubenfeld G., and Wunsch H.. 2014. “Variation of Arterial and Central Venous Catheter Use in United States Intensive Care Units.” Anesthesiology 120 (3): 650–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Goldman, L. E. , Chu P. W., Osmond D., and Bindman A.. 2013. “Accuracy of Do Not Resuscitate (DNR) in Administrative Data.” Medical Care Research and Review: MCRR 70 (1): 98–112. [DOI] [PubMed] [Google Scholar]
  21. Gould, A. L . 1998. “Multi‐Centre Trial Analysis Revisited.” Statistics in Medicine 17 (15‐16): 1779–97; discussion 1799–800. [DOI] [PubMed] [Google Scholar]
  22. Halpern, S. D. , Loewenstein G., Volpp K. G., Cooney E., Vranas K., Quill C. M., McKenzie M. S., Harhay M. O., Gabler N. B., Silva T., Arnold R., Angus D. C., and Bryce C.. 2013. “Default Options in Advance Directives Influence How Patients Set Goals for End‐of‐Life Care.” Health Affairs (Project Hope) 32 (2): 408–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Harrell, F. E. 2001. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer. [Google Scholar]
  24. Hart, J. L. , Harhay M. O., Gabler N. B., Ratcliffe S. J., Quill C. M., and Halpern S. D.. 2015. “Variability Among US Intensive Care Units in Managing the Care of Patients Admitted with Preexisting Limits on Life‐Sustaining Therapies.” JAMA Internal Medicine 175 (6): 1019–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. HCUP California State Inpatient Database (SID) . Healthcare Cost and Utilization Project (HCUP). 2011. Rockville, MD: Agency for Healthcare Research and Quality. Available at http://www.hcup-us.ahrq.gov/sidoverview.jsp [Google Scholar]
  26. Hemphill 3rd, J. C. , Newman J., Zhao S., and Johnston S. C.. 2004. “Hospital Usage of Early Do‐Not‐Resuscitate Orders and Outcome after Intracerebral Hemorrhage.” Stroke: A Journal of Cerebral Circulation 35 (5): 1130–4. [DOI] [PubMed] [Google Scholar]
  27. Kelly, A. G. , Zahuranec D. B., Holloway R. G., Morgenstern L. B., and Burke J. F.. 2014. “Variation in Do‐Not‐Resuscitate Orders for Patients with Ischemic Stroke: Implications for National Hospital Comparisons.” Stroke: A Journal of Cerebral Circulation 45 (3): 822–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. La Puma, J. , Silverstein M. D., Stocking C. B., Roland D., and Siegler M.. 1988. “Life‐Sustaining Treatment. A Prospective Study of Patients with DNR Orders in a Teaching Hospital.” Archives of Internal Medicine 148 (10): 2193–8. [DOI] [PubMed] [Google Scholar]
  29. Localio, A. R. , Berlin J. A., Ten Have T. R., and Kimmel S. E.. 2001. “Adjustments for Center in Multicenter Studies: An Overview.” Annals of Internal Medicine 135 (2): 112–23. [DOI] [PubMed] [Google Scholar]
  30. Marrie, T. J. , Fine M. J., Kapoor W. N., Coley C. M., Singer D. E., and Obrosky D. S.. 2002. “Community‐Acquired Pneumonia and Do Not Resuscitate Orders.” Journal of the American Geriatrics Society 50 (2): 290–9. [DOI] [PubMed] [Google Scholar]
  31. Martin, G. S. , Mannino D. M., Eaton S., and Moss M.. 2003. “The Epidemiology of Sepsis in the United States from 1979 Through 2000.” New England Journal of Medicine 348 (16): 1546–54. [DOI] [PubMed] [Google Scholar]
  32. “National POLST: Physician Orders for Life‐Sustaining Treatment Paradigm ” b. [accessed on April 11, 2014]. Available at http://www.polst.org/
  33. Qureshi, A. I. , Adil M. M., and Suri M. F.. 2013. “Rate of Utilization and Determinants of Withdrawal of Care in Acute Ischemic Stroke Treated with Thrombolytics in USA.” Medical Care 51 (12): 1094–100. [DOI] [PubMed] [Google Scholar]
  34. Sjoding, M. W. , Luo K., Miller M. A., and Iwashyna T. J.. 2015. “When Do Confounding by Indication and Inadequate Risk Adjustment Bias Critical Care Studies? A Simulation Study.” Critical Care (London, England) 19: 195. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Tabak, Y. P. , Johannes R. S., Silber J. H., and Kurtz S. G.. 2005. “Should Do‐Not‐Resuscitate Status Be Included as a Mortality Risk Adjustor? The Impact of DNR Variations on Performance Reporting.” Medical Care 43 (7): 658–66. [DOI] [PubMed] [Google Scholar]
  36. Verrill, S. , and Johnson R. A.. 2007. “Confidence Bounds and Hypothesis Tests for Normal Distribution Coefficients of Variation.” Communications in Statistics – Theory and Methods 36 (12): 2187–206. [Google Scholar]
  37. Walkey, A. J. , Weinberg J., Wiener R. S., Cooke C. R., and Lindenauer P. K.. 2016. “Association of Do‐Not‐Resuscitate Orders and Hospital Mortality Rate among Patients with Pneumonia.” JAMA Internal Medicine 176 (1): 97–104. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Wiener, R. S. , and Welch H. G.. 2007. “Trends in the Use of the Pulmonary Artery Catheter in the United States, 1993–2004.” JAMA: The Journal of the American Medical Association 298 (4): 423–9. [DOI] [PubMed] [Google Scholar]
  39. Zingmond, D. S. , and Wenger N. S.. 2005. “Regional and Institutional Variation in the Initiation of Early Do‐Not‐Resuscitate Orders.” Archives of Internal Medicine 165 (15): 1705–12. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Appendix SA1: Author Matrix.

Click here for additional data file. (1.2MB, pdf)

Table S1: ICD‐9 Definitions.

Table S2: Effect Estimates for Associations between Comorbidities and Mortality with Resulting Risk Score Index Calculation.

Table S3: Effect Estimates for Associations between Acute Organ Failures and Mortality with Resulting Risk Score Index Calculation.

Table S4: Characteristics of Patients with Acute Organ Failures, According to Early Do Not Resuscitate (DNR) Order Status.

Table S5: Proportional Contribution to Model Prediction of Organ‐Supportive Interventions Explained for DNR Status, Patient and Hospital Characteristics, and Clustering Within Hospitals.

Table S6: Coefficients of Variation for Between‐Hospital Utilization Rates of Interventions among Decedents.

Table S7: Sensitivity Analysis: Interventions for Patients Admitted to High‐Versus Low‐DNR‐Rate Hospitals, Excluding Patients Transferred or with Rehabilitation Codes.

Table S8: Sensitivity Analysis: Interventions for Patients Admitted to High‐Versus Low‐DNR‐Rate Hospitals, Restricted to Age Greater Than or Equal to Eighty Years Old.

Figure S1: Distribution of Hospital DNR Rates.

Figure S2: (A–D) Risk‐Standardized Hospital Rates of Interventions Among Patients with DNR Orders.

Click here for additional data file. (73.9KB, docx)

Articles from Health Services Research are provided here courtesy of Health Research & Educational Trust

RESOURCES