ethics in cardiopulmonary medicine

Narcotic and Benzodiazepine Use After

Withdrawal of Life Support*
Association With Time to Death?
Jeannie D. Chan, PharmD, MPH; Patsy D. Treece, RN, MN;
Ruth A. Engelberg, PhD; Lauren Crowley, BA;
Gordon D. Rubenfeld, MD, MSc, FCCP; Kenneth P. Steinberg, MD, FCCP; and
J. Randall Curtis, MD, MPH, FCCP

Objective: To determine whether the dose of narcotics and benzodiazepines is associated with
length of time from mechanical ventilation withdrawal to death in the setting of withdrawal of
life-sustaining treatment in the ICU.
Design: Retrospective chart review.
Setting: University-affiliated, level I trauma center.
Patients: Consecutive critically ill patients who had mechanical ventilation withdrawn and
subsequently died in the ICU during two study time periods.
Results: There were 75 eligible patients with a mean age of 59 years. The primary ICU admission
diagnoses included intracranial hemorrhage (37%), trauma (27%), acute respiratory failure
(27%), and acute renal failure (20%). Patients died during a median of 35 min (range, 1 to 890
min) after ventilator withdrawal. On average, 16.2 mg/h opiates in morphine equivalents and
7.5 mg/h benzodiazepine in lorazepam equivalents were administered during the time period
starting 1 h before ventilator withdrawal and ending at death. There was no statistically
significant relationship between the average hourly narcotic and benzodiazepine use during the
1-h period prior to ventilator withdrawal until death, and the time from ventilator withdrawal to
death. The restriction of medication assessment in the last 2 h of life showed an inverse
association between the use of benzodiazepines and time to death. For every 1 mg/h increase in
benzodiazepine use, time to death was increased by 13 min (p ⴝ 0.015). There was no
relationship between narcotic dose and time to death during the last 2 h of life (p ⴝ 0.11).
Conclusions: We found no evidence that the use of narcotics or benzodiazepines to treat
discomfort after the withdrawal of life support hastens death in critically ill patients at our center.
Clinicians should strive to control patient symptoms in this setting and should document the
rationale for escalating drug doses. (CHEST 2004; 126:286 –293)

Key words: benzodiazepines; death; end-of-life care; ICU; mechanical ventilation; narcotics; withdrawing life support

Abbreviations: APACHE ⫽ acute physiology and chronic health evaluation; GCS ⫽ Glasgow coma scale

D espite remarkable advances in critical care tech-

nology and treatment, the mortality of patients
in the ICU remains high. A large proportion of
*From the Department of Epidemiology (Dr. Chan), School of
Public Health and Community Medicine, University of Washing-
ton, Seattle; and Harborview Medical Center (Mss. Treece and
Crowley, and Drs. Engelberg, Rubenfeld, Steinberg, and Curtis),
Division of Pulmonary and Critical Care, Seattle, WA.
This project was funded by a grant from the Greenwall Founda-
tion and the National Institute of Nursing Research (grant No.
R01NR/AG05226-03).
Manuscript received August 11, 2003; revision accepted Febru-
ary 23, 2004.
deaths in the ICU occurs in the context of withhold-
ing or withdrawing life-sustaining treatments.1,2 The
Study to Understand Prognoses and Preferences for
Outcomes and Risks of Treatments3 showed that,
according to seriously ill hospitalized patients and
Reproduction of this article is prohibited without written permis-
sion from the American College of Chest Physicians (e-mail:
permissions@chestnet.org).
Correspondence to: Jeannie D. Chan, PharmD, MPH, Harbor-
view Medical Center, 325 Ninth Ave, Room GEH-74, Department
of Pharmacy, Box 359885, Seattle, WA 98104; e-mail: jdchan@
u.washington.edu

286 Ethics in Cardiopulmonary Medicine

their proxies, approximately 50% of patients were in
moderate or severe pain during the last 3 days of life.
The study also suggested that patients with critical
illness and those cared for by intensivists had equal
or worse symptom control than other seriously ill
hospitalized patients.4 The problem of symptom
assessment in the ICU can be particularly challeng-
ing, because many patients have impaired cognition
and communication. A cohort study5 of cancer pa-
tients suggested that between 55% and 75% of
critically ill patients who could describe their symp-
toms reported experiencing moderate-to-severe lev-
els of pain, discomfort, anxiety, sleep disturbance,
unsatisfied hunger, or thirst while in the ICU. In one
study6 of trauma patients receiving intensive care,
95% of physicians and 81% of nurses reported that
their patients had received adequate analgesia,
whereas 74% of these patients rated their pain as
moderate or severe. These studies suggested that
ICU patients commonly experience distressing
symptoms at levels of severity that are substantial
and may be underestimated by caregivers. However,
despite the need to incorporate the principles of
palliative care into practices in the ICU to provide
better symptom control and end-of-life care, there is
limited research to guide clinicians in understanding
and improving practice in this area.7–10
In addition to these documented shortcomings in
symptom assessment and management, there are
considerable discrepancies and variations among
health-care professionals in the way that life-sustain-
ing therapy is withheld or withdrawn.2,11,12 Although
variation does not necessarily indicate poor quality of
care, large degrees of variation suggest a lack of
consensus in clinicians’ approaches to dying patients.
Furthermore, critical care nurses report frustration
with the end-of-life care provided by physicians and
cite as an example their impression that some phy-
sicians are reluctant to use sedation for fear of
hastening death.13,14 This impression is reinforced by
surveys of physicians that reflect concern about
hastening death with the use of sedation in the
setting of withdrawing life-sustaining treatments.15,16
These studies raise the possibility that well-inten-
tioned but unskilled cessation of life-sustaining treat-
ment may cause distressing symptoms, leading to
significant and unnecessary suffering in the final
hours of life.
Given the lack of consistency in the delivery of
end-of-life care and reports from critical care nurses
implying that some nurses may be using narcotics or
benzodiazepines to hasten death,13–16 concerns
about hastening death by oversedating patients are
understandable. We therefore conducted a retro-
spective study to determine whether the use of
narcotics or benzodiazepines is associated with the
www.chestjournal.org
length of time between the removal of mechanical
ventilation and death among ICU patients for whom
life-sustaining treatments were being withdrawn.
Materials and Methods
The data were collected as part of a hospital-wide quality
improvement project. A total of 2,250 patients were admitted to
the ICU of Harborview Medical Center over an 8-month period
including June to August 2000 and July to November 2001. All
patients (178 patients) who died in the ICU during this 8-month
time period were screened for eligibility for participation in the
study. Eligible patients included those who had received me-
chanical ventilation within 1 week before death and had mechan-
ical ventilation discontinued prior to death (137 patients). We
excluded patients who received cardiopulmonary resuscitation
immediately prior to death (22 patients) and those who had no
documentation of life support withdrawal (2 patients). We also
attempted to exclude patients for whom the use of benzodiaz-
epines or narcotics after withdrawal of mechanical ventilation
would be unnecessary, such as those who were brain dead (29
patients) and those who had a Glasgow coma scale (GCS) score
of 3 and died within 24 h of ICU admission (9 patients). A total
of 75 patients were included in the analysis. The institutional
review board reviewed this study and determined it to be exempt
from requirements for written consent, because anonymous data
were obtained from the previous hospital-wide quality improve-
ment project such that patients and clinicians could not be
identified.
Basic demographic characteristics (ie, age, gender, and race)
and clinical information (ie, ICU diagnoses, GCS score, ICU
length of stay, and IV narcotic and sedative drugs used 24 h prior
to death) had been abstracted previously from electronic chart
review as part of a quality improvement project. Based on a
relative potency scale for benzodiazepines and opiates used by a
previous study,17 doses of opiates (ie, fentanyl and morphine)
were compared according to a dose-equivalent conversion factor
of 15 ␮g fentanyl to 1 mg morphine. Lorazepam and midazolam
were compared based on a dose-equivalent conversion factor of
2.5 mg midazolam to 1 mg lorazepam. The cumulative amounts
of benzodiazepines and narcotics used were calculated during the
following three stages of life support withdrawal: (1) the 24-h
period before death; (2) the time period from 1 h prior to
ventilation withdrawal until the time of death; and (3) the 2-h
period prior to death. In response to comments from reviewers,
we also examined the 4-h and the 8-h periods prior to death.
Cumulative medication use during these different time intervals
was divided by the total amount of time in the interval and was
expressed as an average number of milligrams per hour admin-
istered over these time periods.
We used multivariate linear regression modeling with robust
variance estimates (a conservative approach that does not assume
equal variances of parameters in the model) to determine
whether the use of narcotics and sedatives during these time
periods was associated with the length of time from the with-
drawal of mechanical ventilation to death. Our primary analysis
was to test for an association between the predictor variables of
the narcotic and benzodiazepine doses used during the time
period from 1 h prior to ventilator withdrawal until death and the
outcome variable of length of time from ventilator withdrawal to
death. As secondary analyses, we also examined the predictor
variables of the doses of narcotics and benzodiazepines used
during the last 2 h, 4 h, and 8 h of life in association with the same
outcome variable. Hypothesis-driven linear regression models
were constructed by placing a priori defined potential confound-
CHEST / 126 / 1 / JULY, 2004 287
ers between medication use and time to death into the models Table 1—Characteristics of Study Subjects*
followed by the predictor variables of interest. These potential
confounders included age (years), gender, hospital admission Characteristics Values
diagnosis of an intracranial hemorrhage (yes or no), worst GCS
Age, yr 59 ⫾ 19† (16–91)
score in the 24 h prior to death as indicator for neurologic status,
Male gender, % 53
and ICU length of stay (days) as a surrogate marker for disease
Race, %
severity. Comparisons between average hourly medication dos-
Non-Hispanic white 77
ages at different stages of life support withdrawal were analyzed
African American 5
using the Student t test without equal variance assumption.
Asian 3
Differences were considered to be statistically significant for a
Hispanic 3
two-sided p value of ⬍ 0.05. Data analyses were performed using
Uncoded 8
a statistical software package (STATA, version 7.0; Stata Corpo-
Mean worst GCS score 24 h prior to death‡ 4 (3–10)
ration; College Station, TX).
Median ICU length of stay, d 4 (1–82)
Median time to death from ventilator 35 (1–890)
withdrawal, min
Results ICU admission diagnosis, %
Intracranial hemorrhage 37
The mean age of the study population was 59 years Respiratory arrest/failure 27
(age range, 16 to 91 years). The majority of the Trauma 27
patients were non-Hispanic whites. Intracranial Acute renal failure 20
hemorrhage was the most common primary ICU Myocardial infarction 12
admission diagnosis (37%), and neurologic status was Pneumonia 12
ARDS 9
poor in this study population, with a mean worst Sepsis/septic shock 9
GCS score of 4 in the 24 h prior to death. The other Coagulopathy 8
frequent ICU admission diagnoses were trauma Anoxic brain injury 7
(27%), acute respiratory failure (27%), and acute Congestive heart failure 7
renal failure (20%). The median length of ICU stay Stroke 7
GI bleed 5
in this study sample was 4 days. Once the process of Neoplasm 5
ventilator withdrawal was initiated, patients died Other 18
within a median of 35 min (25th percentile to 75th
*Values in parentheses are ranges; n ⫽ 75.
percentile, 16 to 146 min) [Table 1]. †Values given as mean ⫾ SD.
Medications were administered by both bolus and ‡Data available in 74 subjects.
continuous infusion during life support withdrawal.
A progressive increase in narcotic dosage was ob-
served starting 8 h prior to death, with the most
evident increase in dosage occurring during the last during the last 2 h of life. There was no statistically
4 h of life. Similarly, there was a steady increase in significant difference by demographic characteris-
benzodiazepine dosage noted throughout the last 8 h tics, primary ICU admission diagnosis, and ICU
of life (Fig 1). However, justifications for medication length of stay between users and nonusers of benzo-
dosage increase were not consistently documented diazepines, although users tended to have a margin-
using sedation or pain scores, and we therefore were ally higher GCS score (median GCS score, 5 vs 4,
not able to report patient comfort or sedation levels. respectively; p ⫽ 0.054). Among patients who re-
Overall, five patients did not receive any opiate or ceived these medications, morphine (90%) and
benzodiazepine during the 24-h period preceding lorazepam (80%) were the primary agents used
death. Narcotic medications were commonly used compared to other opiates and benzodiazepines,
during the two time periods (ie, 1 h before ventilator respectively.
withdrawal until death and the last 2 h of life) in 84% In this study population, an average of 4 mg/h
of patients (63 patients). Users and nonusers of narcotic (in morphine equivalents), and 1.6 mg/h
opiates were similar in demographic characteristics benzodiazepines (in lorazepam equivalents) were
and neurologic status as assessed by worst GCS administered during the last 24 h of life. Medication
score. However, those who received opiates were dosages were escalated when the initiation of life-
less likely to have trauma as a primary admission sustaining therapy withdrawal was anticipated. Dur-
diagnosis (p ⫽ 0.020) and tended to have longer ICU ing the time interval from 1 h prior to ventilator
lengths of stay (mean length of stay, 9.4 vs 3.7 days, withdrawal until death, average doses of 16.2 mg/h
respectively; p ⫽ 0.012) than did those who did not narcotics and 7.5 mg/h benzodiazepines were admin-
receive opiates. Benzodiazepine use was less fre- istered. These drug dosages represented marked
quent, with only 40% of the patients (30 patients) increases compared to the average hourly dosages of
receiving these sedatives during the time periods opiates and benzodiazepines used during the time
from 1 h before ventilator withdrawal until death and interval from the 24th h before death to 1 h prior to

288 Ethics in Cardiopulmonary Medicine

 Figure 1. Mean dose of narcotic and sedative medication administered to study subjects over the final
24 h of life. For each patient receiving a narcotic or sedative drug corresponding to the particular hour
of “time to death,” the cumulative amount was calculated, and the average hourly dose was tabulated
based on the number of patients receiving a narcotic or sedative drug at that particular hour.

ventilator withdrawal, respectively (p ⬍ 0.001 for
both opiates and benzodiazepines). During the last
2 h of life, average doses of 18.1 mg/h narcotics
and 9.2 mg/h benzodiazepines were administered
(Table 2). These medication dosages were signifi-
cantly higher than the average hourly dosages of
opiates and benzodiazepines used during the 22 h
before death after excluding the final 2 h of life,
respectively (p ⬍ 0.001 for both opiates and benzo-
diazepines).
Multivariate linear regression models examining
the association between narcotic and benzodiazepine
medication used (expressed in milligrams per hour,
the predictor variable) and the time from ventilator
withdrawal to death (the outcome variable) are
presented in Tables 3 and 4. Age, gender, and
neurologic status did not appear to be predictive of
time to death after adjustment for the other covari-
ates, but ICU length of stay was inversely associated
with time to death. Every 1-day increase in ICU stay
corresponded to an approximate 2.5-min reduction
from the time of ventilator withdrawal to death
(p ⱕ 0.03). Adjustment for baseline medication use
(defined as a narcotic or benzodiazepine received in
the 24th h before death), extubation prior to death,
and time periods of data collection (June to August
2000 and July to November 2001) did not change
these estimates appreciably (data not presented).
Our results demonstrated no statistically signifi-
cant relationship between either benzodiazepine or
narcotic dose during the time interval from 1 h prior
to ventilator withdrawal until death and the outcome
of time to death from ventilator discontinuation
(Table 3). However, there was a significant dose
effect of hourly medication use in the last 2 h of life
that was associated with the time from ventilator
withdrawal to death (Table 4). On average, every
1 mg/h increase in benzodiazepine use corresponded
to a statistically significant 13-min increase in the
duration of time between ventilator withdrawal and
death (p ⫽ 0.015). On the contrary, for every 1 mg/h
increase in narcotic use, there was a 2-min reduction
in the time from ventilator withdrawal to death,
although statistical significance was not achieved
(p ⫽ 0.11). Further analyses with the average hourly
amount of medication used in the last 4 and 8 h of
life as the predictors of interest yielded similar
results, with no statistical association seen between
narcotic use and time to death but with a trend
toward increased benzodiazepine use being associ-
ated with an increased time to death (Table 4).
Discussion
The principle of double effect states that it is
acceptable for medications to be used that may have
the potential to hasten death in the setting of
terminal illness, provided that the purpose of the
medications is to relieve suffering. Although there is
some debate about the philosophy of this principle, it
is generally well-accepted in the practice of palliative
medicine.18 Nonetheless, surveys of physicians show
that a substantial minority of physicians withhold or

www.chestjournal.org CHEST / 126 / 1 / JULY, 2004 289

 Table 2—Administration of Narcotic and Benzodiazepine Medications*

Variables Cumulative Amount Hourly Amount

Average medication usage 24 h before death

Morphine, mg 81.4 (2–1,052) 3.4 (0.1–44)
Fentanyl, ␮g 2267.0 (50–14,400) 94.5 (2–600)
Total narcotic in morphine equivalents,† mg 97.0 (2–1,052) 4.0 (0.1–44)
Lorazepam, mg 39.3 (1–379) 1.6 (0.04–16)
Midazolam, mg 34.3 (3–214) 1.4 (0.1–9)
Total benzodiazepine in lorazepam equivalents,‡ mg 37.3 (1–379) 1.6 (0.04–16)
Average medication usage 1 h before ventilator withdrawal to death
Morphine, mg 52.9 (2–399) 15.6 (1–106)
Fentanyl, ␮g 628.3 (25–3,054) 322.3 (25–1222)
Total narcotic in morphine equivalents,§ mg 51.7 (2–399) 16.2 (1–106)
Lorazepam, mg 37.3 (0.5–261) 8.6 (0.1–35)
Midazolam, mg 12.8 (2–28) 8.2 (1.6–18.6)
Total benzodiazepine in lorazepam equivalents,㛳 mg 30.9 (0.5–261) 7.5 (0.1–35)
Average medication usage 2 h before death
Morphine, mg 35.1 (2–286) 17.6 (1–143)
Fentanyl, ␮g 606.0 (25–2,554) 303.0 (13–1,277)
Total narcotic in morphine equivalents,§ mg 36.2 (2–286) 18.1 (1–143)
Lorazepam, mg 20.7 (1–72) 10.3 (0.5–36)
Midazolam, mg 14.8 (2–30) 7.4 (1–15)
Total benzodiazepine in lorazepam equivalents,㛳 mg 18.5 (1–72) 9.2 (0.4–36)
*Values in parentheses are ranges. Narcotic use is calculated based on the combined amount of morphine and fentanyl (fentanyl, 15 ␮g; morphine,
1 mg). Benzodiazepine use is calculated based on the combined amount of lorazepam and midazolam (midazolam, 2.5 mg; lorazepam, 1 mg).
†Calculation based on 70 patients.
‡Calculation based on 42 patients.
§Calculation based on 63 patients.
㛳Calculation based on 30 patients.

limit narcotics and benzodiazepines at the end of life benzodiazepines are not being used in ways that
out of fear that they will be perceived to be hastening significantly hasten death. This is consistent with the
death.15 In this context, it is useful to know whether results of a previous report19 stating that survival
narcotics and benzodiazepines are being used in a duration was unrelated to morphine dosage in pa-
way that hastens death. tients undergoing mechanical ventilation withdrawal
Our study suggested that in our ICU, after the as part of end-of-life care. Other non-ICU stud-
withdrawal of mechanical ventilation, narcotics and ies20 –23 also have documented the lack of relation-
ship between opiate and sedative use during end-of-
life care and time to death. In a prospective study of
Table 3—Dose-Response Relationship Between Average 120 patients with terminal cancer assisted by a home
Hourly Narcotic and Sedative Medication Use 1 h care team, there was no detectable survival differ-
Prior to Ventilator Withdrawal and Time From
ence between sedated and nonsedated patients.20
Ventilator Withdrawal to Death*
There was also no significant difference found in
Time to another study21 examining survival time between the
Variables Death, min 95% CI p Value
30 sedated patients and the 85 nonsedated patients
Hourly narcotic dose,† ⫺ 2.2 ⫺ 7.4–3.0 0.41 in a hospice inpatient unit. In a retrospective study of
mg/h 238 patients receiving palliative care during the last
Hourly benzodiazepine 10.4 ⫺ 8.7–29.5 0.28
dose,‡ mg/h
week of life, patients who received a marked increase
Age, yr 0.6 ⫺ 2.1–3.3 0.64 in opiate dosage did not have a shorter survival time
Sex ⫺ 5.2 ⫺ 95.2–84.7 0.91 compared to those who received no increases.22
Worst GCS score 25.5 ⫺ 4.3–55.4 0.093 Finally, among the 227 patients with terminal cancer
Intracranial hemorrhage ⫺ 11.6 ⫺ 98.4–75.2 0.79
who were admitted to a hospice unit in Taiwan,23
ICU length of stay, d ⫺ 2.5 ⫺ 4.7–⫺ 0.2 0.030
there was no statistically significant difference in
*CI ⫽ confidence interval. survival time between sedated and nonsedated pa-
†Estimate for hourly narcotic adjusted for benzodiazepine use, age,
sex, GCS, intracranial hemorrhage, and ICU length of stay.
tients. All but one of these studies was performed in
‡Estimate for hourly benzodiazepine adjusted for narcotic use, age, the hospice setting, and medication assessment was
sex, GCS, intracranial hemorrhage, and ICU length of stay. completed over a much longer spectrum of time

290 Ethics in Cardiopulmonary Medicine

 Table 4 —Dose-Response Relationship Between sumption and coagulability, may promote protein
Average Hourly Narcotic and Sedative Medication Use catabolism and proinflammatory cytokine expres-
and Time From Ventilator Withdrawal to Death*
sion, and thus may trigger the systemic inflammatory
Time to response.24 –27 The anxiolytic property of the benzo-
Death, diazepines may have achieved a calming effect in
Variables min 95% CI p Value
these dying patients without the compromise of
Medication use 2 h before death hastening death. It is, however, important to empha-
Hourly narcotic dose,† mg/h ⫺ 2.3 ⫺ 5.0–0.5 0.11
size that the positive association between benzodiaz-
Hourly benzodiazepine 13.0 2.7–23.3 0.015
dose,‡ mg/h epine use and increasing time to death from me-
Age, yr 0.7 ⫺ 2.0–3.3 0.62 chanical ventilation withdrawal in this study can only
Sex 10.5 ⫺ 78.2–99.2 0.81 be considered as being hypothesis-generating until
Worst GCS score 25.8 ⫺ 4.0–55.5 0.089
these results can be confirmed with a larger study
Intracranial hemorrhage ⫺ 15.9 ⫺ 89.8–58.0 0.67
ICU length of stay, d ⫺ 2.6 ⫺ 4.8–⫺ 0.4 0.024 population.
Medication use 4 h before death The use of an average hourly amount of narcotics
Hourly narcotic dose,† mg/h ⫺ 1.4 ⫺ 6.1–3.3 0.56 and benzodiazepines from 1 h prior to the with-
Hourly benzodiazepine 11.7 ⫺ 2.4–25.8 0.10
drawal of mechanical ventilation until death as the
dose,‡ mg/h
Age, yr 0.8 ⫺ 1.9–3.4 0.57 marker of exposure to these medications is an arbi-
Sex 3.8 ⫺ 83.0–90.5 0.93 trary time period. We, however, believe that the 1-h
Worst GCS score 24.4 ⫺ 5.7–54.5 0.11 window prior to the withdrawal of mechanical ven-
Intracranial hemorrhage ⫺ 9.2 ⫺ 81.8–63.4 0.80
tilation allowed us to capture the possibility of
ICU length of stay, d ⫺ 2.4 ⫺ 4.5–⫺ 0.2 0.034
Medication use 8 h before death anticipatory dosing before withdrawal and that it is a
Hourly narcotic dose,† mg/h ⫺ 0.3 ⫺ 5.9–5.2 0.91 reasonable proxy for the “true” effect of these med-
Hourly benzodiazepine 13.7 ⫺ 0.7–28.1 0.062 ications preceding death. Furthermore, we postu-
dose,‡ mg/h
lated that medication use during the last 2 h of life is
Age, yr 1.0 ⫺ 1.6–3.7 0.45
Sex 3.9 ⫺ 81.2–89.1 0.93 the most clinically sensitive measure to detect an
Worst GCS score 24.2 ⫺ 5.6–54.1 0.11 association with hastening of death, as patients would
Intracranial hemorrhage ⫺ 0.5 ⫺ 71.9–70.9 0.99 be more susceptible to this effect of opiates and
ICU length of stay, d ⫺ 2.4 ⫺ 4.6–⫺ 0.3 0.029
benzodiazepines during the time immediately pre-
*See Table 3 for abbreviations not used in text. ceding death.
†Estimate for hourly narcotic adjusted for benzodiazepine use, age,
sex, GCS, intracranial hemorrhage, and ICU length of stay.
We documented an increase in medication use
‡Estimate for hourly benzodiazepine adjusted for narcotic use, age, during the last 4 to 8 h of life, which was consistent
sex, GCS, intracranial hemorrhage, and ICU length of stay. with that reported by others. Hall and Rocker11
showed that the increase in morphine administration
was most evident in the final 4 h before death, a time
frame that is consistent with the average time from
prior to death, but, nonetheless, our results are
the withdrawal of life support to death in their study.
consistent with these prior findings that opiate and
Although the medication dosages that we reported
sedative use is not associated with the hastening of
are similar to the findings from earlier studies,17,28
death.
Our results demonstrate no statistically significant there is tremendous variability in the range of these
dose-response relationship between either benzodi- medications. This variability is likely due to the fact
azepine or narcotic use during the time interval from that symptom burden is highly unpredictable, the
1 h prior to ventilator withdrawal until death and the assessment of symptoms can be difficult in the dying
outcome of time to death after ventilator withdrawal. critically ill population, and tolerance to narcotics
However, we found a statistically significant associ- and benzodiazepines is common, leading to large
ation between increased benzodiazepine dose during variations in medication administration.
the last 2 h of life and increased time from ventilator Our study has several important limitations. First,
withdrawal to death. Perhaps the most likely expla- the study was limited by a small sample size and may
nation for this finding is that those patients who not have sufficient power to detect small differences.
survived longer developed tolerance to benzodiaz- As this is an anonymous database obtained for quality
epines and therefore required a higher dose to improvement purposes, we are unable to add to the
maintain comfort. Alternatively, symptom distress sample size. The confirmation of our results with a
may hasten death in the setting of withdrawal of larger population would be useful. In addition, we
life-sustaining treatment. Physiologic stress from un- were unable to collect additional data such as acute
relieved pain and anxiety may increase oxygen con- physiology and chronic health evaluation (APACHE)

www.chestjournal.org CHEST / 126 / 1 / JULY, 2004 291

scores because of the anonymous nature of the data.
However, it is not clear that severity of illness,
independent of neurologic status, would be associ-
ated with opiate or benzodiazepine use and therefore
would contribute as a confounder. Second, the re-
striction to medication use during the last 2 h prior to
death may be a more clinically sensitive measure for
detecting an association with hastening of death, but
it is problematic for the subset of patients (25% of
the study population) who took longer than 2 h to die
after ventilator withdrawal. This is because the as-
sessment of exposure (drug use in the last 2 h of life)
occurs after the initiation of the outcome measure-
ment (ie, time from ventilator withdrawal to death)
in these patients. Since an important criterion for a
causal relationship is that the exposure must precede
the outcome, inferences of causality cannot be de-
finitively made. Third, an observational study cannot
dismiss the possibility of residual confounding de-
spite analytic efforts through the adjustment of
potential confounders. However, an observational
study is the only ethical approach to this study
question. Fourth, patient comfort or sedation level
was not routinely recorded, and therefore we were
not able to assess whether the levels of sedation and
analgesia were appropriate. Finally, our findings may
have limited generalizability to other ICUs as the
study population consisted of a substantial number of
intracranial hemorrhage and trauma patients. Strat-
ification by diagnosis of intracranial hemorrhage was
considered, but our sample size was too small to
explore effect modification by neurologic status.
Concerns about hastening death by oversedating
patients are understandable, and the judicious use of
narcotics and benzodiazepines is necessary. Never-
theless, clinicians should be aware of the difficulties
of assessing discomfort in critically ill patients and
should understand that many patients develop toler-
ance to sedative and opiate medications. When death
is imminent, these medications should be given in a
way that alleviates suffering and achieves optimal
patient comfort. The principle of double effect,
along with our data suggesting that increasing doses
of narcotics and benzodiazepines were not associated
with hastened death, should reassure critical care
clinicians that they need not limit pain and symptom
management in the setting of discomfort during the
withdrawal of life-sustaining treatments. However,
when narcotic and benzodiazepine medications are
used to manage symptoms during end-of-life care, it
is important for clinicians to record the reasons for
the escalating doses in order to document that these
drugs are being used for the purpose of controlling
symptoms.9,29
292
References
1 Prendergast TJ, Luce JM. Increasing incidence of withhold-
ing and withdrawal of life support from the critically ill. Am J
Respir Crit Care Med 1997; 155:15–20
2 Prendergast TJ, Claessens MT, Luce JM. A national survey of
end-of-life care for critically ill patients. Am J Respir Crit
Care Med 1998; 158:1163–1167
3 The SUPPORT Principal Investigators. 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). JAMA 1996; 274:1591–
1598
4 Desbiens NA, Wu AW. Pain and suffering in seriously ill
hospitalized patients. J Am Geriatr Soc 2000; 48:S183–S186
5 Nelson JE, Meier D, Oei EJ, et al. Self-reported symptom
experience of critically ill cancer patients receiving intensive
care. Crit Care Med 2001; 29:277–282
6 Whipple JK, Lewis KS, Quebbeman EJ, et al. Analysis of pain
management in critically ill patients. Pharmacotherapy 1995;
15:592–599
7 Rubenfeld GD, Curtis JR, End-of-Life Care in the ICU
Working Group. End of life care in the intensive care: a
research agenda. Crit Care Med 2001; 29:2001–2006
8 Danis M, Federman D, Fins JJ, et al. Incorporating palliative
care into critical care education: principles, challenges, and
opportunities. Crit Care Med 1999; 27:2005–2013
9 Truog RD, Cist AFM, Brackett SE, et al. Recommendations
for end-of-life care in the intensive care unit: the Ethics
Committee of the Society of Critical Care Medicine. Crit
Care Med 2001; 29:2332–2347
10 Faber-Langendoen K, Lanken PN. Dying patients in the
intensive care unit: forgoing treatment, maintaining care. Ann
Intern Med 2000; 133:886 – 893
11 Hall RI, Rocker GM. End-of-life care in the ICU: treatments
provided when life support was or was not withdrawn. Chest
2000; 118:1424 –1430
12 Cook DJ, Guyatt GH, Jaeschke R, et al. Determinants in
Canadian health care workers of the decision to withdraw life
support from the critically ill. JAMA 1995; 273:703–708
13 Asch DA, DeKay ML. Euthanasia among US critical care
nurses: practices, attitudes, and social and professional corre-
lates. Med Care 1997; 35:890 –900
14 Asch DA, Shea JA, Jedrziewski MK, et al. The limits of
suffering: critical care nurses’ views of hospital care at the end
of life. Soc Sci Med 1997; 45:1661–1668
15 Faber-Langendoen K. The clinical management of dying
patients receiving mechanical ventilation: a survey of physi-
cian practice. Chest 1994; 106:880 – 888
16 Asch DA, Christakis NA. Why do physicians prefer to with-
draw some forms of life support over others? Intrinsic
attributes of life-sustaining treatments are associated with
physicians’ preferences. Med Care 1996; 34:103–111
17 Wilson WC, Smedira NG, Fink C, et al. Ordering and
administration of sedatives and analgesics during the with-
holding and withdrawal of life support from critically ill
patients. JAMA 1992; 267:949 –953
18 Sulmasy DP, Pellegrino ED. The rule of double effect:
clearing up the double talk. Arch Intern Med 1999; 159:545–
550
19 Daly BJ, Thomas D, Dyer MA. Procedures used in the
withdrawal of mechanical ventilation. Am J Crit Care 1996;
5:331–338
20 Ventafridda V, Ripamonti C, De Conno F, et al. Symptom
prevalence and control during cancer patients’ last days of
life. J Palliat Care 1990; 6:7–11
21 Stone P, Phillips C, Spruyt O, et al. A comparison of the use
Ethics in Cardiopulmonary Medicine
 of sedatives in a hospital support team and in a hospice. Palliat
Med 1997; 11:140 –144
22 Thorns A, Sykes N. Opioid use in last week of life and
implications for end-of-life decision-making. Lancet 2000;
356:398 –399
23 Chiu TY, Hu WY, Lue BH, et al. Sedation for refractory
symptoms of terminal cancer patients in Taiwan. J Pain
Symptom Manage 2001; 21:467– 472
24 Epstein J, Breslow MJ. The stress response of critical illness.
Crit Care Clin 1999; 15:17–33
25 O’Gara PT. The hemodynamic consequences of pain and its
management. J Intensive Care Med 1988; 3:3–5
26 Srivastava S, Chatila W, Amoateng-Adjepong Y. Myocar-
dial ischemia and weaning failure in patients with coronary
www.chestjournal.org
artery disease: an update. Crit Care Med 1999; 27:2109 –
2112
27 Swinamer DL, Phang PT, Jones RL, et al. Effect of routine
administration of analgesia on energy expenditure in critically
ill patients. Chest 93:4 –10
28 Keenan SP, Busche KD, Chen LM, et al. A retrospective
review of a large cohort of patients undergoing the process of
withholding or withdrawal of life support. Crit Care Med
1997; 22:1020 –1025
29 Rubenfeld GD, Crawford S. Principles and practice of with-
drawing life-sustaining treatments in the ICU. In: Curtis JR,
Rubenfeld GD, eds. Managing death in the intensive care
unit: the transition from cure to comfort. New York, NY:
Oxford University Press, 2001; 127–148
CHEST / 126 / 1 / JULY, 2004 293