Laveena MunshiSonny DhananiSam D. ShemieLaura HornbyGenevieve GoreJason Shahin

Predicting time to death after withdrawalof life-sustaining therapy

Received: 6 January 2015Accepted: 17 March 2015Published online: 6 May 2015� Springer-Verlag Berlin Heidelberg andESICM 2015

Take-home message: There currentlyexists significant heterogeneity in studiesevaluating time to death after thewithdrawal of life-sustaining therapy.However, our evaluation of current datarevealed that controlled ventilation,oxygenation, vasopressor use, GlasgowComa Scale/Score and brain stem reflexeswere most consistently associated with ahastened time to death. More rigorousresearch evaluating these important riskfactors is needed in the general intensivecare unit population as well in patients inspecialized units and in donation aftercardiac death candidates.

Electronic supplementary materialThe online version of this article(doi:10.1007/s00134-015-3762-9) containssupplementary material, which is availableto authorized users.

L. MunshiInterdepartmental Division of Critical CareMedicine, and Department of Medicine,University of Toronto, University HealthNetwork and Mount Sinai Hospital,Toronto, Canada

S. DhananiDivision of Pediatric Critical Care, Facultyof Medicine, University of Ottawa, Ottawa,ON, Canada

S. Dhanani � L. HornbyChildren’s Hospital of Eastern OntarioResearch Institute, Ottawa, ON, Canada

S. D. ShemiePediatric Critical Care, Montreal Children’sHospital, McGill University Health Centre,McGill University, Montreal, QC, Canada

G. GoreSchulich Library of Science andEngineering, McGill University, 809Sherbrooke Street West, Montreal H3A0C1, Canada

J. Shahin ())Department of Critical Care, Department ofMedicine, Respiratory Division, RespiratoryEpidemiology Clinical Research Unit,McGill University Health Centre, 3650St-Urbain, Montreal, QC H2X 2P4, Canadae-mail: jason.shahin@mcgill.caTel.: ?1-514-9341934

Abstract Purpose: Predictingtime to death following the with-drawal of life-sustaining therapy isdifficult. Accurate predictions maybetter prepare families and improvethe process of donation after circula-tory death. Methods: Wesystematically reviewed any predic-tive factors for time to death afterwithdrawal of life support therapy.Results: Fifteen observational stud-ies met our inclusion criteria. Theprimary outcome was time to death,which was evaluated to be within 60min in the majority of studies (13/15).Additional time endpoints evaluatedincluded time to death within 30,

120 min, and 10 h, respectively.While most studies evaluated riskfactors associated with time to death,a few derived or validated predictiontools. Consistent predictors of time todeath that were identified in five ormore studies included the followingrisk factors: controlled ventilation,oxygenation, vasopressor use, Glas-gow Coma Scale/Score, and brainstem reflexes. Seven unique predic-tion tools were derived, validated, orboth across some of the studies.These tools, at best, had only mod-erate sensitivity to predicting the timeto death. Simultaneous withdrawal ofall support and physician opinionwere only evaluated in more recentstudies and demonstrated promisingpredictor capabilities. Conclu-sions: While the risk factorscontrolled ventilation, oxygenation,vasopressors, level of consciousness,and brainstem reflexes have beenmost consistently found to be associ-ated with time to death, the additionof novel predictors, such as physicianopinion and simultaneous withdrawalof all support, warrant further inves-tigation. The currently existingprediction tools are not highly sensi-tive. A more accurate andgeneralizable tool is needed to informend-of-life care and enhance the pre-dictions of donation after circulatorydeath eligibility.

Intensive Care Med (2015) 41:1014–1028DOI 10.1007/s00134-015-3762-9 SYSTEMATIC REVIEW

Keywords Withholding treatment �Life support care � Death �

Time factors �Donation after cardiac death

Introduction

The decision to withdraw life-sustaining therapy (WLST)is the most common event prior to death in the intensivecare unit (ICU) [1]. A recent analysis of withdrawalpractices demonstrated that while the majority of patientsdie within 24 h, the ability to predict time to death may notbe accurate and occasionally is unexpectedly prolonged[2]. Accurate prediction of the time to death would allowfamilies to prepare for the loss of their loved one duringterminal care in the ICU and can help improve the processof donation after circulatory death (DCD). ControlledDCD is a method by which patients can become organdonors after WLST, with death determined on the basis ofcirculatory criteria. Approximately 20–30 % of consenteddonors do not die within the time limits required to permitDCD, resulting in false expectations for families andconsumption of hospital resources [3]. The ability to ac-curately predict time to death is crucial to identifyingsuitable donors, as a protracted time to death exposes theorgans to a prolonged warm ischemic time, rendering themless desirable for transplantation [4]. Having the ability toestimate warm ischemic time would assist healthcareproviders to determine donation potential, inform familyon the likelihood of successful donation, minimize emo-tional distress for families and for healthcare providersthemselves, and optimize hospital resource use.

A variety of prediction tools have attempted to predicttime to death following WLST. However, given con-flicting results and lack of external validity, these have notbeen widely adopted. The aim of this study was to per-form a systematic review to identify factors and riskprediction tools associated with time to death in thegeneral ICU population as well as in the setting of con-trolled DCD.

Methods

Search strategy

We searched MEDLINE via OVID (1946 up to present)and PubMed, EMBASE via OVID (1947 up to present),and CENTRAL (inception up to August 2014). We ex-amined reference lists of all included articles and relevantreview articles to identify any additional inclusions.Clinical experts, unpublished studies, and ongoing trialswere explored for inclusion; however, conference ab-stracts and gray literature were not reviewed. Our searchcombined Medical Subject Headings (MeSH; or appro-priate controlled vocabulary) and keywords for time,

death, and withdrawal [Electronic Supplementary Mate-rial (ESM) Appendix 1]. There were no languagerestrictions.

Study selection

Randomized controlled trials, observational cohort, andcase–control studies were included. The patient popula-tions included any patient beyond the neonatal periodwho underwent WLST in the ICU. The minimumdefinition of life-sustaining therapy included any invasiveor non-invasive mechanical ventilation or hemodynamicsupport. Studies examining any variables associated withtime to death were evaluated. Specifically, studies ex-amining risk factors, risk prediction models, clinicaldecision-making tools, or validation of any existing toolswere included. Given the large number of confoundersavailable, only studies that adjusted for confounding wereincluded.

Two reviewers (LM, JS) independently reviewed thetitles and abstracts identified in the literature searches.Full texts were retrieved and reviewed. Any disagree-ments were resolved by overall group consensus (LM,SD, JS). The final list of included studies was assessed forquality of methods.

Data extraction and quality assessment

The two reviewers (LM, JS) independently extractedrelevant data using a customized form developed by ourgroup. Study methodology and reporting were assessedindependently. Given that no gold standard exists for themethodological assessment of risk factor studies, thequestions were drawn from a previously published riskfactor/risk prediction tool systematic review [5]. Tenquestions assessed the methodology; study objectives,outcome description, patient characteristics, multicenterrecruitment, existence of an a priori analysis plan, ad-justment for known risk factors, rationale behind riskfactor inclusion and definition of risk factors, adequacy ofsample size, risk factor selection, and model strategychoice were all considered.

The primary outcome was time to death followingWLST in the ICU. Our aim was to explore factors foundto be statistically significant on multivariable analysis tobe associated with time to death. We also assessed thesuccessful prediction of time to death using a clinicaldecision support tool. Subgroup analysis was performedin studies focusing on DCD. A DCD-like study was

1015

defined as any study in which the patients were identifiedas either DCD candidates or as being similar to those whowould be DCD candidates according to the criteria oforgan procurement agencies.

Data analysis

Given that the data were derived exclusively from ob-servational studies with heterogeneous approaches toanalysis, we did not pool the data or subject it to meta-analysis. Rather, we summarized the literature, and dataanalysis consisted of a tabulation of study characteristics.

Results

The electronic search retrieved 1842 citations resulting in41 full-text articles for review (Fig. 1). Fifteen observa-tional studies fulfilled our inclusion criteria, 13 of whichevaluated an adult population and two a pediatric popula-tion [1, 6–19]. In all studies, with one exception [2], time todeath as an outcome was dichotomized. Time of death

within 60 min was the primary outcome in 13 of the 15included studies, with assessments of 5173 patients [6–18].Three studies additionally evaluated time to death within120 min [6, 12, 14]. Two studies evaluated factors associ-ated with time to death without specifying a time cutoff [2,19], one study included time to death within 30 min [13],and one study included time to death within 10 h [10].Baseline characteristics are summarized in Table 1. Twostudies were derived from one single cohort, with one ofthese studies reporting the results of the entire cohort andthe second study focusing on the potential DCD group [9,11]. Median age among the study population ranged be-tween 45 and 71 years in the adult studies and between 10and 24 months in the pediatric studies. The percentage ofadult patients who died within 60 and within 120 minranged between 44 and 71 % and between 54 and 83 %,respectively. Data for the other time-points were notavailable. One pediatric study reported on the percentage ofpatients who died within 30 and within 60 min, which wasfound to be 72 and 85 %, respectively [13]. The studieswere clinically heterogeneous with regard to patientpopulation and method of WLST. Methodologically, thestudies were also heterogeneous. Although most studiesperformed an exploratory analysis of risk factors of time todeath, only a few studies derived prediction tools of time todeath, and even a smaller subset validated previouslyderived prediction tools (Table 1). Seven studies evaluatedDCD-like patients [6–9, 13, 14, 16], and one studyevaluated identified actual DCD candidates [17]. Thedefinition for a DCD-like patient varied with each study(Table 1). Etiology of admission to ICU was pre-dominantly catastrophic neurologic disorders, followed bycirculatory and respiratory failure. Withdrawal of life-sustaining therapy consisted of withdrawal of ventilatorsupport in all studies (extubation or disconnection) andcessation of vasoactive medications in most studies. Adetailed overview of the specific components of the with-drawal process and whether it occurred simultaneously orin a staged fashion was not described in detail in moststudies (Table 1). Study quality across most of the studieswas noted to be generally moderate to high with respect toclearly identified objectives, multicenter nature, incorpo-ration of major known risk factors, and description of thestatistical model employed. The methodological assess-ment is summarized in Table 2.

Risk factors associated with time to death

Independent risk factors associated with time to deathusing multivariable logistic regression are summarized inTable 3. The majority of these studies focusedspecifically on time to death within 60 min [7, 8, 10–18].There was considerable variability between studies re-garding key predictor risk factors associated with time todeath. Risk factors that were persistently found to beFig. 1 Flow diagram of study selection. ICU Intensive care unit

1016

Tab

le1

Baselinestudycharacteristics

Studya

Studytype

Population

DCDbvs.

general

ICU

population

Age(SD

orIQ

R)

ICU

type

(npatients)

Componentsof

withdrawalc

Methodsof

withdrawald

(sim

ultaneous,

staged

ornot

specified)

Cause

for

withdrawal

Outcome

assessed

Percentage

death

within

60min

Adultpopulation

Brieva[9]

Australia

Predictiontool

derivation

DCDe

55(17)f,

54(22)

Mixed

ICU

(318)d

Mechanical

ventilator

Vasopressors/inotropes

Cardiacpacing

ECMO

IABP

RRT

Notspecified

NA

Death

within

60min

50%

Huynh[10]

USA

Riskfactoranalysis

General

62(16)

Mixed

(157)

Mechanical

ventilator

Notspecified

NA

Predictors

oftimeto

death

andtimeto

death

within

60min

and10h

Mediantimeto

death

(54min;IQ

R12–408)

NA

Brieva[11]

Australia

Riskfactoranalysis

andpredictiontool

derivation

General

66(20)

Mixed

ICU

(765)

Mechanical

ventilator

Vasopressors/inotropes

Cardiacpacing

ECMO

IABP

RRT

Notspecified

NA

Death

within

60min

49.3

%

DaV

ila[17]UK

Riskfactoranalysis

andpredictiontool

derivation

DCD

45(16)

NA

(420)

Mechanical

ventilator

Supportivedrugs

Sim

ultaneous

withdrawal

ofall

support

CNS(63%)

Hypoxia

(13%)

Cardiac(11%)

Death

within

60min

71%

deG

root[8]

Netherlands

ValidationofYee

etal.[15]prediction

tool

DCD

56(15–18)

Mixed

ICU(82)

Mechanical

ventilator

Antibiotics

Inotropic

agents

Notspecified

CNS

Death

within

60min

61%

Rabinstein[12]

USA/

Netherlands

Riskfactoranalysis

(evaluationof

predictors

identified

inYee

etal.tool[15]

ValidationofYee

etal.[15]prediction

tool

General

63(17)

NA

(178)

Mechanical

ventilator

Sedation

Opiate

analgesia

Notspecified

CNS

Death

within

60min

and120min

46%

(60min)

54%

(120min)

Wind[14]

Netherlands

Riskfactoranalysis

andpredictiontool

derivation

DCD

52(13)

Mixed

ICU

(211)

Mechanical

ventilator

Analgesia

andsedation

Inotropic

support

Mechanical

ventilatorand

inotropes

simultaneously

Analgesicsand

sedatives

not

specified

CNS

Death

within

60min

and120min

76%

(60min)

83%

(120min)

Yee

[15]USA

Riskfactoranalysis

General

NA

Neuro

ICU

(149)

Mechanical

ventilator

Notspecified

CNS

Death

within

60min

50%

1017

Table

1continued

Studya

Studytype

Population

DCDbvs.

general

ICU

population

Age(SD

orIQ

R)

ICU

type

(npatients)

Componentsof

withdrawalc

Methodsof

withdrawald

(sim

ultaneous,

staged

ornot

specified)

Cause

for

withdrawal

Outcome

assessed

Percentage

death

within

60min

Cooke[2]USA

Riskfactoranalysis

General

71

Mixed

ICU

(1505)

Mechanical

ventilator

Notspecified

Respiratory

Neurologic

Cardiac

Infectious

Trauma

Gastrointestinal

Malignancy

Predictors

oftimeto

death

Mediantimeto

death

(55.8

min;IQ

R15–330)

Suntharalingam

[16]UK

Riskfactoranalysis

DCD

44(16-65)

Renal

transplant

centers/

General

ICU

(191)

Mechanical

ventilator

Inotropes/vasopressors

Sim

ultaneous

withdrawal

CNS

Death

within

60min

56.5

%

Coleman

[18]

Australia

Evaluationof

previousprediction

tools(W

isconsin

Tool,UNOSTool,

HunterNew

EnglandArea

Score,ICU

staff

prediction)

General

66(15)

Mixed

ICU(81)

Mechanical

ventilator

Oxygen

supplementation

Vasopressors/inotropes

CardiacPacing

ECMO

IABP

RRT

Sedatives

andanalgesia

initiatedfollowing

withdrawal

Notspecified

Cardiac

Multiorgan

dysfunction

syndrome

CNS

Death

within

60min

44%

DeV

ita[7]USA

Riskfactoranalysis,

predictiontool

derivation,

validationof

UNOS

General

DCD

subgroup

57(54-77)

Mixed

ICU

(505)

Mechanical

ventilator

Vasopressors/inotropes

Analgesia

andsedation

ECMO

Ventricularassistdevice

IABP

Pacem

aker

Notspecified

Respiratory

failure

(96%)

CNSfailure

(55%)

Circulatory

failure

(49%)

Postcardiac

arrest

(23%)

Death

within

60min

45.0

%

Lew

is[6]

(Wisconsin

Tool)

USA

Predictiontool

derivation

DCD

NA

Mixed

ICU(43)

Mechanical

ventilator

Vasopressors/inotropes

ECMO

(veno-arterial)

Berlinheart

Notspecified

CNS

Death

within

60min

and120min

56%

(60min)

65%

(120min)

Pediatric

population

Shore

[13]USA

Riskfactoranalysis

andpredictiontool

derivation

DCD

10months

(1 month–

6years)

Mixed

(Pediatric)

ICU

(518)

Mechanical

ventilator

Vasopressors/Inotropes

Notspecified

CNS

Respiratory

failure

Congenital

cardiacdisease

Death

within

30min

and60min

72%

(30min)

87%

(60min)

1018

Table

1continued

Studya

Studytype

Population

DCDbvs.

general

ICU

population

Age(SD

orIQ

R)

ICU

type

(npatients)

Componentsof

withdrawalc

Methodsof

withdrawald

(sim

ultaneous,

staged

ornot

specified)

Cause

for

withdrawal

Outcome

assessed

Percentage

death

within

60min

Zaw

istowski

[19]

USA

Riskfactoranalysis

General

24months

Mixed

(Pediatric)

ICU

(50)

Mechanical

ventilator

Vasopressors/Inotropes

ECMO

Ventricularassistdevice

IABP

Pacem

aker

RRT

Horm

onetherapy

Antimicrobials

Analgesia

andsedation

80%

underwent

simultaneous

withdrawal

within

5min

Respiratory

Cardiac

Liver

disease

Predictors

oftimeto

death

Mediantimeto

death

[15min;IQ

R5,3)

NA

CNSCentral

nervoussystem

,DCDdonationaftercardiacdeath,ECMOextracorporealmem

braneoxygenation,IABPintraaorticballoonpump,ICUintensivecare

unit,IQ

Rinterquartilerange,NA

notavailable,NOSnototherwisespecified,RRTrenal

replacementtherapy,SD

standarddeviation,UNOSUnited

Network

forOrgan

Sharing

aInform

ationispresentedas

thefirstauthorofthestudy,reference

number

inreference

list,andcountryin

whichthestudywas

conducted

bDCDpotentialaccordingBrievaetal.[9]:agegroup(notspecified)andlack

ofmalignancy.DCDpotentialaccordingto

DeV

ilaetal.[11]:DCDacceptedliver

transplantofferspriorto

withdrawal

oflife-sustainingtherapy(LST),detailsnotprovided

onselectioncriteria.DCDpotentialaccordingto

DeG

rootetal.[8]:patientswhodiedas

aconsequence

ofsubarachnoid

hem

orrhage,traumatic

brain

injury,orintracerebralhem

orrhage;DCDpotentialaccordingto

Shore

etal.[13]:pediatricpatientsundergoingwithdrawalofLST.DCDpotentialaccordingto

Windetal.[14]:those

patients

whousually

suffer

from

irreversible

neurologic

injury.DCD

potential

accordingto

Suntharalingam

etal.[16]:aged

16–65years,history

ofhuman

immunodeficiency

virus,suspectednew

-variant

Creutzfeld-Jacobdisease,untreatedmajorsepsis.

DCD

potential

accordingto

DeV

itaet

al.[7]:

DCD

subgroup,age\60years,creatinine\2mg/dL,nooverwhelmingsepsis.

DCD

potential

accordingto

Lew

iset

al.[6]:severebrain

injury

receivingmechanical

ventilationforwhom

aphysician

isevaluatingforbrain

death

withaGlasgow

ComaScore

(GCS)of\5,planforLST

withdrawal

dTotalnumber

indevelopmentalsetandvalidationset

cNotallpapersincluded

adetailedoverview

ofthecomponentswithdrawnormethodsofwithdrawal;therefore

thecomponentsandmethodsofwithdrawal

may

notbelimited

tothelistsabove

eDCD

subgroupofcohortincluded

inthestudyofBrievaet

al.[11](2013)

fNote:foradultstudies,themedianagefordeath

ingiven

inyears;forpediatric

studies,medianagefordeath

isas

indicated.Dataarepresentedas

medianagefortimeto

death

within

60min,

followed

bymedianagefortimeto

death

of[60min

1019

Table

2Methodologyassessment

Studya

Objectives

clearly

described

Main

out-

come

mea-

sures

clearly

identified

Multicenter

nature

Analysis

defined

apriori

Did

the

analysis

accountforma-

jor

known

risk

factors

Rationale

behind

the

inclusion

of

the

risk

factors

included

Were

risk

factors

clearly

defined

Riskfactorselectionformodel

Statistical

model

Huynh[10]

USA

YY

NY

YY

YA

prioriselection,previous

literature,investigator

Allvariableskeptin

Cox

proportional

hazardsmodel

and

logisticregressionmodels

Brieva[9,11]

Australia

YY

YY

YY

YPreviousliterature,investigator,

univariable

analysis

Unclear

Davila[17]UK

YY

NY

YY

YPreviousliterature,investigator,

univariable

analysis

Forw

ardselection

deG

root[8]

Netherlands

YY

NY

YY

YPreviousliterature,investigator,

univariable

analysis

Allvariableskeptin

model

Rabinstein[12]

USA/

Netherlands

YY

YY

YY

YPreviousliterature,investigator,

univariable

analysis

Allvariableskeptin

model

Shore

[13]USA

YY

NY

YY

YUnivariable

analysis

Forw

ardselection

Wind[14]

Netherlands

YY

YY

YN

YUnivariable

analysis

Forw

ardselection

Yee

[15]USA

YY

NY

YY

YUnivariable

analysis

Unclear

Cooke[2]USA

YN

YY

YN

YA

prioriselection

Allvariableskeptin

model

Suntharalingam

[16]UK

YY

YY

YN

YA

prioriselection,previous

literature,investigator

Backwardandforw

ardselection

Coleman

[18]

Australia

YY

YY

YY

YNA

Unclear

Devita[7]USA

YY

YY

YY

YPreviousliterature,investigator,

univariable

analysis

Backwardandforw

ardseletion

Zaw

istowski

[19]

USA

YY

NN

YN

NNA

Unclear

Lew

is[6]USA

YUnclear

Unclear

YY

YY

Investigator

NA

NNo,NAnotavailable,Yyes

aInform

ationis

presentedas

thefirstauthorofthestudy,reference

number

inreference

list,andcountryin

whichthestudywas

conducted

1020

Table 3 Risk factors associated with time to death in multivariable models

Studya Risk factors associated with time to deathon multivariable analysis

Effect estimate (95 % CI) p value

Huynh [10] (60 min) Vasopresssors within 12 h of MVwithdrawal

2.05 (1.37–3.07)

FiO2[70 % 1.92 (1.24–2.98)Huynh [10] (10 h) Vasopresssors within 12 h of MV

withdrawal4.53 (1.52–13.45)

Primary service neurology/neurosurgical 0.30 (0.09–0.95)Brieva [11] (including ICU specialistopinion) (60 min)

Physician opinion 8.44 (4.3, 16.6)c \0.001pH 0.67 (0.47, 0.94) 0.03Systolic blood pressure 0.99 (0.98, 1.00) 0.01GCS 0.85 (0.74, 0.98) 0.03Positive end expiratory pressure 1.17 (1.1, 1.3) \0.01Analgesia 0.32 (0.15, 0.70) \0.01

Brieva [11] (excluding ICU specialistopinion) (60 min)

pH 0.57 (0.41, 0.81)c \0.01Systolic blood pressure 0.99 (0.98, 1.00) \0.01GCS 0.82 (0.72, 0.93) \0.01Positive end expiratory pressure 1.19 (1.08, 1.30) \0.001Analgesia 0.42 (0.21, 0.84) 0.02Spontaneous respiration rate 0.96 (0.94, 0.99) \0.01

Devila [17] (60 min) Age\40 years 3.62 (1.15, 11.4) 0.002No inotropes 0.19 (0.07, 0.52) 0.001No cough/gag 0.15 (0.05, 0.46) 0.001

deGroot [8] (Yee et al. validation [15],absent corneal, absent cough,extensor or absent motor response,oxygenation index[4.2) (60 min)

No other variables found to beindependently predictive of death within60 min on multivariable analysis

Rabinstein [12] (60 min) Absent corneal 2.67 (1.19, 6.01) 0.02Absent cough 4.16 (1.79, 9.70) \0.001Extensor or absent motor 2.99 (1.22, 7.34) 0.02Oxygenation index[3.0 2.31 (1.10, 4.85) 0.03

Shore [13] (60 min) Age\1 month 0.13Norepinephrine, epinephrine,phenylephrine

9.25

Extracorporeal life support 51.9PEEP[10 cmH2O 13.7[0 % spontaneous ventilation 0.59

Shore [13] (30 min) Age\1 month 0.12Norepinephrine, epinephrine,phenylephrine

12.0

Extracorporeal life support 12.3PEEP[10 cmH2O 3.0[0 % spontaneous ventilation 0.2

Wind [14] (60 min) Controlled mechanical ventilation 2.50 (1.18, 5.258) 0.02Wind [14] (120 min) Controlled mechanical ventilation 2.59 (1.17, 5.73) 0.02

Norepinephrine use 4.03 (1.43, 11.32) 0.01Cardiovascular comorbidity 0.32 (0.14, 0.69) 0.004

Yee [15] (60 min) Absent corneal 4.24 (1.57, 11.5) 0.005Absent cough 4.47 (1.93, 10.3) 0.0005Extensor or absent motor 2.83 (1.01, 7.91) 0.05Oxygenation Index 3.36 (1.33, 8.50) 0.01

Cooke [2] Age 0.95 (0.90, 0.99)d

Female 0.86 (0.77, 0.97)Nonwhite 1.17 (1.01, 1.35)4-year college degree 1.21 (1.01, 1.44)Surgical service 1.29 (1.06, 1.56)Vasopressors 1.67 (1.49, 1.88)Intravenous fluids 1.16 (1.01, 1.32)

Suntharalingam [16] (60 min) Age 31–40 years 0.70 (0.38, 1.28) 0.001Age 41–50 years 0.46 (0.29, 0.76) 0.008Age[50 years 0.37 (0.23, 0.59) 0.006Increased FiO2 1.01 (1.00, 1.02)Mode of ventilation: not pressure support 1.67 (1.16, 2.41)

Coleman [18] (60 min) Modified UNOS tool 4.90 (1.38, 17.60)

1021

predictive of time to death within 60 min in more thantwo studies are summarized in Table 4 and include age,female gender, controlled mode of ventilation, oxygena-tion [positive end expiratory pressure (PEEP)], partialpressure of oxygen or fraction of inspired oxygenthreshold, oxygenation index, hemodynamics, neurologicexam [Glasgow Coma Scale/Score (GCS) or brain stemreflexes), pH, simultaneous withdrawal of all supports,low analgesia, and physician opinion. Even though theabove risk factors were significant predictors in multiplestudies, the results were not consistent across all studiesthat examined the risk factors. Consistent predictors ofdeath which were identified in five or more studies werecontrolled mode of ventilation, risk factors associatedwith higher oxygenation requirements, use of vasopres-sors, and lower GCS or impaired brain stem reflexes.

Only one of the 15 studies reported risk factors asso-ciated with time to death within 120 min according tomultivariable logistic regression [14], namely, controlledmechanical ventilation, the use of norepinephrine, and thepresence of a cardiovascular comorbidity. One study re-ported on time to death within 10 h [10]; the authorsreported that use of vasopressors within 12 h prior towithdrawal and neurology/neurosurgery service not beingthe primary service at time of withdrawal were associatedwith a hastened death.

No consistent variables were found in the two pedi-atric studies to be associated with time to death; however,

age, vasopressor use, extracorporeal life support, higherPEEP, controlled mechanical ventilation, simultaneouswithdrawal, female sex, and the absence of continuousrenal replacement therapy were found to be associatedwith time to death in each individual study [13, 17].

When evaluating only the studies of DCD, the pre-dictors of death were similar. The most significantpredictors included controlled mode of ventilation (3studies), oxygenation (4), and neurologic function (3).

Ten studies evaluated the sensitivity and specificity oftheir derived models [6–9, 11–14, 17, 18], and fourstudies [8, 12, 17, 18] validated previously publishedmodels (summarized in Table 5).

The Wisconsin tool is an evaluation tool intended forpotential DCDs with severe neurologic injury who do notfulfill brain death criteria. The tool incorporates respira-tory rate and oxygenation after 10 min of disconnectionfrom the ventilator, tidal volume, negative inspiratoryforce, use of vasopressors, age, airway type, and bodymass index. A score between 10 and 23 is assigned basedupon these variables, with a higher score presumed to beassociated with a higher probability of death. In the initialvalidation study, the tool was found to have a sensitivityand specificity for predicting death within 60 min of 0.83and 0.84 and for predicting time to death within 120 minof 0.85 and 0.45; however when validated in a subsequentcohort, the results were not robust, with a lower sensi-tivity (0.42) and specificity (0.76) [18].

Table 3 continued

Studya Risk factors associated with time to deathon multivariable analysis

Effect estimate (95 % CI) p value

Devita [7] (general population)(60 min)

GCS = 3 2.83 (1.79, 4.46) \0.001SaO2/FiO2\230 1.78 (1.09, 2.90) \0.05Peak inspiratory pressure[35 2.58 (1.49, 4.48) \0.001Respiratory rate off ventilator\8 6.01 (2.29, 4.48) \0.001Diastolic blood pressure (10 mmHg) 0.80 (0.69, 0.93) \0.01PaO2\72 3.10 (1.53, 6.30) \0.01Epinephrine, norepinephrine orphenylephrine[0.2

3.02 (1.75, 5.21) \0.001

All treatments withdrawn within 10 min 8.55 (4.23, 17.30) \0.001Endotracheal tube withdrawn 2.28 (1.33, 3.90) \0.01Comfort medications given during firsthour after withdrawal of life sustainingtherapy

0.35 (0.21, 0.59) \0.001

Devita [7] (UNOS criteria)b 1 2.72e

2 4.62C3 10.60

Zawistowski [19] Simultaneous withdrawal of interventions \0.001Female 0.01No continuous RRT 0.02

CI Confidence interval, FiO2 fraction of inspired oxygen, GCSGlasgow Coma Scale/Score, MV mechanical ventilation, PaO2,partial pressure oxygen, PEEP positive end expiratory pressure,SPO2 Saturation level of oxygen in hemoglobina Information is presented as the first author of the study, followedby time to death criterion or other criteria

b UNOS criteria: see ESM Appendix 2 for full criteriac Relative risk (95 % CI)d Hazard ratios (95 % CI)e Unadjusted rates

1022

The UNOS tool was evaluated in two studies and is anumeric tool created to assess the likelihood of deathwithin 60 min; it relies upon a trial of disconnection fromventilation to evaluate spontaneous respirations andoxygenation [7, 18]. In addition, it incorporates hemo-dynamic variables and the presence of a series ofpharmacologic and mechanical hemodynamic supportdevices. Based upon the number of UNOS criteria pre-sent, a score is assigned between 0 and 5, with 5presumed to be associated with a higher likelihood ofdeath within 60 min. Coleman et al. [18] externallyvalidated an adapted version of the UNOS tool (i.e., theydid not disconnect patients from the ventilator and rede-fined ventilator dependence and oxygen disruption); theseauthors reported the UNOS tool to have a sensitivity of0.61 and specificity of 0.84. In the study by Devita et al.[7] among potential DCD patients, 33 % of those who hada UNOS score of 0 still died within 60 min. While thespecificity of the UNOS score was much better than thatreported by Coleman et al. [18] (88 % of those with 4criteria dying within 60 min), only two of the 505 patientshad a score of 4 or 5.

The Yee score was the first score to explore detailedneurologic predictor variables in a population of patientswith catastrophic neurologic disease [15]. The authorsidentified absent corneal reflex, absent cough reflex, ab-sent or extensor motor response, and an oxygenation

index of [4.2 as risk factors that were associated withtime to death within 60 min. The more features that arepresent, the higher the probability of death within 60 min.This model was further validated by de Groot et al. [8] ina similar population where it demonstrated moderatesensitivity (0.64) and specificity (0.72). These variableswere further evaluated and modified to create the DCD-Nscore by Rabinstein et al. [12] who used a differentoxygenation index cutoff. When evaluated in theirpopulation of neurologically impaired patients, itdemonstrated moderate sensitivity and specificity (0.72and 0.78, respectively); the specificity improved (0.91)when the oxygenation index was modeled as a continuousvariable.

In the Hunter New England Area Composite score, thecombination of ventilator dependence, oxygenation im-pairment, and systolic blood pressure of\100 mmHg wasassociated with a poor sensitivity (0.38) but high speci-ficity (0.96). This score has not been further validated[18].

In the Donors after cardiac death: Validating Identi-fication Criteria (DVIC) study, DaVita at al. [7] evaluatedpredictors of death in a general ICU population andconstructed decision rules for predicting death within60 min using the nonparametric Classification and Re-gression Tree Analyses (CART) model [7]. Of the modelsthey constructed, the best model incorporated depressed

Table 4 Common risk factors associated with time to death

Risk factor Studies evaluatingrisk factor in univariableanalysis (n)

Studies where risk factor wasstatistically significant inmultivariable analysis (n)

Demographic variablesAge[40 12 3Female 10 2Respiratory variablesControlled mode of ventilation, no spontaneous respirationsa 8 5OxygenationPositive end expiratory pressure 7 4PaO2 or FiO2 threshold 10 3Oxygenation indexb 3 3Hemodynamic variablesSystolic blood pressure 5 2Vasopressors 7 5Neurologic variablesGlasgow coma scale 6 3No cough/gag 5 3No corneal 4 2Absent or extensor motor 3 2Metabolic variablespH 4 2OtherSimultaneous WLST or within 10 min 2 2Physician opinion 3 2Less analgesia 3 2

Summary of variables that were statistically significant on multi-variable analysis in two of more studies for time to deathWLST withdrawal of life sustaining therapya Devita et al., respiratory rate\8 off ventilator

b Used as continuous variable for 1 study, threshold of[3.0 and 4.2used in two other studies

1023

Table

5Riskmodels,predictiontools

andperform

ance

VPN

VPPyticfiice pS

ytivitisneSC

UA

tnempoleve

Dled o

Mselbaira

Vcfiicep S

yd utS Brievaet

al.[9]

•ICU

specialistopinionof

death

New

lyDerived

?Validation(halfcohort)

Classificatio

nandRegression

Treeanalysis

(CART

Salfordsystem

s)

0.82

(v)

0.76

(v)

0.80

(v)

0.79

(v)

•Glasgow

comascale

•Sy

stolic

Blood

Pressure

•Po

sitiv

eendexpiratory

pressure

•Sp

ontaneousrespiratoryrate

Brievaet

al.[9]

•Excluding

ICU

specialistopinion

New

lyDerived

?Validation(halfcohort)

Classificatio

nandRegression

Treeanalysis

(CART

Salfordsystem

s)

0.62

(v)

0.78

(v)

0.77

(v)

0.65

(v)

•Glasgow

Com

aScale

•Sy

stolic

Blood

Pressure

•Po

sitiv

eendexpiratory

pressure

•Sp

ontaneousrespiratoryrate

Brievaet

al.[ 11]

•IncludingICU

specialistopinion

New

lyDerived

?Validation(halfcohort)

Multiv

ariableexplanatory

models

0.84

(v)

0.84

(v)

0.72

(v)

0.77

(v)

0.81

(v)

•Analgesia

•pH

•Glasgow

Com

aScale

•Sy

stolic

Blood

Pressure

•Po

sitiv

eendexpiratory

pressure

Brievaet

al.[11]

•Excluding

ICU

specialistop

inion

New

lyDerived

?Validation(halfcohort)

Multiv

ariableexplanatory

models

0.78

(v)

0.82

(v)

0.59

(v)

0.68

(v)

0.75

(v)

•Analgesia

•pH

•Glasgow

Com

aScale

•Sy

stolic

Blood

Pressure

•Po

sitiv

eendexpiratory

pressure

•Sp

ontaneousrespiratoryrate

DeV

ilaet

al.[7]

•citsigol

elbairavitluM

devireD

ylwe

Neg a

ronoD

regression

model

0.83

•Gag

orcoughreflex

•Inotropicsupport

•Bodymassindex

•ICU

length

ofstay

•Liver

enzymes

deGroot

etal.[ 8]

•Corneal

reflex

Validation

Re-evaluatedYee

etal

variablesusingmultiv

ariable

logistic

regression

analysis

0.77

•Cough

reflex

•Motor

reflex

•Oxygenatio

nindex

Rabinsteinet

al.[12]

•Corneal

reflex

Validation

Re-evaluatedYee

etal

variablesusingmultiv

ariable

logistic

regression

analysis

0.81

0.72

0.78

•Cough

reflex

•Motor

reflex

•Oxygenatio

nindex

Windet

al.[14 ]

(60min)

•Controlledmechanicalventilatio

nNew

lyDerived

Multiv

ariablelogistic

regression

analysis

0.74

0.70

0.74

Windet

al.[14]

(120

min)

•Controlledmechanicalventilatio

nNew

lyDerived

Multiv

ariablelogistic

regression

analysis

0.76

0.84

0.52

•Norepinephrineadministration

•cardiovascular

comorbidity

Colem

anet

al.[18]

•09.0

67.098.0

87.0devire

Dyl

weN

n oini potsil aiceps

UC I

1024

Table

5continued

VPN

VPPyticfiicepS

ytivitisneSC

UA

tnempoleve

Dled o

Mselbaira

Vcfiice pS

y dutS Colem

anet

al.[ 18

26.085.0

67.024.0

n oita dilaV

lo oT

nis noc siW

]•Bodymassindex

•Vasopressor

use

•Age

•Route

ofintubatio

n•Sp

ontaneousrespiratoryrate

•Negativeinspiratoryforce

•Tidal

volume

•Oxygensaturatio

nColem

anet

al.[ 18

37.067.0

48.016.0

n oitadilaV

er ocSS

ON

U]

Colem

anet

al.[ 18]

•66.0

78.069.0

93.0noitadila

Vecne dn epe d

rotalitneV

•Oxygenatio

n•Sy

stolic

bloodpressure

DeV

itaet

al.[7]

•With

draw

alprocessvariablesa

New

lyDerived

Multiv

ariableexploratory

analysis,Nonparametric

Classificatio

nand

RegressionTreeanalysis

(CARTSalfordsystem

s)

0.79

0.63

0.63

0.78

•Glasgow

comascale

•SaO2/FiO2

•Peak

inspiratorypressure

Lew

iset

al.[6]

(60min)

Wisconsin

tool

bNew

lyDerived

Variables

included

basedupon

clinical

ratio

nale

0.73

0.84

0.84

•Bodymassindex

•Vasopressor

use

•Age

•Route

ofintubatio

n•Sp

ontaneousrespiratoryrate

•Negativeinspiratoryforce

•Tidal

volume

•Oxygensaturatio

nLew

iset

al.[6]

(120

min)

Wisconsin

tool

bNew

lyDerived

Variables

included

basedupon

clinical

ratio

nale

0.85

0.45

•Bodymassindex

•Vasopressor

use

•Age

•Route

ofintubatio

n•Sp

ontaneousrespiratoryrate

•Negativeinspiratoryforce

•Tidal

volume

•Oxygensaturatio

n

ALTalanineam

inotransferase,

AUC

area

under

thecurve,

ICU

intensivecare

unit,NPVnegativepredictivevalue,

OIoxygenationindex,PIP

peakinspiratory

pressure,PPV

positivepredictivevalue,

(v)validationcohort,UNOSUnited

Network

forOrgan

Sharing

aWithdrawal

within

10min,ETTremoval,comfortmeasuresgiveduringfirsthourafterwithdrawal

bRespiratory

variablesmeasuredafterdiscontinuationofmechanical

ventilationfor10min

1025

GCS, impaired saturation level of oxygen in hemoglobin/fraction of inspired oxygen (SaO2/FiO2), peak elevatedinspiratory pressure, and all treatments withdrawn within10 min. The more features present, the higher likelihoodof death within 60 min. This yielded a sensitivity andspecificity of 0.75 and 0.73, respectively. This model wasnot externally validated in another population.

In the logistic regression model by Brieva et al. [11](the PREDICT study), applied to the validation cohort ofa general ICU population, the best model included ICUspecialist opinion, pH, systolic blood pressure, GCS,PEEP and lower analgesia. These factors were found to beassociated with time to death with a sensitivity andspecificity of 0.84 and 0.72, respectively. In a subsequentstudy which focused on the potential DCD subgroupwithin their original cohort study, these authors developedCART models to create decision tree analysis for time todeath within 60 min. Their best model included ICUspecialist opinion, PEEP, GCS, spontaneous respiratoryrate, and systolic blood pressure with a sensitivity andspecificity of 0.82 and 0.76, respectively [9].

Discussion

This is the first systematic review to evaluate time todeath after WLST. Across 15 studies we identified asignificant amount of heterogeneity with respect to patientpopulation and variables incorporated into the exploratoryprediction models. While this heterogeneity prevented ameta-analysis, we were able to gain significant insightinto the most consistent key predictor variables associatedwith time to death across different populations whichprimarily focused on time to death within 60 min,namely, controlled mechanical ventilation, risk factorsrelated to oxygenation impairment, vasopressor use, andneurologic function. These predictor variables were con-sistent across the studies involving general ICU patientsand those focusing primarily on DCD candidates.

Neurologic- and respiratory-related risk factors werethe most consistent variables across all studies associatedwith time to death. A lower GCS, absence of cough/gag,and absent corneal reflexes/extensor motor reflexes wereassociated with faster time to death. Clinically, patientswith these risk factors likely have a catastrophic neuro-logic injury that does not progress to brain death. Thepathway by which these variables lead to a hastened deathis possibly associated with the loss of sufficient respira-tory effort due to brainstem injury. Thus, the removal ofventilator support likely leads to a rapid rise in the partialpressure of CO2, leading to a significant respiratory aci-dosis and hemodynamic collapse. Additionally, thosepatients with a significant amount of neurologic ischemia,clinically manifested by a depressed GCS or loss ofbrainstem reflexes, may develop hemodynamic instability

as a result of a cascade of inflammatory mediator releaseassociated with cellular necrosis [23].

Respiratory-related risk factors are also strongly cor-related with time to death, with the most consistentvariables being controlled mode of ventilation and/or ahigher level of oxygen support. The etiology behind thecessation of spontaneous respiration, necessitating a con-trolled mode of ventilation, is often linked to catastrophicneurologic brainstem injury. Similar to the rationale forhastened death in the setting of neurologic risk factors, thewithdrawal of ventilator support may often lead to a rapiddeath associated with a rapid development of respiratoryacidosis. In addition, the withdrawal of high levels ofoxygen support (FiO2 or PEEP) has also been associatedwith a shorter time to death across a subset of studies.Ensuing hypoxia likely results in anaerobic metabolism,severe lactic acidosis, cellular death, and electrolyte andhemodynamic instability, ultimately leading to death: thehigher the oxygen support, presumably the more rapid thedevelopment of these changes.

Sedation and symptom management is a cornerstoneof palliative care management. While it may be postu-lated that increased sedation could hasten death byblunting the respiratory drive and inducing hemodynamicinstability, this has not been demonstrated in the reviewedstudies [5, 9, 12]. In fact, in some of the reviews, themanagement of sedation and symptoms has been associ-ated with an increased time to death [18]. One possibleexplanation for this relationship may be the inverse re-lationship between the depth of coma and symptomswarranting sedation after WLST.

Physician opinion has not been evaluated extensively,in part because traditionally physician opinion has notbeen considered to accurately predict time to death,thereby leading to the search for more accurate predictiontools [12]. However, the findings by Brieva and col-leagues [9] are intriguing as in their study physicianopinion was strongly associated with time to death(relative risk 8.44; 95 % confidence interval 4.3, 16.6).Given this significant association, the role of physicianopinion in predicting time to death warrants furtherevaluation.

Few studies included in this review described thestepwise approach to WLST, which includes actionssurrounding extubation, sedation management, and va-sopressor weaning and cessation, and whether they occursimultaneously or in a staged fashion. Differences inpractice could have an important impact on time to death.Removal of the endotracheal tube during WLST wasevaluated in a few studies but did not consistently cor-relate with time to death [7, 14].

Across the seven prediction tools proposed, only a fewwere validated in a separate cohort, and most had onlymoderate sensitivity. The University of Wisconsin toolunderwent external validation in the study by Colemanet al. [18] and only demonstrated moderate sensitivity and

1026

specificity. The major difference between the initial tooldevelopment study and the validation study was that thepopulation in the development study suffered from severeneurological injury while the validation study included ageneral ICU population, which may make the tool lessreliable. While the UNOS tool and a modified formatwere externally validated in two studies of a general ICUpopulation [7, 18], one possible explanation for the lackof external validity could stem from the tool’s heavy fo-cus on hemodynamic support devices, suggesting that itmay be most applicable in units which frequently employthese devices, such as cardiac units. The Yee tool, whichwas originally derived from a neurologic ICU population,was further externally validated by two studies using asimilar population. Brieva et al. [11] validated theirderived CART models in their two studies using a sepa-rate validation cohort; these models have not undergonefurther external validation. The Hunter New EnglandArea Composite score and the DVIC CART models havenot yet undergone further external validation.

There are a series of limitations in our study worthnoting. First, this review focused heavily on time to deathwithin 60 min due to the state of the current literature; thisreview therefore provides useful information for practi-tioners interested in that time-point. However, regardingthe general ICU population, it is not uncommon forfamilies to inquire if death will occur ‘‘immediately’’following withdrawal. Additionally, for the purpose ofDCD, new standard wait times are approaching 120 minas opposed to the traditional 60 min (depending upon or-gan of interest). Therefore, different time-points ofevaluation should be considered in future studies. Second,this review highlights predictors of time to death that wereconsistent across many of the studies included in the re-view. However, not every study evaluated the same list ofrisk factors, thereby limiting our ability to draw any con-clusions about those risk predictors found to be stronglyassociated with our outcome of interest yet only evaluatedin a small number of studies. Third, these studies wereheterogeneous in terms of the patient populationsevaluated. Predictors of time to death in the catastrophicneurologic disease population may be very different thanthose in the population with catastrophic medical disease.

While this review is the first to highlight the mostconsistent and important risk factors associated with timeto death and has identified promising novel features

demonstrating significance but for which reproducibilityis pending, it also highlights the paucity and heterogeneityof the data that currently exist on this topic. Moving forth,subsequent research must not only include all of thevariables that were found here to be consistently associ-ated with time to death but must also evaluate theirpredictive ability in the general medical population aswell as in specific subgroups, including patients withcatastrophic neurologic injury and donation after cardiacdeath candidates. In addition, different time-points shouldbe explored, and there should be a more detailed overviewof withdrawal practices and the differential effect of si-multaneous withdrawal compared to a staged process ifboth options are available.

Conclusions

Knowledge of time to death can help inform decisions onpalliation as well as assist with family preparedness andease the grieving process. In addition, time to death is im-portant for DCD as accurate prediction can have an impacton resource use and can reliably inform family discussions.This study highlights the significant work that has beenaccomplished thus far in the field in identifying consistentvariables associated with time to death, namely, controlledventilation, oxygenation, vasopressor use, GCS, and brainstem reflexes. It can inform future research endeavors fo-cused on the development ofmore accurate prediction toolsfor generalized ICU, neurologic ICU, and DCD popula-tions. Important questions that remain and need to be morerigorously studied include various time-points of death aswell as method of withdrawal. These have important im-plications for end-of-life care in the ICU.

Acknowledgments Laveena Munshi receives funding from theEliot Phillipson Clinician Scientist. Training Program, Departmentof Medicine, University of Toronto. Jason Shahin receives fundingsupport from the Department of Medicine, McGill UniversityHealth Centre.

Conflicts of interest The authors declare that they have no con-flicts of interest. Sonny Dhanani is the Chief Medical Officer ofTrillium Gift of Life Network. Sam Shemie is a medical advisor tothe Canadian Blood Services.

References

1. Carlet J, Thijs LG, Antonelli M et al(2004) Challenges in end-of-life care inthe ICU. Statement of the 5thinternational consensus conference incritical care: Brussels, Belgium, April2003. Intensive Care Med30(5):770–784

2. Cooke C, Hotchkin D, Engelberg R et al(2010) Predictors of time to death afterterminal withdrawal of mechanicalventilation in the ICU. Chest138(2):289–297

3. Dhanani S (2014) Chief MedicalOfficer, Trillium Gift of Life Network.Personal correspondence

4. Shemie SD, Baker AJ, Knoll G et al(2006) National recommendations fordonation after cardiocirculatory death inCanada: donation after

1027

cardiocirculatory death in Canada. CanMed Assoc J 175(8):S1

5. Muskett H, Shahin K, Eyres G et al(2011) Risk factors for invasive fungaldisease in critically ill adult patients: asystematic review. Crit Care15(6):R287

6. Lewis J, Peltier J, Nelson H et al (2003)Development of the University ofWisconsin Donation after cardiac deathevaluation tool. Prog Transplant13:265–273

7. DeVita M, Mori Brooks M,Zawistowski C et al (2008) Donors aftercardiac death: validation ofidentification criteria (DVIC) study forpredictors of rapid death. Am JTransplant 8:432–441

8. de Groot Y, Lingsma H, Bakker J et al(2012) External validation of aprognostic model predicting time ofdeath after withdrawal of life support inneurocritical patients. Crit Care Med40(3):233–238

9. Brieva J, Coleman N, Lacey J, et al(2014) Prediction of death in less than60 minutes after withdrawal ofcardiorespiratory support in potentialorgan donors after circulatory death.Transplantation 8(10):1112–1118

10. Huynh T, Walling AM, Le TX et al(2013) Factors associated withpalliative withdrawal of mechanicalventilation and time to death afterwithdrawal. J Palliat Med16(11):1368–1374

11. Brieva J, Coleman N, Lacey J et al(2013) Prediction of death in less than60 minutes following withdrawal ofcardiorespiratory support in ICUs. CritCare Med 41(12):2677–2687

12. Rabinstein AA, Yee AH, Mandrekar Jet al (2012) Prediction of potential fororgan donation after cardiac death inpatients in neurocritical state: aprospective observational study. LancetNeurol 22:414–419

13. Shore P, Huang R, Roy L et al (2012)Development of a bedside tool topredict time to death after withdrawalof life-sustaining therapies in infantsand children. Pediatr Crit Care Med13:415–422

14. Wind J, Snoeijs M, Brugman CA et al(2012) Prediction of time of death afterwithdrawal of life-sustaining treatmentin potential donors after cardiac death.Crit Care Med 40:766–769

15. Yee AH, Rabinstein AA, Thapa P et al(2010) Factors influencing time to deathafter withdrawal of life support inneurocritical patients. Neurology74:1380–1385

16. Suntharalingam C, Sharples L, DudleyC et al (2009) Time to cardiac deathafter withdrawal of life-sustainingtreatment in potential organ donors. AmJ Transplant 9:2157–2165

17. Davila D, Ciria R, Jassem W et al(2012) Prediction models of donorarrest and graft utilization in livertransplantation from Maastricht-3

donors after circulatory death. Am JTransplant 12(12):3414–3424

18. Coleman N, Brieva J, Crowfoot E(2008) Prediction of death afterwithdrawal of life-sustainingtreatments. Crit Care Resusc10:278–284

19. Zawistowski C, DeVita M (2004) Adescriptive study of children dying inthe pediatric intensive care unit afterwithdrawal of life-sustaining treatment.Pediat Crit Care Med 5(3):216–223

20. Chan JD, Treece PD, Engelberg RAet al (2004) Narcotic andbenzodiazepine use after withdrawal oflife support: association with time todeath? Chest 126:286–293

21. NSW Health Department (2007) Organdonation after cardiac death. Health andresearch and ethics. File No 04/3586-2.NSW Department of Health, Sydney

22. Levvey B, Westall G, Kotsimbos T et al(2008) Definitions of warm ischemictime when using controlled donationafter cardiac death lung donors.Transplantation 86(12):1702–1706

23. Powner DJ, Hendrich A, Nyhuis A,Strate R (1992) Changes in serumcatecholamine levels in patients whoare brain dead. J Heart Lung Transplant11(6):1046–1053

1028