Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=infd20

Download by: [Tufts University] Date: 27 October 2017, At: 11:55

Infectious Diseases

ISSN: 2374-4235 (Print) 2374-4243 (Online) Journal homepage: http://www.tandfonline.com/loi/infd20

Pitfalls in studies of eosinopenia and neutrophil-to-lymphocyte count ratio

Stamatis Karakonstantis, Dimitra Kalemaki, Emmanouil Tzagkarakis &Charalampos Lydakis

To cite this article: Stamatis Karakonstantis, Dimitra Kalemaki, Emmanouil Tzagkarakis &Charalampos Lydakis (2017): Pitfalls in studies of eosinopenia and neutrophil-to-lymphocyte countratio, Infectious Diseases, DOI: 10.1080/23744235.2017.1388537

To link to this article: http://dx.doi.org/10.1080/23744235.2017.1388537

Published online: 26 Oct 2017.

Submit your article to this journal

View related articles

View Crossmark data

INFECTIOUS DISEASES,2017; VOL. 0,NO. 0, 1–12

https://doi.org/10.1080/23744235.2017.1388537

REVIEW ARTICLE

Pitfalls in studies of eosinopenia andneutrophil-to-lymphocyte count ratio

Stamatis Karakonstantisa , Dimitra Kalemakib, Emmanouil Tzagkarakisc and Charalampos Lydakisd

aResident of Internal Medicine, Second Department of Internal Medicine, General Hospital of Heraklion ‘Venizeleio-Pananeio’,Heraklion, Greece; bResident of General Medicine, University Hospital of Heraklion, Voutes, Heraklion, Greece; cConsultant inInternal Medicine. Second Department of Internal Medicine, General Hospital of Heraklion ‘Venizeleio-Pananeio’, Heraklion,Greece; dHead of the Second Department of Internal Medicine, Second Department of Internal Medicine, General Hospital ofHeraklion ‘Venizeleio-Pananeio’, Heraklion, Greece

ABSTRACTThere is a number of publications evaluating the eosinophil count and the neutrophil-to-lymphocyte count ratio for diagno-sis, prognosis or monitoring of patients. Of special interest is the use of these parameters for discrimination betweenthe different causes of fever (e.g. bacterial versus viral vs. non-infectious causes of fever) and for monitoring the efficacy oftherapy and predict the course of the patient. However, pitfalls in previous study designs prevent applicability to clinicalpractice. Here, we provide a short review of the relevant literature and summarize important factors that should be takeninto account when designing studies, with special attention to the selection of a proper and clinically meaningful studypopulation and the effects of the stress response and of corticosteroids.

KEYWORDSEosinopeniaEosinophil countNeutrophil-to-lymphocyte count ratioPitfalls

ARTICLE HISTORYReceived 22 August 2017Revised 25 September 2017Accepted 30 September 2017

CONTACTStamatis Karakonstantis

stamkar2003@gmail.comResident of Internal Medicine, 2nd

Department of Internal Medicine, General Hospitalof Heraklion ‘Venizeleio-Pananeio’, LeoforosKnossou, Heraklion, Greece, Postal code 71409

� 2017 Society for Scandinavian Journal of Infectious Diseases

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

Introduction

Eosinopenia, neutrophilia and lymphopenia are com-monly observed in patients with infections. Eosinopeniaof acute infection is thought to be the result of seques-tration of eosinophils at the site of infection [1,2],although other stress responses may also result in eosi-nopenia, mediated by adrenal corticosteroids and epi-nephrine [3–5]. Neutrophilia during acute infection is theresult of mobilization of neutrophils from the bone mar-row mediated by chemokines [6], while lymphopeniaresults from re-distribution of lymphocytes within thelymphatic system and apoptosis [7].

There has been some interest in using the eosinophilcount (EC) and the neutrophil-to-lymphocyte-count ratio(NLCR) for diagnostic, prognostic or monitoring purposesin patients with infections. More specifically EC and/orNLCR have been studied for the following purposes: (1)for the differentiation of bacterial from viral infection[8,9], (2) for the differentiation of infectious from non-infectious causes of fever [8–19], (3) as prognosticmarkers (e.g. [14,20–26]), (4) as an early predictor ofresponse to antimicrobial therapy [27], (5) as predictorsof bloodstream infections [28–32] and (6) for the diagno-sis of particular micro-organisms [33–36]. However, mostof these studies have significant limitations. Here, wepresent a short review of the literature and then focuson the pitfalls of previous study designs, in order to aidwith the design of future studies of EC and NLCR.

Short review of the literature

Table 1 summarizes the studies that have evaluatedeosinopenia and NLCR as a diagnostic marker and as apredictor of bacteraemia.

Gil et al [11], in a study conducted in an internalmedicine department, found that an EC of less than 40/ll was a very specific marker in differentiating infectiousfrom non-infectious causes of elevated C-reactive protein(CRP). Later, Efstathiou et al [17] found that an EC <40/ll(in combination with CRP >6mg/dl and ferritin< 500 lg/l) was a good independent marker of infectionin patients with fever of unknown origin. Subsequently,Abidi et al [10] demonstrated that eosinopenia (<50/ll)was a very sensitive and specific marker for identifyingsepsis in patients admitted to the ICU, although otherstudies in the ICU setting failed to reproduce theseresults [12–15]. More recently, profound eosinopenia(<10/ul) was found to be a very specific marker of sepsisin patients presenting to the Emergency Department[16]. As a predictor of bacteraemia eosinopenia has both

low sensitivity and low specificity [28–30]. Of note is thevariable EC cut-off value used in different studies, whichcomplicates the interpretation of studies of EC. Thislikely reflects the variability in the populations studied,as discussed later.

Similar to eosinopenia, the NLCR has been studied asa marker of sepsis, and as a predictor of bacteraemia.For example, Ljungstr€om et al. [19] found that inpatients with suspected sepsis a high NLCR may be auseful biomarker of bacterial sepsis, while a low NRCL(<3) may be useful to rule out bacterial sepsis, althoughthe over diagnostic value of NLCR was low (area underthe curve =0.63). Naess et al. [8] showed that patientswith bacterial infections have a significantly higher NLCRcompared to patients with viral infection or non-infec-tious causes of fever. As a marker of bacteraemia theNLCR has a moderate sensitivity and specificity [31,32].

Eosinopenia and NLCR for prognostic purposes havebeen studied with conflicting results [14,20–26]. Forexample, Abidi et al. [21] found that EC <40/ul at admis-sion to the ICU and during the first 7 days was inde-pendently associated with higher risk of mortality, whileYip et al found that eosinopenia at discharge from theICU was associated with unexpected re-admissions [37].Others have failed to confirm eosinopenia as a predictorof mortality [26], and Merino et al found that the clinicalvalue of eosinopenia as a predictor of mortality is low[20]. Of more interest, perhaps, is persistence of eosino-penia (and high NLCR) in non-survivors compared torapid resolution of eosinopenia (and reduction of theNLCR) in survivors [20–22,26], suggesting that EC andNLCR may be useful monitoring parameters. Forexample, Davido et al. [27] recently reported that initi-ation of correct antimicrobial therapy results in promptresolution of eosinopenia, while eosinopenia persisted inpatients with infections caused by bacteria resistant tothe initial antimicrobial therapy. Despite some limitationsof this study, in our opinion, this is the most promisingclinical application of eosinopenia and warrants confirm-ation in a larger prospective study.

The effect of the stress response and ofcorticosteroids

It is well established that corticosteroids can reduce eosi-nophils, raise neutrophils and reduce lymphocytes[38,39]. Therefore, it is an important confounding factorand receipt of corticosteroids before presentation or inthe emergency department or during the hospital stayshould be taken into account when assessing EC

2 S. KARAKONSTANTIS ET AL.

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

Table1.

Summaryof

previous

stud

ieson

eosino

peniaandNLCRas

markers

ofinfectionandbacteraemia.

Stud

ies

Setting-design

Inclusion/exclusioncriteria

Stud

ypo

pulatio

n(num

berof

patients)

Results

Gilet

al.2

003[11]

Internal

MedicineDepartm

ent

– Prospective

– n¼138

Inclusioncriteria:C

RP>

2mg/dl

attwosuccessive

measurements.

Exclusioncriteria:

Corticosteroidsat

admission

,Cushing

synd

rome,

parasitic

disease,

eosino

philicvasculitis,adrenal

insufficiency,lym

phom

a

Infected

(83),

Non

-infected

(55):

Polymyalgiarheumatic/giant

cell

arteritis(21),cancer(10),systemic

vasculitis(5),veno

usthrombo

-em

bolism

(4)sarcoido

sis(3),anky-

losing

spon

dylitis(3),Croh

n’s(2),

fecolith(2),rheumatoidarthritis

(1),thyroiditis

(1),remittingsero-

negativesymmetrical

syno

vitis

with

pittingoedema(1),chon

dro-

calcinosis(1),Sharp(1)

Foreosino

peniaat

acut-off<4

0/ll:

Sensitivity

56%,specificity

92%,

PPV92%,N

PV57%,LRþ

=7

ForCR

Pat

acut-off>1

0mg/dl:

Sensitivity

75%,specificity

72%,P

PV81%,N

PV64%

Efstathiou

etal.2

010[17]

Internal

MedicineDepartm

ent

– Prospective

– n¼213

Inclusioncriteria:

Feverof

unknow

norigin

according

tothecriteria

ofDurackand

Street.

Exclusioncriteria:

Immun

osup

pression

,neutrop

enia,

nosocomialF

UO,m

issing

data,

insufficientbasicwork-up

.

Infections

(69)

Endo

carditis(21),tub

erculosis(17),

brucellosis(7),osteom

yelitis(2),

prostheticjointinfection(1),

leptospirosis(2),pelvicinflamma-

tory

disease(1),intra-abdo

minal

abscess(1),psoasabscess(1),Q

fever(2),Whipp

le’sdisease(1),

Lyme’sdisease(2),cytomegalo-

virus(3),leishm

aniasis(7),sub-

diaphragmaticam

oebadic

abscess(1)

Non

-infections

(144)

Lymph

omas

(14),solid

tumou

rs(8),

system

iclupu

serythematosus

(18),

Still’sdisease(16),vasculiticsyn-

drom

es(28),sarcoidosis(5),

Croh

n’sdisease(1),familial

medi-

terraneanfever(1),TRAP

S(1),

subacute

thyroiditis

(5),recurrent

pulmon

aryem

bolism

(2)

Castleman’sdisease(1),Kikuchi’s

necrotizinglymph

adenitis(1),no

diagno

sis(39)

Eosino

penia(<

40/ll),

CRP>6mg/dl

andferritin<500lg/lw

ereinde-

pend

entmarkers

ofinfectionin

themultivariate

analysis.For

the

diagno

sisof

infection,

thecom-

binedpresence

oftwoor

moreof

thesemarkers

hadasensitivity

of91.4%,specificity

of92.3%,P

PVof

86.5%,N

PV95.2%,p

ositive

likeli-

hood

ratio

11.9

andnegativelike-

lihoodratio

0.09.

Abidie

tal.2

008[10]

Medical

ICU

– Prospective

– n¼177

Inclusioncriteria:C

onsecutive

patientsadmitted

toamedical

ICU

Exclusioncriteria:

Patientswho

died

(n¼12)or

were

discharged

with

in24

hafter

admission

(n¼9)

Infected

(120):

Sepsis(41),severesepsis(55),septic

shock(24)

Non

-infected

1SIRS

(20):

COPD

exacerbatio

n(6),asthma(4),

diabeticketoacidosis(4),acute

poison

ing(3),cardiogenicshock

(2),gastrointestinal

bleeding

(1)

Non

-infected

1no

tSIRS

(37):

Acutepo

ison

ing(30),scorpion

enveno

mation(3),acuteischaemic

stroke

(2),hypercalcaem

ia(2)

Forinfectionversus

non-infection

(ECcut-off<5

0/ll):sensitivity

80%

(95%

CI,7

1–86),specificity

91%

(95%

CI,7

9–96),LRþ

9.12

(95%

CI,3

.9–21),LR-

0.21

(95%

CI,

0.15–0.31)

ForinfectionSIRS

versus

non-infec-

tion

SIRS

(ECcut-off<4

0/ll):

Sensitivity

80%

(95%

CI,7

1–86),

specificity

80%

(95%

CI,5

5–93),

LRþ

4(95%

CI,1

.65–9.65),LR-

0.25

(95%

CI,0

.17–

0.36)

Eosino

peniaversus

CRP(cut-off

>7mg/dl):

(continued)

INFECTIOUS DISEASES 3

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

Table1.

Continued

Stud

ies

Setting-design

Inclusion/exclusioncriteria

Stud

ypo

pulatio

n(num

berof

patients)

Results

Forinfectionversus

non-infection:

ROCAU

C0.89

(95%

CI,0

.8–0.94)

versus

0.77

(95%

CI,0

.70–0.84)

p<.001

ForinfectionSIRS

vsno

n-infection

SIRS:R

OCAU

C0.84

(95%

CI,

0.74–0.94)

foreosino

phils

versus

0.77

(95%

CI,0

.67–0.87)

p¼.175

Smith

sonet

al.2

009[12]

MedicalICU

– Retrospective

– n¼191

Inclusioncriteria:p

atientsadmitted

tothemedicalICU

Exclusioncriteria:H

IV,h

aematolog

icmalignancies

Infectious

SIRS

(142)

Non

-infectious

SIRS

(49)

Differencesin

eosino

phils

coun

tbetweenthegrou

pswereno

tstatistical

sign

ificant.Eosinop

enia

<40/llw

aspresentin

49.3%

oftheinfected

grou

pandin

36.7%

oftheno

n-infected

grou

pShaabanet

al.2

010[13]

MedicalICU

– Prospective

– n¼68

Inclusioncriteria:C

onsecutive

patientsadmitted

toamedical

ICU

Exclusioncriteria:

Patientswho

died

orweredis-

chargedwith

in24

hafter

admission

Infectiongrou

p(31):

Sepsis(15),severesepsis(11),septic

shock(5)

Non

-infectiongrou

p(37):

Cardiaccond

ition

s(4),tracheal

sten-

osis(1),massive

non-parapn

eu-

mon

icpleurale

ffusion

(2),

hypercapnicrespiratory

failure

(3),

COPD

exacerbatio

n(2),PE

(1),

severe

electrolyteimbalance-

alteredmentalstatus(4),anaemia

(3),drug

intoxicatio

n(3),GIb

leed-

ing(4),DKA

(2),acutecerebrovas-

cularaccidents(3),status

epilepticus

(3)

Atacut-offof

<50/lle

osinop

enia

yieldedasensitivity

of81%,speci-

ficity

65%,P

PV66%,N

PV80%.

AUCof

ROC0.72

(95%

CI0.6–0.83)

Atacut-offof

7mg/dl

CRPyielded

asensitivity

of94%,specificity

84%,P

PV83%,N

PV94%,A

UCof

ROC0.92

(95%

CI0.83–0.97)

Procalcitoninat

acut-offof

1.5ng

/mlyieldedasensitivity

of84%,

specificity

92%,P

PV90%,N

PV80%,A

UCof

ROC0.89

(95%

CI0.79–0.95

Garnacho-Mon

tero

Jet

al.2

014[14]

Medico-surgical

ICU

– Prospective

– n¼163

Inclusioncriteria:

Alla

dultpatientsadmitted

toamed-

ico-surgical

ICUandfulfilling

the

SIRS

criteria.

Exclusioncriteria:

Activemalignancy,acutemyocardial

infarctio

nin

thelast

mon

th.

Infectious

SIRS

(117):

Severe

sepsis(44),septic

shock(73)

Non

-infectious

SIRS

(43):

acuterespiratory

failure

(13),early

post-operativecourse

ofabdo

m-

inalsurgery(8),acutepancreatitis

(8),multip

letrauma(2),febrile

synd

romein

conn

ectivetissuedis-

ease

(2)

Eosino

phils

<25/ll:sensitivity

65%,

specificity

29%,P

PV62%,N

PV23%,A

UCRO

C0.54

CRP>1

0.38

mg/dl:sensitivity

91%,

specificity

41%,P

PV81%,N

PV61%

Procalcitonin>1

.39ng

/ml:sensitivity

91,specificity

67%,P

PV88%,N

PV74%

Anandet

al.2

016[15]

MedicalICU

– Prospective

– n¼170

Inclusioncriteria:

Alla

dultpatientsadmitted

toamed-

icalICUwith

SIRS.

Exclusioncriteria:

Patientstransferredfrom

otherICUs,

post-operative,immun

ocom

prom

-ised,p

regn

ant,activemalignancy,

thosewho

died

orweredis-

chargedin

thenext

24h,

patients

with

bilateralp

neum

onia

(sus-

pected

viralinfectio

n)andpatients

Infectious

SIRS

(125):

Non

-infectious

SIRS

(45):

Acutepancreatitis(25),traum

a(20)

Nosign

ificant

diffe

rencein

the

eosino

philcoun

tbetweenthetwo

grou

ps.

Eosino

peniaat

acut-offof

<50/ll:

sensitivity

23%,specificity

69%,

PPV67%,N

PV24%,A

UCRO

C0.454

Procalcitoninat

cut-offof

>3.13ng

/ml:sensitivity

82.4%,specificity

82.2%,P

PV93%,N

PV63%,A

UC

ROC0.907

(continued)

4 S. KARAKONSTANTIS ET AL.

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

Table1.

Continued

Stud

ies

Setting-design

Inclusion/exclusioncriteria

Stud

ypo

pulatio

n(num

berof

patients)

Results

with

trop

icaldiseases

(e.g.m

alaria,

deng

ue,leptospira,rickettesiae)

LavoignetCE

etal.2

016[16]

EmergencyDepartm

ent

– Retrospective

– n¼692

Inclusioncriteria:A

llpatientspre-

sentingto

theem

ergencydepart-

ment.

Exclusioncriteria:immun

odeficiency,

immun

ossupressive

therapy,

patientson

corticosteroids,

chem

otherapy,activemalignancy,

haem

atolog

icmalignancies,

ongo

ingantib

iotic

therapyat

presentatio

n

Sepsisgrou

p(125)

Sepsis(108),severe

sepsis(16),septic

shock(1).

Nosepsisgrou

p(567)

Other

patientsvisitin

gtheED

Eosino

peniaat

acut-offof

<10/ll

yieldedaspecificity

of91%

for

sepsis.Sensitivity

46.4%,P

PV53%,

NPV

88.5%

Meaneosino

philcoun

ts:

Non

-infected:1

30/ul

Infected

with

outSIRS:1

48/ul

Sepsistotal:54/ul

Severe

sepsis:2

7/ul

Gucyetm

ezet

al.2

016[18]

Medico-surgical

ICU

– Retrospective

– n¼1257

Inclusioncriteria:

Adultpatientsadmitted

toamedico-

surgical

ICUandfulfilling

SIRS

cri-

teria.

Exclusioncriteria:

Age<18

yearsold(334),re-adm

itted

(53),d

iagn

osed

with

haem

ato-

logicald

isease,o

ncorticosteroid

andimmun

osup

pressive

therapy

(n¼345),SIRSnegative(318),cul-

ture

negativesepsis(45),u

ndocu-

mentedlabo

ratory

values

and

outcom

es(301)

Sepsisgrou

p(441)

Non

-sep

sisSIRS

(816)

Electivesurgery(540),em

ergency

surgery(36),m

edical

diseases

(240)

Eosino

philcoun

tdidno

tdiffe

rsig-

nificantly

betweenthetwo

grou

ps.

Ljun

gstr€ om

etal.2

017[19]

Emergencydepartment

– Prospective

– n=1572

Inclusioncriteria:

Consecutiveadultpatientsadmitted

totheED

forsuspectedcommu-

nity-onset

sepsis.

Exclusioncriteria:

Noexclusioncriteria

Bacterialinfections

(n=874)

Bacterialsepsis(Sep

sis-2criteria)

(n=667)

Severe

sepsis/sep

ticshock(Sep

sis-2

criteria)(n=169)

Sepsisaccordingto

Sepsis-3

criteria

(n=560)

Others(n=698)

Non

-verified

bacterialinfectio

nor

aviralinfectio

nor

ano

n-infectious

disease

AUCof

NLCRfordiagno

sisbacter-

aemia:0

.71;

95%

CI0.67–0.75.

AUCof

NLCRfordiscrim

inatingbac-

terialinfectio

nsfrom

‘others’:0

.63;

95%

CI0.61–0.66

AUCof

NLCRfordiagno

sing

sepsis

(according

toSepsis3):0

.67;

95%

CI0.64–0.69

Procalcitoninwas

notsign

ificantly

better

comparedto

NLCR.

Naess

etal.2

013[9]and2017

[8]

Internal

MedicineDepartm

ent

– Retrospective

– n¼299

Inclusioncriteria:

Retrospectiveselectionof

patients

admitted

totheho

spitalw

itha

diagno

sisof

‘fever’with

outany

otherdiagno

sissugg

estin

gthe

causeof

fever.

Exclusioncriteria:

Patientswith

leukaemia.P

atients

with

adiagno

sisat

admission

(e.g.

‘fever,pn

eumon

ia’).

Infectionconfirmed

withmicrobiol-

ogy,

serology

orradiolog

y(n=185)

Clinically

diag

nosedinfection

(n=73)

Bacterial(n=150)

Viralinfection(n=14)

Infectious

mon

onucleosis(n¼9)

NLCRwas

sign

ificantlyhigh

erin

patientswith

bacterialinfectio

n(com

paredto

viralinfectio

nsor

noinfection).Thiswas

morepro-

noun

cedin

patientswith

feverof

less

than

oneweek’sdu

ratio

nNLCRwas

sign

ificantlyhigh

erin

patientswith

‘septicaemia’com-

paredto

patientswith

otherbac-

terialinfectio

ns.

NLCRwas

anindepend

entpredictor

(continued)

INFECTIOUS DISEASES 5

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

Table1.

Continued

Stud

ies

Setting-design

Inclusion/exclusioncriteria

Stud

ypo

pulatio

n(num

berof

patients)

Results

Autoimmun

ediseases

(n=8)

Maligna

ncy(n=5)

Miscellane

ous(n=16)

Nodiag

nosis(n=12)

ofthediagno

sticcatego

ry(bacter-

ialinfectio

nversus

viralinfectio

nvs.clinicallydiagno

sedinfection)

inpatientswith

feverdu

ratio

n<7

days.

deJageret

al.2

010[32]

EmergencyDepartm

ent

– Retrospective

– n¼184

Inclusioncriteria:

Consecutiveadultpatientsadmitted

totheED

with

suspectedcommu-

nity-acquiredbacteraemia.

Exclusioncriteria:

Haematolog

ical

disease,patients

receivingchem

otherapy

orglucocorticoids

Withba

cteraemia

(n=92)

Witho

utba

cteraemia

(n=92)

Ninety-twoage-

andgend

er-m

atched

patientswith

suspectedbacter-

aemia.Eighty-fivehadan

infec-

tious

diagno

sisafterclinical

and

microbiolog

ical

assessment.

NLCRwas

abetter

predictorof

bac-

teraem

iacomparedto

CRP,

WBC

and

neutroph

ilcoun

t.At

acut-off>10:

Sensitivity

77.2%,specificity

63%,

PPV67.6%,N

PV73.4%,A

UC0.73

(95%

CI,0

.66–0.81).

Hoet

al.2

009[28]

ICU

– Retrospective

– n¼66

Inclusioncriteria:

Adultpatientsadmitted

toan

ICU

with

bacteraemia.Twocontrols,

thepatient

who

was

admitted

totheICUimmediatelybefore

and

afterthe‘case’,w

ereselected

con-

currently

foreach

bloodstream

infection‘case’.W

hether

controls

hadbloodcultu

restakenisno

tdescrib

ed.

Exclusioncriteria:

Positivebloodcultu

redu

eto

con-

tamination.

Leukop

enia

dueto

haem

atolog

icdiseaseor

chem

otherapy.

Patien

tswithba

cterem

ia(n=22)

Infected

pancreatitis(n¼1),n

ecrotiz-

ingfasciitis(n¼2),p

neum

onia

(n¼7),catheter-relatedsepsis

(n¼2),w

ound

infection(n¼2),

infectiveendo

carditis(n¼1),p

eri-

tonitis

(n¼1),cellulitis(n¼1),

urinarytractinfection(n¼5)

Controls(n=44)

Pneumon

ia(n¼4),m

ultip

letrauma

(n¼4),cardiog

enicshock(n¼1),

drug

overdo

se(n¼5),valve

replacem

ent(n¼3),sub

arachn

oid

haem

orrhage(n¼4),m

eningitis

(n¼1),coron

aryartery

bypass

(n¼4),g

astrointestin

alob

struc-

tion(n¼1),ischaem

iclegam

pu-

tatio

n(n¼1),intracranial

haem

atom

a(n¼2),asthm

a(n¼1),o

ralabscess

(n¼3),h

yper-

osmolar

non-ketotic

coma(n¼1),

spinal

injuries(n¼1),cardiac

arrest

(n¼1),ischaem

icbo

wel

disease(n¼1),encephalitis

(n¼1),acute

pulmon

aryoedema

(n¼1),intra-abd

ominal

sepsis

(n¼2),b

urns

(n¼2)

Althou

ghEC

was

sign

ificantlylower

inpatientswith

bacteraemia,and

eosino

peniawas

notan

independ

-entpredictorof

bacteraemia.

Setterberg

etal.2

004[29]

Internal

medicine

– Retrospective

– n¼17

Inclusioncriteria:

Retrospectiveselectionof

patients

with

bloodcultu

respo

sitivefor

Gram-negativebacilli,

Staphylococcus

aureus,and

Streptococcusspecies.Ag

e-and

sex-matched

controlp

atientswith

negativebloodcultu

res.

Exclusioncriteria:

Immun

osup

pressedpatients,

Patien

tswithba

cterem

ia(n=171)

Controls(n=174)

‘Noutility

associated

with

eosino

pe-

nia(defined

asarelativeeosino

-ph

illevelo

f0%

)forpredictin

gbacteraemia.’‘The

absenceof

per-

ipheralb

lood

eosino

phils

cann

otbe

used

asaclinicallyreliable

markerof

bacteraemicinfection.’

(continued)

6 S. KARAKONSTANTIS ET AL.

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

Table1.

Continued

Stud

ies

Setting-design

Inclusion/exclusioncriteria

Stud

ypo

pulatio

n(num

berof

patients)

Results

neutropenia

Wibrow

etal.2

011[30]

Teaching

hospital

– Retrospective

– n¼531

Inclusioncriteria:

Consecutivepatientswith

apo

sitive

bloodcultu

re(‘cases’)andconcur-

rent

rand

omlyselected

‘con

trols’

(negativebloodcultu

re)with

sus-

picion

ofbloodstream

infection.

Exclusioncriteria:

coagulase-negativestaphylococcus

species,antib

iotic

usethepast

week,haem

atolog

icdisorder,

immun

odeficiency,chem

otherapy,

immun

osup

pressive

therapy(e.g.

corticosteroids,metho

trexate,

cyclospo

rin)

Adu

ltpa

tien

tswithbloo

dstream

infection(n=157)

Paed

iatricpa

tien

tswithbloo

d-stream

infection(n=85)

Adu

ltcontrols(n=195)

Paed

iatriccontrols(n=94)

EC<10/lla

sapredictorof

blood-

stream

infectionin

adultpatients:

sensitivity

47%,specificity

79%,

AUC0.651(95%

CI0.712–0.589).

ECwas

notsign

ificantlydiffe

rent

betweencasesandcontrolsin

paediatricpatients)

Lowsbyet

al.2

015[31]

EmergencyDepartm

ent

– Retrospective

– n¼1954

Inclusioncriteria

Adultpatientswith

pyrexial

illness

(tym

panictemperature

>37.9

� C,

ormeetin

gcriteria

forsepsis,that

is,SIRSdu

eto

suspectedinfec-

tion).P

atientswereselected

ifthey

hadbloodcultu

resdraw

n.Exclusioncriteria:

Haematolog

icalmalignancy,chem

o-therapytreatm

ent,patientson

corticosteroid

therapy.False-po

si-

tivebloodcultu

res,attributed

toskin

contam

ination,

wereexclud

edfrom

thefin

alanalysis.

True

-positivebloo

dcultures

(n=270)

Forpredictin

gbacteraemiatheNLCR

atacut-off>10

hadasensitivity

of70%,specificity

57%,P

PV20%,

NPV

92%.

‘Alth

ough

NLCRou

tperform

sconven-

tional

markers

ofinfection,

itisinsufficient

initselfto

guideclinical

manage-

mentof

patientswith

suspected

bloodstream

infectionandit

offers

noadvantageover

lymph

o-cyte

coun

t’

AUC:

area

underthecurve;

CI:confidence

interval;LRþ:

positivelikelihoodratio

n;LR�:

negativelikelihoodratio

;NPV

:negativepredictivevalue;

PPV:

positivepredictivevalue;

ROC:

receiver

operatingcharacteristic

curve.

INFECTIOUS DISEASES 7

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

and NLCR. Some interesting studies of EC and NLCRhave failed to take this into consideration. For example,Davido et al. [27], who described the value of monitor-ing eosinophil counts to assess the efficacy of antimicro-bial therapy, only excluded patients receivingcorticosteroids at a dose >60mg/day and did not reportwhether patients had been given corticosteroids at pres-entation to the emergency department or during hos-pital stay. Similarly, Naess et al. [8] who described NLCRas a useful marker to discriminate between differentcauses of fever, did not comment on whether includedpatients were taking corticosteroids, although a clarifica-tion was later published in response to our correspond-ence [40,41]. Other studies evaluating eosinopenia as adiagnostic marker have also failed to take this factorinto account [10,12–15,17,36] and the same applies tomany of the prognostic studies [14,24–26], and some ofthe studies of EC and NLCR as predictors of bacteraemia[28,29]. Some studies do comment on the use of cortico-steroids in included patients but did not exclude suchpatients from analysis [20–22].

Other conditions that may elicit a systemic inflamma-tory response or any acute stress may have a similareffect on EC and NLCR as in patients with infection. Forexample, we have found eosinopenia (<50/ll) in a sig-nificant percentage of patients with acute pancreatitis

and acute upper gastrointestinal bleeding (23 of 39patients and 29 of 55 patients, respectively – unpub-lished retrospective data). Others have described eosino-penia or high NLCR in other acute conditions, forexample, acute myocardial infarction (e.g. [42,43]), stroke(e.g. [44–46]), acute trauma (e.g. [47,48]) and even dur-ing emotional stress [5]. Therefore, as previously sug-gested [12–14,18], eosinopenia cannot be considered asa specific marker of sepsis, and any concurrent acutestress may affect EC or NLCR and should be taken intoaccount if possible.

Pitfalls in using EC and NLRC to differentiateinfections from non-infectious diseases

For this purpose of differentiating infectious from non-infectious illness, an ideal study sample would bepatients with an unclear diagnosis at presentation,including both infectious and non-infectious diseases inthe initial differential diagnosis (Figure 1). Includingpatients with an obvious ‘non-infection’ diagnosis mayresult in an inaccurate estimation of the diagnostic spe-cificity of EC or NLCR, while including patients with anobvious infectious disease would most likely result in anoverestimation of the diagnostic sensitivity (Figure 1).Results from a study, which includes patients with obvi-ous infectious or non-infectious diagnoses, would not be

Figure 1. The ideal study population for studying the diagnostic accuracy of eosinophil count (EC) or neutrophil-to-lymphocyte count ratio(NLCR) in differentiating infection from non-infection.

8 S. KARAKONSTANTIS ET AL.

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

clinically useful. For example, a physician would notneed to check the EC or the NLCR to confirm pneumo-nia in a patient with fever, productive cough and a newlobar infiltrate on chest x-ray. The same applies to apatient with an obvious, typical exacerbation of a knownautoimmune or rheumatologic disease or a patientadmitted due to drug overdose.

Reviewing the relevant publications [10–16] (Table1), the study populations were far from ideal. Forexample, Abidi et al. [10], who reported that eosinope-nia is a reliable marker of sepsis in patients admittedto the intensive care unit, included several patientswith obvious non-infectious diagnoses and without areason to suspect an infection (e.g. 33 of 57 patientsin the non-infection group were admitted due toacute poisoning, two patients had hypercalcaemia andthree patients were admitted due to scorpionenvenomation). Similarly, the inclusion criterion for thestudy by Gil et al. [11] was a CRP above 2mg/dl,while the non-sepsis group in the study by Lavoignetet al. [16] included all patients presenting to theemergency department irrespective of the reason.Therefore, none of these three studies that favouredthe use of eosinopenia as a diagnostic markerincluded a clinically meaningful study population. Onthe other hand, studies that did not support eosinope-nia as a diagnostic marker of sepsis [12–14,18] alsofailed to include an appropriate study population. Forexample, including in the non-sepsis group acutetrauma patients, post-surgical patients, patients withpancreatitis or patients with acute bleeding (i.e. non-infectious conditions that may also produce an eosino-penic response and a high NLCR) in the absence ofsuspected infection, will result in underestimated diag-nostic specificity of eosinopenia or NLCR. On the con-trary, a study of patients with fever of unknown originidentified eosinopenia as a good independent pre-dictor of infection [17].

Pitfalls in using EC and NLCR to differentiatebacterial from viral infection

A study on the differentiation of bacterial from viralinfections should include a clear description of how aviral or bacterial infection is defined, as well as adetailed description of patients included in the study.This is important to allow proper interpretation of theresults. For example, Naess et al. [8,9] described NLCR asa useful marker to discriminate viral from bacterial infec-tion. However, a significant portion of patients in the

‘viral infection’ group had infectious mononucleosis (9 of14 patients), that is, a viral infection typically associatedwith lymphocytosis, therefore limiting the applicability oftheir results to other more common viral infections. Also,patients with a confirmed viral infection (e.g. H1N1 con-firmed with RT-PCR), may have concurrent bacterialsuperinfections, further complicating the interpretationof the results.

Considering that the value of differentiating a viralfrom a bacterial infection would be to guide treatmentwith antibiotics, it might be more useful to compare ECor NLCR of patients successfully managed without antibi-otics to patients that required antibiotic treatment(instead of comparing and trying to define, viral vs. bac-terial infection). Other parameters, like the severity ofthe disease or the presence of SIRS criteria, should betaken into account. Whether the EC or NLRC simplyreflect these parameters or may have additional valuerequires further investigation.

Pitfalls in monitoring EC or NLCR to assessresponse to antimicrobial therapy

As mentioned above, Davido et al. [27] recentlyreported that monitoring the EC may be useful to pre-dict the efficacy of the selected antimicrobial regimen.For such a study, ideally the two groups of patients(correct vs. incorrect antibiotic regimen) should bematched in terms of diagnoses and initial antibioticregimens, and patients receiving corticosteroids (anydose) at presentation or during hospital stay shouldbe excluded. The development of eosinophilic drugreactions during antibiotic treatment should also berecorded as this may cause a rise of the EC, irrespect-ive of response to therapy. Of note is that asymptom-atic eosinophilia (>500/ll) is not uncommon duringintravenous antibiotic treatment, and although it usu-ally occurs later in the course of therapy [49], it mayoccur during the first days of treatment [50],which can be an important confounding factor.Furthermore, although early-onset eosinophilia is usu-ally accompanied by signs of allergy, a milder asymp-tomatic rise of EC as a reaction to the administrationof antibiotics cannot be ruled out. Moreover, eosino-philia occurs more often with some antibiotics thanothers [49] and some antibiotics may act faster thanother. For these reasons, it is important that the twogroups of patients should be matched in terms of theinitial antibiotic regimen. The route of administrationmay also be important and should be matched

INFECTIOUS DISEASES 9

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

between the two groups, as allergic reactions (and theassociated rise of the EC) may be more common fol-lowing parenteral compared to oral administration [51].

Other considerations

Some other issues that should be taken into accountwhen designing studies of EC and NLCR are thefollowing:

1. Patients with infection but with a high baselineeosinophil count due to other concurrent conditions(e.g. chronic rhinitis, asthma, atopic dermatitis, para-sitic disease, some malignancies, autoimmune dis-eases, etc. [52]) may be less likely to present with anEC lower than the defined cut-off. In such patients,the change of EC from the baseline (if known) maybe a more accurate, but impractical, marker.

2. The accuracy of haematology instruments in measur-ing low EC may be low [53].

3. The presence of active haematological malignancies,cytotoxic chemotherapy and HIV infection may alterthe complete blood count parameters and suchpatients should be excluded from studies evaluatingEC or NLCR or studied as a separate group.

4. Antibiotic use before presentation to the hospitalshould be taken into account, and it may be appro-priate to exclude such patients [16], as an appropri-ate antibiotic regimen might result in fast resolutionof eosinopenia [27].

Conclusions

Although there are many publications regarding thevalue of EC and NLCR for diagnostic, prognostic or moni-toring purposes, significant pitfalls limit the applicabilityof these results to clinical practice. Considering the lowcost and the fact that a complete blood count is avail-able in all patients admitted to hospital, it may be worthinvestigating the value of EC or NLCR in properlydesigned prospective studies. Monitoring EC to predictthe efficacy of antimicrobial therapy seems to be apromising clinical application that requires validation inlarger studies.

Disclosure statement

The authors report no conflicts of interest

ORCID

Stamatis Karakonstantis http://orcid.org/0000-0002-2643-3184

References

[1] Bass DA. Behavior of eosinophil leukocytes in acute inflam-mation. II. eosinophil dynamics during acute inflammation.J Clin Invest. 1975;56:870–879.

[2] Bass DA, Gonwa TA, Szejda P, et al. Eosinopenia of acuteinfection: production of eosinopenia by chemotactic factorsof acute inflammation. J Clin Invest. 1980;65:1265–1271.

[3] Best WR, Muehrcke RC, Kark RM. Studies on adrenocorticaleosinopenia: a clinical and statistical evaluation of four-hour eosinophil response test. J Clin Invest. 1952;31:733–742.

[4] Hills AG, Forsham PH, Finch CA. Changes in circulating leu-kocytes induced by the administration of pituitary adreno-corticotrophic hormone in man. Blood. 1948;3:755–768.

[5] Dreyfuss F, Feldman S. Eosinopenia induced by emotionalstress. Acta Med Scand. 1952;144:107–113.

[6] Summers C, Rankin SM, Condliffe AM, et al. Neutrophil kin-etics in health and disease. Trends Immunol.2010;31:318–324.

[7] Zahorec R. Ratio of neutrophil to lymphocyte counts-rapid and simple parameter of systemic inflammationand stress in critically ill. Bratisl Lek Listy. 2001;102:5–14.

[8] Naess A, Nilssen SS, Mo R, et al. Role of neutrophil tolymphocyte and monocyte to lymphocyte ratios in thediagnosis of bacterial infection in patients with fever.Infection. 2017;45:299–307.

[9] Naess A, Mo R, Nilssen SS, et al. Infections in patientshospitalized for fever as related to duration andother predictors at admittance. Infection. 2014;42:485–492.

[10] Abidi K, Khoudri I, Belayachi J, et al. Eosinopenia is a reli-able marker of sepsis on admission to medical intensivecare units. Crit Care. 2008;12:R59.

[11] Gil H, Magy N, Mauny F, et al. Value of eosinopenia ininflammatory disorders: an "old" marker revisited. Rev MedInterne. 2003;24:431–435.

[12] Smithson A, Perello R, Nicolas JM. Is eosinopenia a reliablemarker of sepsis? Crit Care. 2009;13:409.

[13] Shaaban H, Daniel S, Sison R, et al. Eosinopenia: is it agood marker of sepsis in comparison to procalcitonin andC-reactive protein levels for patients admitted to a criticalcare unit in an urban hospital? J Crit Care. 2010;25:570–575.

[14] Garnacho-Montero J, Huici-Moreno MJ, Gutierrez-PizarrayaA, et al. Prognostic and diagnostic value of eosinopenia, C-reactive protein, procalcitonin, and circulating cell-free DNAin critically ill patients admitted with suspicion of sepsis.Crit Care. 2014;18:R116.

[15] Anand D, Ray S, Bhargava S, et al. Exploration of eosinope-nia as a diagnostic parameter to differentiate sepsis fromsystemic inflammatory response syndrome: results from anobservational study. Indian J Crit Care Med.2016;20:285–290.

[16] Lavoignet CE, Le Borgne P, Slimani H, et al. Relevance ofeosinopenia as marker of sepsis in the EmergencyDepartment. Rev Med Interne. 2016;37:730–734.

10 S. KARAKONSTANTIS ET AL.

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

[17] Efstathiou SP, Pefanis AV, Tsiakou AG, et al. Fever ofunknown origin: discrimination between infectious andnon-infectious causes. Eur J Intern Med. 2010;21:137–143.

[18] Gucyetmez B, Atalan HK. C-reactive protein and hemogramparameters for the non-sepsis systemic inflammatoryresponse syndrome and sepsis: what do they mean? PLoSOne. 2016;11:e0148699.

[19] Ljungstrom L, Pernestig AK, Jacobsson G, et al. Diagnosticaccuracy of procalcitonin, neutrophil-lymphocyte countratio, C-reactive protein, and lactate in patients with sus-pected bacterial sepsis. PLoS One. 2017;12:e0181704.

[20] Merino CA, Martinez FT, Cardemil F, et al. Absolute eosino-phils count as a marker of mortality in patients with severesepsis and septic shock in an intensive care unit. J CritCare. 2012;27:394–399.

[21] Abidi K, Belayachi J, Derras Y, et al. Eosinopenia, an earlymarker of increased mortality in critically ill medicalpatients. Intensive Care Med. 2011;37:1136–1142.

[22] Terradas R, Grau S, Blanch J, et al. Eosinophil count andneutrophil-lymphocyte count ratio as prognostic markers inpatients with bacteremia: a retrospective cohort study.PLoS One. 2012;7:e42860.

[23] Rahimi-Rad MH, Asgari B, Hosseinzadeh N, et al.Eosinopenia as a marker of outcome in acute exacerbationsof chronic obstructive pulmonary sisease. Maedica (Buchar).2015;10:10–13.

[24] Steer J, Gibson J, Bourke SC. The DECAF Score: predictinghospital mortality in exacerbations of chronic obstructivepulmonary disease. Thorax. 2012;67:970–976.

[25] Hwang SY, Shin TG, Jo IJ, et al. Neutrophil-to-lymphocyteratio as a prognostic marker in critically-ill septic patients.Am J Emerg Med. 2017;35:234–239.

[26] Escobar-Valdivia EJ, Gonzalez-Aguirre JE, Carrillo-CisnerosER, et al. Eosinophil count at intensive care unit admissionwas not predictor of hospital mortality: results of a casecontrol study. J Intensive Care. 2015;3:27.

[27] Davido B, Makhloufi S, Matt M, et al. Changes in eosinophilcount during bacterial infection: revisiting an old marker toassess the efficacy of antimicrobial therapy. Int J Infect Dis.2017;61:62–66.

[28] Ho KM, Towler SC. A comparison of eosinopenia and C-reactive protein as a marker of bloodstream infections incritically ill patients: a case control study. Anaesth IntensiveCare. 2009;37:450–456.

[29] Setterberg MJ, Newman W, Potti A, et al. Utility of eosino-phil count as predictor of bacteremia. Clin Infect Dis.2004;38:460–461.

[30] Wibrow BA, Ho KM, Flexman JP, et al. Eosinopenia as adiagnostic marker of bloodstream infection in hospitalisedpaediatric and adult patients: a case-control study. AnaesthIntensive Care. 2011;39:224–230.

[31] Lowsby R, Gomes C, Jarman I, et al. Neutrophil to lympho-cyte count ratio as an early indicator of blood stream infec-tion in the emergency department. Emerg Med J.2015;32:531–534.

[32] de Jager CP, Wever PC, Gemen EF, et al. The neutrophil-lymphocyte count ratio in patients with community-acquired pneumonia. PLoS One. 2012;7:e46561.

[33] Matono T, Kutsuna S, Kato Y, et al. Role of classic signs asdiagnostic predictors for enteric fever among returned trav-ellers: relative bradycardia and eosinopenia. PLoS One.2017;12:e0179814.

[34] Farmakiotis D, Varughese J, Sue P, et al. Typhoid Fever inan inner city hospital: a 5-year retrospective review.J Travel Med. 2013;20:17–21.

[35] Farmakiotis D, Ntouloulis C, Mihailidis N, et al. Diagnosticvalue of eosinopenia in non-Typhi Salmonella enteritis. ClinInfect Dis. 2014;59:1197–1198.

[36] Flick H, Drescher M, Prattes J, et al. Predictors of H1N1influenza in the emergency department: proposition for amodified H1N1 case definition. Clin Microbiol Infect.2014;20:105–108.

[37] Yip B, Ho KM. Eosinopenia as a predictor of unexpected re-admission and mortality after intensive care unit discharge.Anaesth Intensive Care. 2013;41:231–241.

[38] Altman LC, Hill JS, Hairfield WM, et al. Effects of corticoste-roids on eosinophil chemotaxis and adherence. J ClinInvest. 1981;67:28–36.

[39] Shoenfeld Y, Gurewich Y, Gallant LA, et al. Prednisone-induced leukocytosis. Influence of dosage, method and dur-ation of administration on the degree of leukocytosis. Am JMed. 1981;71:773–778.

[40] Karakonstantis S, Kalemaki D. Correspondence to "Role ofneutrophil to lymphocyte and monocyte to lymphocyteratios in the diagnosis of bacterial infection in patients withfever". Infection. 2017 Sep 8. [Epub ahead of print].

[41] Naess A, Eide GE, Sjursen H. Reply to Drs. Karakonstantisand Kalemaki. Infection. 2017.

[42] Kirkeby K. Diagnostic and prognostic significance of eosino-penia in acute myocardial infarction. Am J Med Sci.1956;232:50–56.

[43] Lee MJ, Park SD, Kwon SW, et al. Relation between neutro-phil-to-lymphocyte ratio and index of microcirculatoryresistance in patients with ST-segment elevation myocardialinfarction undergoing primary percutaneous coronary inter-vention. Am J Cardiol. 2016;118:1323–1328.

[44] Bolayir A, Cigdem B, Gokce SF, et al. The effect ofeosinopenia on mortality in patients with intracerebralhemorrhage. J Stroke Cerebrovasc Dis. 2017;26:2248–2255.

[45] Wang F, Hu S, Ding Y, et al. Neutrophil-to-lymphocyteratio and 30-day mortality in patients with acute intracere-bral hemorrhage. J Stroke Cerebrovasc Dis. 2016;25:182–187.

[46] Sun Y, You S, Zhong C, et al. Neutrophil to lymphocyteratio and the hematoma volume and stroke severity inacute intracerebral hemorrhage patients. Am J Emerg Med.2017;35:429–433.

[47] Dilektasli E, Inaba K, Haltmeier T, et al. The prognostic valueof neutrophil-to-lymphocyte ratio on mortality in criticallyill trauma patients. J Trauma Acute Care Surg.2016;81:882–888.

[48] Chen W, Yang J, Li B, etet al. Neutrophil to lymphocyteratio as a novel predictor of outcome in patients withsevere traumatic brain injury. J Head Trauma Rehabil. 2017.[Epub ahead of print]. doi: 10.1097/HTR.0000000000000320

INFECTIOUS DISEASES 11

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

[49] Blumenthal KG, Youngster I, Rabideau DJ, et al. Peripheralblood eosinophilia and hypersensitivity reactions amongpatients receiving outpatient parenteral antibiotics.J Allergy Clin Immunol. 2015;136:1288–1294 e1.

[50] Ramirez E, Medrano-Casique N, Tong HY, et al. Eosinophilicdrug reactions detected by a prospective pharmacovigi-lance programme in a tertiary hospital. Br J Clin Pharmacol.2017;83:400–415.

[51] Warrington R, Silviu-Dan F. Drug allergy. Allergy AsthmaClin Immunol. 2011;7:S10.

[52] Kovalszki A, Weller PF. Eosinophilia. Prim Care.2016;43:607–617.

[53] Amundsen EK, Urdal P, Henriksson CE. Poor accuracy of theeosinophil count of hematology instruments limits the useof eosinopenia as a marker of bacterial infection orincreased mortality. Int Jnl Lab Hem. 2015;37:e119–e122.

12 S. KARAKONSTANTIS ET AL.

Dow

nloa

ded

by [

Tuf

ts U

nive

rsity

] at

11:

55 2

7 O

ctob

er 2

017

本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP

图书馆。

图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页 文献云下载 图书馆入口 外文数据库大全 疑难文献辅助工具