Acute mental stress assessment via short term HRV analysis in healthy adults: A systematic review with meta-analysis

(1)

ContentslistsavailableatScienceDirect

Biomedical

Signal

Processing

and

Control

j ou rn a l h o m e pa g e :w w w . e l s e v i e r . c o m / l o c a t e / b s p c

Review

Acute

mental

stress

assessment

via

short

term

HRV

analysis

in

healthy

adults:

A

systematic

review

with

meta-analysis

R. Castaldo

a

_,

_P.

_Melillo

b

_,

_U.

_Bracale

c

_,

_M.

_Caserta

a,c

_,

_M.

_Triassi

c

_,

_L.

_Pecchia

a,∗ a_School_of_Engineering,_University_of_Warwick,_Coventry_CV4_7AL,_UK

b_{Multidisciplinary}_Department_of_Medical,_Surgical_and_Dental_Sciences,_Second_University_of_Naples,_Italy c_Department_of_Public_Health,_University_Federico_II_Naples,_Naples,_Italy

a

r

t

i

c

l

e

i

n

f

o

Articlehistory:

Received28November2014

Receivedinrevisedform6February2015 Accepted18February2015

Availableonline7March2015 Keywords:

Acutementalstress HeartRateVariability

Systematicreviewwithmeta-analysis

a

b

s

t

r

a

c

t

Mentalstressreducesperformances,ontheworkplaceandindailylife,andisoneofthefirstcauses ofcognitivedysfunctions,cardiovasculardisordersanddepression.Thisstudysystematicallyreviewed existingliteratureinvestigating,inhealthysubjects,theassociationsbetweenacutementalstressand shorttermHeartRateVariability(HRV)measuresintime,frequencyandnon-lineardomain.Thegoalof thisstudywastoprovidereliableinformationaboutthetrendsandthepivotvaluesofHRVmeasures duringmentalstress.Asystematicreviewandmeta-analysisoftheevidencewasconducted,performing anexhaustiveresearchofelectronicrepositoriesandlinearresearchingreferencesofpapersresponding totheinclusioncriteria.Afterremovingduplicatesandnotpertinentpapers,journalpapersdescribing well-designedstudiesthatanalyzedrigorouslyHRVwereincludedifanalyzedthesamepopulationof healthysubjectsatrestandduringmentalstress.12paperswereshortlisted,enrollingoverall758 volun-teersandinvestigating22differentHRVmeasures,9ofwhichreportedbyatleast2studiesandtherefore meta-analyzedinthisreview.Fourmeasuresintimeandnon-lineardomains,associatedwithanormal degreeofHRVvariationsresultedsignificantlydepressedduringstress.ThepowerofHRVfluctuations athighfrequencieswassignificantlydepressedduringstress,whiletheratiobetweenlowandhigh frequencyresultedsignificantlyincreased,suggestingasympatheticactivationandaparasympathetic withdrawalduringacutementalstress.Finally,amongthe15non-linearmeasuresextracted,only2were reportedbyatleast2studies,thereforepooled,andonlyoneresultedsignificantlydepressed,suggesting areducedchaoticbehaviourduringmentalstress.HRVresultedsignificantlydepressedduringmental stress,showingareducedvariabilityandlesschaoticbehaviour.Thepooledfrequencydomainmeasures demonstratedasignificantautonomicbalanceshiftduringacutementalstresstowardsthesympathetic activationandtheparasympatheticwithdrawal.PivotvaluesforthepooledmeandifferencesofHRV measuresareprovided.FurtherstudiesinvestigatingHRVnon-linearmeasuresduringmentalstressare stillrequired.However,themethodproposedtotransformandthenmeta-analyzetheHRVmeasures canbeappliedtootherfieldswhereHRVprovedtobeclinicallysignificant.

Contents

1. Introduction... 371

2. Methodsandmaterials... 371

2.1. Searchstrategy... 371

2.2. Inclusionandexclusioncriteriaforpaperselection... 371

2.3. Papershortlisting,dataextractionandoutcomesofinterest... 372

2.4. HRVmeasures... 372

2.5. Statisticalanalysisandsoftwaretools... 372

∗ Correspondingauthorat:SchoolofEngineering,UniversityofWarwick,CoventryCV47AL,UK.Tel.:+4402476573383;mobile:+4407505716490.

E-mailaddresses:R.Castaldo@warwick.ac.uk(R.Castaldo),paolo.melillo@unina2.it (P.Melillo),umberto.bracale@unina.it(U.Bracale),triassi@unina.it(M.Triassi), l.pecchia@warwick.ac.uk(L.Pecchia).

(2)

3. Results... 372

3.1. Characteristicsoftheincludedstudies... 372

3.2. TrendsoftheHRVmeasures... 373

3.2.1. Timedomainfeatures ... 373

3.2.2. Frequenciesdomainfeatures... 373

3.2.3. Non-linearfeatures... 373

3.3. PooledpivotvaluesofHRVmeasuresmeandifferences... 373

3.4. MathematicalmodellingofHRVmeasuresduringmentalstress... 374

4. Discussion... 375

5. Conclusion... 376

References... 376

1. Introduction

Mental stress hasbeeninvestigated in various ﬁeldsdue to

itsdestructiveeffectsonthedailyroutine.Stressmaycause

cog-nitivedysfunctions,cardiovasculardiseases,depression[1,2]and

maylead to deathand illness [3]. Thereis, in fact, an average

of 50% of employees, which suffer of ‘work stress’ [4].

More-over,stressreducesperformancesindailylife[5]andparticularly

in the work places[1,6]. Thismay be particularly relevant for

those jobs that expose the workers or other persons to risky

situations. This is the case of surgeons performing long

surg-eries,inwhichthelossofattentionorconcentrationmaycause

severeeffects onthepatient,orpilots ﬂyingonlongdistances,

whose stress may be dangerous for them and the passengers

[7].

Mental stress inﬂuences the Autonomous Nervous System

(ANS),whichcontrolsourcapabilitytoreacttoexternalstimuli

[3,5]. Therefore, the acute stress may be evaluated with

non-invasivebiomarkermeasurements,whichareconsideredreliable

estimatorsofANS statuses.Thisis thecase of HeartRate

Vari-ability(HRV),whichisconsidereda reliablemeanstoindirectly

observeANS[8],alsoinreallifesettings.HRVreferstothe

varia-tionsofbothinstantaneousheartrateandtheseriesofinter-times

betweenconsecutivepeaksoftheR-waveoftheECG(RRseries)

[9].ThisvariationisunderthecontroloftheANS,whichthrough

theparasympatheticandthesympatheticbranches,isresponsible

foradjustingtheHRVinresponsetoexternalorinternalphysicalor

emotionalstimuli.Anormalsubjectshowsagooddegreeof

varia-tionoftheheartrate,reﬂectingagoodcapabilitytoreacttothose

stimuli[9].

SeveralstudiesinvestigatedinthelastyearshowHRV could

beusedtoassessacuteorchronicmentalstress,inhealthy

sub-jects or in speciﬁc groups of patients, of different ages. These

studiescollectedandanalyzedHRVwithdifferentmethodsand

tools: long or short term registrations were utilized;HRV was

extracted fromECG orSpO2 or viaothermeans; a wide range

ofdeviceswasemployed(i.e.fromcommercialsmartphonesto

CEmarkedmedicaldevices forECGmonitoring,includingsome

advancedbiomedicalampliﬁers);HRVwasanalyzedwith

commer-cialsoftwaretools,sometimeswellvalidatedones;HRVfeaturesin

timeand/orfrequencywereextractedandbothlinearand/or

non-lineardomainswereexplored;differentstatisticaltests,pattern

recognitionordataminingmethodswereusedtoexploreresults

[4,9,10].

Thegoalofthisstudywastoreviewhomogenouslydesigned

studies,inordertoprovidereliableinformationaboutthetrends

and the pivot values of HRV measures during mental stress.

Therefore, thisreviews investigatedsystematically studies

pub-lishedinpeer-reviewedjournalsandfocusingontheassociations

between acute mental stress and HRV, using short term HRV

excerpts,extractedfromECGs,inhealthysubjectsabove18years

old.

2. Methodsandmaterials

2.1. Searchstrategy

Relevantstudiesonthedetectionofacutementalstressthrough

theanalysisoftheHRVwereidentiﬁedandselectedbysearching

onPubMedandOvidSPdatabases.

PertinentarticlesweresearchedusingBooleancombinations

of the following keywords or their equivalent Medical Subject

Heading(MeSH)terms:HeartRateVariability,HRV,mentalstress,

psychologicalstress,andemotionalassessment.

Thefollowingcriteriawereutilizedtolimittheresearch:papers

publishedinthelast10years(since2004),studyonadulthumans

(i.e.notanimals),notchildren,notcancer,notpregnancy.

2.2. Inclusionandexclusioncriteriaforpaperselection

Afteraﬁrstscreeningofthetitlesandabstracts,studies

pub-lished beforeNovember2014wereconsidered suitablefor this

reviewiftheymetallofthefollowingcriteria:

• Paperpublishedonscientiﬁcjournalswithpeerreview;

• PaperfocusedmainlyonmentalstressinvestigationusingHRV;

• The experiments were conducted witha robust design, well

described, includingrepeatedmeasuresin thesame groupof

healthysubjectsatrestandduringstresssection.Studieswere

consideredrobustiftheacquisitionoftheHRVwasadequately

standardized(i.e.stressandrestsessionsinthemomentofthe

daytominimizethecircadianeffectandwiththesubjectsinthe

sameposition).

• Thestudywasnot focusingonchronicstress,and onlyacute

stressorswhereused,asdeﬁnedin[11].

• Thesubjectswerehumansbeingsover18yearsold;

Studieswereexcludedifthey:

• focusedonchronicstress;

• utilizedHRVanalysisonexcerptswerelongerthan10min;

• enrolledprofessionalathletesandobservedthemduringsport

trainingsessions;

• investigatedpainperception;

• reportedHRVmeasureswithnotsufﬁcientquality(i.e.nounit

measuresstated,notappropriatestatisticaldescriptors);

• didnotreportinclusion/exclusioncriteriaforsampleselection.

Twoauthors(RCandPM)independentlyassessedthe

method-ologicalqualityofthepapers,giventheinclusion/exclusioncriteria.

(3)

2.3. Papershortlisting,dataextractionandoutcomesofinterest

Followingtheresearchstrategydescribedabove,allthetitles

respondingtothechosenkeywordswereidentiﬁed.After

exclud-ingrepetitions(titlesindexedbothinPubMedandOvidSp),studies

wereshortlistedaccordingtoinclusions/exclusionscriteria,

inves-tigatingtitles,abstractsandfull-papers.

HRVmeasuresintimeandfrequencydomains,bothlinearand

non-linear,wereextractedbythepapersincludedinthisreview

[1,5,6,12–20].FurtherdetailsaboutHRVmethodological

conven-tionsadoptedinthisreviewcanbefoundintwoopenaccesspapers

[21,22].

ThechangesofHRVmeasuresduringstressandrestsessions

wereinvestigatedobservingtheirtrends.Atrendwasrepresented

witharrowspointedup,ifthemeanvalueofaHRVmeasurewas

increasingduringthestresssection,withrespecttothevalueof

thesamemeasureregisteredduringtherestingsession.Anarrow

pointeddownwasusedifthemeanvalueof theHRV measure

wasdecreasing.Twoarrowswereutilizedforsigniﬁcantchanges

(p-values<0.05).Sincepapersreportedonlyparametersof

mea-suredistributions(i.e.meansandstandarddeviations)thep-values

werecalculatedwithat-testandnotwithparametrictests,which

insomecaseswouldhavebeenmoreappropriated.

2.4. HRVmeasures

RegardinglinearHRVmeasuresintimeandfrequencydomains,

therecommendationsoftheEuropeanSocietyofCardiologyand

theNorthAmericanSocietyofPacingandElectrophysiology

guide-lines[9]werefollowed. Therefore,thepowerspectrum density

measures were considered if reported in normalized units or

ms2_,_for_the_low_frequency_bands_(LF,_0.04–0.15_Hz)_and_the_high

frequencybands (HF, 0.15–0.4Hz). Whenpapers shortlisted for

meta-analysisreportedfeaturesinanon-conventionalway,these

measureswereexcludedfromthepoolingorconvertedinms2_,_if

possible.Forinstance,threestudies[16,17,19]reportedfrequency

domainfeatures log-transformed(LFLog and HFLog). Thesewere

untransformedaccordingto[23],usingthefollowingformula:

meanLF(ms2)=exp(meanLFLog)

stdLF(ms2₎₌_std_LF

Log×[exp(meanLFLog)]2

Onestudyreported frequencymeasures inms2_/Hz _[14] _and

thereforeitsresultswerenotreportedinthisreview.

2.5. Statisticalanalysisandsoftwaretools

HRVmeasureswerepooledifreportedinmorethanonepaper.

Standardmethodsforsystematicreviewswithmeta-analyseswere

employedinthispapertopooltheHRVmeasures[24]:mean

differ-ence(MD)with95%conﬁdenceintervals(95%CI)andp-values(p).

Randomorﬁxedeffectmodelswereusedaccordingtostudy

het-erogeneitymeasuredwithaQstatistictest.AsigniﬁcantQstatistic

isindicativeofdissimilareffectsizesacrossstudiesandto

comple-menttheQtest,wealsocalculatedtheI2_statistic,_which_provides_an

indexofthedegreeofheterogeneityacrossstudies.Inparticular,I2

indicatesthepercentageofthetotalvariabilityineffectsizesdueto

between-studiesvariability,andnotduetosamplingerrorwithin

studies[25].Percentagesofaround25%(I2₌_25),_50%_(I2₌_50),_and

75%(I2₌₇₅₎_may_be_interpreted_as_low,_medium,_and_high

hetero-geneity,respectively[25].Differencesand95%CIwereconsidered

signiﬁcantifthep-valuewaslessthan0.05.TheOpenMeta-analyst

Softwaretoolwasutilizedtocomputethesestatistics.

Fig. 1. Flow chart of literature search: titles, abstract and full-papers included/excluded.

3. Results

Accordingtothesearchstrategydescribedabove,894titleswere

identiﬁed,345inPubMedand549inOvidSp.Afterremoving279

duplicates,615titleswereconsidered.Ofthese,510wereexcluded

afterreadingtheabstractsastheydidnotmeettheinclusion

crite-ria.Fromtheremaining105abstracts,77wereremovedduetothe

exclusioncriteria.Finally,28full-textswereanalyzedandamong

these,16wereexcludedduetoinclusion/exclusioncriteria.

There-fore,12studieswereﬁnallyconsideredappropriateforinclusionin

thissystematicreview.Aﬂowchartoftheliteraturesearchresults

isshowninFig.1.

3.1. Characteristicsoftheincludedstudies

The12studiesenrolledfrom12to399subjecteach,fora

cumu-lativepopulationof785subjects,whichwasﬁnallypooledinthis

review.Detailsaboutpopulation,HRVanalysisreportedineach

study,HRVlengthand statisticalmethods employedtoexplore

signiﬁcantvariationsarereportedinTable1.

Onlythreestudiesusednon-parametricstatistictests(NPST),

which are speciﬁcally designed to investigate signiﬁcant

dif-ference of features non-normally distributed [26]. Ninestudies

[1,6,12–14,16,17,19,20]usedANOVA,ANCOVAort-studenttestto

investigatehowHRVmeasureschangedbeforeandafterthestress

section.Ofthese,Trainaetal.,Papouseketal.,andLackeneretal.

[12,13,15]convertedfrequenciesdomainHRVfeatureswithalog

transform,whiletheremainingapplieddirectlystandard

statisti-calsigniﬁcanceteststofrequencydomainmeasures,whicharenot

normallydistributed,beingasymmetricdistributions ofpositive

(4)

Table1

Characteristicsofstudies. No.of

paper

Author/year Subjects(total/women) HRVanalysis HRVlength(min) Signiﬁcancetests

1 Hjortskov2004[6] 12/12 Timeandfrequency 3 ANOVA

2 Kofman2006[1] 30/– Frequency 10 t-Tests,ANOVA

3 Vuksanovic2007[12] 23/13 Frequencyandnon-linear 5 t-Tests

4 Li2009[13] 399/209 Timeandfrequency 0.5 ANOVA

5 Schubert2009[14] 50/28 Time,frequencyandnon-linear 3 ANCOVA

6 Tharion2009[15] 18/9 Timeandfrequency 5 Non-parametricstatisticaltests

(Wilcoxonsignedranktest)

7 Papousek2010[17] 65/53 Timeandfrequency 3 ANOVA

8 Lackner2011[16] 20/20 Timeandfrequency 5 ANOVA

9 Melillo2011[5] 42/– Non-linear 5 Non-parametricstatisticaltests

(Wilcoxonsignedranktest) 10 Taelman2011[18] 43/22 Timeandfrequency 6 Non-parametricstatisticaltests

(WilcoxonandFriedmantests)

11 Traina2011[19] 13/6 Frequency 5 t-Test

12 Visnovcova2014[20] 70/39 Timeandfrequency 6 t-Test,non-parametricstatistical tests(notspeciﬁed)

TheselectedstudiesreportedECGrecordingsatrest(resting

session, RS) and during induced mental stress sessions (stress

session,StS).Onlyonestudy[15]didnotreportECGrecordings

duringtheStS,buttheECGwasregisteredonthedayofa

univer-sityexamination,whichwasassumedtobethestressingevent.This

maygenerateheterogeneity,astheresponsetostressmaychange

duringorafterStS[27].

Mentalstresswasinducedaskingvolunteerstoperform,under

controlled circumstances, one or more of the following tasks:

ComputerWork Task(CWT), StroopColourWordTask(SCWT),

Arithmetictask(AT),Gametask(GT),publicSpeechtask(ST),

Aca-demicexamination(AE),otherPhysical-MentalTask(PMT).Abrief

descriptionofeachmentaltaskisgiveninTable2.Additionally,

inonestudy[18]wereaskedtoperformalsoaphysicaltask.HRV

measuresregisteredduringthosephysicaltaskswerenotincluded

inthisreviewsinceproducingadifferentkindofresponsethan

mentalstresstasks[28].

3.2. TrendsoftheHRVmeasures

Tables3–5reportedthetrendsandthevaluesofHRVmeasure

inRSandStSintime,frequencyandnon-lineardomains

respec-tively.Additionally,thesetablesreportedthenumberofsubjects

enrolledineachstudyandtherelativeweightofeachstudyduring

thepooling.

3.2.1. Timedomainfeatures

TimedomainHRVmeasuresreportedbythepapersshortlisted

forthismeta-analysisincluded:themeanofR-Rintervals(RR),a

standarddeviationofanR-R interval(SDRR),thesquarerootof

meansquareddifferenceofsuccessiveR-Rs(RMSSD),anda

pro-portionofNN50dividedbytotalnumberofNNthatdiffermore

than50ms(pNN50)asshowninTable3.

InallthestudiestherewasconsensusthatthemeanRR,pNN50

andRMSSDdecreasedduringStS,althoughsomestudiesdidnot

demonstratedasigniﬁcantreductionofthesefeatures.Adecreased

SDRRwasobservedinthemajorityofstudies[15,18,20]withahigh

levelofsigniﬁcance.Onlyonestudyreportedcontradictoryresults

[14].InfactSchubertetal.[14]reportedadiscordantincreaseof

SDRR,whichtheauthorsjustiﬁesasduetoaslowrespirationrate

andarelativereductioninventilationcausedbythespeechtask

usedintoinducestressinthisstudy[14].

3.2.2. Frequenciesdomainfeatures

FrequencydomainHRVmeasuresextractedfromtheshortlisted

papersincluded:thelowfrequencypower(LF),thehighfrequency

power(HF)andtheLF/HFratioasshowninTable4.

RegardingLF, among the8papers reportingthis measure,5

studiesagreedthat LFincreased(3withstatisticalsigniﬁcance),

while3reportedanoppositetrend.Alsothiscase,theﬁndingsby

Tharionetal.[15]werenotconsistentwiththegeneraltrend,

prob-ablybecauseinthisstudythestresssectionwasrecordedintheday

oftheexaminationandnotduringtheexaminationsession.In

addi-tion,Hjortskovetal.[6]alsoshowedcontroversialresults,which

inthiscasemightbeexplainedconsideringthatthiswastheonly

studyinwhichanintroductionsessionwasranbeforethe

registra-tionoftheRSandStS.Finally,Taelmanetal.[18]alsoreporteda

decreasedLFduringStS,whichmightbealsoduetothedifferent

protocoladopted,whichconsistedinaphysicaltasksbeforethe

metalstresssession.

RegardingHF,thereisconsensus,amongthedifferentstudies,

thatthismeasuredecreasedduringacutementalstress.Onlyone

studyindicatedacontroversialtrend,whichhoweverwasnot

sta-tisticalsigniﬁcant[12].

RegardingLF/HFratio,therewasageneralconsensus(5studies

outof7)thatitincreasedduringStS.Theremaining2studies[12,15]

reportedanopposite trend,whichhoweverisnotsupportedby

statisticalsigniﬁcance.

3.2.3. Non-linearfeatures

Non-linearHRVmeasuresextractedfromtheshortlistedpapers

included:theShannonandtheSampleEntropy(respectivelyShEn

andSampEn),thelargestLyapunovexponent(LLE),the

correla-tiondimension(D2),short-andlongtermﬂuctuationslope(˛1and

˛2),thestandarddeviationofthePoincare’plotperpendicularand

alongtothelineofidentity(respectivelySD1andSD2),recurrence

plotdeterminism(DET)andrecurrencerate(REC),maximallength

oflines(lmax)andmeanlengthoflines(lmean).Finally,the

Approxi-mateEntropy(En)computedwiththethresholdrsetto0.2×SDNN,

tothevaluemaximizingtheentropyandtothevaluecomputedby

Chonetal.[29],andreportedrespectivelyasEn(0.2),En(rmax)and

En(chon)asshowninTable5.

Onlytwofeatureswereinvestigatedbymorethanonestudy:D2

[5,14]wasreportedasconsistentlyreducedduringStS(academic

examinationinMelilloetal.[5]andarithmeticaltaskinShubert

etal.[14]);˛1[5,12]whichwasreportedwithsigniﬁcantlyopposite

trends.

3.3. PooledpivotvaluesofHRVmeasuresmeandifferences

Fromtheincludedstudies,8HRVmeasureswerepooledinthis

systematicmeta-analysisbecausereportedatleastintwopapers:4

measuresinthetimedomain(RR,SDRR,pNN50%,rMSSD),3inthe

(5)

Table2

Descriptionofstudydesign.

No.ofpaper Author/year Stressor Description

1 Hjortskov2004[6] ComputerWorkTask (CWT)

Thetaskconsistedofkeyinginrandomnumberswiththedominanthand usingthenumericalpartofthekeyboard.

2 Kofman2006[1] StroopColourWord

Task(SCWT)

Thetaskrequirednamingtheinkcolourinwhichcolours’nameswereprinted onascreedbeingeithercongruentorincongruentwithrespecttothecolour words.

3 Vuksanovic2007[12] Arithmeticaloudtask (AT)

Thetaskrequiredsubtracting17from1000.Subjectswereaskedtoansweras accurateaspossible.Theytoldtheresultsafter5s.

4 Li2009[13] Gametask(GT) Theinstructionsforvideogametask‘Breakout’havebeenstandardizedvia video.Thesubjectswillliesupineonahospitalbedandamonitor25inch colourTVwasplaced2maway.

5 Schubert2009[14] Speechtask(ST) Thistaskinvolvedpreparingandpresentingaspeechinresponsetooneoftwo situations.Subjectshad3mintoprepareand3mintopresenttheirspeech. 6 Tharion2009[15] Academicexamination

(AE)

Theﬁrstrecordingwasdoneonthedayoftheacademicexaminationandthe secondduringholiday.

7 Papousek2010[17] Academicexamination (AE)

Participantsweretoldtonotmovelegsandhandswhilespeaking,toreadout thequestionsloudbeforeansweringit.

8 Lackner2011[16] ArithmeticalTask(AT) Thetaskconsistedoftwotripletofone-digitnumbers,whichweretobeadded orsubtracted.

9 Melillo2011[5] AcademicExamination (AE)

Theﬁrstrecordwasperformedduringverbalexaminationandthesecondone wasperformedafterholiday.

10 Taelman2011[18] Physical-MentalTask (PMT)

Thetaskconsistedofcontinuousmentalcalculationofﬁveoperationswitha twoorthreedigitnumber,whichhadtobeperformedwithoutverbal stimulation.Participantsusedthemousecursortoindicatethecorrectanswer choosingbetweenthreealternatives.Theparticipantswouldhavebeen rewardedwithamovieticket

11 Traina2011[19] ArithmeticalTask(AT) Thetaskrequiredtosubtractingthenumber17startingfromthenumber986. 12 Visnovcova2014[20] StroopColourWord

Test(SCWT) ArithmeticTest(AT)

Thetaskrequiredtoreadthecolours(green,yellow,orange,red,blue,purple) onthewordsdisplayedonascreen,whichwerecongruentorincongruent withthewrittenword.

(Resultsnotreportedinthisreview)Thetaskrequiredtocalculatethree-digit numbersdisplayedindifferentrandomplacesonthescreenintoonedigit numbers.Subsequently,participantsdecidedthattheﬁnalresultwasevenor oddbypushingthekeyboardarrow.

(˛1,D2).Therelativepoolingweightofeachstudywasreportedin

Tables3–5,respectivelyfortime,frequencyandnon-lineardomain.

TheresultsofthepoolingwerereportedinTable6,wherealsothe

trendofthepooledHRVmeasuresisreported.

3.4. MathematicalmodellingofHRVmeasuresduringmental

stress.

Twostudies[5,19]proposedamathematicalmodelto

automat-icallydetectmentalstress.

Melilloetal.[5]proposedamodelbasedonLinearDiscriminant

Analysis(LDA),employingthreeHRVnon-linearmeasures:SD1,

SD2andEn(0.2).Theproposedmodelachievedaccuracy,

sensitiv-ityandspeciﬁcityrespectivelyof90%,86%and95%inautomatically

detectingsubjectsunderstress.Theseperformanceswereachieved

testing thedeveloped classiﬁerusing a 10-foldcross validation

technique,basedonsubjects’exclusions.

Trainaetal.[19]studiedthePearsoncorrelationbetween

fre-quencydomainmeasures(highandlowcomponentsofthepower

spectrum)beforeand afterthestressingsession, demonstrating

Table3

Timedomainfeatures.

Timedomainfeatures Authors Featurestrend Weightofmeta-analysis N Stress Rest

Mean SD Mean SD MeanRR(ms) Lackner2011[16] ↓ 1.82% 20 765.31 314.3 837.1 324.5 Papousek2010[17] ↓↓ 11.65% 65 617.92 210.0 819.7 244.1 Schubert2009[14] ↓↓ 9.25% 50 686.49 240.8 808.6 206.2 Taelman2011[18] ↓↓ 19.98% 43 755.44 134.5 863.5 147.1 Tharion2009[15] ↓↓ 12.83% 18 777.40 114.3 867.3 114.0 Visnovcova2014[20] ↓↓ 41.22% 70 675.99 120.7 847.8 130.4 Vuksanovic2007[12] ↓ 3.25% 23 740.74 263.2 806.5 249.5 SDRR(ms) Schubert2009[14] ↑↑ 3.50% 50 96.00 86.6 33.20 23.83 Taelman2011[18] ↓↓ 37.56% 43 35.40 16.4 46.73 19.48 Tharion2009[15] ↓↓ 8.90% 18 52.40 21.7 74.20 25.93 Visnovcova2014[20] ↓↓ 50.05% 70 48.98 17.9 56.23 21.67 RMSSD(ms) Li2009[13] ↓↓ 26.85% 105 55.50 29.6 68.40 37.70 Li2009[13] ↓↓ 14.61% 84 57.20 36.9 74.20 44.90 Taelman2011[18] ↓↓ 54.38% 43 19.39 13.8 28.74 16.58 Tharion2009[15] ↓ 4.17% 18 49.99 31.07 74.03 39.64 pNN50(%) Taelman2011[18] ↓ 78.43% 43 26.82 16.10 31.83 18.73 Tharion2009[15] ↓↓ 21.57% 18 20.57 19.04 39.37 23.79

↓↓(↑↑):signiﬁcantlylower(higher)understress(p<.05). ↓(↑):lower(higher)duringstresssection(p>.05).

(6)

Table4

Frequencydomainfeatures.

Frequencydomainfeatures Authors Feat.trend Weightofmeta-analysis N Stress Rest

Mean SD Mean SD LF(ms2₎ _Hjortskov₂₀₀₄_[6] _↓↓ _8.52% ₁₂ ₁₃₉₁ ₁₀₂₈ ₁₆₆₄ ₈₀₈ Lackner2011[16] ↑ 10.16% 20 1339 1205 812.4 649.9 Papousek2010[17] ↑ 14.33% 65 1644 987.0 997.3 598.6 Taelman2011[18] ↓↓ 14.83% 43 466.9 460.2 868.4 641.0 Tharion2009[15] ↓↓ 5.74% 18 1192 723.6 2155 2157 Traina2011[19] ↑↑ 15.36% 13 1241 312.0 511.0 114.5 Visnovcova2014[20] ↑↑ 15.69% 70 607.9 457.7 454.9 380.6 Vuksanovic2007[12] ↑↑ 15.37% 23 464.0 356.1 387.6 260.3 HF(ms2₎ _Hjortskov₂₀₀₄_[6] _↓↓ _5.45% ₁₂ ₁₁₃₁ ₇₁₈ ₁₇₇₆ ₁₀₉₂ Lifemale2009[13] ↓↓ 8.39% 84 1200 1454 2001 2066 Limale2009[13] ↓↓ 11.82% 105 1072 1058 1677 1751 Papousek2010[17] ↓ 18.14% 65 668.5 401.5 1097 665.1 Taelman2011[18] ↓↓ 15.62% 43 552.5 428.0 1005 782.6 Tharion2009[15] ↓↓ 1.54% 18 1691 2096 2892 2622 Traina2011[19] ↓ 17.91% 13 252.1 263.6 273.1 246.2 Visnovcova2014[20] ↓↓ 11.50% 70 445.9 1082 639.1 1337 Vuksanovic2007[12] ↑ 9.66% 23 665.1 925.1 595.9 714.4 LF/HF(–) Hjortskov2004[6] ↑ 15.12% 12 1.57 1.09 1.16 0.74 Papousek2010[17] ↑ 22.90% 65 1.11 0.59 0.01 0.80 Schubert2009[14] ↑ 19.88% 50 1.80 1.20 1.50 1.11 Tharion2009[15] ↓ 12.74% 18 1.30 0.80 1.40 1.80 Traina2011[19] ↑↑ 4.61% 13 5.81 2.88 2.57 2.10 Vuksanovic2007[12] ↓ 4.61% 23 1.07 3.83 1.08 2.92 Kofman2006[1] ↑ 20.14% 30 1.48 1.02 0.97 0.68

thatthosecorrelationsweresigniﬁcant.However,thisispartially

arguableasthePearsoncorrelationwouldeventuallylayonthe

assumptionthattheHRVmeasuresarenormallydistributed,while

HRVfrequency measuresare not.Thispaperdidnot develop a

predictivemodelanddidnotvalidatethecorrelationwithacross

validationtechniques.

4. Discussion

This paper presented the results of a systematic literature

reviewwithmeta-analysisofthearticlesinvestigatinghow

short-HRVmeasureschangedduringinducedacutementalstress.Inthis

review, 12studies wereshortlisted and changesduring mental

stressof22HRVmeasuresweresystematicallyreported.Finally,

9HRVmeasureswerepooled.

Theresultsdemonstrated that4 HRVmeasures(RR,RMSSD,

pNN50andD2)decreasedduringstress[13–15,17,18,20].

Themajorityofstudies[6,13,15,17–20]agreedthatSDRRandHF

decreasedduringstress,whileLF/HFandLFincreasedduringstress

[1,6,12,14,16,17,19,20].RegardingSDRR,onlyonestudy[14]outof the4[15,18,20]consideringthismeasure,reportedanincreasing

valueduring stress.Thismaybe due tothefactthat [14]

ana-lyzedHRVexcerptsof3min,while[15,18,20]analyzed5minHRV

excerpts.

Regarding LF, 3 studies [6,15,18] out of 8 [12,16,17,19,20]

consideringthismeasure,reportedadecreasingvalueduringstress.

However,thesestudiesadoptedstudydesignssigniﬁcantly

differ-entfromtheothers.Differentlyfromtheother,theﬁrsttwostudies

usedphysicalmovementsduringthestresssection:inHjortskov

etal.[6],participantsusedthedominanthandtodigitrandomly

Table5

Non-linearfeatures.

Non-lineardomainfeatures Authors Featurestrend Weightofmeta-analysis N Stress Rest

Mean SD Mean SD ˛1(–) Melillo2011[5] ↓↓ 49.41% 42 1.05 0.44 1.41 0.16 Vuksanovic2007[12] ↑↑ 50.59% 23 0.97 0.19 0.85 0.19 D2(–) Melillo2011[5] ↓↓ 45.89% 42 1.65 1.28 2.83 1.09 Schubert2009[14] ↓↓ 54.11% 50 3.2 0.33 3.5 0.27 SD1(ms) Melillo2011[5] ↓↓ – 42 0.02 0.01 0.02 0.01 SD2(ms) Melillo2011[5] ↓↓ – 42 0.05 0.02 0.08 0.02 En(0.2)(–) Melillo2011[5] ↓↓ – 42 0.99 0.24 1.09 0.13 En(rchon)(–) Melillo2011[5] ↓↓ – 42 0.98 0.24 1.11 0.11 En(rmax)(–) Melillo2011[5] ↓ – 42 1.09 0.17 1.12 0.10 ˛2(–) Melillo2011[5] ↓ – 42 0.76 0.13 0.78 0.18 ShEn(–) Melillo2011[5] ↑↑ – 42 3.42 0.39 3.17 0.23 DET(%) Melillo2011[5] ↑ – 42 98.75 1.28 98.61 0.86 REC(%) Melillo2011[5] ↑↑ – 42 42.24 12.05 33.46 6.27

lmean(beats) Melillo2011[5] ↑↑ – 42 14.88 6.77 11.09 2.48

lmax(beats) Melillo2011[5] ↓↓ – 42 213.4 136.5 286.7 111.2

LLE Vuksanovic2007[12] ↑ – 23 0.06 0.019 0.06 0.019

SampEn(–) Vuksanovic2007[12] ↓↓ – 23 1.65 0.06 1.77 0.04

(7)

Table6

PooledHRVmeasures.

Outcome Subjects Heterogeneity(I2_,_p) _Model _Units _MD _CI95% _p-value _Trend

Time RR 289 33%,0.17 Fixed ms −142.2 (−168.9;−115.47) <0.01 ↓↓ SDRR 181 92%,<0.01 Fixed ms −7.627 (−12.20;−2.97) <0.01 ↓↓ RMSSD 250 0%,0.50 Fixed ms −12.03 (−16.78;−7.28) <0.01 ↓↓ pNN50 61 65%,0.09 Fixed – −7.98 (−14.52;−1.45) <0.05 ↓↓ Frequency LF 264 91%,<0.01 Random ms2 _156.1 _(157.6;_469.8) _0.33 _↑ HF 433 62%,<0.01 Random ms2 _−359.7 _(−559.20;_−160.25) _<0.01 _↓↓ LF/HF 211 75%,<0.01 Random – 0.61 (0.14;1.08) <0.01 ↑↑ Non-linear ˛1 65 96%,<0.01 Random – −0.12 (−0.59;0.35) 0.63 ↓ D2 92 91%,<0.01 Fixed – −0.35 (−0.46;−0.23) <0.01 ↓↓

MD:meandifference;CI95%:conﬁdenceintervalat95%;DL:dimensionless. Boldvaluesindicatesigniﬁcantchanges.

numbersontheleftpartof a keyboard;inTaelman etal. [18],

participantsusedthemousecursortoindicatethecorrectanswer

choosingbetweenthreealternatives.Differentlyfromtheother

protocols,theuseofthehandactivatedacorticalareathatisnot

triggeredbythedesignsadoptedbytheremainingstudies,where

thereisnophysicalactivity.Theseresultswerealsoconsistedwith

theﬁndingsreportedinYuetal.[30]witharithmetictest,inwhich

theparticipantswererequiredtousethekeyboard.Theotherstudy

reportingadecreasedvalueofLFduringthestresssessionwas

Thar-ionetal.[15],whichhowevermeasuredthephysiologicalresponse

tothementalstressduringthedayoftheUniversityexamination,

andnotduringtheexamination,asdonein[5,17],whichalsoused

Universityexaminationasstressor.Atthis regard,there is

con-sensusthatthereactiontoastressingsituationis composedby

severalphases[27],eachimplyingadifferentresponseoftheANS

andthereforeadifferentHRVmodulation.Finally,onlyoneHRV

measure,˛1,wasreportedbytwostudies[5,12]achieving

com-pletelyoppositeresults,andbothwithstatisticalsigniﬁcance.No

clearexplanationforsuchheterogeneousresultscanbeinferredby

thetwopapers.

Regardingthemeta-analysis,thepooledvaluesof7HRV

meas-ures(RR,SDRR,RMSSD,pNN50,D2,HFandLF/HF)outofthe9

meta-analyzedchangedsigniﬁcantlyduringmentalstress.Thepooled

valuesofthose7measuresshouldberegardedastheyaremore

reli-ablethanthosepresentedbyeachpaperandshouldbeconsidered

aspossiblepivotvaluesforfuturestudies.Forinstance,theincrease

ofthepolledLF/HFwasstatisticallysigniﬁcantwhileitwasnot

sta-tisticallysigniﬁcantin6outofthe7papersreportingthisHRV

mea-sure.Sincethep-valueisstronglydependentbythesamplesize,a

possibleexplanationisthatthenumberofvolunteersenrolledin

eachofthose6studieswastoosmallcompared withtheLF/HF

meandifferencesmeasured,which,inturn,weretoosmall

com-paredwiththestandarddeviationsmeasuredineachgroupduring

stressandrest[24].RegardingtheLF,thepooledresultswould

havebecomestatisticalsigniﬁcantifthesethreestudiesemploying

adifferentdesign(asdiscussedabove[18,6]movinghandsand[15]

measuringstressinthedayofexaminationandnotduringit)were

excluded.Infact,removingthese3studiesthepooledmean

dif-ferenceforLFduringstresswas286.56ms2_(CI95%_{183.89–389.23,}

p<0.01,ﬁxedeffectmodel),calculatedon164subjectswithavery

smallheterogeneity(I2_94%,_p_<_0.001)._Finally,_it_was_hard_to

dis-cusstheresultsabout˛1,sincethetwostudiescalculatingthisHRV

measuresadoptedsimilarprotocolsandcomparablesamplesizes.

Melilloetal.[5]enrolledaboutthedoubleofsubjects,buttheSD

measuredinthestressingsessionwastoohigh,affectingtheweight

ofthestudy.Therefore,futureresearchesinvestigatingnon-linear

HRVfeaturesduringmentalstressarestillneeded.

ThedecreasesofRR,RMSSD,pNN50,SDRRreﬂectedadepressed

HRVduringstress.Thisisconsistentlyconﬁrmedbypooled

val-uesofHRVfrequencymeasures,amongwhichHFwasprovedto

decreasesigniﬁcantly,reﬂectingadecreasedHRVvariability.

More-over,theobserveddecreaseofD2duringstressingsectionscanbe

interpretedasareducedcomplexityoftheHRV,reﬂectingalower

adaptabilityandﬁtnessofthecardiacpacemakerandafunctional

restrictionoftheparticipatingcardiovascularelements[14].

Finally,thefrequencydomainmeasuresconsistentlysupported

theideathatduringstressthereisageneraldepressionofHRV

with a relative displacement of the vago-sympathetic balance,

duringwhichsympatheticactivationrelativelyovercome

parasym-patheticonce.Infact,theLFaccountingforanactivationofboth

parasympathetic and sympathetic system,increased, while the

HF,whichisassociatedtotheparasympatheticsystemactivation,

decreased.ThisresultwasconﬁrmedbytheincreaseofLF/HF.These

outcomesconﬁrmedtheinductedshiftoftheANSbalancetowards

thesympatheticactivationandtheparasympatheticwithdrawal

duringacutementalstress[20].Thisphenomenonwasexplained

throughthetheoryoftheﬁghtorﬂightresponse,whichsupport

theideathatthereisaninhibitionofthevagoandaprevalence

ofthesympatheticsystemduringastressingsituation[27].This

resultcouldchangeifthestressingsession wasmeasuredafter

(andnotduring)thestressingevent,asdemonstratedin[15]and

consistentlywiththephasesofthestressasdescribedin[27].

5. Conclusion

In conclusion, these review proved that there was

consen-susabouttheHRVmeasureschangingconsistentlyduringmental

stress.Particularly,theresultsofthepoolingoftheHRVmeasures,

providedpivotvaluesforatleast7HRVmeasuresthatchanged

signiﬁcantlyduringstressingsections.Thesesigniﬁcantchanges

conﬁrmedpreviousresultsabouttheinducted shiftoftheANS

balancetowardsthesympatheticactivationandthe

parasympa-theticwithdrawal duringacute mental stress.However,a huge

heterogeneityamongstudiesinvestigatingphysiologicalresponse

tomentalstresswasobservedinthisreview.Moreover,thisreview

provedthatfuturestudiesareneededtoconﬁrmthebehaviourof

non-linearHRVmeasuresduringstress.

Finally,futurestudiesarerecommendedto:deﬁnethestudy

design(i.e.lengthofHRVmeasures)andthestudyprotocol(i.e.

deﬁnitionofstressor,avoidphysicalactivityifnotrequired)

accord-ingtothebestavailableevidence;toanalysestheHRVfeatures

(i.e.usingmeasurement units accordingtointernational

(8)

distribution(i.e.checkifHRVmeasuresarenormallydistributed

ofnot);tocheckHRVstationaryifHRVmeasuresarelongerthan

5min;tomeasurestressingsessionaccordingtothegoalofthe

study(i.e.during,beforeorafterthestressingsession).

References

[1]O.Kofman,N.Meiran,E.Greenberg,M.Balas,H.Cohen,Enhancedperformance onexecutivefunctionsassociatedwithexaminationstress:evidencefrom task-switchingandStroopparadigms,Cognit.Emot.20(2006)577–595. [2]O.V.Crowley,P.S.McKinley,M.M.Burg,J.E.Schwartz,C.D.Ryff,M.Weinstein,

etal.,Theinteractiveeffectofchangeinperceivedstressandtraitanxietyon vagalrecoveryfromcognitivechallenge,Int.J.Psychophysiol.82(December) (2011)225–232.

[3]R.L.Pan,J.K.Li,Anoninvasiveparametricevaluationofstresseffectsonglobal cardiovascularfunction,Cardiovasc.Eng.7(June)(2007)74–80.

[4]T.Chandola,A.Heraclides,M.Kumari,Psychophysiologicalbiomarkersof work-placestressors,Neurosci.Biobehav.Rev.35(September)(2010)51–57. [5]P.Melillo,M.Bracale,L.Pecchia,NonlinearHeartRateVariabilityfeaturesfor

real-lifestressdetection.Casestudy:studentsunderstressduetouniversity examination,Biomed.Eng.Online10(2011)1–13.

[6]N.Hjortskov,D.Rissén,A.Blangsted,N.Fallentin,U.Lundberg,K.Søgaard, Theeffectofmentalstressonheartratevariabilityandbloodpressureduring computerwork,Eur.J.Appl.Physiol.92(2004)84–89.

[7]R.R.Singh,S.Conjeti,R.Banerjee,Acomparativeevaluationofneuralnetwork classiﬁersforstresslevelanalysisofautomotivedriversusingphysiological signals,Biomed.SignalProcess.Control8(2013)740–754.

[8]A.H.Marques,M.N.Silverman,E.M.Sternberg,Evaluationofstresssystems by applying noninvasive methodologies:measurements of neuroimmune biomarkersinthesweat,heartratevariabilityandsalivarycortisol, Neuroim-munomodulation17(2010)205–208.

[9]T.Force,Heartratevariabilityguidelines:standardsofmeasurement, physio-logicalinterpretation,andclinicaluse,Eur.HeartJ.17(1996)354–381. [10]U.R.Acharya,K.P.Joseph,N.Kannathal,C.M.Lim,J.S.Suri,Heartratevariability:

areview,Med.Biol.Eng.Comput.44(2006)1031–1051.

[11]S.C.Segerstrom,G.E.Miller,Psychologicalstressandthehumanimmune sys-tem:ameta-analyticstudyof30yearsofinquiry,Psychol.Bull.130(2004) 601–630.

[12]V.Vuksanovi ´c,V.Gal,Heartratevariabilityinmentalstressaloud,Med.Eng. Phys.29(2007)344–349.

[13]Z.Li,H.Snieder,S.Su,X.Ding,J.F.Thayer,F.A.Treiber,etal.,Alongitudinal studyinyouthofheartratevariabilityatrestandinresponsetostress,Int.J. Psychophysiol.73(September)(2009)212–217.

[14]C. Schubert, et al.,Effects of stress on heart rate complexity – a com-parison betweenshort-term and chronic stress,Biol. Psychol. 80 (2009) 325–332.

[15]E.Tharion,S.Parthasarathy,N.Neelakantan,Short-termheartratevariability measuresinstudentsduringexaminations,Natl.Med.J.India22(Mar-Apr(2)) (2009)63–66.

[16]H.K. Lackner, I. Papousek, J.J. Batzel, A. Roessler, H. Scharfetter, H. Hinghofer-Szalkay,Phasesynchronizationofhemodynamicvariablesand res-pirationduringmentalchallenge,Int.J.Psychophysiol.79(March)(2011) 401–409.

[17]I.Papousek,K.Nauschnegg,M.Paechter,H.K.Lackner,N.Goswami,G.Schulter, Traitandstatepositiveaffectandcardiovascularrecoveryfromexperimental academicstress,Biol.Psychol.83(February)(2010)108–115.

[18]J.Taelman,S.Vandeput,E.Vlemincx,A.Spaepen,S.VanHuffel,Instantaneous changesinheartrateregulationduetomentalloadinsimulatedofﬁcework, Eur.J.Appl.Physiol.111(July)(2011)1497–1505.

[19]C.A.Traina,M.Galullo,F.G.Russo,Effectsofanxietyduetomentalstress onheartratevariabilityinhealthysubjects,MinervaPsichiatr.52(2011) 227–231.

[20]Z.Visnovcova,M.Mestanik,M.Javorka,D.Mokra,M.Gala,A.Jurko,etal., Com-plexityandtimeasymmetryofheartratevariabilityarealteredinacutemental stress,Physiol.Meas.35(2014)1319.

[21]P.Melillo, N.De Luca, M.Bracale, L.Pecchia, Classiﬁcation tree forrisk assessment in patients sufferingfrom congestiveheart failure via long-term heart rate variability, IEEEJ. Biomed. HealthInf. 17(May) (2013) 727–733.

[22]P.Melillo,R.Izzo,N.DeLuca,L.Pecchia,Heartratevariabilityandtargetorgan damageinhypertensivepatients,BMCCardiovasc.Disord.12(2012)105. [23]J.M.Bland,D.G.Altman,Transformations,means,andconﬁdenceintervals,BMJ:

Br.Med.J.312(1996)1079.

[24]A.J.Sutton,K.R.Abrams,D.R.Jones,D.R.Jones,T.A.Sheldon,F.Song,Methods forMeta-analysisinMedicalResearch,JohnWiley,2000.

[25]J.A.Chalmers,D.S.Quintana,J.Maree,A.Abbott,A.H.Kemp,Anxiety disor-dersareassociatedwithreducedheartratevariability:ameta-analysis,Front. Psychiatry5(2014).

[26]S.Siegel,NonparametricStatisticsfortheBehavioralSciences,1956. [27]H.S.Bracha,Freeze,ﬂight,ﬁght,fright,faint:adaptationistperspectivesonthe

acutestressresponsespectrum,CNSSpectr.9(2004)679–685.

[28]S.A.Paritala,EffectsofPhysicalandMentalTasksonHeartRateVariability, KakatiyaUniversity,India,2009.

[29]K.Chon,C.G.Scully,S.Lu,Approximateentropyforallsignals,IEEEEng.Med. Biol.Mag.28(November–December)(2009)18–23.

[30]X.Yu,J.Zhang,Estimatingthecortexandautonomicnervousactivity dur-ing a mental arithmetic task, Biomed. Signal Process. Control 7 (2012) 303–308.