Mette KragAnders PernerJørn WetterslevMatt P. WiseMorten Hylander Møller

Stress ulcer prophylaxis versus placeboor no prophylaxis in critically ill patientsA systematic review of randomised clinical trials withmeta-analysis and trial sequential analysis

Received: 30 July 2013Accepted: 25 September 2013Published online: 19 October 2013� Springer-Verlag Berlin Heidelberg andESICM 2013

Systematic review registrationThe International Prospective Register ofSystematic Reviews (PROSPERO), No.CRD42013004142.

Electronic supplementary materialThe online version of this article(doi:10.1007/s00134-013-3125-3) containssupplementary material, which is availableto authorized users.

M. Krag � A. Perner �M. Hylander Møller ())Department of Intensive Care, CopenhagenUniversity Hospital Rigshospitalet,Blegdamsvej 9, 2100 Copenhagen,Denmarke-mail: mortenhylander@gmail.comTel.: ?45-22555343URL: www.sup-icu.com

J. WetterslevCopenhagen Trial Unit, Centre for ClinicalIntervention Research, CopenhagenUniversity Hospital Rigshospitalet,Copenhagen, Denmark

M. P. WiseDepartment of Adult Critical Care,University Hospital of Wales, Cardiff, UK

Abstract Purpose: To assess theeffects of stress ulcer prophylaxis(SUP) versus placebo or no prophy-laxis on all-cause mortality,gastrointestinal (GI) bleeding andhospital-acquired pneumonia in adultcritically ill patients in the intensivecare unit (ICU). Methods: We per-formed a systematic review usingmeta-analysis and trial sequentialanalysis (TSA). Eligible trials wererandomised clinical trials comparingproton pump inhibitors or histamine 2receptor antagonists with either pla-cebo or no prophylaxis. Tworeviewers independently assessedstudies for inclusion and extracteddata. The Cochrane Collaborationmethodology was used. Risk ratios/relative risks (RR) with 95 % confi-dence intervals (CI) were estimated.The predefined outcome measureswere all-cause mortality, GI bleeding,and hospital-acquired pneumonia.Results: Twenty trials (n = 1,971)were included; all were judged ashaving a high risk of bias. There was

no statistically significant differencein mortality (fixed effect: RR 1.00,95 % CI 0.84–1.20; P = 0.87;I2 = 0 %) or hospital-acquired pneu-monia (random effects: RR 1.23,95 % CI 0.86–1.78; P = 0.28;I2 = 19 %) between SUP patients andthe no prophylaxis/placebo patients.These findings were confirmed in theTSA. With respect to GI bleeding, astatistically significant difference wasfound in the conventional meta-ana-lysis (random effects: RR 0.44, 95 %CI 0.28–0.68; P = 0.01; I2 = 48 %);however, TSA (TSA adjusted 95 %CI 0.18–1.11) and subgroup analysescould not confirm this finding. Con-clusions: This systematic reviewusing meta-analysis and TSA dem-onstrated that both the quality and thequantity of evidence supporting theuse of SUP in adult ICU patients islow. Consequently, large randomisedclinical trials are warranted.

Keywords Stress ulceration �Gastrointestinal bleeding �All-cause mortality � Meta-analysis �Trial sequential analysis �Stress ulcer prophylaxis

Introduction

Critically ill patients are at risk of stress-related gastro-intestinal (GI) bleeding [1]. The reported incidence of GI

bleeding in the intensive care unit (ICU) ranges from 2 to15 %, however this data derives from research published15–20 years ago [2, 3]. Intensive care practice haschanged substantially over recent decades and,

Intensive Care Med (2014) 40:11–22DOI 10.1007/s00134-013-3125-3 SYSTEMATIC REVIEW

consequently, the incidence of GI bleeding in critically illpatients may also have changed.

GI bleeding due to stress ulceration has been associ-ated with increased mortality and a prolonged length ofICU stay of 4–8 days [3, 4]. The results of older clinicaltrials indicate that stress ulcer prophylaxis (SUP) reducesthe frequency of GI bleeding in ICU patients, and SUP istherefore regarded as a standard of care in ICU as outlinedby the Surviving Sepsis Campaign guidelines [5]. How-ever, the rationale and level of evidence of SUP in ICUpatients has been questioned because of limited data,methodological flaws in some trials, possible increasedincidence of hospital-acquired pneumonia and Clostrid-ium difficile enteritis following the use of SUP andgeneral improvements in intensive care [1, 6–9]. Fur-thermore, there is a lack of studies comparing the use ofSUP versus no prophylaxis or placebo. Uncertainty overwhether routine SUP is indicated in critically ill patientstherefore exists amongst clinicians. As a result, there is aneed to weigh the risks of SUP in ICU patients against thebenefits of this approach in this patient group using up-to-date rigorous evidence-based methodology. The objectiveof our systematic review was to assess the effects of SUPversus placebo or no prophylaxis on all-cause mortality,GI bleeding and hospital-acquired pneumonia in criticallyill patients using strict bias evaluation, cumulative meta-analysis and trial sequential analysis (TSA).

Methods

This systematic review is based on the methodologyrecommended by the Cochrane Collaboration [10], andthe review has been prepared according to the PRISMAstatement [11]. The protocol is published in the Interna-tional Prospective Register of Systematic Reviews(PROSPERO), no. CRD42013004142.

Eligibility criteria

Potentially eligible trials had to be randomised, includeadult patients admitted to the ICU, have an interventiongroup that received SUP with proton pump inhibitors(PPIs) or histamine 2 receptor antagonists (H2RAs) andinclude a control group that received placebo or noprophylaxis. We included trials irrespective of languageand publication status. Trials were permitted to havemore than one intervention group. Exclusion criteriawere studies in animals, trials in patients aged\18 years, trials in patients not admitted to the ICUand trials only reporting non-patient-centred outcomemeasures [12], such as gastric pH and gastriccolonisation.

Search strategy

We framed the following focused research question: ‘‘IsSUP with PPIs or H2RAs in critically ill patients in theICU superior to placebo or no prophylaxis?’’

A population, intervention, comparator and outcomes-based question and literature search was created [13][Electronic Supplementary Material (ESM) 1].

The following databases were searched for literature:MEDLINE, including MeSH (January 1966 to March2013), EMBASE (January 1980 to March 2013) and theCochrane Library (Issue 2, March 2013). We also handsearched the reference lists of included trials and othersystematic reviews of SUP in the critically ill patients.The electronic literature search was last updated March20, 2013.

Study selection

Two authors (MK and MHM) independently reviewed alltitles and abstracts identified in the literature search andexcluded trials that were obviously not relevant. Theremaining trials were evaluated in full text. Disagree-ments were resolved by JW.

Data extraction

Two authors (MK and MHM) independently extractedinformation from each included trial using a dataextraction form. The extracted information included trialcharacteristics (year of publication, duration, country),characteristics of the trial participants (inclusion criteria,type of nutrition used), exclusion criteria, type of inter-vention/control (name, dosing, duration, route ofadministration, comparator) and outcomes. Trials werecategorised as using enteral nutrition—if patients receivedany volume of enteral nutrition. The predefined primaryoutcome measure of this review was all-cause mortality,and the predefined secondary outcomes were GI bleedingand pneumonia. The outcome measures were the same asthose defined by the authors of the original trials.

Risk of bias assessment

To determine the validity of the included trials, twoauthors (MK and MHM) independently assessed the riskof bias as advised by the Cochrane Collaboration [10],including the domains of random sequence generation,allocation concealment, blinding, incomplete outcomedata, selective outcome reporting, baseline imbalance andbias due to vested financial interest. If one or moredomains were judged as being high or unclear, we clas-sified the trial as having a high risk of bias [10].

12

Statistical analyses

Review Manager 5.1.7 was used for statistical analyses, andfor the TSA we used the TSA program version 0.9 beta(www.ctu.dk/tsa). For each included trial we calculated therelative risks (RR) with 95 % confidence intervals (CI) forthe three dichotomous outcome measures, and we pooledthese measures in the meta-analysis. Heterogeneity amongtrials was quantified with inconsistency factor (I2) [14] andthe Diversity (D2) statistics [15]. If the I2 statistic was 0, wereported the results from a fixed-effect model, and if the I2

statistic was [0, we reported results from both a fixed-effectand random-effects model. The cumulative meta-analysiswas challenged with the application of TSA—a sensitivityanalysis that widens the confidence intervals in case the dataare too sparse to draw firm conclusions [15–17] (ESM 2). Inaddition, the sensitivity analysis included application ofcontinuity correction in trials of zero events [18]. Risk ofsmall trial bias was assessed by Funnel plot asymmetry [19].

Subgroup analyses

We performed five predefined subgroup analyses and onepost hoc subgroup analysis: (1) high versus low risk ofbias trials (a possible increased intervention effect in trialswith a high risk of bias); (2) adequate versus inadequaterandom sequence generation, allocation concealment andblinding (a possible increased intervention effect in trialswith an inadequate random sequence generation, alloca-tion concealment and blinding); (3) use of PPIs versusH2RAs (a possible increased intervention effect in the PPIgroup); (4) medical versus surgical versus mixed ICU (apossible increased intervention effect in surgical patients);(5) use of enteral nutrition versus no enteral nutrition (apossible increased risk of hospital-acquired pneumonia inpatients receiving enteral nutrition); (6) placebo trialsversus no prophylaxis trials (a possible increased inter-vention effect in the no prophylaxis trials).

Results

Figure 1 summarises the results of the search: 20 trialswere included, all of which were published in English(Table 1) [20–39]. The main reasons for exclusion oftrials were (1) no relevant patient-centred outcome mea-sures were reported [40, 41] and (2) the trials wereconducted in a non-ICU setting (ESM 3) [42, 43].

Characteristics of trials

Of the 20 trials included in this review, 16 (80 %) weresingle-centre trials [20–29, 31, 34–38]. The four

multicentre trials were all from the USA and comprised atotal of 585 patients (30 % of the included patients) [30,32, 33, 39]. Twelve trials (60 %) used placebo as com-parator [20–22, 24, 31, 36, 37, 39], whereas the remainingtrials used no prophylaxis. In seven trials, patients werefed enterally [20, 22, 27, 32, 34, 36, 38].

Participants

The 20 included trials enrolled 1,971 adult patients in theICU. Seven trials included patients from mixed ICUs [21,

Fig. 1 Study flow diagram

13

Tab

le1

Char

acte

rist

ics

of

the

incl

uded

tria

ls

Fir

stau

thor

of

publi

shed

tria

l

No.

of

pat

ients

intr

ial

(n)

Set

ting/

countr

yT

rial

dura

tion

(month

s)

Inte

nsi

ve

care

unit

Ente

ral

nutr

itio

nIn

clusi

on

crit

eria

(popula

tion)

Excl

usi

on

crit

eria

Inte

rven

tion

Com

par

ator

Outc

om

es

Contr

ol

gro

up:

no

pro

phyla

xis

Apte

[20]

34

Sin

gle

centr

e/In

dia

Unknow

nM

edic

alY

esT

rach

eoto

mis

edpat

ients

wit

hte

tatu

sP

neu

monia

;S

UP

Ran

itid

ine

50

mg/e

ver

y6

hIV

No

pro

phyla

xis

Mort

alit

yN

oso

com

ial

pneu

monia

Over

tG

Ible

edin

gb

Bas

so[2

1]

116

Sin

gle

centr

e/It

aly

14

Surg

ical

No

Hig

hri

skpla

stic

-an

dneu

rosu

rger

ya

GI

ble

edin

g;

GI

surg

ery;

age

\12

yea

rs;

coag

ulo

pat

hy.

Cim

etid

ine

200

mg/e

ver

y6

hIV

or

ora

lly

inat

leas

t10

day

s

No

pro

phyla

xis

Over

tG

Ible

edin

g

Ben

-Men

achem

[22

]200

Sin

gle

centr

e/U

SA

10

Med

ical

Yes

Any

GI

ble

edin

g;

GI

surg

ery;

age

\18

yea

rs;

expec

ted

ICU

stay

\24

h;

SU

P;

anti

coag

ula

nts

;re

cent

surg

ery/

anae

sthes

ia;

hea

din

jury

;ra

ised

ICP

;gra

de

4hep

atic

ence

phal

opat

hy;

pre

gnan

cyor

lact

atat

ion

Cim

etid

ine

load

ing

dose

of

300

mg

IVan

din

fusi

on

titr

ated

tom

ainta

ingas

tric

pH

[4

No

pro

phyla

xis

Hosp

ital

mort

alit

yN

oso

com

ial

pneu

monia

Cli

nic

ally

signifi

cant

GI

ble

edin

gc

Dar

long

[24

]31

Sin

gle

centr

e/In

dia

Unknow

nM

ixed

No

MV

([24

hex

pec

ted

dura

tion)

and

NG

tube

insi

tu

GI

ble

edin

g;

SU

P;

coag

ulo

pat

hy;

anti

coag

ula

nts

Ran

itid

ine

50

mg/

ever

y8

hIV

No

pro

phyla

xis

Over

tG

Ible

edin

g

Mac

Dougal

l[3

1]

62

Sin

gle

centr

e/U

K18

Med

ical

No

Fulm

inan

thep

atic

fail

ure

GI

ble

edin

gM

etia

mid

e150

mg/

h(n

=10)

or

cim

etid

ine

100

mg/h

(n=

16)

IVy

giv

ing

pH

[5

No

pro

phyla

xis

Mort

alit

yO

ver

tG

Ible

edin

g

Reu

sser

[36

]40

Sin

gle

centr

e/S

wit

zerl

and

26

Surg

ical

No

Acu

tetr

aum

atic

or

sponta

neo

us

bra

inin

jury

and

neu

rosu

rger

yan

dM

V[

48

h

GI

ble

edin

g;

age

\15

yea

rs;

GI

surg

ery;

PU

D;

SU

P

Ran

itid

ine

50

mg/

ever

y8/6

hIV

titr

ated

tom

ainta

ingas

tric

pH

[4

No

pro

phyla

xis

Mort

alit

yO

ver

tG

Ible

edin

g

Ruiz

-San

tana

[37

]49

Sin

gle

centr

e/S

pai

n14

Mix

edY

esM

V(e

xpec

ted

dura

tion

[6

day

s)an

dm

etab

oli

cst

ress

;no

shock

and

no

renal

or

hep

atic

dis

ease

;to

tal

par

ente

ral

nutr

itio

n

GI

ble

edin

g;

GI

surg

ery;

PU

D;

SC

I;A

KI;

hep

atic

fail

ure

,ca

taboli

cin

dex

score

B0;

SU

P

Ran

itid

ine

50

mg/

ever

y8

hIV

No

pro

phyla

xis

Mort

alit

yO

ver

tG

Ible

edin

g

Zin

ner

[39

]200

Mult

icen

tre/

US

A24

Surg

ical

No

Expec

ted

ICU

dura

tion

[48

hG

Ible

edin

g;

GI

surg

ery;

PU

DC

imet

idin

e300

mg/e

ver

y6

hIV

duri

ng

ICU

stay

No

pro

phyla

xis

Mort

alit

yO

ver

tG

Ible

edin

g

14

Ta

ble

1co

nti

nu

ed

Fir

stau

thor

of

publi

shed

tria

l

No.

of

pat

ients

intr

ial

(n)

Set

ting/

countr

yT

rial

dura

tion

(month

s)

Inte

nsi

ve

care

unit

Ente

ral

nutr

itio

nIn

clusi

on

crit

eria

(popula

tion)

Excl

usi

on

crit

eria

Inte

rven

tion

Com

par

ator

Outc

om

es

Contr

ol

gro

up:

pla

cebo

Burg

ess

[23

]34

Sin

gle

centr

e/U

SA

9S

urg

ical

No

Sev

ere

hea

din

jury

and

GC

SB

10

PU

D;

GI

inju

ry;

SU

P;

ora

lin

take

Ran

itid

ine

6.2

5m

g/h

IVfo

rup

to72

h

Pla

cebo

Mort

alit

yO

ver

tG

Ible

edin

g

Fri

edm

an[2

5]

25

Sin

gle

centr

e/U

SA

18

Med

ical

No

MV

GI

ble

edin

g;

AK

I;S

UP

;pre

gnan

cyC

imet

idin

e300

mg/e

ver

y6

hIV

Pla

cebo

Over

tG

Ible

edin

g

Gro

ll[2

6]

221

Sin

gle

centr

e/C

anad

a21

Mix

edN

oA

ny

GI

ble

edin

g;

AK

I;S

UP

;pre

gnan

cy;

dru

gover

dosa

ge;

AM

I

Cim

etid

ine

300

mg/e

ver

y6

hIV

Pla

cebo

Mort

alit

yO

ver

tG

Ible

edin

g

Hal

lora

n[2

7]

50

Sin

gle

centr

e/U

SA

21

Surg

ical

Yes

Sev

ere

hea

din

jury

Bra

indea

th;

PU

D;

pre

gnan

cy;

GI

inju

ry;

sever

ehep

atic

or

renal

dis

ease

Cim

etid

ine

300

mg/e

ver

y4

hIV

for

up

to3

wee

ks

Pla

cebo

Over

tG

Ible

edin

g

Han

isch

[28

]114

Sin

gle

centr

e/G

erm

any

12

Surg

ical

No

MV

PU

D;

GI

ble

edin

g;

age\

18

yea

rs;

tran

spla

nte

dpat

ients

(kid

ney

,li

ver

,hea

rt);

pneu

monia

;G

Isu

rger

y

Ran

itid

ine

50

mg

93

IVP

lace

bo

Mort

alit

yN

oso

com

ial

pneu

monia

Cli

nic

ally

signifi

cant

GI

ble

edin

g

Kan

toro

va

[29

]208

Sin

gle

centr

e/C

zech

Rep

ubli

c

29

Surg

ical

Yes

MV

(expec

ted

dura

tion

[48

h)

or

coag

ulo

pat

hy;

age

[18

yea

rs;

NG

tube

insi

tu

GI

ble

edin

g;

GI

surg

ery;

pneu

monia

;S

UP

;P

UD

;an

tico

agula

nts

;A

KI,

coag

ulo

pat

hy;

life

expec

tancy

\3

month

s

Om

epra

zole

40

mg

IV;

Fam

oti

din

e40

mg

92

IV

Pla

cebo

Hosp

ital

mort

alit

yN

oso

com

ial

pneu

monia

Cli

nic

ally

signifi

cant

GI

ble

edin

g

Kar

lsta

dt

[30

]87

Mult

icen

tre/

US

AU

nknow

nM

ixed

No

At

leas

tone

of

the

foll

ow

ing:

maj

or

thora

cic

or

abdom

inal

surg

ery;

mult

iple

trau

ma;

hypote

nsi

on;

hypovole

mic

shock

;se

psi

s;ac

ute

resp

irat

ory

fail

ure

GI

ble

edin

g;

hep

atic

fail

ure

;A

KI;

SU

P;

pre

gnan

cyor

lact

atat

ion;

age

\16

yea

rs;

hyper

secr

etory

dis

ord

ers

Cim

etid

ine

300

mg

load

ing

dose

,fo

llow

edby

infu

sion

at50

mg/e

ver

yhour

Pla

cebo

Mort

alit

yN

oso

com

ial

pneu

monia

Cli

nic

ally

signifi

cant

GI

ble

edin

g

15

Ta

ble

1co

nti

nu

ed

Fir

stau

thor

of

publi

shed

tria

l

No.

of

pat

ients

intr

ial

(n)

Set

ting/

countr

yT

rial

dura

tion

(month

s)

Inte

nsi

ve

care

unit

Ente

ral

nutr

itio

nIn

clusi

on

crit

eria

(popula

tion)

Excl

usi

on

crit

eria

Inte

rven

tion

Com

par

ator

Outc

om

es

Mar

tin

[32

]131

Mult

icen

tre/

US

A7

Mix

edY

esE

xpec

ted

ICU

stay

[36

h;

NG

tube

insi

tu;

atle

ast

one

of

the

foll

ow

ing:

maj

or

surg

ery;

mult

iple

trau

ma;

hypote

nsi

on;

hypovole

mic

shock

;se

psi

s;ac

ute

resp

irat

ory

fail

ure

;ja

undic

e;burn

[30%

GI

ble

edin

g;

pre

gnan

cyor

lact

atat

ion;

age

\16

yea

rs;

GI

surg

ery;

SU

P;

anti

coag

ula

nts

Cim

etid

ine

300

mg

load

ing

dose

,fo

llow

edby

50

mg/h

IVfo

rup

to7

day

s

Pla

cebo

30-d

aym

ort

alit

yN

oso

com

ial

pneu

monia

Over

tG

Ible

edin

g

Met

z[3

3]

167

Mult

icen

tre/

US

A20

Surg

ical

No

Sev

ere

hea

din

jury

;G

CS

B10;

NG

tube

inpla

ce

GI

ble

edin

g;

age

\18

yea

rs;

expec

ted

ICU

stay

\72

h;

burn

s[20

%;

AK

I;P

UD

;co

agulo

pat

hy;

SU

P

Ran

itid

ine

6.2

5m

g/h

IVfo

rup

to5

day

s

Pla

cebo

Mort

alit

yN

oso

com

ial

pneu

monia

Over

tG

Ible

edin

g

Peu

ra[3

4]

39

Sin

gle

centr

e/U

SA

Unknow

nM

edic

alY

esE

xpec

ted

ICU

stay

[5

day

sG

Ible

edin

g;

age\

18

yea

rs;

GI

surg

ery;

AM

I;pre

gnan

cy

Cim

etid

ine

300

mg/e

ver

y6

hIV

for

3–14

day

s

Pla

cebo

Mort

alit

yO

ver

tG

Ible

edin

g

Pow

ell

[35]

41

Sin

gle

centr

e/U

KU

nknow

nS

urg

ical

No

Sch

edule

dco

ronar

yar

tery

bypas

sgra

ftpat

ients

inIC

U

PU

D;

GI

surg

ery;

SU

P;

sever

eal

lerg

y;

AK

I;hep

atic

dis

ease

;an

tico

agula

nts

Ran

itid

ine

50

mg/

ever

y8

hIV

;om

epra

zole

80

mg

load

ing

dose

foll

ow

edby

40

mg/e

ver

y8

hby

bolu

sIV

;om

epra

zole

80

mg

load

ing

dose

foll

ow

edby

40

mg/e

ver

y8

hby

infu

sion

Pla

cebo

Mort

alit

yO

ver

tG

Ible

edin

g

Van

den

Ber

g[3

8]

28

Sin

gle

centr

e/N

ether

lands

Unknow

nM

ixed

Yes

MV

Unknow

nC

imet

idin

e20

mg/

kg/e

ver

y24

hIV

Pla

cebo

Over

tG

Ible

edin

g

AK

Iac

ute

kid

ney

inju

ry,

AM

Iac

ute

myoca

rdia

lin

farc

tion,

EN

ente

ral

nutr

itio

n,

GI

gas

troin

test

inal

,G

CS

Gla

sgow

Com

aS

core

;IC

Pin

trac

rania

lpre

ssure

,IC

Uin

tensi

ve

care

unit

,IV

intr

aven

ousl

y,

MV

mec

han

ical

ven

tila

tion,

NG

nas

ogas

tric

,P

UD

pep

tic

ulc

erdis

ease

,SC

Isp

inal

cord

inju

ry,

SU

Pst

ress

ulc

erpro

phyla

xis

aN

euro

surg

ery;

hea

din

jury

,re

spir

atory

fail

ure

;se

psi

s;burn

s[

20

%;

hypote

nsi

on;

post

oper

ativ

eco

mpli

cati

ons;

acute

renal

fail

ure

;m

ult

iple

trau

ma

bO

ver

tG

Ible

edin

gin

cluded

hae

mat

emes

is,

coff

eegro

und

emes

is,

mel

ena,

hae

mat

och

ezia

or

blo

ody

nas

ogas

tric

aspir

ate

cC

linic

ally

signifi

cant

GI

ble

edin

gin

cludes

over

tG

Ible

edin

gan

dcl

inic

ally

rele

van

tch

anges

invit

alsi

gns

or

hae

moglo

bin

16

24, 26, 30, 32, 37, 38], eight trials included surgical ICUpatients only (including trauma and neurosurgery) [23, 27–29, 33, 35, 36, 39] and five trials included patients inmedical ICUs [20, 22, 25, 31, 34]. Inclusion and exclusioncriteria varied considerably between trials (Table 1).

Interventions

Twenty trials evaluated H2RAs and two trials evaluatedPPIs. Seven trials assessed ranitidine [20, 23, 24, 28, 33, 36,37], ten trials assessed cimetidine [21, 22, 25–27, 30, 32, 34,38, 39], two trials assessed omeprazole [29, 35] and threetrials evaluated more than one intervention (e.g. both PPIand H2RA) [29, 31, 35]. The route of administration wasintravenously in 19 trials [20, 22–39] and either orally orintravenously in one trial [21]. SUP dosing and duration oftreatment varied across trials (Table 1).

Bias risk assessment

No trials were judged to be of low risk of bias in all sixdomains (Fig. 2). The main reasons for high risk of biaswere inadequate random sequence generation, allocationconcealment or blinding (ESM 4). One trial had adequaterandom sequence generation, allocation concealment andblinding [28] and was included in the subgroup analysisof adequate versus inadequate random sequence genera-tion, allocation concealment and blinding. Eight trials hadpotential financial bias because of sponsorship by phar-maceutical companies [20, 23, 25, 27, 30–33].

Outcome measures

All-cause mortality

Mortality data were obtained from 15 trials including1,604 patients [20, 22, 23, 26–32, 34–37, 39]. The meta-analysis of all 15 trials showed no significant difference inmortality in patients treated with SUP compared withthose treated with placebo or no prophylaxis (fixed effect:RR 1.00, 95 % CI 0.84–1.20; P = 0.87; I2 = 0 %;Fig. 3). Application of an empirical continuity correctionof 0.01 in the two no-event trials did not change the result.The subgroup analysis of trials using PPIs versus trialsusing H2RAs showed no increased intervention effect inthe PPI group (test of interaction P = 0.68). The sub-group analysis of trials with a low risk of bias versus ahigh risk of bias could not be done because there were notrials with a low risk of bias. In the subgroup analysis oftrials with adequate versus inadequate random sequencegeneration, allocation concealment and blinding and inthe placebo versus no prophylaxis subgroup analysis, noincreased intervention effect was found in the inadequate

Fig. 2 Risk of bias summary. Review of authors’ judgements abouteach risk of bias item for each included study. Red High risk, greenlow risk, yellow unclear

17

group or in the no prophylaxis group. The subgroupanalysis according to type of ICU, showed no increasedintervention effect in the surgical ICU trials (test ofinteraction P = 0.11). Finally, no subgroup differencewas found between enterally fed patients and non-enter-ally fed patients (test of interaction P = 0.11). TheFunnel plot raised concern about small trial bias (ESM 8).

TSA showed that 57 % (1,594 patients) of the requiredinformation size to detect or reject a 20 % relative riskreduction (RRR) corresponding to 2,794 patients wasaccrued. The cumulative Z curve did not even touch theconventional boundary for harm or benefit (P [ 0.05) orthe trial sequential monitoring boundary for harm orbenefit (ESM 5). However, the Z curve did reach thefutility area, hereby excluding a 20 % RRR in mortalityby using PPIs or H2RA.

GI bleeding

All 20 trials (n = 1,971) had data on GI bleeding [20–39].The conventional meta-analysis showed a statisticallysignificant difference in GI bleeding in patients treatedwith SUP compared with those treated with placebo or noprophylaxis (fixed effect: RR 0.41, 95 % CI 0.31–0.53;

P = 0.01; I2 = 48 %; random effects: RR 0.44, 95 % CI0.28–0.68) (Fig. 4). Application of an empirical conti-nuity correction of 0.01 in the four no-event trials did notchange the result. In the subgroup analysis of trials usingPPIs versus trials using H2RAs, no increased interventioneffect in the PPI group was found (test of interactionP = 0.54). No trials had a low risk of bias, so the sub-group analysis of low versus high risk of bias trials couldnot be done. The adequate versus inadequate randomsequence generation, allocation concealment and blindingsubgroup analysis, and the placebo versus no prophylaxissubgroup analysis showed no signs of an increasedintervention effect in the inadequate group or in the noprophylaxis group. The subgroup analysis according totype of ICU showed no increased intervention effect inthe surgical ICU trials (test of interaction P = 0.92). Nostatistically significant subgroup difference betweenenterally fed patients and parenteral nutrition was found(test of interaction P = 0.15). The risk of smaller trialbias was low according to the Funnel plot (ESM 8).

TSA showed that only 22 % (1,881 patients) of therequired information size of 8,707 patients was accrued.The cumulative Z curve crossed the conventionalboundary for benefit (P \ 0.05), but not the trialsequential monitoring boundary for benefit (ESM 9).

Fig. 3 Stress ulcer prophylaxis (SUP) and all-cause mortality. Size of squares for risk ratio (RR) reflects the weight of the trial in thepooled analyses. Horizontal bars 95 % Confidence intervals (CI)

18

Hospital-acquired pneumonia

Seven trials comprising 1,008 patients reported data onhospital-acquired pneumonia. The meta-analysis showedno statistically significant difference in pneumonia inpatients treated with SUP compared with those treatedwith placebo or no prophylaxis (fixed effect: RR 1.16,95 % CI 0.84–1.58; P = 0.28; I2 = 19 %; randomeffects: RR 1.23, 95 % CI 0.86–1.78) (ESM 6). Appli-cation of an empirical continuity correction of 0.01 inthe two no-event trials did not change the result. Thesubgroup analysis of trials using PPIs versus trials usingH2RAs showed no increased intervention effect in thePPI group (test of interaction P = 0.56). The subgroupanalysis of trials with a low risk of bias versus a highrisk of bias could not be done. In the subgroup analysisof adequate versus inadequate random sequence gener-ation, allocation concealment and blinding, no signs ofan increased intervention effect in the inadequate groupwas found. An increased intervention effect in the pla-cebo group was found in the placebo versus noprophylaxis subgroup analysis (test of interaction

P = 0.02); however, significant statistical heterogeneitywas present (I2 = 80 %). The subgroup analysisaccording to type of ICU showed no increased inter-vention effect in the surgical ICU trials (test ofinteraction P = 0.11). Finally, no statistically significantincreased risk of hospital-acquired pneumonia in theenteral nutrition group was found (test of interactionP = 0.06). No evidence of smaller trial bias was presentin the Funnel plot (ESM 8).

TSA showed that a mere 12 % (1,008 patients) of therequired information size of 8,694 patients was accrued.The cumulative Z curve did not touch the conventionalboundary for harm or benefit, or the trial sequentialmonitoring boundary for harm or benefit (ESM 7).

Discussion

In the present systematic review using meta-analysis andTSA on SUP in adult critically ill patients in the ICU,SUP was not statistically significantly different from

Fig. 4 SUP and gastrointestinal (GI) bleeding. Size of squares for RR reflects the weight of the trial in pooled analyses. Horizontal bars95 % CI

19

placebo or no prophylaxis in terms of mortality, GIbleeding and pneumonia.

Mortality

The pooled analysis of mortality showed neither benefitnor harm of SUP with PPIs or H2RAs. No subgroupdifferences were present. The sensitivity analysis withTSA confirmed the finding in the conventional meta-analysis. Importantly, the TSA showed that it is unlikelythat SUP will result in a relative mortality reduction of20 % if further trials are conducted in adult ICU patients.

According to the risk of bias assessment [10], all trialshad a high risk of bias. Thus, the pooled analyses may beinfluenced by the poor quality of existing trials, which couldresult in inflated point estimates and thus make interpretationdifficult. Furthermore, the Funnel plot asymmetry with theabsence of small negative trials increases the risk of over-estimating the effect of SUP [44, 45].

GI bleeding

The conventional pooled analysis of GI bleeding showeda benefit of H2RAs. However, this finding could not beconfirmed in the analysis of trials with adequate randomsequence generation, allocation concealment and blinding(n = 1), and in the TSA. Consequently, an inflated pointestimate in the conventional pooled analysis can be sus-pected. No subgroup differences were present.Considering the high risk of bias and sparse data, a gen-uine benefit of SUP on the risk of GI bleeding in adultICU patients may be questioned.

Hospital-acquired pneumonia

No statistically significant benefit or harm of SUP on the riskof hospital-acquired pneumonia was demonstrated in theconventional meta-analysis or TSA. This was confirmed inthe subgroup analyses. The overall high risk of bias in thetrials warrants careful interpretation of the results because ofan increased risk of falsely inflated estimates [44, 45].

Strengths and limitations of the review

The compliance with the recommendations of the Coch-rane Collaboration is a strength of the present systematicreview. The recommendations implemented in our reviewinclude a published protocol, an up-to-date literaturesearch with no language restrictions, an independent lit-erature search, data extraction and bias risk assessment bytwo authors and the inclusion of trials irrespective ofpublication and language status. In addition, we reduced

the risk of random error in the meta-analysis with theapplication of TSA to increase the robustness of theanalyses; this methodology has not been used in existingmeta-analyses on SUP in the ICU. We excluded trialsmerely reporting non-patient-centered outcomes in orderto make the results relevant for clinical practice. Thesubgroup analysis of trials with adequate versus inade-quate random sequence generation, allocationconcealment and blinding might have resulted in spuriousfindings. However, this analysis was introduced with theaim of estimating a possible bias effect. The heterogeneityof the included trials was considerable. We did not definethe three outcome measures evaluated; rather, we used thedefinitions proposed by the authors, which may haveresulted in trial heterogeneity. Most of the included trialshave been conducted in high-risk patients, which must bekept in mind when interpreting the results. Overall, sta-tistical heterogeneity did not seem to be a big issue, andwe have reported both fixed-effect and random-effectspooled estimates when heterogeneity was present.

Relation to other reviews and implication for futureresearch

No previous systematic reviews have been published onPPIs versus placebo or no prophylaxis, and only a fewsystematic reviews have evaluated H2RAs versus placeboor no prophylaxis. In 2010, Marik and colleagues sug-gested that in patients who are fed enterally, SUP does notreduce the risk of GI bleeding from stress ulcers and mayeven increase the risk of pneumonia and death [1].However, there are a number of limitations to the reviewof these authors as three published trials were not iden-tified or included [24, 25, 31], the risk of bias andprecision assessment was not conducted as recommendedby GRADE [46], no study protocol was published orregistered and the increased risk of random errors in theconventional meta-analysis was not evaluated. Theseissues may have contributed to the discrepancies in rela-tion to the present review. In 1996, Cook and colleaguesconducted a thorough and comprehensive systematicreview of SUP in critically ill patients [47]. However, theuse of PPIs in the treatment of peptic ulcer disease beganafter 1996; consequently, the effect of PPIs was notevaluated in this meta-analysis. A number of methodo-logical discrepancies between the Cook review and ourreview do exist. Not surprisingly, the placebo/no pro-phylaxis part of the Cook review included fewer trials andpatients. Secondly, a different risk of bias assessment wasused. Thirdly, no published protocol was identified.Finally, the increased risk of random errors due to pos-sible multiple updating and sparse data was not assessed.Despite these dissimilarities, the pooled estimates in theconventional meta-analysis were of the same magnitudeas those observed in our study.

20

The authors of recently published systematic reviewshave suggested that PPIs significantly lower the risk of GIbleeding, without influencing the risk of hospital-acquiredpneumonia, mortality or length of stay, as compared toH2RAs [6, 48]. However, whether PPIs are superior toH2RAs may not be relevant when H2RAs have notunequivocally been shown to be superior to placebo. Thissituation has the imminent risk that even though PPIs maybe superior to H2RAs, they may not be better than pla-cebo or no prophylaxis.

The TSA included in the present review adds impor-tant information to the area of SUP in adult ICU patients,as they highlight the lack of firm evidence for the use ofSUP in this population. To ensure patient safety, well-powered trials with a low risk of bias are urgently needed.

Conclusion

This systematic review using meta-analysis and TSA hasdemonstrated that the quality and quantity of evidence forthe use of SUP in adult ICU patients is low and that thereis no firm evidence for benefit or harm of SUP as com-pared to placebo or no prophylaxis. Consequently, agenuine benefit of SUP in adult ICU patients may bequestioned, and large randomised clinical trials should beconducted to answer the question of whether critically illpatients in the ICU should be treated with SUP or not.

Conflicts of interest None.

References

1. Marik PE, Vasu T, Hirani A,Pachinburavan M (2010) Stress ulcerprophylaxis in the new millennium: asystematic review and meta-analysis.Crit Care Med 38(11):2222–2228. doi:10.1097/CCM.0b013e3181f17adf

2. Cook D, Heyland D, Griffith L, Cook R,Marshall J, Pagliarello J (1999) Riskfactors for clinically important uppergastrointestinal bleeding in patientsrequiring mechanical ventilation.Canadian Critical Care Trials Group.Crit Care Med 27(12):2812–2817

3. Cook DJ, Fuller HD, Guyatt GH,Marshall JC, Leasa D, Hall R, WintonTL, Rutledge F, Todd TJ, Roy P et al(1994) Risk factors for gastrointestinalbleeding in critically ill patients.Canadian Critical Care Trials Group.N Engl J Med 330(6):377–381. doi:10.1056/NEJM199402103300601

4. Reintam Blaser A, Poeze M, MalbrainML, Bjorck M, Oudemans-van StraatenHM, Starkopf J (2013) Gastrointestinalsymptoms during the first week ofintensive care are associated with pooroutcome: a prospective multicentrestudy. Intensive Care Med39(5):899–909. doi:10.1007/s00134-013-2831-1

5. Dellinger RP, Levy MM, Carlet JM,Bion J, Parker MM, Jaeschke R,Reinhart K, Angus DC, Brun-BuissonC, Beale R, Calandra T, Dhainaut JF,Gerlach H, Harvey M, Marini JJ,Marshall J, Ranieri M, Ramsay G,Sevransky J, Thompson BT, TownsendS, Vender JS, Zimmerman JL, VincentJL (2008) Surviving Sepsis Campaign:international guidelines formanagement of severe sepsis and septicshock: 2008. Intensive Care Med34(1):17–60. doi:10.1007/s00134-007-0934-2

6. Alhazzani W, Alenezi F, Jaeschke RZ,Moayyedi P, Cook DJ (2013) Protonpump inhibitors versus histamine 2receptor antagonists for stress ulcerprophylaxis in critically ill patients: asystematic review and meta-analysis.Crit Care Med. doi:10.1097/CCM.0b013e3182758734

7. Rubulotta F, Gullo A, Iscra F (2007)Recommendations for ulcer prophylaxisin the treatment of patients with severesepsis and septic shock: a dog chasingits tail? Intensive Care Med33(4):718–720. doi:10.1007/s00134-007-0551-0

8. Krag M, Perner A, Wetterslev J, MollerMH (2013) Stress ulcer prophylaxis inthe intensive care unit: is it indicated? Atopical systematic review. ActaAnaesthesiol Scand. doi:10.1111/aas.12099

9. Lin PC, Chang CH, Hsu PI, Tseng PL,Huang YB (2010) The efficacy andsafety of proton pump inhibitors vshistamine-2 receptor antagonists forstress ulcer bleeding prophylaxis amongcritical care patients: a meta-analysis.Crit Care Med 38(4):1197–1205. doi:10.1097/CCM.0b013e3181d69ccf

10. Higgins JPT, Green S (2012) Cochranehandbook for systematic reviews ofinterventions, version 5.1.0. CochraneCollaboration, Oxford

11. Moher D, Liberati A, Tetzlaff J, AltmanDG (2009) Preferred reporting items forsystematic reviews and meta-analyses:the PRISMA statement. BMJ 339:b2535

12. Ciani O, Buyse M, Garside R, Pavey T,Stein K, Sterne JA, Taylor RS (2013)Comparison of treatment effect sizesassociated with surrogate and finalpatient relevant outcomes inrandomised controlled trials: meta-epidemiological study. BMJ 346:f457

13. Guyatt GH, Oxman AD, Kunz R,Atkins D, Brozek J, Vist G, Alderson P,Glasziou P, Falck-Ytter Y, SchunemannHJ (2011) GRADE guidelines: 2.Framing the question and deciding onimportant outcomes. J Clin Epidemiol64(4):395–400. doi:10.1016/j.jclinepi.2010.09.012

14. Higgins JP, Thompson SG (2002)Quantifying heterogeneity in a meta-analysis. Stat Med 21(11):1539–1558.doi:10.1002/sim.1186

15. Wetterslev J, Thorlund K, Brok J,Gluud C (2009) Estimating requiredinformation size by quantifyingdiversity in random-effects model meta-analyses. BMC Med Res Methodol9:86. doi:10.1186/1471-2288-9-86

16. Wetterslev J, Thorlund K, Brok J,Gluud C (2008) Trial sequentialanalysis may establish when firmevidence is reached in cumulative meta-analysis. J Clin Epidemiol 61(1):64–75.doi:10.1016/j.jclinepi.2007.03.013

17. Higgins JP, Whitehead A, Simmonds M(2011) Sequential methods for random-effects meta-analysis. Stat Med30(9):903–921. doi:10.1002/sim.4088

18. Sweeting MJ, Sutton AJ, Lambert PC(2004) What to add to nothing? Use andavoidance of continuity corrections inmeta-analysis of sparse data. Stat Med23(9):1351–1375. doi:10.1002/sim.1761

19. Egger M, Davey Smith G, Schneider M,Minder C (1997) Bias in meta-analysisdetected by a simple, graphical test.BMJ 315(7109):629–634

20. Apte NM, Karnad DR, Medhekar TP,Tilve GH, Morye S, Bhave GG (1992)Gastric colonization and pneumonia inintubated critically ill patients receivingstress ulcer prophylaxis: a randomized,controlled trial. Crit Care Med20(5):590–593

21

21. Basso N, Bagarani M, Materia A,Fiorani S, Lunardi P, Speranza V(1981) Cimetidine and antacidprophylaxis of acute uppergastrointestinal bleeding in high riskpatients. Controlled, randomized trial.Am J Surg 141(3):339–3410002-9610(81)90191-4

22. Ben-Menachem T, Fogel R, Patel RV,Touchette M, Zarowitz BJ, HadzijahicN, Divine G, Verter J, Bresalier RS(1994) Prophylaxis for stress-relatedgastric hemorrhage in the medicalintensive care unit. A randomized,controlled, single-blind study. AnnIntern Med 121(8):568–575

23. Burgess P, Larson GM, Davidson P,Brown J, Metz CA (1995) Effect ofranitidine on intragastric pH and stress-related upper gastrointestinal bleedingin patients with severe head injury. DigDis Sci 40(3):645–650

24. Darlong V, Jayalakhsmi TS, Kaul HL,Tandon R (2003) Stress ulcerprophylaxis in patients on ventilator.Trop Gastroenterol 24(3):124–128

25. Friedman CJ, Oblinger MJ, Suratt PM,Bowers J, Goldberg SK, Sperling MH,Blitzer AH (1982) Prophylaxis of uppergastrointestinal hemorrhage in patientsrequiring mechanical ventilation. CritCare Med 10(5):316–319

26. Groll A, Simon JB, Wigle RD, TaguchiK, Todd RJ, Depew WT (1986)Cimetidine prophylaxis forgastrointestinal bleeding in an intensivecare unit. Gut 27(2):135–140

27. Halloran LG, Zfass AM, Gayle WE,Wheeler CB, Miller JD (1980)Prevention of acute gastrointestinalcomplications after severe head injury:a controlled trial of cimetidineprophylaxis. Am J Surg 139(1):44–480002-9610(80)90228-7

28. Hanisch EW, Encke A, Naujoks F,Windolf J (1998) A randomized,double-blind trial for stress ulcerprophylaxis shows no evidence ofincreased pneumonia. Am J Surg176(5):453–457 S0002961098002396

29. Kantorova I, Svoboda P, Scheer P,Doubek J, Rehorkova D, Bosakova H,Ochmann J (2004) Stress ulcerprophylaxis in critically ill patients: arandomized controlled trial.Hepatogastroenterology51(57):757–761

30. Karlstadt RG, Iberti TJ, Silverstein J,Lindenberg L, Rright-Asare P,Rockhold F, Young MD (1990)Comparison of cimitidine and placebofor the prophylaxis of uppergastrointestinal bleeding due to stress-related gastric mucosal damage in theintensive care unit. J Intensive CareMed 5:26–32

31. Macdougall BR, Bailey RJ, Williams R(1977) H2-receptor antagonists andantacids in the prevention of acutegastrointestinal haemorrhage infulminant hepatic failure. Twocontrolled trials. Lancet1(8012):617–619 S0140-6736(77)92055-4

32. Martin LF, Booth FV, Karlstadt RG,Silverstein JH, Jacobs DM, Hampsey J,Bowman SC, D’Ambrosio CA,Rockhold FW (1993) Continuousintravenous cimetidine decreases stress-related upper gastrointestinalhemorrhage without promotingpneumonia. Crit Care Med 21(1):19–30

33. Metz CA, Livingston DH, Smith JS,Larson GM, Wilson TH (1993) Impactof multiple risk factors and ranitidineprophylaxis on the development ofstress-related upper gastrointestinalbleeding: a prospective, multicenter,double-blind, randomized trial. TheRanitidine Head Injury Study Group.Crit Care Med 21(12):1844–1849

34. Peura DA, Johnson LF (1985)Cimetidine for prevention and treatmentof gastroduodenal mucosal lesions inpatients in an intensive care unit. AnnIntern Med 103(2):173–177

35. Powell H, Morgan M, Li SK (1993)Inhibition of gastric acid secretion inthe intensive care unit after coronaryartery bypass graft. A pilot controlstudy of intravenous omeprazole bybolus and infusion, ranitidine andplacebo. Theor Surg 8:125–130

36. Reusser P, Gyr K, Scheidegger D,Buchmann B, Buser M, Zimmerli W(1990) Prospective endoscopic study ofstress erosions and ulcers in critically illneurosurgical patients: currentincidence and effect of acid-reducingprophylaxis. Crit Care Med18(3):270–274

37. Ruiz-Santana S, Ortiz E, Gonzalez B,Bolanos J, Ruiz-Santana AJ, ManzanoJL (1991) Stress-inducedgastroduodenal lesions and totalparenteral nutrition in critically illpatients: frequency, complications, andthe value of prophylactic treatment. Aprospective, randomized study. CritCare Med 19(7):887–891

38. van den Berg B, van Blankenstein M(1985) Prevention of stress-inducedupper gastrointestinal bleeding bycimetidine in patients on assistedventilation. Digestion 31(1):1–8

39. Zinner MJ, Zuidema GD, Smith P,Mignosa M (1981) The prevention ofupper gastrointestinal tract bleeding inpatients in an intensive care unit. SurgGynecol Obstet 153(2):214–220

40. Calvet X, Baigorri F, Duarte M, SauraP, Royo C, Joseph D, Mas A, Artigas A(1998) Effect of ranitidine on gastricintramucosal pH in critically illpatients. Intensive Care Med24(1):12–17

41. Gursoy O, Memis D, Sut N (2008)Effect of proton pump inhibitors ongastric juice volume, gastric pH andgastric intramucosal pH in critically illpatients: a randomized, double-blind,placebo-controlled study. Clin DrugInvestig 28(12):777–782 28125

42. Chan KH, Lai EC, Tuen H, Ngan JH,Mok F, Fan YW, Fung CF, Yu WC(1995) Prospective double-blindplacebo-controlled randomized trial onthe use of ranitidine in preventingpostoperative gastroduodenalcomplications in high-riskneurosurgical patients. J Neurosurg82(3):413–417. doi:10.3171/jns.1995.82.3.0413

43. Cheadle WG, Vitale GC, Mackie CR,Cuschieri A (1985) Prophylacticpostoperative nasogastricdecompression. A prospective study ofits requirement and the influence ofcimetidine in 200 patients. Ann Surg202(3):361–366

44. Zhang Z, Xu X, Ni H (2013) Smallstudies may overestimate the effectsizes in critical care meta-analyses: ameta-epidemiological study. Crit Care17(1):R2. doi:10.1186/cc11919

45. Thorlund K, Imberger G, Walsh M, ChuR, Gluud C, Wetterslev J, Guyatt G,Devereaux PJ, Thabane L (2011) Thenumber of patients and events requiredto limit the risk of overestimation ofintervention effects in meta-analysis–asimulation study. PLoS One6(10):e25491. doi:10.1371/journal.pone.0025491

46. Guyatt GH, Oxman AD, Vist GE, KunzR, Falck-Ytter Y, Alonso-Coello P,Schunemann HJ (2008) GRADE: anemerging consensus on rating quality ofevidence and strength ofrecommendations. BMJ336(7650):924–926. doi:10.1136/bmj.39489.470347.AD

47. Cook DJ, Reeve BK, Guyatt GH,Heyland DK, Griffith LE, BuckinghamL, Tryba M (1996) Stress ulcerprophylaxis in critically ill patients.Resolving discordant meta-analyses.JAMA 275(4):308–314

48. Barkun AN, Bardou M, Pham CQ,Martel M (2012) Proton pumpinhibitors vs. histamine 2 receptorantagonists for stress-related mucosalbleeding prophylaxis in critically illpatients: a meta-analysis. Am JGastroenterol 107(4):507–520. doi:10.1038/ajg.2011.474

22