The American Journal of Medicine (2005) 118, 400-408

CLINICAL RESEARCH STUDY

Adjustment for do-not-resuscitate orders reverses the

apparent in-hospital mortality advantage for minorities
Naomi Bardach, MD,a Shoujun Zhao, MD, PhD,a Steven Pantilat, MD,b
S. Claiborne Johnston, MD, PhDc

a
Department of Neurology
b
Division of General Internal Medicine, Department of Medicine, and
c
Department of Neurology and Epidemiology, University of California.

KEYWORDS: ABSTRACT
Do-not-resuscitate; PURPOSE: The use of do-not-resuscitate (DNR) orders may differ by sex or ethnicity, and DNR
Ethnicity; status may be associated with outcomes for hospitalized patients. Thus, we sought to determine whether
Sex; differences in rates of DNR by sex and ethnicity influenced differences in mortality.
Health care outcomes; SUBJECTS AND METHODS: We included all patients admitted to nonfederal California hospitals in
Disparity; 1999 with stroke, congestive heart failure, pneumonia, chronic obstructive pulmonary disease, chronic
Mortality renal failure, angina, or diabetes mellitus. Rates of physician orders for DNR written within 24 hours
of hospital admission and in-hospital mortality were compared between sexes and ethnicities after
adjustment for age, admission source and diagnosis, payment type, and comorbidity scores in multi-
variable logistic regression models.
RESULTS: Of 327 890 patients included, 25 196 (7.7%) had DNR orders. In adjusted models,
women were more likely to have DNR orders than men (odds ratio [OR] 1.19; 95% confidence interval
1.16 –1.23; P ⬍0.001) and non-Hispanic whites were more likely to have DNR orders than other
ethnicities (OR 1.75; 1.69 –1.82; P ⬍0.001). Overall, 13 549 (4.1%) patients died in the hospital. Risk
of death was greater in those with a DNR order (OR 7.0; 6.7–7.3; P ⬍0.001). Non-Hispanic whites
appeared to have a greater risk of in-hospital death in adjusted models (OR 1.09; 1.04 –1.12; P ⬍0.001)
when DNR status was ignored; however, the risk of death appeared to be lower in non-Hispanic whites
in the complete model with DNR included (OR 0.94; 0.90 – 0.99; P ⫽ 0.01). A survival advantage for
women was also more apparent after including DNR status in the adjusted model.
CONCLUSIONS: Women and non-Hispanic whites are more likely to have DNR orders. DNR status
affected the measurement of sex-ethnic differences in mortality risk.
© 2005 Elsevier Inc. All rights reserved.

Patients increasingly eschew resuscitation in the event of

Ms. Bardach is supported by the Doris Duke Charitable Foundation’s
Medical Student Research Fellowship; Dr. Johnston is supported by NIH/
NINDS NS 02254.
Requests for reprints should be addressed to S. Claiborne Johnston, MD,
PhD, Department of Neurology, Box 0114, University of California, San
Francisco, 505 Parnassus Ave., M-798, San Francisco, CA 94143-0114.
E-mail address: clay.johnston@ucsfmedctr.org.
cardiac or pulmonary arrest, resulting in more frequent
physician do-not-resuscitate (DNR) orders.1-4 In
1985–1986, 13% of a large sample of hospitalized Medicare
patients received DNR orders during hospital admission.2
Attention to patient autonomy5 and concerns about the cost
of end-of-life care5,6 have contributed to increased use of
DNR orders.

0002-9343/$ -see front matter © 2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.amjmed.2005.01.008
Bardach et al DNR Orders and Minorities 401

Table 1 California patients admitted in 1999 with 1 of 7 common diagnoses (total admissions, N ⫽ 327 890; patients with DNR
orders, n ⫽ 25 196)

Associated odds of DNR

Univariate Adjusted*

Characteristic No. DNR (%) OR 95% CI P value† OR 95% CI P value†

Age
⬍65 127 095 2.24 1.00 — — 1.00 — —
ⱖ65 200 795 11.13 5.47 5.26–5.69 ⬍0.001 3.33 3.17–3.49 ⬍0.001
Sex
Male 153 881 6.51 1.00 — — 1.00 — —
Female 174 009 8.72 1.37 1.34–1.41 ⬍0.001 1.19 1.16–1.23 ⬍0.001
Ethnicity
White 210 075 9.64 1.00 — — 1.00 — —
Asian 22 029 4.84 0.48 0.45–0.51 ⬍0.001 0.52 0.48–0.55 ⬍0.001
Hispanic 38 513 4.15 0.41 0.39–0.43 ⬍0.001 0.60 0.57–0.64 ⬍0.001
African-American 33 373 3.95 0.39 0.36–0.41 ⬍0.001 0.58 0.55–0.62 ⬍0.001
Admission Source:
Non-nursing home 322 739 7.51 1.00 — — 1.00 — —
Nursing home 5151 18.35 2.77 2.57–2.97 ⬍0.001 1.71 1.58–1.84 ⬍0.001
Payment source
Medicare 186 761 10.84 1.00 — — 1.00 — —
Medi-Cal/indigent 60 110 2.91 0.25 0.23–0.26 ⬍0.001 0.98 0.93–1.04 0.45
HMO 48 444 3.99 0.34 0.33–0.36 ⬍0.001 1.01 0.96–1.07 0.58
Non-HMO private 29 007 4.01 0.34 0.32–0.37 ⬍0.001 1.01 0.94–1.07 0.86
Comorbidity score‡
0 132 850 4.79 1.00 — — 1.00 — —
1 54 390 8.72 1.90 1.83–1.98 ⬍0.001 1.34 1.29–1.40 ⬍0.001
2 61 960 7.88 1.70 1.64–1.77 ⬍0.001 1.25 1.20–1.30 ⬍0.001
3 or 4 52 772 10.54 2.35 2.26–2.43 ⬍0.001 1.50 1.44–1.57 ⬍0.001
ⱖ5 25 918 14.06 3.25 3.12–3.40 ⬍0.001 2.27 2.15–2.39 ⬍0.001
Admission diagnosis
Angina 54 953 1.91 1.00 — — 1.00 — —
Diabetes 17 077 3.12 1.65 1.48–1.83 ⬍0.001 2.31 2.07–2.57 ⬍0.001
COPD§ 34 934 6.97 3.84 3.57–4.13 ⬍0.001 2.80 2.59–3.02 ⬍0.001
Congestive heart failure 53 113 8.83 4.97 4.64–5.32 ⬍0.001 2.55 2.37–2.74 ⬍0.001
Chronic renal failure 12 174 9.20 5.20 4.77–5.67 ⬍0.001 3.93 3.57–4.33 ⬍0.001
Pneumonia 79 207 9.34 5.28 4.95–5.64 ⬍0.001 3.79 3.54–4.06 ⬍0.001
Stroke 76 432 10.43 5.97 5.59–6.37 ⬍0.001 2.83 2.63–3.04 ⬍0.001
*Adjusted by age, sex, ethnicity, admission source, payment source, comorbidity scores, and admission diagnosis, as categorized above. DNR ⫽
do-not-resuscitate orders, OR ⫽ odds ratio, CI ⫽ confidence interval.
†P values calculated using logistic regression.
‡Comorbidity scores were based on the Charlson comorbidity index adapted for administrative databases.
§Chronic obstructive pulmonary disease.

Few studies have evaluated the predictors and impact
of DNR orders. Several studies have found that utiliza-
tion of DNR orders varies by sex and ethnicity,2,7-11 but
others have not demonstrated an association.9,12 Two
studies demonstrated that patients with DNR orders are
more likely to die in the hospital, even after adjustment
for characteristics such as age and severity of illness.13,14
DNR appears to influence outcomes because care is less
aggressive and because it is a measure of disease severity
not captured with other variables.3,15,16
Ethnicity is an important predictor of death after diag-
nosis or admission for several diseases.17-23 Differences in
access to care23,24 and socioeconomic status24,25 have been
hypothesized as possible explanations, but the cause of
ethnic differences in mortality is largely unknown. Interest-
ingly, 2 recent studies found lower in-hospital mortality in
African-Americans than in non-Hispanic whites, but the
influence of DNR was not evaluated.18,23 Similarly, without
consideration of DNR, rates of in-hospital mortality are
often found to be lower for women.26
Because patients with DNR orders upon admission are
more likely to die during the hospital stay, a higher rate of
DNR orders among white patients could at least partially
explain their higher hospital mortality rate. Furthermore,
correcting for the influence of higher DNR rates among
white patients might actually show that in-hospital mortality
rates for minority patients are higher, not lower, than for
white patients. We sought to test this hypothesis; thus, we
asked, is DNR a confounder in the association of sex and
ethnicity with in-hospital mortality?
402 The American Journal of Medicine, Vol 118, No 4, April 2005

Table 2 The association of sex and ethnicity with a physician order for DNR for 7 common diagnoses

Adjusted odds of DNR*

Women vs. men Non-whites vs. whites

Admission diagnosis OR 95% CI P value† OR 95% CI P value†

Overall 1.19 1.16–1.23 ⬍0.001 0.57 0.55–0.59 ⬍0.001
Angina 1.21 1.07–1.38 ⬍0.001 0.74 0.64–0.86 ⬍0.001
Diabetes 1.20 1.00–1.44 0.04 0.55 0.45–0.67 ⬍0.001
COPD‡ 1.12 1.03–1.21 0.01 0.44 0.38–0.50 ⬍0.001
Congestive heart failure 1.16 1.09–1.24 ⬍0.001 0.50 0.46–0.55 ⬍0.001
Chronic renal failure 1.27 1.12–1.44 ⬍0.001 0.59 0.50–0.68 ⬍0.001
Pneumonia 1.15 1.09–1.21 ⬍0.001 0.53 0.49–0.57 ⬍0.001
Stroke 1.28 1.21–1.34 ⬍0.001 0.60 0.57–0.64 ⬍0.001
*All models adjusted by age, sex, ethnicity, admission source, payment source, comorbidity scores. DNR ⫽ do-not-resuscitate orders, OR ⫽ odds
ratio, CI ⫽ confidence interval.
†P values calculated using logistic regression.
‡Chronic obstructive pulmonary disease.

Methods as all others. The nonwhite group consisted of African-

Subjects
The Human Subjects Committee of the University of
California, San Francisco, approved this study. California’s
Office of State Health Planning and Development (OSHPD)
maintains a statewide database of inpatient discharge ab-
stracts for all nonfederal hospitals. Based on a list of com-
mon diagnoses included in prior studies of ethnicity and
mortality,23 we selected a cohort of patients discharged
between January 1 and December 31, 1999, with 1 of 7
primary diagnoses: stroke (International Classification of
Diseases, Ninth Revision, Clinical Modification [ICD-9-
CM] code 430 – 431, 432.0, 432.1, 433– 434.9, 436 – 437.9),
angina (diagnosis-related group [DRG] 140, 143), conges-
tive heart failure (DRG 127), pneumonia (DRG 89 –91),
chronic obstructive pulmonary disease (DRG 88), chronic
renal failure (ICD-9 code 585), and diabetes mellitus (DRG
294, 295). To mimic a population-based study, patients
residing outside the state of California were excluded. For
patients with multiple admissions, only the first admission
was included so that one patient could not contribute mul-
tiple data points.
Predictor variables and outcomes
Ethnicity was based on patient self-report of race, which
is selected from the following categories on hospital admis-
sion—white, black, Asian/Pacific Islander, Native American/
Alaskan/Aleutian, and other—and on self report of Hispanic
origin.27 For this analysis, we defined white as whites not of
Hispanic origin, African-American as blacks not of His-
panic origin, Hispanic as all those patients who identified
themselves as being of Hispanic ethnicity, Asian as Asian/
Pacific Islanders not of Hispanic origin, and Other/unknown
Americans, Asians, and Hispanics.
Beginning in 1999, OSHPD required hospitals to indi-
cate the presence of a physician DNR order within 24 hours
of hospital admission. The DNR variable is abstracted by
staff in hospitals and indicates whether the inpatient medical
chart within 24 hours of admission contains a physician
directive to limit resuscitative efforts (such as chest com-
pressions, intubation, assisted ventilation, or defibrillation)
in the event of a patient’s cardiac or pulmonary arrest.27
For payment type categorization, we combined self-pay-
ers with those listed as indigent or receiving Medicaid and
combined worker’s compensation with private, non-HMO
insurance in order to simplify listing of results. Comorbidity
scores were developed using a database version of the
Charlson comorbidity index and are a summary of major
secondary diagnoses weighted by severity.28 The index pre-
dicts 1-year mortality based on the presence of coded co-
morbidities. We categorized scores as 0, 1, 2, 3– 4, or ⬎4 to
simplify presentation of results. OSHPD provides ongoing
training and quality-improvement projects to hospital med-
ical record staff to enhance reliability of coding.27
Statistical analysis
We defined predictors of DNR in univariate and multi-
variable analyses with logistic regression, assessing the fol-
lowing characteristics: age, sex, ethnicity, admission source,
payment type, and comorbidity scores. All of these factors
were included in each multivariable logistic regression
model. Analyses were performed for the entire cohort and
for each diagnosis separately; in the overall cohort analysis,
a variable for diagnostic group was included in the multi-
variable model.
To determine whether DNR influenced the association of
in-hospital mortality with sex and ethnicity, we evaluated
DNR as a confounder in stratified analysis and by compar-
 Bardach et al
Table 3 Predictors of in-hospital mortality ignoring, adjusting for, and stratifying for DNR status*

Full cohort Stratified by DNR

Ignoring DNR Adjusted for DNR Those with DNR Those without DNR
(n ⫽ 327 890) (n ⫽ 327 890) (n ⫽ 25 169) (n ⫽ 302 694)

DNR Orders and Minorities

Characteristic OR 95% CI P value OR 95% CI P value OR 95% CI P value OR 95% CI P value
Age, per decade 1.30 1.28–1.32 ⬍0.001 1.17 1.16–1.19 ⬍0.001 1.01 0.99–1.04 0.33 1.22 1.20–1.24 ⬍0.001
Female 0.92 0.88–0.95 ⬍0.001 0.85 0.82–0.88 ⬍0.001 0.76 0.72–0.81 ⬍0.001 0.89 0.85–0.94 ⬍0.001
Ethnicity (reference group: white)
African-American 0.87 0.81–0.93 ⬍0.001 0.98 0.92–1.05 0.62 0.78 0.67–0.91 0.002 1.07 0.99–1.15 0.10
Hispanic 0.87 0.81–0.92 ⬍0.001 0.97 0.91–1.04 0.43 1.11 0.98–1.26 0.12 0.94 0.86–1.01 0.10
Asian 1.04 0.97–1.11 0.31 1.24 1.15–1.33 ⬍0.001 1.37 1.19–1.58 ⬍0.001 1.18 1.09–1.28 ⬍0.001
Payment (reference group:
Medicare)
Medi-Cal/indigent 1.35 1.27–1.44 ⬍0.001 1.32 1.24–1.42 ⬍0.001 1.42 1.25–1.63 ⬍0.001 1.28 1.18–1.38 ⬍0.001
HMO 1.01 0.94–1.08 0.82 0.96 0.90–1.03 0.30 1.00 0.88–1.14 0.97 1.02 0.94–1.10 0.65
Non-HMO private 1.20 1.11–1.29 ⬍0.001 1.15 1.06–1.24 0.001 1.26 1.09–1.47 0.002 1.11 1.01–1.22 0.03
Admission source (reference
group: non-nursing home)
Nursing Home 1.75 1.59–1.92 ⬍0.001 1.52 1.38–1.69 ⬍0.001 1.15 0.99–1.36 0.07 1.86 1.63–2.11 ⬍0.001
Comorbidity score (reference
group: 0)†
1 1.14 1.07–1.21 ⬍0.001 1.05 0.99–1.12 0.08 0.78 0.70–0.87 ⬍0.001 1.14 1.06–1.24 0.001
2 0.91 0.86–0.96 0.001 0.85 0.81–0.90 ⬍0.001 0.90 0.81–0.99 0.03 0.99 0.92–1.06 0.76
3 or 4 1.02 0.99–1.11 0.57 0.90 0.85–0.96 0.02 0.72 0.65–0.79 ⬍0.001 1.19 1.11–1.28 ⬍0.001
5 or greater 1.51 1.42–1.61 ⬍0.001 1.24 1.16–1.33 ⬍0.001 0.88 0.79–0.98 0.03 1.86 1.72–2.00 ⬍0.001
Admission diagnosis (reference
group: angina)
Diabetes 10.1 7.49–13.7 ⬍0.001 8.49 6.25–11.5 ⬍0.001 5.33 2.90–9.79 ⬍0.001 9.39 6.53–13.5 ⬍0.001
COPD‡ 10.5 8.01–13.8 ⬍0.001 8.31 6.32–10.9 ⬍0.001 7.28 4.29–12.3 ⬍0.001 9.22 6.63–12.8 ⬍0.001
Pneumonia 20.5 15.9–26.4 ⬍0.001 16.0 12.4–20.7 ⬍0.001 15.4 9.23–25.8 ⬍0.001 22.7 16.6–31.0 ⬍0.001
Congestive heart failure 20.5 15.8–26.5 ⬍0.001 18.2 14.0–23.6 ⬍0.001 34.4 26.7–44.5 ⬍0.001 23.7 17.4–32.5 ⬍0.001
Stroke 91.2 70.7–118 ⬍0.001 83.8 65.0–108 ⬍0.001 32.5 19.4–54.4 ⬍0.001 58.5 42.9–79.7 ⬍0.001
Chronic renal failure 73.8 56.7–96.0 ⬍0.001 62.7 48.1–81.6 ⬍0.001 35.5 20.9–60.1 ⬍0.001 89.9 65.6–123 ⬍0.001
DNR — — — 6.98 6.71–7.27 ⬍0.001 — — — — — —
*All models adjusted by age, sex, race, admission source, payment source, comorbidity scores, and admission diagnosis, as categorized above. DNR ⫽ do-not-resuscitate orders, OR ⫽ odds ratio, CI ⫽
confidence interval.
†Comorbidity scores were based on the Charlson comorbidity index adapted for administrative databases.
‡Chronic obstructive pulmonary disease.

403
404
ing associations with and without DNR included in multi-
variable models that included all variables. We compared
sex-mortality and ethnicity-mortality associations in those
with and without a DNR order; P values of these associa-
tions were determined using the chi-squared test. In multi-
variable analysis, we used logistic regression to define pre-
dictors of in-hospital mortality with and without including
DNR in the models. Multivariable analysis was performed
for the entire cohort and in groups stratified by DNR status.
All demographic variables were included in the models, and
separate analyses were performed for each diagnostic group,
as described above. We did not test for statistical differences
in the mortality odds ratios before or after DNR was in-
cluded in the model, as statistically insignificant differences
could have clinical significance.
We were concerned that adjustment for prognostic fac-
tors was incomplete in multivariable models of in-hospital
mortality, so we performed a sensitivity analysis using pro-
pensity scores for DNR in the mortality model, using both
continuous and categorical scores.29 Propensity scores sum-
marize the association of multiple potential confounders in
a single variable; in this study, the propensity score sum-
marized the association of demographics and other patient
characteristics with DNR status. The Statistical Analysis
System (SAS) software (8.1 release, SAS Institute Inc.;
Cary, NC) and the Stata statistical package (version 7.0,
Stata Corporation; College Station, TX) were used for sta-
tistical analysis.
Results
Cohort
In 1999, there were 466 539 admissions to nonfederal
hospitals in California for 1 of the 7 diagnoses studied,
130 329 of which were not first admissions during the
period of study. A cohort of 327 890 patients remained after
excluding admissions for non-California residents (n ⫽
8320). DNR orders were written within 24 hours for 7.7%
of admissions (n ⫽ 25 195). The average age was 64.5 years
(median 70 years, intraquartile range, 54 – 80 years); women
constituted 53.0% of the cohort, whites 64.0%, African-Amer-
icans 10.2%, Hispanics 11.7%, and Asians 6.7%. Comorbidity
scores ranged from 0 to 28, with 57.1% of patients having a
score of 0 –1, consistent with limited life-threatening comor-
bidity.
Predictors of DNR
In unadjusted analyses, patients who were white, female,
and older were more likely to have DNR orders (Table 1).
Patients with higher comorbidity scores and admission from
a nursing home were also more likely to have DNR orders.
These associations persisted in the analyses adjusted by age,
sex, ethnicity, admission source, payment source, comor-
The American Journal of Medicine, Vol 118, No 4, April 2005
bidity scores, and admission diagnosis (Table 1). In univar-
iate analysis, rates of DNR orders were higher in those with
Medicare coverage, but this association was not apparent
after adjustment. Rates of DNR orders varied by admission
diagnosis (Table 1). Women and non-Hispanic whites were
more likely to be DNR for all admission diagnoses (Table 2).
Predictors of mortality
Overall mortality for the entire cohort was 4.1% (n ⫽
13 549). Those with a DNR order were more likely to die
(odds ratio [OR] 9.55, 95% confidence intervals [CI], 9.20 –
9.92, P ⬍0.001), and the association persisted in the mul-
tivariable analysis (Table 3). When the score for propensity
of DNR orders based on demographic and other patient
characteristics was used in the mortality model instead of
the individual demographic variables, DNR remained
strongly associated with mortality (overall OR 7.17,
P ⬍0.001; OR 4.48 –15.7 among the different diagnostic
groups, all with P ⬍0.001).
Influence of DNR on predictors of mortality
Among those without a DNR order, in-hospital mortality
was similar for women compared with men (2.6% vs. 2.8%,
P ⬍0.001). Among those with a DNR order, in-hospital
mortality was lower in women than in men (19.4% vs.
23.7%, P ⬍0.001). Similarly, in comparing in-hospital mor-
tality between whites and other ethnicities, mortality rates
were similar (non-whites vs. whites, 2.6% vs. 2.8%, P ⫽
0.005) among those without DNR orders. However, among
those with DNR orders, in-hospital deaths were more fre-
quent in non-whites (22.8% vs. 20.8%, P ⫽ 0.003). The
directions of these associations were similar after adjust-
ment (Table 3).
After adjustment, women were less likely to die in the
hospital overall and in most diagnostic groups (overall
women vs. men: OR 0.92, 95% CI 0.88 – 0.95, P ⬍0.001;
Figure 1). The difference between women and men was gen-
erally greater when DNR was included in the analysis, and
women appeared to survive more frequently than before DNR
was included in the model (overall women vs. men: OR 0.85,
95% CI 0.82– 0.88, P ⬍0.001; Table 3 and Figure 1).
When DNR was omitted from multivariable models,
in-hospital mortality appeared lower in non-whites com-
pared with whites (overall non-whites vs. whites: OR 0.92,
95% CI 0.89 – 0.96, P ⬍0.001). However, when DNR status
was included, the apparent survival advantage for non-whites
was reduced in every diagnostic group (Figure 2) and was
eliminated overall (overall non-whites vs. whites: OR 1.06,
95% CI 1.01–1.11, P ⫽ 0.01; Table 3).
Multivariable analyses stratified by those with DNR or-
ders and those without (Table 3) showed that women were
more likely than men to survive in both groups (women vs.
men with DNR orders: OR 0.76, 95% CI 0.72– 0.81,
P ⬍0.001; without DNR orders: OR 0.89, 95% CI 0.85–
Bardach et al DNR Orders and Minorities 405

Figure 1 Adjusted odds ratios for in-hospital mortality, women vs. men, in the 7 diagnostic groups and the total cohort before (gray) and
after (black) inclusion of DNR status in the model. Odds ratios were consistently lower when DNR status was included in the model. DNR
⫽ do-not-resuscitate orders, COPD ⫽ chronic obstructive pulmonary disease.

0.94, P ⬍0.001) and that non-whites were as likely to die as
whites in both groups (non-whites vs. whites with DNR
orders: OR 1.08, 95% CI 0.99 –1.17, P ⫽ 0.08; without
DNR orders: OR 1.01, 95% CI 0.96 –1.06, P ⫽ 0.66).
Discussion
In this study of Californian in-patients, we found that a
physician DNR order within 24 hours of admission was
associated with in-hospital death even after adjustment for
age, demographic characteristics, and comorbidity score,
and in an analysis with propensity scores. Because cardio-
pulmonary resuscitation has a limited impact on survival,30
we assume that DNR reflects less aggressive care overall, as
demonstrated in other studies.3 Although the presence of a
DNR order within 24 hours of admission would be expected
to be associated with disease severity as well, it also indi-
cates a patient’s preferences for care.15,16 Thus, the presence
of a DNR order likely captures information not otherwise
represented in administrative databases.
After adjustment, we found that odds of having a DNR
order were 19% greater in women compared with men, and
43% greater in non-Hispanic whites compared with other
ethnicities. These findings confirm results of several previ-
ous studies.7-11,31 The one large study that did not show a
difference in DNR usage between whites and non-
whites12,32 required informed consent to participate,33 and
patients who are willing to participate in a study may not be
representative of all patients. Ethnic differences in use of
advanced directives likely reflect cultural preferences, such
as whether death should occur in the hospital. Although
details of patient management are limited in our study, it is
the largest to evaluate DNR use in a well-defined, socio-
economically diverse multiethnic population, and provides
new data on frequency of DNR orders in Hispanics and
Asian-Americans.
We found that DNR status was a confounder in the
association of ethnicity and mortality. Non-whites were
less likely to die in the hospital in an analysis ignoring
DNR status, but this apparent mortality benefit disap-
peared after adjustment for DNR status. Thus, higher
rates of DNR orders in whites may create the appearance
406 The American Journal of Medicine, Vol 118, No 4, April 2005

Figure 2 Adjusted odds ratios for in-hospital mortality, non-whites vs. whites, in the 7 diagnostic groups and the total cohort before (gray)
and after (black) inclusion of DNR status in the model. Odds ratios were consistently greater when DNR status was included in the model.
DNR ⫽ do-not-resuscitate orders, COPD ⫽ chronic obstructive pulmonary disease.

of a survival advantage for other ethnicities, and unmea-
sured differences in use of DNR orders may mask ethnic
and sex disparities. In addition, odds of in-hospital mor-
tality tended to be lower for women, and the survival
advantage for women was even greater after DNR status
was included in the models. Failure to capture DNR in
the model could misrepresent the component of outcome
that is due to disparities in the delivery of medical care.
Furthermore, DNR may be a confounder in other studies
of risk factors for mortality.
Previous studies have ignored DNR status when evalu-
ating outcome differences among ethnicities. In these stud-
ies of a variety of diseases and populations, some have
documented lower mortality rates for whites,19-21,34-39 some
found no difference,39-43 and some reported higher mortal-
ity rates for whites.18,43-45 Differences in follow-up dura-
tion, study settings, and included diseases may explain these
variable results. In this study, we found that ethnic dispar-
ities varied broadly among the diagnostic categories, but
that whites tended to have higher overall rates of in-hospital
mortality when DNR status was ignored. However, when
DNR was included, the apparent benefit of being nonwhite
was reduced for every disease category studied and, overall,
whites had lower mortality rates than other ethnicities.
This study is limited by its use of administrative data.
Although coding of most variables included in the analysis
is probably reliable, the DNR item, new in 1999, may be
more vulnerable to errors as abstractors gain experience
with it. However, random errors in coding tend to reduce the
impact of the variable. Another major limitation of this
study is incomplete knowledge about severity of disease,
which is inadequately captured by demographic character-
istics and the comorbidity score we employed. In fact, we
found that both DNR and mortality were associated with
measures of disease severity, including age and a comor-
bidity score, so DNR may be acting as a surrogate for
unmeasured disease severity. However, this would not fully
explain the shift in ethnic disparity we demonstrated be-
cause rates of DNR were greater in whites, a group that
generally has lower disease severity at admission.46-48 Fur-
thermore, if DNR was simply a marker for unmeasured
disease severity, adjusting for it should always reduce the
apparent sex-mortality and ethnicity-mortality associa-
tions,49 and this was not the case. Finally, we considered
Bardach et al DNR Orders and Minorities
DNR orders only within the first 24 hours of admission,
when a previously established advanced directive, rather
than disease course, is likely to be more important. Thus,
unmeasured disease severity is not likely to explain fully the
association of DNR with sex and ethnicity or the shifting of
sex-mortality and ethnicity-mortality associations when
DNR is included in the models.
In this study, consideration of DNR status was important
in estimations of sex and ethnic differences. If we had not
adjusted for DNR status, we would have concluded that
non-whites had lower rates of in-hospital mortality than
whites. However, with adjustment for DNR, we found just
the opposite. With adjustment, we found higher rates of
death in Asians but similar rates in African-Americans and
Hispanics compared with whites. Administrative data sim-
ilar to that in the OSHPD database has been an important
source of information about outcome differences between
ethnicities and sexes. Failure to include DNR in previous
studies of mortality could have led to a misrepresentation of
ethnic and sex differences in outcomes. In addition, hospital
mortality rates generated from administrative data are often
used to assess hospital quality. The adjustment for DNR
may be important in evaluating hospital quality because
frequency of use is likely to vary between institutions and to
contribute to differences in adjusted mortality rates. Future
studies should routinely consider the association of DNR
status with patient outcomes.
Acknowledgments
Many thanks to Eugene Bardach, Heather Fullerton, and
Mendocino Steele for their editorial contributions.
References
1. Jonsson PV, McNamee M, Campion EW. The ‘Do not resuscitate’
order. A profile of its changing use. Arch Intern Med. 1988;148:2373–
2375.
2. Wenger NS, Pearson ML, Desmond KA, Kahn KL. Changes over time
in the use of do not resuscitate orders and the outcomes of patients
receiving them. Med Care. 1997;35:311–319.
3. Rubenfeld GD. Do-not-resuscitate orders: a critical review of the
literature. Respir Care. 1995;40:528 –535; discussion 35–37.
4. Jayes RL, Zimmerman JE, Wagner DP, et al. Do-not-resuscitate orders
in intensive care units. Current practices and recent changes. JAMA.
1993;270:2213–2217.
5. Emanuel EJ, Emanuel LL. The economics of dying. The illusion of
cost savings at the end of life. N Engl J Med. 1994;330:540 –544.
6. Lundberg GD. American health care system management objectives.
The aura of inevitability becomes incarnate. JAMA. 1993;269:2554 –
2555.
7. Wenger NS, Pearson ML, Desmond KA, et al. Epidemiology of
do-not-resuscitate orders. Disparity by age, diagnosis, gender, race,
and functional impairment. Arch Intern Med. 1995;155:2056 –2062.
8. Shepardson LB, Gordon HS, Ibrahim SA, et al. Racial variation in the
use of do-not-resuscitate orders. J Gen Intern Med. 1999;14:15–20.
407
9. Jayes RL, Zimmerman JE, Wagner DP, Knaus WA. Variations in the
use of do-not-resuscitate orders in ICUS. Findings from a national
study. Chest. 1996;110:1332–1339.
10. García JA, Romano PS, Chan BK, et al. Sociodemographic factors and
the assignment of do-not-resuscitate orders in patients with acute
myocardial infarctions. Med Care. 2000;38:670 – 678.
11. Gan SC, Beaver SK, Houck PM, et al. Treatment of acute myocardial
infarction and 30-day mortality among women and men. N Engl
J Med. 2000;343:8 –15.
12. Phillips RS, Hamel MB, Teno JM, et al. Patient race and decisions to
withhold or withdraw life-sustaining treatments for seriously ill hos-
pitalized adults. SUPPORT Investigators. Study to Understand Prog-
noses and Preferences for Outcomes and Risks of Treatments. Am J
Med. 2000;108:14 –19.
13. Shepardson LB, Youngner SJ, Speroff T, Rosenthal GE. Increased risk
of death in patients with do-not-resuscitate orders. Med Care. 1999;
37:727–737.
14. Wenger NS, Pearson ML, Desmond KA, et al. Outcomes of patients
with do-not-resuscitate orders. Toward an understanding of what do-
not-resuscitate orders mean and how they affect patients. Arch Intern
Med. 1995;155:2063–2068.
15. Hakim RB, Teno JM, Harrell FE Jr, et al. Factors associated with
do-not-resuscitate orders: patients’ preferences, prognoses, and physi-
cians’ judgments. SUPPORT Investigators. Study to Understand Prog-
noses and Preferences for Outcomes and Risks of Treatment. Ann
Intern Med. 1996;125:284 –293.
16. Hamel MB, Teno JM, Goldman L, et al. Patient age and decisions to
withhold life-sustaining treatments from seriously ill, hospitalized
adults. SUPPORT Investigators. Study to Understand Prognoses and
Preferences for Outcomes and Risks of Treatment. Ann Intern Med.
1999;130:116 –125.
17. Eley JW, Hill HA, Chen VW, et al. Racial differences in survival from
breast cancer. Results of the National Cancer Institute Black/White
Cancer Survival Study. JAMA. 1994;272:947–954.
18. Gordon HS, Harper DL, Rosenthal GE. Racial variation in predicted
and observed in-hospital death. A regional analysis. JAMA. 1996;276:
1639 –1644.
19. Castaner A, Simmons BE, Mar M, Cooper R. Myocardial infarction
among black patients: poor prognosis after hospital discharge. Ann
Intern Med. 1988;109:33–35.
20. Gray RJ, Nessim S, Khan SS, et al. Adverse 5-year outcome after
coronary artery bypass surgery in blacks. Arch Intern Med. 1996;156:
769 –773.
21. Greenwald HP, Polissar NL, Borgatta EF, et al. Social factors, treat-
ment, and survival in early-stage non-small cell lung cancer. Am J
Public Health. 1998;88:1681–1684.
22. Kasiske BL, Neylan JF 3rd, Riggio RR, et al. The effect of race on
access and outcome in transplantation. N Engl J Med. 1991;324:302–
307.
23. Jha AK, Shlipak MG, Hosmer W, et al. Racial differences in mortality
among men hospitalized in the Veterans Affairs health care system.
JAMA. 2001;285:297–303.
24. Kahn KL, Pearson ML, Harrison ER, et al. Health care for black and
poor hospitalized Medicare patients. JAMA. 1994;271:1169 –1174.
25. Guralnik JM, Land KC, Blazer D, et al. Educational status and active
life expectancy among older blacks and whites. N Engl J Med. 1993;
329:110 –116.
26. OPHS. US Office of Public Health and Science. Healthy People 2010:
Understanding and Improving Health. Washington, DC: OPHS: 2001.
27. Office of State Health Planning and Development (OSHPD). The
California Hospital Discharge Data Reporting Manual, 3rd edition.
Sacramento, CA: Office of Statewide Planning and Development;
1999-2000.
28. D’Hoore W, Sicotte C, Tilquin C. Risk adjustment in outcome assess-
ment: the Charlson comorbidity index. Methods Inf Med. 1993;32:
382–387.
408
29. D’Agostino RB, Jr. Propensity score methods for bias reduction in the
comparison of a treatment to a non-randomized control group. Stat
Med. 1998;17:2265–2281.
30. Peberdy MA, Kaye W, Ornato JP, et al. Cardiopulmonary resuscitation
of adults in the hospital: a report of 14720 cardiac arrests from the
National Registry of Cardiopulmonary Resuscitation. Resuscitation.
2003;58:297–308.
31. Koch KA, Rodeffer HD, Wears RL. Changing patterns of terminal
care management in an intensive care unit. Crit Care Med. 1994;22:
233–243.
32. Borum ML, Lynn J, Zhong Z. The effects of patient race on outcomes
in seriously ill patients in SUPPORT: an overview of economic im-
pact, medical intervention, and end-of-life decisions. Study to Under-
stand Prognoses and Preferences for Outcomes and Risks of Treat-
ments. J Am Geriatr Soc. 2000;48(5 suppl):S194 –S198.
33. A controlled trial to improve care for seriously ill hospitalized patients.
The study to understand prognoses and preferences for outcomes and
risks of treatments (SUPPORT). The SUPPORT Principal Investiga-
tors. JAMA. 1995;274:1591–1598.
34. Bach PB, Cramer LD, Warren JL, Begg CB. Racial differences in the
treatment of early-stage lung cancer. N Engl J Med. 1999;341:1198 –
1205.
35. Ebell MH, Smith M, Kruse JA, et al. Effect of race on survival
following in-hospital cardiopulmonary resuscitation. J Fam Pract.
1995;40:571–577.
36. Siddique RM, Siddique MI, Connors AF Jr, Rimm AA. Thirty-day
case-fatality rates for pulmonary embolism in the elderly. Arch Intern
Med. 1996;156:2343–2347.
37. Horner RD, Lawler FH, Hainer BL. Relationship between patient race
and survival following admission to intensive care among patients of
primary care physicians. Health Serv Res. 1991;26:531–542.
38. Dries DL, Exner DV, Gersh BJ, et al. Racial differences in the outcome
of left ventricular dysfunction. N Engl J Med. 1999;340:609 – 616.
39. Bennett CL, Horner RD, Weinstein RA, et al. Racial differences in
care among hospitalized patients with Pneumocystis carinii pneumonia
The American Journal of Medicine, Vol 118, No 4, April 2005
in Chicago, New York, Los Angeles, Miami, and Raleigh-Durham.
Arch Intern Med. 1995;155:1586 –1592.
40. Conigliaro J, Whittle J, Good CB, et al. Understanding racial variation
in the use of coronary revascularization procedures: the role of clinical
factors. Arch Intern Med. 2000;160:1329 –1335.
41. Taylor HA Jr, Canto JG, Sanderson B, et al. Management and out-
comes for black patients with acute myocardial infarction in the reper-
fusion era. National Registry of Myocardial Infarction 2 Investigators.
Am J Cardiol. 1998;82:1019 –1023.
42. Taylor AJ, Meyer GS, Morse RW, Pearson CE. Can characteristics of
a health care system mitigate ethnic bias in access to cardiovascular
procedures? Experience from the Military Health Services System
J Am Coll Cardiol. 1997;30:901–907.
43. Peterson ED, Wright SM, Daley J, Thibault GE. Racial variation in
cardiac procedure use and survival following acute myocardial infarc-
tion in the Department of Veterans Affairs. JAMA. 1994;271:1175–
1180.
44. Bloembergen WE, Port FK, Mauger EA, Wolfe RA. Causes of death
in dialysis patients: racial and gender differences. J Am Soc Nephrol.
1994;5:1231–1242.
45. Phillips RS, Hamel MB, Teno JM, et al. Race, resource use, and
survival in seriously ill hospitalized adults. The SUPPORT Investiga-
tors. J Gen Intern Med. 1996;11:387–396.
46. Horner RD, Oddone EZ, Matchar DB. Theories explaining racial
differences in the utilization of diagnostic and therapeutic procedures
for cerebrovascular disease. Milbank Q. 1995;73:443– 462.
47. Ghali JK, Cooper RS, Kowatly I, Liao Y. Delay between onset of chest
pain and arrival to the coronary care unit among minority and disad-
vantaged patients. J Natl Med Assoc. 1993;85:180 –184.
48. Cooper RS, Simmons B, Castaner A, et al. Survival rates and prehos-
pital delay during myocardial infarction among black persons. Am J
Cardiol. 1986;57:208 –211.
49. Weinberg CR. Toward a clearer definition of confounding. Am J
Epidemiol. 1993;137:1– 8.