Seriously Ill Hospitalized Patients

A

Controlled

Trial

to

_Improve

Care

for

_Seriously

III

_Hospitalized

Patients

The

_Study

to

Understand

_Prognoses

and

Preferences

for Outcomes and Risks of

Treatments

_(SUPPORT)

The SUPPORT

_{Principal Investigators}

Objectives.\p=m-\To

improve

end-of-life decision

_making

and reduce the

_frequency

ofa

mechanically

_supported,

painful,

and

prolonged

_{process of}

dying.

Design.\p=m-\A2-year prospective

observational

_{study (phase I)}

with 4301

_patients

followed

_by

a

_2-year

controlledclinical trial

_(phase

_II)

with 4804

_patients

and their

physicians

randomized

_{by specialty}

_group tothe intervention _group

_(n=2652)

or

control _group

_(n=2152).

Setting.\p=m-\Fiveteaching

hospitals

in the UnitedStates.

Patients.\p=m-\A

total of 9105 adults

_hospitalized

with one or more of nine

life-threatening diagnoses;

an overall 6-month

_mortality

rateof47%.

Intervention.\p=m-\Physicians

in the _{intervention group} received estimates of the

likelihoodof 6-monthsurvivalfor_every

_day

_upto6

months,

outcomesof

cardiopul-monaryresuscitation

(CPR),

and functional

disability

at2 months. A

_specially

trained

nursehad

multiple

contacts with the

_{patient, family, physician,}

and

_hospital

staffto

elicit

_preferences,

_improve

_{understanding}

of

outcomes,

encourageattention to

pain

control,

andfacilitateadvancecare

planning

and

patient-physician

communication.

Results.\p=m-\The

_phase

I observation documented

_shortcomings

in

communica-tion,

frequency

of

aggressive

treatment,

and thecharacteristics of

hospital

death:

only

47% of

_physicians

knew when their

_{patients preferred}

toavoid

_CPR;

46% of do-not-resuscitate

_(DNR)

orderswerewritten within 2

days

of

death;

38% of

patients

who died

_spent

atleast 10

_days

inanintensivecareunit

(ICU);

and for 50% of

con-scious

_patients

_{who died}inthe

_hospital,

_family

members

_reported

moderateto se-vere

pain

atleast half thetime.

During

the

_phase

II

_{intervention,}

_{patients experienced}

no

improvement

in

patient-physician

communication

(eg,

37%ofcontrol

_patients

and40%ofintervention

_patients

discussed CPR

_preferences)

orin the five

targeted

outcomes,

ie,

incidenceor

timing

ofwritten DNR orders

(adjusted

_{ratio, 1.02;}

95%

confidenceinterval

[Cl],

0.90to

_{1.15), physicians' knowledge}

oftheir

patients'pref-erencesnot toberesuscitated

_(adjusted

_ratio,

1.22;

95%

_Cl,

0.99 to

_1.49),

number

of

_days

_spent

inan

_ICU,

_receiving

mechanical

_ventilation,

orcomatosebeforedeath

(adjusted

ratio,

0.97;

95%

_Cl,

0.87 to

1.07),

orlevel of

reported pain (adjusted

ratio,

1.15;

95%

_Cl,

1.00 to

_1.33).

The intervention alsodidnotreduceuseof

hospital

re-sources

(adjusted

ratio,

1.05;

95%

Cl,

0.99to

_1.12).

Conclusions.\p=m-\The

_phase

I observation of SUPPORT confirmed substantial

shortcomings

incarefor

_seriously

ill

_hospitalized

adults. The

phase

II intervention

failed to

_improve

care or

_patient

outcomes.

_Enhancing

_{opportunities}

for more

patient-physician

communication,

although

advocatedasthe

_major

methodfor

im-proving patient

outcomes,

may be

inadequate

to

_change

established

practices.

To

improve

the

_experience

of

_seriously

ill and

dying patients,

greater

individual and societal commitment and more

proactive

andforcefulmeasures _{may be needed.}

(JAMA.1995;274:1591-1598)

A_complete list of the SUPPORTprincipal

investi-gators and collaborators appearsatthe end of this article.

Theopinionsandfindingscontained in this articleare

those of the authors and donot_{necessarily represent}

the views of the Robert Wood Johnson Foundationor

their Board of Trustees.

Reprintrequests toICU Research Unit,Box600,

Universityof_VirginiaHealthSciencesCenter,

Char-lottesville,VA 22908_(WilliamA.Knaus,MD).

PUBLIC HEALTH and clinical medi¬ cine

_during

this

_century

have

_given

Americans the

_opportunity

tolive

_longer

andmore

productive

lives,

despite

_pro¬

gressive

illness. Forsome

patients,

how¬

ever,this progress has resulted in pro¬

longed dying, accompanied

by

substan¬ tial emotional and financial

_expense.1

Many

Americans

_today

fear

_they

will lose controlovertheir lives if

they

be¬ come

critically

ill,

and their

dying

will

be

_prolonged

and

_impersonal.2

This has led to an

increasingly

visible

right-to-diemovement. Two_years aftervoters inCalifornia and

_Washington

Statenar¬

rowly

defeatedreferendaon

physician-assisted

_euthanasia,

_Oregon

voters _ap¬

proved physician prescription

of lethal medications for persons withaterminal

disease.3·4

_Physicians

and ethicistshave

debated whentousecardiac resuscita¬

tion and other

_aggressive

treatments

for

_patients

with advanced illnesses.6,6

Many

worryabout the economic and hu¬

man cost of

_providing

_{life-sustaining}

treatment near the end oflife.79

Foreditorialcommentsee 1634.

In _response,

_{professional organiza¬}

tions,

the

_judiciary,

consumer

organiza¬

tions,

anda

president's

commission have

all advocated more

emphasis

onrealis¬

tically forecasting

outcomes of

life-sustaining

treatment andon

improved

communication between

_physician

and

patient.21016

Statutes

_requiring

informed

consentand

_{communication,}

like the Pa¬ tient Self-determination

_Act,17

have been

passed.

Advancecare

planning

and ef¬

fective

_ongoing

communication _among

clinicians,

patients,

and familiesare es¬

sentialtoachieve these

_goals.

Previous

studies

_{indicate, however,}

thatcommu¬

nication is often absentor occurs

only

during

a crisis.1820

Physicians today

of¬

ten

_perceive

deathas

failure,1

_they

tend

prog-noses,21

and

_{they provide}

more exten¬

sivetreatment to

_seriously

ill

_patients

than

_they

would choose forthemselves.22

PhaseI of the

_Study

toUnderstand

Prognoses

and Preferences for Outcomes and Risks of Treatments

(SUPPORT)

confirmed barriersto

_optimal

_manage¬ mentand shortfallsin

_{patient-physician}

communication.23·24The

_phase

II inter¬ vention

_sought

to address these defi¬ ciencies

_{by providing physicians}

with

accurate

_predictive

information on fu¬

ture functional

_ability,26

survival

_prob¬

ability

for each

_day

_upto6

_months,26

and

patient preferences

for end-of-life_care;

askillednurse

augmented

thecareteam to elicit

_{patient preferences, provide}

prognoses,enhance

understanding,

en¬

able

_palliative

_care, and facilitate ad¬

vance

planning.

We

hypothesized

that

increased communication and under¬

standing

of_prognosesand

_preferences

would resultin earliertreatmentdeci¬

sions,

reductions in time

_spent

inunde¬ sirablestatesbefore

_death,

and reduced

resourceuse.Thisarticle describes the

effect of theSUPPORTinterventionon

five

_specific

outcomes:

physician

under¬

standing

of

_{patient preferences;}

inci¬ dence and timeof documentation of do-not-resuscitate

(DNR)

orders;

pain;

time

spent

in an intensive care unit

(ICU),

comatose,or

receiving

mechanicalven¬

tilation before

_death;

and

_hospital

re¬

source use

(Figure

1).

METHODS

Phase I was a

prospective

observa¬

tional

_study

that described the_process of decision

_making

and

_patient

outcomes.

Phase II was acluster randomizedcon¬

trolled clinical trialto testthe effect of the intervention.

_Enrollment,

data col¬

lection,

and

_interviewing

were

virtually

identical

_during

thetwo

_{phases.21,23"28}

Enrollment

Qualified patients

were in the ad¬

vanced

_stages

ofone or more ofnine

illnesses:acute

_respiratory

_failure,

mul¬

tiple

organ

system

failure with

_sepsis,

multiple

organ

system

failure withma¬

lignancy,

coma,chronic obstructive

lung

disease,

congestive

heart

_failure,

cirrho¬

sis,

metastatic colon _cancer, and

non-small cell

_lung

cancer. Patients were

excluded if

_they

were _youngerthan 18

years, were

discharged

or died within

48hours of

_qualifying

for the

_study,

were

admitted with a scheduled

discharge

within 72

_hours,

didnot

_speak

_English,

wereadmittedtothe

_psychiatric

_ward,

had

_{acquired immunodeficiency}

_syn¬

drome,

or were

pregnant

or sustained an acute

_{burn, head,}

or other trauma

(unless

they

later

_developed

acuteres¬

piratory

failure or

multiple

_{organ sys¬}

tem

failure).26·28

Nurses trained in the

SUPPORT

_eligibility

criteriareviewed

hospital

admissions and ICU

_patients

daily

to

_identify

_newly

_{qualified patients.}

Phase I enrolled

_patients

fromJune

1989toJune

1991,

and

_phase

II enrolled

patients

from

_January

1992

_through

January

1994. Patientswererecruited

fromfive medical centers: Beth Israel

Hospital,

Boston, Mass;

MetroHealth Medical

_{Center, Cleveland, Ohio;}

Duke

University

Medical

_{Center, Durham,}

NC;

Marshfield Clinic/St

_Joseph's

Hos¬

pital,

Marshfield, Wis;

and the Univer¬

sity

of CaliforniaatLos

_Angeles

Medi¬ cal Center. An

_independent

committee monitored

_potential

adverse

_events,

in¬

cluding

6-month

_mortality

for interven¬ tion

_patients

and

_changes

in

_patient

sat¬

isfaction with medical care.

Mortality

follow-up

to6monthswas

complete

for

all

_phase

I

_patients.

In

_{phase II,}

_{22 pa¬}

tients

(0.5%)

wereunavailable for

follow-up atamedian of 80

days.

DataCollectionMethods

Data collectionwasbasedonbothcon¬

current and

_{retrospective}

medical re¬

cord reviews andoninterviews with_pa¬

tients,

patient

surrogates (defined

asthe

person who would make decisionsif the

patient

was unable to do

_so),

and _pa¬ tients'

_physicians.

Medical Record-Based Data.—We collected

_{physiological}

indicators of disease

_{severity,29,30}

_length

of

stay,

a

modified version of the

_Therapeutic

In¬ tervention

_Scoring

_System,31

and

comor-bidities from the medical recordson

days

1,3, 7,14,

and25.The

_permanent

medi¬ cal recordwas

retrospectively

reviewed

for discussionsordecisions

concerning

18

_important

_issues,

suchastheuseof

dialysis,

withdrawal from a

ventilator,

and DNR orders.

_{Reliability testing}

on

10% of the medical records showed at

least90%

agreement

onabstracted data.

Interview Data.—Patients and their

designated

surrogates

wereinterviewed

in the

_hospital

between

_days

2 and 7

(median,

day

4)

and

_again

between

_days

6 and 15

_(median,

_day

12)

after

_study

enrollment,

whetherornotthe

_patient

remained

_{hospitalized.}

The

_surrogate

wasinterviewed4to10weeks after the

patient's

death.

_Among

the45% of_pa¬ tients who wereable to

_communicate,

the_{response rate for the}firstinterview

was85%. The

_surrogate

_{response rate}

for the first interviewwas87%.Forthe

second-week

_interviews,

the

_patient

re¬

sponseratewas71%and the

_surrogate

response ratewas78%.The interviews

collectedinformation on

patient

demo¬

graphics,

functional

_status,

self-assessed

quality

of

_life,

communicationwith

_phy¬

sicians,

frequency

and

_severity

of

_pain,

satisfaction with medical

_care,32

and the

patient's preferences

for

cardiopulmo-nary resuscitation

(CPR).

When a _pa¬

tient interviewwasnot

_possible,

thesur¬

rogate's

responseswere

substituted,

a

strategy

that mirrors clinical

_practice.

Important

elements of the

_{patient/sur¬}

rogate

questionnaires

wereretested for

reliability.

Initial and

_repeat

_responses

had

_greater

than80%

_agreement.

The most senior available

_physician

acknowledging responsibility

forthe_pa¬ tient's medical decisionswasinterviewed

in thefirst and second weeks after_pa¬ tient enrollment

(median

_days,

3and

_11,

respectively).

In both

_interviews,

we

asked the

_{physician's understanding}

of the

_{patient's preferences}

for CPR. In the second

_interview,

_physicians

as¬

signed

tothe interventionwere

queried

aboutitsinfluenceonthe

patient's

care.

Physician

response rateswere86%for

the first interview and82%for thesec¬

ond interview.

Phase IIIntervention

Presented with

_{early findings}

from

phase

I

_documenting

substantial short¬

comings

in

_{communication,}

decision mak¬

ing,

and

_{outcomes,23,24 physicians}

atthe

participating

institutions voiced inter¬

estin

_attempting

_{change. Physician}

lead¬

ersand

study investigators

atthe sites

met todiscusshowdecision

_making

could be

_improved

tomore

closely

reflectboth

probable

outcomesand

_{patient prefer¬}

encesand_{ways to}

improve patient,

fam¬

ily,

and

_physician

communication.

_Phy¬

sicians

_suggested

that communication could

_improve

if there weremorereli¬

able and

_prompt

information

_generated

by

the

_study

and if

_study

_personnel

would make it more efficient to have

conversations. In_{response to}these_sug¬

gestions,

the

_phase

IIintervention aimed

to

_improve

communication and decision

making by providing timely

and reliable

prognostic

information,

by eliciting

and

documenting patient

and

_family

_prefer¬

encesand

understanding

of disease_prog¬

nosis and

_treatment,

and

_{by providing}

a

skillednurseto

_help

_{carry out}the needed

discussions,

convenethe

meetings,

and

bring

tobear the relevant information. The elements of the intervention and their

_timing

are

presented

inTable 1.

In each _case, the nurse was free to

shape

her roleso astoachieve the best

possible

careandoutcome. For

_example,

she sometimes

_engaged

in extensive emotional

_support.

Other

_times,

she

mainly provided

information and en¬

sured that all

_parties

heardoneanother

effectively.

All of the nurses' involve¬

ment

_{required approval}

of the attend¬

ing

physicians.

In

_virtually

all_cases,the

physician approval

camewithnolimits.

Physicians

were

_{free, however,}

tolimit the intervention in _{any way}that

_they

feltwasbest for the

patient,

and there

Objectivesand_OrganizationofSUPPORT

PhaseI(1989to1991)

Study4301 PatientsatFiveTeaching Hospitals

• _{Describe Outcomes}

• Develop PrognosticModels • Identify Shortcomingsof Care • _{EstablishAdjustmentMethods} • Design_Intervention

Control

11Physician Groups

2152 Patients(45%)

Phasell(1992to₁₉₉₄₎

ApplyInterventionto4804 Patients

Randomized_by27Physician Groups

Intervention

16PhysicianGroups

2652Patients_(55%)

Adjusted Analysesof InterventionvsControl forFiveOutcomes

• _{Incidence and}TimingofWritten_DNROrders

• Patient-PhysicianAgreementonCPR Preferences

• _Daysinan_{ICU, Comatose,}orReceivingMechanical VentilationBeforeDeath

•_Pain

• Hospital_{Resource Use}

Figure1.—Overall schematic_presentationof_phasesI andII ofthe_Studyto UnderstandPrognosesand

Preferences forOutcomesand Risks ofTreatment(SUPPORT) project,1989to1994. DNR indicates do

notresuscitate;CPR, cardlopulmonaryresuscitation;andICU,Intensivecareunit.

was no

requirement

for themtoshareor

discuss the information with the

_patient

or

family

ortoallow the nurse's involve¬

mentto continue. Thenursewasiden¬

tified on her

badge

andintheconsent processas

_part

ofa_research

_effort,

but

she had the role and appearance ofa

typical

clinical

_specialist.

Randomization.—To limit contami¬

nation,

patients

were

assigned

tointer¬ vention or control

(usual care)

status

basedonthe

specialty

oftheir

attending

physician.

Physician

specialties

weredi¬

vided into five _groups: internal medi¬

cine,

pulmonology/medical

ICU,

oncol¬

ogy, surgery,and

cardiology.

We useda

cluster randomization schemeto

assign

the intervention

_randomly

to27

_physi¬

cian

_group-site

_{combinations,}

restricted

by

the conditions that 50% to 60% of

patients

would be

_assigned

intervention

status,

and that atleast one interven¬

tion andonecontrol

physician

specialty

group be at each of thefive

_study

in¬

stitutions. This resulted in11

_physician

specialty

groups

assigned

tocontrol and 16

_assigned

totheintervention

_(Figure

1).

Analyses

werebasedonallocationto

intervention

_(ie,

intention to

treat),

ir¬

respective

of whether· a

given patient

received the intervention.

_{Investigators}

wereblindedtothe

_phase

IIresults dur¬

ing

data collection.

Analytic

Methods.—Five measures were chosen to evaluate the interven¬ tion:

(1)

The

_timing

ofwrittenDNRor¬

derswas

analyzed

witha

log-normal

re-gression

modeltoprepare

Kaplan-Meier

predicted

median time untilthefirst DNR

orderwaswritten. Ifa DNRorderwas

not

_written,

DNR order

_timing

was

censoredatthe

_day

of deathor

hospital

discharge.

(2)

Patient and

_physician

_agree¬

menton

preferences

to withhold resus¬

citationwasbasedonthefirst interview

of the

_patient

(or

surrogate

ifthe

_patient

was unable tobe

interviewed)

and the

responsible physician. Agreement

wasde¬

finedasa_responseto

_forgo

resuscitation from both

_patient

and

_{physician, analyzed}

with

_{binary logistic regression,}

and_ap¬

plied

toall interviewed

_patients

or sur¬

rogates

who had

_{matching physician}

interviews.

(3)

_{Days spent}

in an

ICU,

receiving

mechanical

_ventilation,

or co¬

matosebefore deathwere

analyzed

using

ordinary least-squares

regression

(after

taking

the

_log

of0.5

_plus

the number of

days)

and

_only

included

_phase

II

_patients

who died

_during

the index

hospitaliza-tion.(4)

Frequency

and

_severity

of

_pain

analyses

were based on all

patients

or

surrogates

interviewed in the second week withacombinedmeasure

_(moder¬

ate or severe

pain

all,

most,

orhalf the

time)

and

_{analyzed using}

a

single,

ordinal

logistic regression

model.

₍₅₎

_Hospital

re¬

sourceuse wasdefinedasthe

log

of the

product

of the_average

_Therapeutic

In¬

tervention

_{Scoring System}

_rating

and

length

of

_hospital

stay

after the second

day

of the

_study.

In

_{regression analyses}

on

phase

I

data,

thismeasure

_closely

es¬

timated

_hospital

billsacrossthe five

study

institutions

(Pearson

ñ2=0.93on

log prod¬

uct).Weused

_ordinary

_{least-squares}

re¬

gression

tomodel the

_log

ofresource_use,

which was then converted to 1993 dol¬ lars. This method allows

comparisons

of

groupsandinstitutionsacrosstime with¬

out

_having

to

_adjust

for

_{varying hospital}

billing practices.

Power and

_Safety

Calculations Power calculations basedon

phase

I

data indicated

_greater

_{than 90% power}

(a=.05)

to detect a

1-day

decrease in

days

untilaDNR orderwas

written,

a

5%increasein the

_proportion

of

_physi¬

cians and

_{patients agreeing}

on aDNR

order,

a 20% decrease in undesirable

days,

a 10% decreasein

reported pain,

anda5% decrease inresource use.Ef¬

fects of the intervention on

mortality

rateswere

quantified

by

the estimated

intervention,

controlhazard ratio from

adjusted

Cox models.33

Adjustment

Methods for Phase II Results

Because

_patients

were

assigned

toin¬ terventionorcontrolstatusbasedona

limited number of

_specialty

_groups,the

resulting

cohorts

_might

be unbalanced

in

_patient

baseline risk factors. Further¬

more,

practice

patterns

amongthe

phy¬

sician

_specialty

groups in

_phase

I dif¬ fered

_{substantially.}

We controlled for these

_{expected preintervention}

differ¬

ences

using

baseline multivariable risk

scoresthatwerederived

by generating

modelsto

_{predict phase}

Ioutcomes,each of which

_incorporated

interactions be¬

tween

_{physician specialty}

and

_hospitals.

Observed imbalances in

_phase

II base¬ line

_patient

characteristicswerealso ad¬

justed

using

a

propensity

scorethatcor¬

rectedforselection bias associated with

being assigned

tointerventionstatus.34

Further details onthe construction of

both of these riskscores areavailableon

request.

Imputation

methods for miss¬

ing

data have been

_published.25

_Finally,

we simulated the

phase

II randomiza¬

tion schemeonthe

phase

I datatoevalu¬

ate secular trends. To

_adjust

for mul¬

tiple

outcomes, ourmethods

prespeci-fied

_adjusting

confidence intervals

(CIs)

using

the method of

_Hochberg

and

Ben-jamini35

ifmore thanone value was

less than.05.Allstatistical

_analyses

were

done with UNIX

_S-Plus,

version 3.2 soft¬ ware36 and the

_Design

_library.37

RESULTS

Phase I Observations

Phase I enrolled4301

patients (Fig¬

ure₁₎withamedian_ageof65 yearsand

other characteristics summarized in Table 2. The mean

predicted

6-month

survival

_probability

was 52% with an

Table1.—Content, Recipient,andTimingof PhaseII SUPPORTIntervention*

Content Providedto _Timing

Feedback of_phaseIresults

Benchmarkinginformation

describingphaseIIncidence of_{patient-physician}

communication,pain,

and_timingof DNR order

All intervention_{physicians Early phase}I

Prognosticinformation Survival time estimates for up

to6mo26

Intervention_physiciansand

medical record Study daysand 26 2, 4, 8, 15,

Prognosisfor outcome for CPR ifneeded26

Intervention_physiciansand

medical record Study day

2

Survivalestimates,enhanced

by physician26

Intervention_physicians

and medical record Study day

4

Prognosis, probabilityofsevere

disability,at 2mo25 Interventionphysicians

and

medical record Studyday 8 Interview information

Patient and_{surrogate report}of

prognosis, preferencesabout

CPR,advance_{directives,quality} oflife,informationdesires,and

pain

Intervention_physiciansand medical record

First and second_studyweeks Interviewon_knowledgeof

preferencest

Intervention_{physicians Study day}10

Nurse involvement

Explainingprognosticestimates

and interviewreports Patient,family,staff,

intervention

physicians,andmedical records Studyday

3 and_continuously until deathor6mo

Enhancingunderstandingof_likely

outcomes/preferences Patient, family,staff,

intervention

physicians,and medical records Studyday

3 and_continuously until deathor6mo

Elicitingand_documenting

preferences/advancedirectives Patient,family,staff,

intervention

physicians,and medical records Studyday

3 and_continuously until deathor6mo

Assessingpainand_enabling

treatment

Patient,family,staff,intervention

physicians,and medical records Studyday

3 and_continuously until deathor6mo

Voicingpatient/familypreferences

and values Patient,family,staff,

Intervention

physicians,and medical record Study day

3and_continuously until deathor6mo

Conveningmeetings, negotiating agreements

Patient,family,staff,intervention

physicians,and medical records Study dayuntil death3 andor6continuouslymo

Encouragingplanningfor future

decisions Patient,family,staff,

intervention

physicians,andmedical records Studyday

3 and_continuously until deathor6mo "SUPPORTindicatesStudytoUnderstand_Prognosesand Preferences for Outcomes and Risks ofTreatment;

DNR,donotresuscitate;CPR, cardiopulmonaryresuscitation;andICU,intensivecareunit.

tPhysicianinterviewon_day10wasforevaluation,notpartoftheintervention.

48%

(Table 2).

Thirty-one

percent

of

phase

I

_patients

with interviews _pre¬ ferred that CPR be

_withheld,

but

_only

47% of their

_physicians

_accurately

re¬

ported

this

_{preference during}

the first

interview.

_Nearly

half

(49%)

of the960

phase

I

_patients

whoindicatedadesire

for CPRtobe withheld didnothave a

DNR order written

_during

that

hospi-talization.

_Nearly

onethird of these_pa¬

tients

₍₂₇₈

[29%])

died before

_discharge.

Among

all

_phase

I

_patients

who died

during

the index

_{hospitalization}

(n=1150),

79%died withaDNR

order,

but 46% of

these orderswerewrittenwithin2

_days

of death.

_Among

all

_phase

I

_deaths,

the median number of

_days

_spent

inan

ICU,

comatose,

or

receiving

mechanical venti¬

lation was

8,

morethanonethird

(38%)

spent

atleast10

_days

inan

ICU,

and 46%

received mechanicalventilationwithin 3

days

of death. In the second

_week,

22%of

patients reported being

in moderate to

severe

pain

atleast half the time. In in¬ terviews conducted aftera

patient

died,

surrogates

indicated that 50% of allcon¬

scious

_phase

I

_patients

whodied in the

hospital experienced

moderateorsevere

pain

atleast halfthetime

_during

their last

3

_days

of life.

We found substantial variation in the fiveoutcomesamong

physician specialty

groups and across the five institutions.

Across

_{institutions,}

the median number of

_{days spent}

inanICU before deathvar¬

ied from5 to 9.The

_proportion

of

_patients

reporting

moderatetosevere

pain

atleast halfthe time varied

_by

afactor of

2.7,

from

12%to32%across

study

institutions. The

predicted

median number of

_days

untila

DNR order waswritten fora standard

patient

varied

_by

afactor of

3.5,

from 73

days

for

_patients

on a

surgical

serviceto 22for

_oncology.

One

_study

institution had

a

predicted

median time until DNRwas

written forastandard

patient

of28

days,

and another institution had a

predicted

median time of 49

_{days. Agreement}

on

DNR varied from8%for

_cardiology

_pa¬ tientsto 24% for

_{oncology patients}

and fromalow of8% atone

study

institution

toa

high

of 27%atanother. The median

number of

_days

_spent

in anICU before

death

_ranged

from14inthe

_surgical

_spe¬ cialties to 5 for

_patients

in

_pulmonary/

ICU and

_oncology

services. Phase II

_Demographics

PhaseIIenrolled4804

_patients,

2152

assigned

tousual medicalcareand 2652

assigned

tointerventionstatus

_(Figure

1).

Their characteristicswere

generally

similar to those of

_phase

I

_patients

(Table 2).

Delivery

of theIntervention

Ninety-five

percent

ofintervention_pa¬ tients receivedoneormore

patient-spe¬

cific

_components

oftheintervention. The SUPPORT nurse was involved in the care of all but 133

patients,

and 75 of

thesewere

patients

who died or were

discharged

on the

day

of enrollment.

The SUPPORT nurse communicated

with the

_physician

in

_virtually

allcases.

She talked

_directly

with the

_patient

or

family

in mostcases

(eg,

with 84%con¬

cerning prognosis,

77% about

_pain,

63% about

_likely

outcomes or

resuscitation,

and 73%

_concerning

written advance

directives).

For

_patients

inthe

_hospital

for 7 or more

days

after

qualifying

for the

_study,

the median number of SUPPORTnursecontactswith the_pa¬

tient,

family,

or

physician

wasfour and

themean wassix.Documentation in the

progressnotesofdiscussionsabout_pa¬ tients'

_preferences

with

_regard

to re¬

suscitationwasincreased from 38.5% in

phase

I_{to 50.3% among}

_phase

II inter¬ vention

_patients.

The

_{patient's physician}

received at

least one

prognostic

_report

for 94% of

patients,

and the

_report

was

_put

in the

medical record of80%. The

_{patient's phy¬}

sician receivedatleastone

printed

re¬

port

of

_patient

or

_surrogate

_understand¬

ing

and

_preferences

in 78% ofcases.

No

_physician

refused toreceive the

printed

reports

ortohavethemshared with other

_professional

staff. The

_phy¬

sicians for43

_patients

refused toallow the SUPPORT nurse to have contact

with the

_patient

and

_family,

and seven

patients

or

surrogates

refusedto

speak

with the SUPPORTnurse.

Effect of InterventiononOutcomes

The

_prevalence

or

timing

of documen¬

tation of DNR orders for the 2534 in¬

tervention

_patients

wasthesameasfor

the2208control

_{patients (adjusted}

ratio

of median

time,

1.02;

95%

_CI,

0.90 to

1.15) (Table 3).

Therewas asmallasso¬

ciation of the intervention with

_improved

patient-physician

DNR

_agreement

for the1480intervention

_patients

who had

patient

or

_surrogate

and

matching phy¬

sician

_interviews,

_compared

with 1159 control

_{patients (adjusted}

_{ratio, 1.22;}

95%

CI,

0.99 to

1.49).

The number of

_days

spent

inan

ICU,

_comatose,

or

_receiving

mechanical ventilation before death for the 680intervention

_patients

who died

inthe

_hospital

wasthesameasfor the

530 control

_{patients (adjusted}

ratio of median

_days,

_0.97;

95%

_CI,

0.87to

1.07).

Table 2.—Characteristics of 9105 SUPPORTPatients, 1989to 1994* Phase II Total PhaseI (n=4301) PhaseII (n=4804) I-Control Group (n=2152) Intervention Group (n=2652) Median age, y 65 64 66 Sex,%female 43.0 44.3 43.0 45.4 Race,% White 79.5 79.0 82.1 Black 15.3 15.4 11.5 19 Other 5.3 5.6 6.4 5.0 Annual income<$11000,%

Educationhighschoolormore, % 53.4 59.0 63.2 56

Primaryinsurance,% Private 29.2 29.6 32.7 27.1 Medicare 56.0 53.5 Medicaid 11.8 10.8 11.2 10.6 None 5.3 3.6 2.6 4.4 Diseaseclass,%

Acute organsystemfailuret 42.9 49.6 52.8

Chronicdiseased 34.4 28.8 25.0 31.8

Nontraumaticcoma 5.7 7.3 5.9 8.4

Cancer§ 16.9

MeanADL Scale_scorelffl 1.6 1.5 1.5 1.5 Median APS of APACHElll||# 33 32 32

SUPPORT 6-momeansurvival_estimateli 0.52 0.52 0.51 0.52

Hospital mortality,% 26.7 25.2 24.6 25.6

6-mo_Mortality,%

Median_{hospitalcharges,}$thousands 21 29 33 27

*SUPPORTindicates_StudytoUnderstand_Prognosesand Preferences for Outcomes and RisksofTreatment.

tlncludesacute_respiratoryfailure and_{multipleorgan}systemfailure withorwithout_sepsis.

^Includes

end-stage cirrhosis andacute exacerbations ofsevere _congestiveheart failureor severe chronic

obstructive_pulmonarydisease.

(¡IncludescoloncancerwithlivermétastasesandstageIIIor_stageIVnon-small cell_lungcancer.

IIMeasuredonthe third_{study day.}

UADLindicates activities ofdaily livingscorefrom0 to7,witheach_{point representing}an_impairmentinbasic

function, reflecting patientstatus 2weeks before admission.

#APS indicatesacute_physiologyscore;and APACHEIII,Acute_{Physiology,Age,}Chronic Health EvaluationIII.29

The APS varies from0 to252 witha_higherscore_{indicatinggreaterphysiological instability}and risk of death.

Reported

pain

increased for the 1677in¬ tervention

_patients

and

_surrogates

inter¬ viewed in the second

_week,

_compared

with the control _group

_(adjusted

_ratio,

1.15;

95%

_CI,

1.00to1.33)

(Table 3).

Therewas

no

change

in

_hospital

resourcesused for

2593 intervention

_patients

not dead or

discharged

before the third

_{study day}

com¬

pared

with 2129 control

_{patients (adjusted}

ratio of averageresource_use,

_1.05;

95%

CI,

0.99to

1.12).

The

_unadjusted

differences between in¬ tervention and control

_patients

forme¬

dian

_days

until the first DNR orderwas

writtenwere

large, especially

for

_patients

with coloncancerand non-small cell

lung

cancerfor whom the median number of

days

untilawrittenDNR orderwas80%

lower in intervention

_{patients. Adjust¬}

ment for baseline imbalances reduced

much of the difference in each

category

(Table 4).

The differences that

_persist

in thecancer

_category

areof uncertain im¬

portance,

being

one_among

multiple

com¬

parisons

and

_being

basedon asmallnum¬

ber of

_patients

(Table 4).

Figure

2illustrates the secular trends of each outcomein the

_phase

II inter¬ vention and control_groups,aswellasin

phase

I,

using

simulations of the

_physi¬

cian

_{specialty groupings}

used in

_phase

II.None of the five outcomes

_changed

significantly

during

the 5 years of the

study.

The differences between those who would have been

_assigned

tointer¬

vention and control in

_phase

I

_persisted

throughout

the SUPPORT

_study,

un¬

affected

_by

timeor

by

the intervention.

Communicationand Preferences The intervention didnot

_change

the

unadjusted

proportion

of

_patients

or sur¬

rogates

reporting

a discussion about

CPR;

37%of control

_patients

and40%of intervention

_patients

_reported

discuss¬

ing

their

_preference.

Of

_patients

who did nothave sucha

discussion,

41%of

each_groupsaid

_they

would liketodis¬

cussCPR. Seventeen

percent

ofcontrol

patients

and20%of intervention

_patients

changed

their resuscitation

_preferences

to

_forgo

CPR

_by

the second week after

enrollment,

and39%of control

_patients

and 41% of intervention

_patients

re¬

ported having

a discussion about their

prognosis

witha

physician.

Of those who

didnotdiscuss their

_prognosis,

44% of control

_patients

and42%ofintervention

Table 3.—Effectofthe SUPPORT PhaseII Inter¬ ventionon FiveOutcomes:InterventionGroupvs ControlGroup,1992to1994

Adjusted

Ratio

_(95%

CI)

Median time until DNR order

waswritten,d _{1.02(0.90-1.15)} DNR_agreement,% 1.22_(0.99-1.49)

Undesirable states, median d 0.97_(0.87-1.07)

Pain,% _{1.15(1.00-1.33)}

ResourceUse,median

1993 dollars _{1.05(0.99-1.12)}

'SUPPORTindicates_StudytoUnderstand_Prognoses and Preferences for Outcomes and Risks ofTreatment; DNR,donotresuscitate;andCI,confidence interval.

patients reported

that

_they

would like

tohave such adiscussion.

Physician's

Perspective

onIntervention

In the second

_physician

_interview,

59%

acknowledged

receiving

the

_prognostic

reports

and34%

_acknowledged

receiv¬

ing

the

_preference

_reports.

Fifteen_per¬

cent

_{reported discussing}

this

_specific

in¬ formation with

_patients

or families.

Nearly

a

_quarter

of

respondents

(22%)

said

_{they thought}

theSUPPORT nurses' involvement

_{improved patient}

care.

Safety Monitoring

After

_adjusting

for baseline differ¬

ences,the 6-month

mortality

for

phase

II control

_patients

wasthesameasfor

intervention

_{patients (adjusted}

relative

hazard, 0.95;

95%

_CI,

0.87to

1.04).

Both control

(68%)

and intervention

(69%)

pa¬

tientsor

_surrogates

rated theircare as

excellent or_very

good.

COMMENT

Findings

from

_phase

I of SUPPORT

documented many

shortcomings

ofcare.

The SUPPORT

"patients

were all seri¬

ously

ill,

and their

_{dying proved}

to be

predictable,

yet

discussions anddecisions

substantially

in advance of death were

uncommon.

Nearly

half of all DNR orders

werewritteninthe last2

_days

of life. The final

_{hospitalization}

for half of

_patients

included morethan 8

days

in

generally

undesirablestates:in an

_ICU,

receiving

mechanical

_ventilation,

orcomatose.Fami¬ lies

_reported

that half of the

_patients

who

were able to communicate in their last few

_{days spent}

mostof thetime inmod¬

erateor severe

pain.

Basedon a

study

in adefined

population

atourWisconsin

site,

weestimate that

patients meeting

SUP¬

PORT criteriaaccountfor

_{approximately}

400 000 admissions peryearinthe United

States and that another 925000

_people

are

similarly

ill but wouldnotmeetSUP¬

PORT

_entry

_requirements

of

_being

hos¬

pitalized

or in intensive care.38 Patients

withSUPPORT illnesses and

_severity

ac¬

countfor about40% of_persons

_dying

in the defined

_population.

Table 4.—Effect of theSUPPORTInterventionon_{Five Outcomes Within theMajor}Disease_Categories*

UnadjustedOutcomes AdjustedOutcomes Acute Acute

Respiratory Exacerbation Advanced

or_Multiple ofCirrhosis, LungorColon

SystemFailure COPD,or_{CHF Coma Cancer}

Acute Acute

Respiratory Exacerbation Advanced

or_Multiple ofCirrhosis, LungorColon

SystemFailure COPD,orCHF Coma Cancer Median time until DNRorderwaswritten,d

Control 46 40 10 58 38 34

DNRagreement,%

Control 13 33 17 25

13 14 21 32 Undesirable states,median d

Control 10 9.5 6.5

10

Pain,%

Control 16 20

Intervention 17 18 23 16 12 26

Resourceuse,median 1993 dollars

Control 20 900 5100

19 500 5600 33 400 8900 22 000 6100

*SUPPORTindicates_StudytoUnderstand _Prognosesand Preferences for Outcomes and Risks of_{Treatments; COPD,chronic obstructive pulmonarydisease; CHF,}

congestiveheartfailure;andDNR,donotresuscitate.

Building

on the

findings

in

phase

_I,

observations of

_{others,1"16'23·24·39"'6}

the

_opin¬

ions of

_physicians

atthe five

_sites,

and the marked variation in their baseline

practices,

the

_phase

IIintervention aimed

tomake iteasierto achievebetter deci¬ sion

_making

for these

_seriously

_{ill pa¬} tients._{The intervention gave}

_physicians

reliable

_prognostic

information and

_timely

reports

of

_patient

and

_surrogate

_percep¬

tions,

the two most

_important

factors

cited

_{recently by physicians}

when con¬

sidering life-support

decisions for criti¬

cally

ill

_patients.44

The interventionnurse

also undertook

_{time-consuming}

discus¬

sions,

arranged meetings,

provided

in¬

formation,

supplied

forms,

and did_any¬

thing

else_{to encourage}the

_patient

and

family

to engageinaninformed and col¬

laborative

_{decision-making}

_processwith

awell-informed

physician

(Table 1).

The interventionwaslimited

by

its_ap¬

plication

toadiverse group of

physicians

and

_patients,

all of whom had to

_comply

voluntarily.

The intervention had to be

perceived

as

helpful, polite,

and _appro¬

priate.

As an initial

_attempt

to

_change

outcomesfor

_seriously

ill

_patients,

wedid

notseek

_authority

tobecoerciveormore

than

_minimally

_disruptive.

As

_designed,

however,

the interventionwas

vigorously

applied.

TheSUPPORTnurses werecom¬

mitted,

energetic,

and

_highly

trained.

_They

engaged

in the care of

virtually

all our

patients,

and

_nearly

_everyonehad

_printed

reports

delivered

_promptly.

Because we

thought

that

changes

in

the

_{decision-making}

_processesthatwere

not reflected in

_improved

_patient

out¬

comeswouldnotbe worth much_expense,

we

specified

five

_outcomes,

each indi¬

cating

an

important

improvement

in_pa¬

tient

_experience,

asthe main

_targets

of

the intervention.

The intervention had no

impact

on

anyof these

designated

targets

(Tables

3and

4).

Furthermore,

even

though

the

targeted

outcomes are

objectives

of

much ethical and

_legal

_writing

and of

some

explicit

social

policy

(such

as in¬

formed consent

statutes,

the Patient

Self-determination

_Act,

and

_guidelines

on

pain),1017

therewere nosecular trends

toward

_improvement

for intervention

orcontrol

patients during

the_{5 years}of

SUPPORT data collection

_(Figure

2).

These results raisefundamental_ques¬ tions about the intent and

_design

of this trial.Do

_patients

and

_physicians

seethe

documented

_shortcomings

as

troubling?

Can enhanced decision

_making

_improve

the

_experience

of

_seriously

ill and

_dying

patients?

Were the inevitable limitations of this

_project

too

great

todraw

strong

conclusions?

Because therewas no movement to¬

ward what wouldseemtobe better_prac¬

tices,

onecouldconclude that

physicians,

patients,

and families are

fairly

com¬

fortable with thecurrentsituation. Cer¬

tainly,

most

_patients

and families indi¬ cated

_they

were

_satisfied,

no matter

what

_happened

tothem.

_Physicians

have their established

_patterns

of_care, and

while

_they

were

willing

to have the SUPPORTnurse

_present

and

carrying

on

_{conversations,}

physician

behavior_ap¬

peared unchanged. Perhaps physicians

and

_patients

inthis

_{study acknowledged}

problems

with thecare of

seriously

ill

patients

as a_group.

However,

when in¬

volved with their own situation or en¬

gaged

in thecareof their individual_pa¬

tients,

they

felt

_they

were

doing

the

best

_they

_could,

weresatisfied

they

were

doing

well,

and didnotwishto

_directly

confront

_problems

orface choices.46·47

The

_{study certainly}

castsa

pall

over

anyclaim

that,

ifthe healthcare

_system

is

_given

additionalresourcesfor collabo¬

rative decision

_making

in the form of skilled

_professional

time,

improvements

will occur. In

phase

II of

_SUPPORT,

improved

information,

enhancedconver¬

sation,

andan

explicit

efforttoencour¬

ageuseofoutcomedata and

_preferences

in decision

_making

were

completely

in¬

effectual,

despite

the fact that the

_study

had

_enough

_{power to} detect small ef¬ fects.

It is

_possible

that the intervention would have beenmoreeffective if

imple¬

mented in different

_settings,

earlier in

the courseof

illness,

or with

physician

leaders rather thannurses as

implement-ers.

Perhaps,

it would have been effec¬

tive if continued formoretimeortested

atlaterend

_points.48

_However,

theover¬

all results of this

_study

arenotencour¬

aging.

No

pattern

emerged

that

_implied

that the interventionwassuccessful for

some set of

_patients

or

physicians

or

that its

_impact

increasedovertime. The

five

_hospitals

had been chosen for their

diversity

and their

_willingness

to un¬

dertakeasubstantial and controversial

challenge.

Yetnoneshoweda

tendency

toward

_improvement

in theseoutcomes.

SUPPORT did

_{demonstrate, however,}

that issues this

_complex

canbe studied

withsufficient scientific

_rigor

tobecon¬

fident of the

_findings.

We achieved

_good

interview_{response rates among}

_seriously

ill

_patients,

their

_families,

and

_physicians,

widespread

acceptance

of the interven¬ tion in diverse

_hospitals,

and

_high-qual¬

ity

data. Consent and

_{confidentiality}

is¬

sues were

complex

but amenableto so¬

lution. The

_analytic

issues

_required

_ap¬

plication

of

_relatively

novel

_approaches,

but

_{they proved}

effective. The

_study

also demonstrated the need for such methods

60 50- 40-30 Phase I Phase II -1-1—-1-1-1— 1989 1990 1991 1992 1993 1994 Year 60 40 I 20 Phase I Phase II 1989 1990 1991 1992 Year 1993 1994 2>c 30

oil25

g c 20 «œ >>is Q .

5as

15 10 5- 0-PhaseI -1-1989 1990 Phase I 1991 1992 Year 1993 1994 •a 80 0. £ 60 > en 40 o 20 Phase I PhaseI -1-1—-1-1-1— 1989 1990 1991 1992 1993 1994 Year 80 60- 40-» 20-PhaseI Phase II -1- 1989 1990 1991 1992 1993 Year 1994 Intervention_Group Control_Group

Figure2.—Seculartrends infive_patientoutcomes In theStudyto UnderstandPrognosesandPreferences for OutcomesandRisks of Treatment_(SUPPORT)from June1989through January1994. The horizontalaxes_representthe yearsofSUPPORT₍₁₉₈₉to_1994).The time between_phaseIand_phaseIIis_represented

bya_{space. The intervention and controlgroup lineshave}beensmoothednonparametrically.The_phaseIIresults_representthe actual_impactof thetrial,and the

phaseI resultsare_{the baseline}orhistorical differences. Therewere no_significantdifferencesinIntervention_patientsbetween_phasesIandII,andafter_adjustment, no_significantdifferenceswerenotedbetweenphaseIIcontrol and interventionpatientsfor the five mainoutcomes. DNRIndicates donot_resuscitate,andICU,

intensivecareunit.(Seetextfor detailed definitions andexact_{sample sizes.)}

when

_performing

evaluations of

_complex

interventions in

_seriously

ill

_patients.

We would have concluded that the interven¬

tion

_positively

influenced all outcomes

hadwenothad

_phase

I results for base¬

line

_adjustment

and

_phase

II control_pa¬ tientstoevaluate secular trends

(Table

4

and

_Figure

2).

In

_conclusion,

we areleft withatrou¬

bling

situation. The

_picture

wedescribe

of the careof

seriously

illor

dying

_per¬

sons is not attractive. One would

certainly prefer

to envision

_that,

when confronted with

_{life-threatening}

_illness,

the

_patient

and

_family

would be included

in

_discussions,

realistic estimates ofout¬

come would be

valued,

pain

would be

treated,

and

_dying

would not be _pro¬

longed.

Thatis still a

worthy

vision.2·49

However,

it isnot

_likely

tobe achieved

through

an intervention such as that

implemented by

SUPPORT. Success will

require

reexamination ofourindividual

and collective commitmenttothese

_goals,

morecreative effortsat

_shaping

thetreat¬ ment process,

and,

perhaps,

more_pro¬

active and forceful

_attempts

at

_change.

The SUPPORTprincipalinvestigators:Alfred F.Connors, Jr, MD, and Neal V. Dawson, MD,

MetroHealth Medical _{Center, Cleveland, Ohio;} Norman A.Desbiens, MD,Marshfield(Wis)Medi¬ cal ResearchFoundation;William J.Fulkerson, Jr, MD, Duke _University Medical _{Center, Durham,}

NC;LeeGoldman, MD,MPH,Beth Israel_Hospi¬

tal, Boston, Mass;William A._{Knaus, MD,George}

WashingtonUniversityMedicalCenter,Washing¬

ton, DC;Joanne Lynn, MD, Dartmouth Medical

School, Hanover, NH;and Robert K. Oye, MD,

Universityof CaliforniaatLosAngeles Medical Center.

National Coordinating Center: William A.

Knaus, MD,George WashingtonUniversityMedicai

Center, Washington,DC, and JoanneLynn, MD,

Dartmouth Medical School, Hanover, NH

(co-principal investigators); Marilyn Bergner, PhD

(deceased),and AnneDamiano, ScD,JohnsHopkins University,Baltimore, Md;RosemarieHakim, PhD, George Washington University Medical _Center,

Donald J. Murphy, MD, Presbyterian-St Lukes MedicalCenter, Denver, Colo;JoanTeno, MD,and Beth_Virnig,PhD,Dartmouth Medical_{School;Doug¬} las P._{Wagner,PhD,George Washington University} Medical_Center;and Albert W._{Wu, MD, MPH,}and Yutaka_{Yasui, PhD,}JohnsHopkins University (coin-vestigators);Detra K.Robinson, MA,GeorgeWash¬

ingtonUniversityMedical Center(chartabstraction

supervisor);Barbara_{Kreling, BA,}GeorgeWashing¬

ton_UniversityMedical Center_{(survey coordinator);} Jennie_{Dulac, BSN, RN,}DartmouthMedical School

(intervention implementation coordinator); Rose

Baker, MSHyg, George Washington University

Medical Center (database manager); and Sam

Holayel,BS,ThomasMeeks, BA,MazenMustafa, MS,and Juan_Vegarra,BS(programmers).

National Statistical Center,DukeUniversity

MedicalCenter, Durham,NC: CarlosAlzóla, MS,

and Frank E.Harrell,Jr,PhD.

Beth IsraelHospital,Boston,Mass:LeeGold¬

man,MD,MPH_{(principal investigator);}E. Francis

Cook, ScD,MaryBeth_{Hamel, MD,Lynn}Peterson, MD, Russell S. _{Phillips, MD,} Joel _{Tsevat, MD,} LachlanForrow, MD,LindaLesky,MD,and_Roger

Davis, ScD(coinvestigators); NancyKressin, MS,

andJeanmarieSolzan,BA(interviewsupervisors);

Ann Louise Puopolo,BSN, RN _(chartabstractor

supervisor);Laura_QuimbyBarrett, BSN,RN,Nora

Bucko, BSN, RN, Deborah _{Brown, MSN, RN,}

Maureen Burns,BSN, RN,CathyFoskett, BSN, RN,AmyHozid, BSN, RN,CarolKeohane, BSN, RN, ColleenMartinez, BSN, RN, Dorcie

McWee-ney, BSN, RN, DebraMelia, BSN, RN, Shelley

Otto, MSN, RN,KathySheehan, BSN, RN,Alice

Smith, BSN, RN,and LaurenTofias, MS,RN(chart

abstractors);Bernice_{Arthur, BA,}Carol_{Collins, BA,}

MaryCunnion, BA, DeborahDyer, BA, Corinne

Kulak, BS, Mary Michaels, BA,MaureenO'Keefe, BA,MarianParker, AB, MBA,LaurenTuchin, BA, Dolly Wax, BA,andDianaWeld, A(interviewers);

Liz Hiltunen, MS, RN, CS, Geòrgie Marks, MS,

MEd, RN,Nancy Mazzapica,MSN, RN,andCindy

Medich,MS,RN(SUPPORTnurseclinicians);and JaneSoukup,MS(analyst/datamanager).

Duke_UniversityMedicalCenter, Durham,NC: WilliamJ.Fulkerson, Jr,MD(principal investiga¬

tor); Robert M. Califf, MD, AnthonyN. Galanos,

MD, Peter_{Kussin, MD,}and Lawrence H.

Muhl-baier,PhD_{(coinvestigators);}MariaWinchell, MS

(projectdirector); Lee_Mallatratt, RN_(chartab¬

stractorsupervisor); Ella_Akin, BA_(interviewer

supervisor); LynneBelcher, RN,Elizabeth_Buller,

BSN, RN,Eileen_{Clair, RN,}LauraDrew, BSN,

RN, Libby Fogelman, BSN, RN, Dianna Frye,

BSN, RN,BethFraulo, BSN,RN,DebbieGessner, BSN, RN,Jill_{Hamilton, BSN, RN,}Kendra_Kruse,

BSN, RN,DawnLandis,BSN, RN,LouiseNobles,

BSN, RN,Rene Oliverio, BSN, RN,and Carroll

Wheeler, BSN, RN (chart abstractors); Nancy

Banks, MA, Steven Berry, BA,Monie Clayton,

Patricia Hartwell, MAT, Nan Hubbard, Isabel

Kussin, BA,BarbaraNorman, BA,JackieNoveau, BSN, Heather _{Read, BA,}and Barbara _Warren, MSW_{(interviewers);}JaneCastle, MSN, RN,Beth

Fraulo, BSN, RN,ReneOliverio, BSN, RN,and

KathyTurner, MSN,RN(SUPPORTnurseclini¬

cians);and Rosalie Perdue(datamanager).

MetroHealth Medical_{Center, Cleveland,}Ohio: Alfred F._{Connors, Jr, MD,}and Neal V._Dawson, MD_{(co-principal investigators);}ClaudiaCoulton, PhD, C. SethLandefeld, MD,Theodore_Speroff,

PhD,and StuartYoungner,MD(coinvestigators); Mary J. _{Kennard, MSN,} and _Mary Naccaratto,

MSN (chart abstractor _{supervisors); Mary} Jo

Roach,PhD(interviewersupervisor);Maria

Blink-horn, RN,Cathy Corrigan,RNC,Elsie_{Geric, RN,}

LauraHaas, RN,Jennifer_{Ham, RN,}Julie

Jer-donek, RN, Marilyn Landy, RN, ElaineMarino, RN,PattiOlesen, RN,SherryPatzke, RN,Linda