Assessment of the cortisol awakening response: Expert consensus guidelines
Tobias Stalder
a,∗, Clemens Kirschbaum
a, Brigitte M. Kudielka
b, Emma K. Adam
c, Jens C. Pruessner
d, Stefan Wüst
b, Samantha Dockray
e, Nina Smyth
f, Phil Evans
f,
Dirk H. Hellhammer
g, Robert Miller
a, Mark A. Wetherell
h, Sonia J. Lupien
i, Angela Clow
faDepartmentofPsychology,TUDresden,Dresden,Germany
bDepartmentofPsychology,UniversityofRegensburg,Regensburg,Germany
cSchoolofEducationandSocialPolicy,NorthwesternUniversity,Evanston,USA
dDepartmentofPsychiatry,McGillUniversity,Montreal,Canada
eSchoolofAppliedPsychology,UniversityCollegeCork,Cork,Ireland
fDepartmentofPsychology,UniversityofWestminster,London,UK
gDepartmentofPsychology,TrierUniversity,andStresszentrumTrier,Germany
hDepartmentofPsychology,NorthumbriaUniversity,Newcastle,UK
iDepartmentofPsychiatry,UniversityofMontreal,Montreal,Quebec,Canada
Keywords:
CAR Measurement Saliva Adherence Covariates Guidelines
a b s t r a c t
Thecortisolawakeningresponse(CAR),themarkedincreaseincortisolsecretionoverthefirst30–45min aftermorningawakening,hasbeenrelatedtoawiderangeofpsychosocial,physicalandmentalhealth parameters,makingitakeyvariableforpsychoneuroendocrinologicalresearch.TheCARistypically assessedfromself-collectionofsalivasampleswithinthedomesticsetting.Whilethisconfersecological validity,itlacksdirectresearcheroversightwhichcanbeproblematicasthevalidityofCARmeasure- mentcriticallyreliesonparticipantscloselyfollowingatimedsamplingschedule,beginningwiththe momentofawakening.ResearchersassessingtheCARthusneedtotakeimportantstepstomaximizeand monitorsalivasamplingaccuracyaswellasconsiderarangeofotherrelevantmethodologicalfactors.
Topromotebestpracticeoffutureresearchinthisfield,theInternationalSocietyofPsychoneuroen- docrinologyinitiatedanexpertpanelchargedwith(i)summarizingrelevantevidenceandcollective experienceonmethodologicalfactorsaffectingCARassessmentand(ii)formulatingclearconsensus guidelinesforfutureresearch.Thepresentreportsummarizestheresultsofthisundertaking.Consensus guidelinesarepresentedoncentralaspectsofCARassessment,includingobjectivecontrolofsampling accuracy/adherence,participantinstructions,covariateaccounting,samplingprotocols,quantification strategiesaswellasreportingandinterpretingofCARdata.Meetingthesemethodologicalstandardsin futureresearchwillcreatemorepowerfulresearchdesigns,thusyieldingmorereliableandreproducible resultsandhelpingtofurtheradvanceunderstandinginthisevolvingfieldofresearch.
Contents
1. Introduction...415
2. Cortisolawakeningresponse...416
2.1. Descriptionanddistinctivefeatures...416
2.2. Maincomponents...416
3. Inaccuratesampling:prevalenceandimpact...417
∗ Correspondingauthorat:TechnischeUniversitätDresden,DepartmentofPsy- chology,ZellescherWeg19,01069Dresden,Germany.Fax:+4935146337274.
E-mailaddresses:tobias.stalder@tu-dresden.de,tobias.stalder@googlemail.com (T.Stalder).
Konstanzer Online-Publikations-System (KOPS) URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-2-x4dttsqcxssg4 Erschienen in: Psychoneuroendocrinology ; 63 (2016). - S. 414-432
https://dx.doi.org/10.1016/j.psyneuen.2015.10.010
3.1. Delaybetweenawakeningandinitiationofsampling...417
3.2. Inaccuratepost-awakeningsampling...419
4. Strategiesfordealingwithinaccuratesampling...419
4.1. Objectivemonitoringstrategies...419
4.1.1. Methodsforverifyingawakeningandsamplingtimes...419
4.1.2. Dealingwithverifiedinaccuratedata...420
4.2. Arethereviablestrategieswithouttheuseofobjectivemeasures?...420
4.2.1. Forcedawakening...420
4.2.2. Increasingstatisticalpower...421
4.2.3. ExclusionofCARnon-responders...421
4.2.4. Informationalstrategies...421
5. Maximizingadherence...422
6. Dealingwithcovariates...423
6.1. Instructionsaboutpost-awakeningbehavior...423
6.2. Controlvariables...424
6.2.1. Statecovariates...424
6.2.2. Traitcovariates...425
6.3. Exclusioncriteria...425
7. Proceduralanddesignconsiderations...426
7.1. Samplingtimes...426
7.2. Numberofstudydays(cross-sectionalresearch)...426
7.3. Samplestorageandcortisolanalysis...426
7.4. Statisticalconsiderations...427
8. Summaryandguidelines...427
Conflictofinterest ... 429
Consensusprocessandcontributions ... 429
Roleofthefundingsource...429
Acknowledgement...429
References...429
1. Introduction
Abnormalsecretionoftheglucocorticoidhormonecortisolas the final product of the hypothalamus-pituitary-adrenal (HPA) axis is considered a crucial factor in linking the experience of chronicpsychosocialstresstoadverseeffectsonhealth(Chrousos, 2009).Besidesreactivitytoacutestressors,changestothecirca- dianregulationofcortisolsecretionareconsideredimportantin thiscontext(KondratovaandKondratov,2012;MenetandRosbash, 2011;Naderetal.,2010).Anaspectofcortisolregulationthatis ofspecialinteresttopsychoneuroendocrinological(PNE)inquiry is thecortisol awakening response (CAR), which describesthe markedincreaseincortisollevelsacrossthefirst30–45minfol- lowingmorningawakening(Clowetal.,2004,2010;Elderetal., 2014;KudielkaandWüst,2010).TheCARwasfirstsystematically describedinthemid-1990s(Pruessneretal.,1997)andsoongained attentionasa favorable biomarkerinPNE researchduetosev- eralmethodologicaladvantagesoverpreviouslyemployedcortisol assessmentstrategies(seeSection2).Theseadvantagestogether withevidenceshowinguniqueassociationsoftheCARwithpsy- chosocial,psychiatricandhealth-relatedparametershaveresulted in a rapid increase in publications over the past15 years (see Fig.1a).
TheCARcombinesfeaturesofareactivityindex(responseto awakening)withaspects tiedtocircadian regulation(occurring roughly at the same time every 24h) making it a fascinating researchtopic.However,preciselythesefeaturesalsomakeaccu- rateassessmentoftheCARachallengingtask.WhenrelyingonCAR dataacquiredbyparticipantsthemselves(usuallythroughsaliva sampling),validitycriticallyreliesonparticipantscloselyfollowing atimedsamplingschedule,beginningwiththemomentofawaken- ing.Inaccuratesampletimingcanoccureasilyandcansubstantially biasCARestimates.Furthermore,anumberofothermethodolog- ical factors,such as accounting for covariates,the number and natureofstudydaysand thetimingofsampling,canmarkedly affectCARdata.Whilenotallquestionsregardingtheroleandreg-
99-00 01-02 03-04 05-06 07-08 09-10 11-12 13-14
snoitacilbuP
0 50 100 150 200
(b) CAR research between 2013 and 2014 – methodological aspectsb Assessment days (mean, range)c: 2.0, 1–5
1 day: 31.7% 2 days: 47.1% 3 days: 16.3% 4–5 days: 4.8%
Sampling mes during first 1h (mean,range): 2.7, 1–7
1 sample: 3.6% 2 samp.: 51.1% 3 samp.: 18.7% 4–5 samp.: 25.9%
Reported use of diary log method: Yes: 65.5% No: 34.5%
Objecve controlof awakening med: Yes: 7.9% No: 92.1%
Objecve controlof sampling mese: Yes: 18.6% No: 81.4%
Full objecve control (both methods)f: Yes: 5.7% No: 94.3%
(a) Publicaon count of research using CAR assessments over the past 15 yearsa
Fig.1.(a)DevelopmentofpublicationsusingCARassessmentsoverthepast15 yearsand(b)methodologicalaspectsofresearchfromthelastperiodbetween2013 and2014.
ulationoftheCARhavebeenadequatelysolveduntilnow,several carefulinvestigationshaveexaminedtheimpactofmethodolog- icalfactorsonaccurateCARassessmentandhaverecommended strategiesfordealingwiththem(describedbelow).Unfortunately, suchrecommendationshavenotbeenwidelyimplementedinpub- lishedCARresearch.Fig.1bprovidesanoverviewofmethodological characteristicsofsuchstudies,publishedbetween2013and2014.
Itcanbeseenthattheemployedmethodologicalstandardsvar- iedwidelybetweeninvestigations,withahighnumberofstudies fallingshortofpreviousrecommendationsforbestpracticeinCAR research(e.g.,objectivecontrolofsamplingtimes:Brodericketal., 2004;Kudielkaetal.,2003).
Toaddress this,the InternationalSociety ofPsychoneuroen- docrinology(ISPNE)hasinitiated anexpert paneltosummarize relevantevidenceandcollectiveexperienceonmethodologicalfac- torsaffectingCARassessment. Thegoalofthisinitiative wasto formulateclearconsensusguidelinesbasedoncurrentknowledge forfuture studiesinthisevolvingfield ofresearch.Thepresent report summarizes the results of this undertaking. As a large proportionofCARresearchusessalivarycortisolassessmentsin participants’domesticsetting,aparticularfocusisputonmethod- ologicalchallengesinthisresearchcontext.Giventheimportance ofsampletimeaccuracy,thefirstthreesectionsaredevotedtoan in-depthdiscussionofthistopic,includingstrategiestoincrease samplingaccuracybymaximizing participantadherence. Inthe subsequentsectionsarangeoffurthermethodologicalfactorsare covered.Inthefinalsection,thederivedconsensusguidelinesare outlinedandexplained.
2. Cortisolawakeningresponse
TheCARisexpressedaspartofnormal,healthyhumancircadian physiology.DeviationsfromatypicalCARpatternareassumedto markmaladaptiveneuroendocrineprocesses.Ageneralreviewof psychosocial,psychiatricandhealth-relatedcorrelatesoftheCARis beyondthescopeofthisarticle(reviews:ChidaandSteptoe,2009;
Clowet al.,2004;Fries et al.,2009;Kudielka and Wüst,2010).
However,asaprerequisiteforinterpretingsuchdata,somedis- tinctfeaturesoftheCARneedtobeacknowledged.Itisimportant toemphasizethat,althoughhistoricallytheterm‘CAR’hasbeen usedtodescribedifferentaspectsofpost-awakeningcortisolsecre- tion(includingoveralllevels),onlythedynamicofpost-awakening cortisolsecretionisaccuratelyreferredtoasthe‘CAR’,i.e.,corti- solchangesoccurringduetotheawakeningresponse(Clowetal., 2010).Arationaleforthisrecommendedterminologyislaidoutin Section2.2.
2.1. Descriptionanddistinctivefeatures
TheCARrepresentsasharpincreaseincortisollevelsacrossthe first30–45minfollowingmorningawakening.Inhealthyadults, themagnitudeoftheCARwasfoundtorangebetweena50and 156%increaseinsalivarycortisollevels(Clowetal.,2004).Fig.2a depictspost-awakeningcortisolprofilesofchildren,adolescents andelderlyadultsfromthefirstsystematicdescriptionoftheCAR byPruessneretal.(1997).
Theinitialfindingsuggestingthattheawakeningprocessstim- ulatescortisolsecretionprovidedanexplanationforpreviousdata ofpoortest–retestreliabilityinclock-timebasedcortisolassess- mentsduring theearlymorning hours(e.g., Costeet al., 1994;
SchulzandKnabe,1994)andsuggestedthatalignmentofcortisol samplingwith awakeningwould provide a more reliablemea- sure.Thisnotionwassoonsupportedbydatashowingimproved test–reteststabilityofawakening-alignedpost-awakeningcortisol levelsacrossabroadagerange(rsbetween.39and.67;Pruessner et al., 1997).These initialdata were viewed asindicating that theCARcouldbeusedasareliabletraitbiomarkerand,hence, investigationsoverthefollowingyearsmainlyfocusedoninter- individualvariabilityinCARprofilesusingcross-sectionaldesigns.
Such researchrevealedtheCAR toberelated tovarious physi- caland mentalhealth variables,albeitwithsomeinconsistency (review:ChidaandSteptoe,2009).Morerecentevidenceillustrated thattheCARalsoexhibitsconsiderableintra-individualvariabil- ity(see Section7.2).Indeed,although twinstudiesconsistently foundamoderateheritabilityoftheCAR(Kupperetal.,2005;Wüst etal.,2000a),theexpressionoftheCARonaparticulardayhas beenestimatedtobemoreinfluencedbystatefactorsthanbysta-
ble,trait-likeinfluences,includinggeneticfactors(Almeidaetal., 2009;Hellhammeretal.,2007;Stalderetal.,2010b).Buildingon thesedata,researchalsoincreasinglysetouttoinvestigatestate correlatesoftheCAR(review:Lawetal.,2013).
TheCARperiodisembeddedwithinawell-describedcircadian patternofcortisolsecretion,characterizedbyacortisolincrease priortoawakening,theCARperiodandadeclineofmeancorti- sollevelsovertheremainingdiurnalphase(Veldhuisetal.,1989;
Weitzmanetal.,1971).Importantly,thereisconvergingevidence suggesting that the CAR is relatively distinct from earlier and latercomponentsofcircadiancortisolsecretion.Sleeplaboratory researchrevealedthattheCARisnotamerecontinuationofthe pre-awakeningcortisol increase but comprises a superimposed responsetoawakening(Wilhelmetal.,2007;seeFig.2b).Inaddi- tion,theCARwasfoundtobeunrelatedtocortisollevelsduringthe remainderoftheday(Edwardsetal.,2001a;Mainaetal.,2009)or toalatenttraitcortisolfactorinferredfromvariousdiurnalsamples (Doaneetal.,2015).
Evidence showingthat in healthyhumans the CAR, but not laterdiurnalcortisolsecretion,issensitivetolightexposurefur- therillustratesitsdistinctnature:morningawakeningindarkness ordimlight reducesthedynamic of theCARrelative toawak- eninginlight (Figueiroand Rea,2012;Scheerand Buijs,1999).
Furthermore,winterawakeningusingadawnsimulator(gradually increasinglightlevelsbeforeawakening)hasbeenassociatedwith increasedpost-awakeningcortisolproduction(Thornetal.,2004).
Inrodentstudies,light-inducedeffectsonglucocorticoidsecretion wereabsentfollowinglesionsofthesuprachiasmaticnucleus(SCN) orinSCN-intactmicewithadrenalsympatheticdenervation(Buijs etal.,2003).Thus,anSCN-mediatedextra-pituitarypathwayhas beenimplicatedinregulationoftheCAR,butisunlikelytoaffect cortisolsecretionovertheremainderoftheday(review:Clowetal., 2010).
Besidesanuncouplingfrombasalcircadiancortisolsecretion, researchalso consistently revealed the CAR tobe unrelated to cortisolreactivitytoexperimentally-inducedpsychologicalstress (Bouma et al., 2009; Schmidt-Reinwald et al., 1999). This has important implications for the interpretation of CAR data, i.e., indicating itsdistinctness from cortisolreactivity toacute psy- chological stress. Interestingly,in an earlystudy, the CAR was foundtobecloselyrelated tothecortisolrisefollowing ACTH- challenge(r=.63),suggestingthatitsexpressionmaybeinfluenced bythemaximalcapacityoftheadrenalcortextoproducecortisol (Schmidt-Reinwaldetal.,1999).
Overall,thesedatahighlightthedistinctnatureoftheCARand suggestthatitsassessmentprovidesaddedinformationthatmay notbederivedfromothercortisolmeasures.Thistogetherwith findingsofuniqueassociationswithpsychosocial,cognitiveand health-relatedparametersmakestheCARaninterestingmeasure inPNEresearch.Conversely,thesedataalsoillustratethatwhen interpretingrespectivefindings,researchersneedtobecarefulnot tomistaketheCARforeitheramarkerofgeneralHPAaxisactiv- ity/basalcortisolsecretionorstress-reactivecortisolchanges(Clow etal.,2010).
2.2. Maincomponents
TheCARisa dynamicphenomenon triggeredbytheprocess ofmorningawakening.Strategiesforquantifyingpost-awakening cortisolsecretionneedtoaddresstwomainunderlyingcompo- nents:First,thestartingpointoftheCARperiod,i.e.,thefirstsample synchronizedwiththe momentof awakening(S1).Second,the actualdynamicofthecortisolincrease afterawakening,i.e.,the CARitself,assessedatsetintervalsafterawakening.Importantly, thetwocomponents(S1andCAR)areoftenfoundtobeinversely related,withalowerCAR followingahigher S1,and viceversa
Fig.2.(a)Mean(±SEM)post-awakeningcortisolvaluesforthreeagegroupsfromthefirstsystematicdescriptionoftheCAR(fromPruessneretal.,1997).(b)Sleeplaboratory datashowingtheCARtobesuperimposedonthepre-awakeningcortisolincrease(fromWilhelmetal.,2007).Bothfiguresreproducedwithpermission.
(Adametal.,2006;Bäumleretal.,2013;Huberetal.,2006;Stalder etal.,2009,2010b;Wilhelmetal.,2007;Wüstetal.,2000b).This relationshipcanlikelybeinterpretedasanillustrationofthelaw ofinitialvalue(Wilder,1962).
Clowetal.(2010)reviewedneurophysiologicalevidenceonthe regulationofmorningcortisolsecretionandconcludedthatdiffer- entialprocessesarelikelytobeimportantforthepre-awakening cortisolincrease andfor theCAR(Clowetal., 2010).Theythus suggestedthatseparateresultsshouldbereportedforS1(i.e.,the endpointofthepre-awakeningincrease)andestimatesoftheCAR.
Thisisalsorecommendedaspartofthisconsensusreport(seeSec- tion7.4).Inaddition,quantificationstrategiesareinuse, which combineinformationofS1andtheCAR,thusprovidinganindex ofoverallcortisolsecretionoverthepost-awakeningperiod(e.g., theAUCG,Pruessneretal.,2003).Whenusingsuchmeasures,itis importanttoacknowledgethattheyareinfluencedbybothunder- lyingcomponents(S1andtheCAR).Hence,itisimportanttorefer torespectivemeasuresasreflectingtotal‘post-awakeningcortisol concentrations’orsimilar,butnotasmeasuresoftheCAR(Clow etal.,2010).Inlinewiththisreasoning,graphicalillustrationsin thepresentarticlefocusondepictingS1andameasureoftheCAR, inthisinstancetheareaunderthecurvewithrespecttoincrease (AUCI,Pruessneretal.,2003).Adiscussionofstatisticalapproaches toquantifyingtheCARisprovidedinsection7.4.
3. Inaccuratesampling:prevalenceandimpact
ThevalidityofCARdatacriticallyreliesonthetemporalaccu- racyofsalivasamplingacrossthepost-awakeningperiod.Atypical samplingscheduleinvolvestakingafirstsampleimmediatelyafter awakeningfollowedbyrepeatedassessmentsatspecifiedtimes, e.g.,at10or15minintervalsoverthesubsequent30–60min.Fig.3a illustratesanexemplaryCARsamplingschedule.Failuretocomply withsuchaschedulecanoccurinmultipleways.Inthefollowing, wedistinguishbetweenparticipants(i)failingtocorrectlyreport theirawakeningtimeand/ordelayingtheinitiationofsamplingin relationtothemomentofawakeningand(ii)notadheringtothe specifiedtimeintervalsforlatersampling.
3.1. Delaybetweenawakeningandinitiationofsampling
Thecommencement ofsamplingimmediatelyafter awaken- ingiscrucialforaccuratelycapturingtheCAR.Table1aprovides an overview of studies examining the impact of delayed ini- tiation of samplingafter awakening. Thisresearch employed a rangeofmethods,suchasactigraphy,electrocardiography(ECG) or polysomnography (PSG), to verify participants’ self-reported timesofawakening(seeSection4.2.1foradescriptionofmeth-
ods).Inaddition,tworecentstudiesalsousedelectronicmonitoring devicestoverifytimesofsamplecollection(GriefahnandRobens, 2011;Smythetal.,2013).
Thefirstdescriptionofawakeningtime-relatedsamplinginac- curacywasmadeinapost-hocanalysiscarriedoutonasubgroup of individuals (13.1% of the total sample) who failed to show anyevidence ofa positive CAR (Kupperet al.,2005).By utiliz- ingavailableECGandactigraphydata,itwasrevealedthatthese participantsshowedameandelayof42min(range:10–135min) betweenverifiedandself-reportedawakeningtimes.Bycontrast, participantswithregularCARprofilesmostlyshowedgoodcor- respondencebetweenself-reportedandverifiedawakeningtimes (Kupper et al., 2005). Following these initial data, subsequent researchconfirmedthatfailuretocorrectlyreportthetimeofawak- eningand/ortodelaythebeginningofsamplingafterawakening isrelativelycommonandprofoundlyimpactsCARestimates(see Table1a).Acrossstudies,meanverifiedawakeningtimespreceded meanself-reportedawakeningtimesby3.3–6.2minandmeanself- reportedtimesofcollectingS1by7.1–24.8min(DeSantisetal., 2010;Dockrayetal.,2008;Okunetal.,2010).Aparticularlystriking illustrationofthepotentialextentofsuchinaccuracywasreported byGriefahnandRobens(2011)whoaccumulateddatafromthree studies,eachemployingcarefulobjectiveverificationofawakening andsamplingtimesacross6–8daysperindividual.Theyfoundthat participantsdelayedcollectingS1by3–30minon19.3%ofsampling daysandbyeven>30minon14.0%ofsamplingdays(Griefahnand Robens,2011).
However,delayingthecollectionofS1afterawakeningbymore than15minresultsinfalse-highestimatesofS1andfalse-lowesti- matesoftheCAR.Thispatternemergedbothfromresearchrelying onself-reportsofS1timing(DeSantisetal.,2010;Dockrayetal., 2008;Okunetal.,2010)andfromstudiesobjectivelymonitoring samplingtimes(GriefahnandRobens,2011).Fig.3canddexem- plifytheimpactof20and40minsamplingdelays,respectively,on estimatesofS1andtheCAR(AUCI).
TheimpactofsmallerdelaysinsamplingS1(<15min)hasbeen moredifficulttocapture.Earlierresearchindicatednodifferences inCARestimatesbetweenfullyaccurateindividuals(delays<1min) andthosewith1–15mindelays(Dockrayetal.,2008;Okunetal., 2010).Otherstudies,however,reportedatrendforanattenuated CAR inindividuals with5–15mindelays(DeSantiset al.,2010) orsuggestedthatCARestimatesalreadystartedtodecreasewith delaysexceeding∼10min(GriefahnandRobens,2011).Animpor- tantadditiontothesedatacomesfromrecentresearchbySmyth etal.(2013,2015):employingcarefulcontrolofawakeningand samplingtimesinhealthyparticipants(samplingat5mininter- vals),theirfindingsrevealedthatminordelays(5–15min)yielded estimatesofanincreasedCARandanearlierpeak.Thishasbeen
Table1 OverviewofresearchexaminingtheprevalenceandimpactofCAR-relatedsamplinginaccuracy. StudySample&designObjectivecontrolmethodSamplinginaccuracydescriptivesImpactonCAR (a)Objectivecontrolofawakeningtime(andsamplingtimes): Kupperetal.,2005N=59,subgroupwithnegative CAR,1dayAW:ECG&actigraphy Sampling:-Objectivevs.self-reportedAW:42mindelay (range:10–135min)NegativeCARacrossgroup Dockrayetal.,2008N=83elderly,suspectedCAD patients,1dayAW:Wristactigraphy Sampling:-Objectivevs.self-reportedAW:6.1±14.8min ObjectiveAWvs.S1:12.2±20.3minGreaterS1&no15–30minincreaseinpps. with>15mindelaybtw.objectiveAWandS1 Okunetal.,2010N=207,elderly, heterogeneous,1dayAW:PSG Sampling:-Objectivevs.self-reportedAW:3.3±53.1min ObjectiveAWvs.S1:24.8±32.2minGreaterS1&smallerCARinpps.with>15min delaybtw.objectiveAWandS1 DeSantisetal.,2010N=91,lateadolescents,3daysAW:Wristactigraphy Sampling:-Objectivevs.self-reportedAW:6.2±14.3min ObjectiveAWvs.S1:7.1±15.9minSmallerCARsinpps.with>15mindelaybtw. objectiveandself-reportedAW Griefahn&Robens,2011N=108(from3studies), heterogeneous,6–8daysAW:PSGoractigraphy Sampling:EMD/verifiedby experimenter(sleeplab) ObjectiveAWvs.objectiveS1:≤3mindelay: 66.7%days,>3and30mindelay:19.3%days, >30mindelay:14.0%days
SmallerCARondayswitha>∼10mindelay btw.objectiveAWandobjectiveS1 (descriptiveanalysis) Smythetal.,2013N=50(studyI),healthy females,4daysAW:Actigraphy Sampling:EMDObjectiveAWvs.objectiveS1:4mindelayGreaterCARondayswith5–15mindelay comparedto<5min (b)Objectivecontrolofsamplingtimes: Kudielkaetal.,2003N=47,community-dwelling subjects,1dayAW:- Sampling:EMDObjectivevs.self-reportedsampling:13±4 min(incl.diurnalsamples)aSmallerCARwithinaccuratesampling(S1: ≥10mindelay,30min:±7mindelay) Brodericketal.,2004N=66,femalefibromyalgia patientsandcontrols,7daysAW:- Sampling:EMDDelayedsamplingon80%days(incl.diurnal samples)aPresenceofCARondayswithaccurate sampling;noCARriseoninaccuratedays Kudielkaetal.,2007bN=83,elderly,3daysAW:- Sampling:EMD60%ofparticipantswithinaccuratesampling (anyCARsamplingoutsidespecified10min timewindow)onatleastoneday SmallerCARinparticipantswithinaccurate samplingon2(outof3)days Smith&Dougherty,2014N=81,preschool-agechildren (mainsubjects)andtheir parents,2days
AW:- Sampling:EMD44.3%samplinginaccuracy(anyCARsample outsidespecified10mintimewindow)of parentssamplingintheirchildren
GreaterS1&smallerCARinchildrenofparents withinaccuratesampling Goldenetal.,2014N=935,multi-ethnicsample, mixedgender,3daysAW:- Sampling:EMDSamplingdelay>15minin21.0%ofS1samples and31.2%of30minsamplesSmallerCARinparticipantswhoscoredinthe lowesttertileofoverallcompliance Abbreviations:AW,awakeningtime;btw,between;CAR,cortisolawakeningresponse;ECG,electroencephalography;EMD,electronicmonitoringdevice;pps.,participants;PSG,polysomnography;S1,firstsampleonawakening. aFurtherqualificationoffactorsdeterminingadherencerateswasprovided(seedetaileddiscussioninthetext).
Fig.3.IllustrationofanexemplaryCARprofileandtheimpactofdelayedbeginning ofsamplingafterawakeningontwocommonmeasuresofpost-awakeningcortisol secretion,i.e.,thefirstsampleonawakening(S1)andtheareaunderthecurvewith respecttoincrease(AUCI,Pruessneretal.,2003).Thefigureshowsestimatesgiven (a)correctsamplingandfordelaysof(b)8min,(c)20minand(d)40minbetween awakeningandcollectingS1.Latersamples(S2–S4)areshowntobecollectedinline withthespecified15minintervals.
accountedforbytheobservationthatcortisollevelsremainedrel- ativelyunchangedoverthefirst5–10minpost-awakening(‘latent period’),witha significantincrease firstbeingdetectableinthe 15minsample.Thesedatasuggestthatmoderatesamplingdelays shift the examined time window closer to the actual increase componentbyremovingthelatentperiodfromtheanalysis,thus resultinginhigherCARestimateswithanearlierpeakcomponent (Smythetal.,2013,2015).Fig.3billustratesthisnotionforanexem- plary8minsamplingdelay.
Inaccuracy in the commencement of sampling immediately after awakening can arise from a range of scenarios, includ- ing non-adherence due to motivational reasons (avoidance of discomfort, attending to other responsibilities, etc.). However, observationsbyGriefahnandRobens(2011)aswellasbySmyth etal.(2013)suggestedthatnon-motivationalfactorsmightalso
influenceawakening-relatedinaccuracies.Inbothstudies,consid- erablesamplingdelaysoccurredforS1eventhoughparticipants tooklatersamplesincloseaccordancewiththeprotocol.Thissug- geststhatdelayedsamplingafterawakeningmaybetheprimary causeofinaccurateCARassessmentandariseinwell-intentioned andotherwiseconscientiousparticipants(Smythetal.,2013).A potentialexplanationforthisistheoccurrenceofsleepinertiain theimmediatepost-awakeningperiod,i.e.,astateofreducedcogni- tiveandmotorperformance(TassiandMuzet,2000).Sleepinertia mayincreasethedifficultyofadheringtorequestedtimingsand/or mayimpedetheprecisedeterminationofthemomentwhenoneis fullyawake(Clowetal.,2010;Smythetal.,2013).
In sum,recent evidencesuggeststhat even well-intentioned participantsmay not alwaysbe able toprecisely identify their awakeningmoment.Thiscanleadtomoderatedelaysincollect- ingS1afterawakeningthataresufficienttosubstantiallybiasCAR estimates.Inlightofthis,objectiveverificationofawakeningtime isnecessaryforobtainingvalidCARdata.
3.2. Inaccuratepost-awakeningsampling
BesidesfailuretocollectS1immediatelyonawakening,inac- curacymayalsoarisefromdelaysatsubsequentsamplingtimes.
Table1bsummarizesdataonthecorrespondencebetweenself- reported and objectively verified times of saliva sampling in ambulatoryCARresearch.Twolandmarkstudiesfocusedonsam- pling accuracy during the post-awakening and the remaining diurnal samplingperiod(Broderick etal., 2004;Kudielka etal., 2003).Inaccuratesalivasamplingoccurredrelativelyfrequently andwasassociatedwithanunderestimationoftheCAR.Thisgen- eralpatternwaslaterconfirmedbyresearchspecificallyfocusing ontheCARinadults(Kudielkaetal.,2007b),inparentsobtaining CARsamplesoftheirpreschool-agechildren(SmithandDougherty, 2014)and ina largemulti-ethnicsample (Goldenet al.,2014).
Animportant qualitative extension of thesedata wasprovided byfindingsshowingthattheaccuracyofsalivasamplingcanbe considerablyimproved byinformingparticipantsaboutthefact thattheyarebeingobjectivelymonitored(Brodericketal.,2004;
Kudielkaetal.,2003).Thepotentialimplicationsofthislatterfind- ingarediscussedindetailinSection4.2.4.
4. Strategiesfordealingwithinaccuratesampling
Thefollowingsectionsdescribeavailableobjectivemonitoring strategiesforambulatoryCARresearch,discusswaysfordealing withidentifiedinaccuratedataandlookintopotentialstrategiesin lieuofobjectivemeasures.
4.1. Objectivemonitoringstrategies
AccurateassessmentoftheCARrequiresobjectiveverification ofawakeningandsamplingtimes(AdamandKumari,2009).Ide- ally,suchobjectivemethodsshouldbeemployedincombination withadiarylogsystemtorecordself-reportdataofawakeningand samplingtimes(besidesotherfactors,suchaspotentialcovariates;
seeSection6.3).
4.1.1. Methodsforverifyingawakeningandsamplingtimes
Severalmethods havebeenused toverifyawakeningtimes.
Polysomnography(PSG)isconsideredthegoldstandardinsleep research(VandeWateretal.,2011)andhasbeenusedforveri- fyingawakeningtimesinCARresearch(e.g.,Gribbinetal.,2012;
Griefahnand Robens, 2010,2011; Okunet al.,2010).However, PSGiscostly,labor-intensiveanddisruptivetoparticipants’nor- malroutines(VandeWateretal.,2011).Wristactigraphymight beamorereadilyobtainablemethodasitisminimally-disruptive,
relativelyinexpensiveandwell-validatedagainstPSGforassess- ingsleepparameters(e.g.,Coleetal.,1992;Lichsteinetal.,2006).
Wristactigraphyhasbeensuccessfully usedinambulatory CAR researchacrossseveralstudies(e.g.,Smythetal.,2013).Another actigraphy-basedapproachistheuseofchest-wornmotilitymon- itorsthatadditionallyrecordheart inter-beat-interval(IBI)data (CARresearch:Kupperetal.,2005;Stalderetal.,2011).Asarousing fromsleepisassociatedwithanincreaseinheartrate(e.g.,Huikuri etal.,1994)andrapidcardiovascularactivationlastingaroundten heartbeats(Trinderetal.,2001,2003),availableIBIdata(together withactigraphydata)couldfurtherhelptomorepreciselydeter- minetheawakeningmoment.Still,eachoftheabovedescribed methodsappearssuitablefortheverificationofawakeningtimein CARresearch.Inaddition,futureresearchmayexploretheuseof recentlydevelopedsmartphone-linkedorconsumer-branddevices aspotentiallowcostalternativesforobjectiveawakeningtimever- ification(review:Kellyetal.,2012).However,thisstronglyrestson thesuccessfulvalidationofsuchdevicesagainstawell-established method,suchasPSG,whichtodateismostlystilllacking(Kelly etal.,2012;Meltzeretal.,2015).
Concerningtheverification ofsamplingtimesin ambulatory research,thecommonlyusedelectronicmonitoringsystemshave provenuseful.Thesetypicallyusescrewtopbottles thatrecord timesofbottleopenings,however,boxesthatrecordtimestamps havealsobeendevised.Bystoringsalivasamplingdevicesinside thebottleandinstructingparticipantstorestrictopeningstothe timesofsampletaking,therespectivetimestampsprovideanindi- rectindexofsamplingtimes.Clearly,theuseofsuchsystemscannot fullyprotectagainstintentionalmisuse(e.g.,participantsmaystill takeoutsamplesfromthebottlewithoutperformingthesaliva sampling)butitdoespresentthecurrentbestpractice.Alternatives tothisapproachmightariseasaconsequenceofmoderntechnol- ogy.Forexample,smartphoneswithbuilt-incamerascouldbeused toobtaintime-stampedself-photographs(‘selfies’)byparticipants whencollectinga sampletoverify samplingaccuracyin future studies.Ifadequatelydeveloped,suchastrategycanequallyberec- ommendedfortheverificationofsamplingaccuracyaselectronic monitoringsystems.
Analternativeapproachthatremovestheneedforsampling time verification isthe useof automatedsamplingmethods to assesstheCAR.Forone,intravenousbloodsampling,whencou- pledwithstationaryPSGassessment(e.g.,Wilhelmetal.,2007), ensurestheaccuracyofCARassessment.However,asthisresearch istypicallyrestrictedtothesleeplaboratory,itsartificialsettingis associatedwithreducedecologicalvalidity.Thismaybeprevented byrecentlydevelopedsystemsforautomatedsamplingofsubcuta- neoustissuefreecortisol(Bhakeetal.,2013).Althoughclearlymore demandingandinvasiveforparticipantsthantheself-collectionof salivasamples,thisapproachcouldpotentiallyallowtheassess- ment ofthe CAR in participants’ homesettingsin some future research.
4.1.2. Dealingwithverifiedinaccuratedata
Onceinformationonsamplingaccuracyhasbeenobtained,it canbeusedtoreducebiasonCARestimatesthrough(i)dataexclu- sionstrategiesand(ii)statisticalmodelingapproaches.
Fordataexclusionstrategies,theextentofinaccuracyisusu- allyfirstcalculatedasthediscrepancybetweenthescheduledand theactual/verifiedsamplingtime(t).Thet-valueofindivid- ualsamplingtimesisthencomparedtoapredeterminedaccuracy margin(e.g.,5,10or15min)and,incaseanytexceedsthismar- gin,CARdatafortherespectivesamplingdayareexcludedfrom subsequentanalyses(e.g.,Kudielkaetal.,2003).Whenusingsuch anapproach,decidingonthemostsuitableaccuracymarginfor dataexclusion is difficult.Proceedingfromtheabove reviewed findings(particularly:Smythetal.,2013,2015),evensmalltime
discrepanciesmayentailsubstantialbiasonCARestimates,unless theybecomenegligible(i.e.,t≈0min).However,narrowingaccu- racy margins causes a growing loss of (putativelyinformative) data.Consequently,anyconsensusaboutafixedaccuracymargin isnecessarilyatrade-offbetweenscientificprecisionandpractical feasibility.Forexample,previousresearchemployingawakening timeverificationbywristactigraphysuggeststhatspecifyingan accuracy margin oft=0±5minfor S1 will yielddata loss of 26–46%(DeSantisetal.,2010;Dockrayetal.,2008).Inaddition, theselectiveexclusionofparticipantswithinaccuratesampling mayresultinpotentialselectionbiasandreducedgeneralizability ofresults.Inordertokeepthepercentageofclassifiedinaccurate data(andthusdatalossandpotentialbias)aslowaspossible,itis crucialtoemployafullrangeofmeasurestomaximizeadherence (seeSection5).
Inthesecondgroupofstrategies,verifiedinaccuratedataare notexcludedbutinsteadtheobjectiveinformationonactualsam- plingtimesisutilizedforthecalculationofCARestimates(i.e.,these dataareincorporatedintothestatisticalmodel).Hence,thispro- videsamoreeconomicalapproach,preventingunwanteddataand participantloss,andresultingconcernsregardingreducedgener- alizability.Tousesuchastrategy,statisticalmodelsarerequired thatadequatelydescribethetemporaldynamicsofcortisolsecre- tionacrosstheCARperiod.Section7.4providesadescriptionof suchmodellingapproachesinCARresearch.
Insum,thedecisionaboutthemostadequatestrategyinvolves tradingoffconsiderationsaboutscientificprecisionagainstthose ofpracticalfeasibility.Researchers’primaryconcernshouldbeto obtainvalid,unbiaseddata.Theuseofawell-specifiedstatistical modeloftheCARthatincorporatesverifiedawakeningandsam- pling time datafulfils this criterion and should bethemethod of choice. Whenusing data exclusion strategies, achieving any confidence thatCAR dataare notbiased requiresthe specifica- tionofrelativelystrictaccuracymarginswhich,unfortunately,is associatedwithdataloss.Toachievecomparabilitybetweenstud- ies,werecommend that futureresearchemploysa transparent approach whereby it is clearlystated whetherfindings emerge whenapplyingastrictaccuracymarginoft=0±5minforeach post-awakeningsample,eitherasthesoleapproachorincombina- tionwithresearchers’ownanalyticalstrategy(i.e.,asanadditional sensitivityanalysis).
4.2. Arethereviablestrategieswithouttheuseofobjective measures?
Theuseofobjectivemonitoringstrategiesincreasesthecosts perparticipantandmaythusreducethenumberofparticipants fromwhomendocrinedatacanbeobtained(AdamandKumari, 2009).Thismayleadresearcherstoconsiderwhetheralternative strategiesexistthatyieldvalidCARdatawithouthavingtoemploy objectivemeasures.
4.2.1. Forcedawakening
Adesign-based strategytocounteract problems ofsampling inaccuracyistoexternallyawakenparticipants(usuallythrough study personnel). This is a work-intensive approach that has beenusedwithparticipantsexaminedinahospitalsetting(e.g., Huberetal.,2006;NicolsonandVanDiest,2000),sleeplabora- tory(Wilhelmetal.,2007)orquarantinedaspartofalargerstudy (Polketal.,2005).Recently,avariationofthisapproachhasbeen employedininfantsandyoungchildrenwhoweretooyoungto samplesalivathemselvesandwerethuswokenupbytheirparents toensuretheaccuracyofsamplinginitiation(Bäumleretal.,2013;
Bäumleretal.2014a;Bäumleretal.2014b;Stalderetal.,2013).
Inthelatterstudies,thiswasfurthercomplementedbyobjective verificationofawakeningandsamplingtimes.
Anargumentinfavorofaforcedawakeningapproachisthat currentevidencesuggeststhattheCARis unaffectedbypartici- pants’modeofawakening(spontaneousvs.externallywoken;e.g., byalarmclock;Stalderetal.,2009;Wüstetal.,2000b).Thismakes itunlikelythatforcedawakeningleadstofundamentallydifferent CARprofilesthanspontaneousawakening.However,aremaining dangeristhatparticipantsmaywakeuppriortotheplannedwake uptime.Hence,aforcedawakeningapproachshouldstillbecom- plementedbyobjectiveawakeningtimeverification.Undersuch acondition,forcedawakeningmayhelptoyieldhighqualityCAR data,particularlyifstudypersonnelalsocontinuetomonitorthe accuracyofsubsequentsalivasampling.Thelatterpossibilitymay thensparetheuseofelectronicmonitoringdevicestoverifysam- plingaccuracy(seeSection4.2.1).
Besidesissuesrelatedtosampletiminginaccuracy,however, researchersassessingtheCARinahospitalorsleep-laboratoryset- tingneedtobeawareofthepossibilityofstate-relatedconfounding whichcaninducesignificantbiasinCARanalyses(seeSection6.2).
Further,asmentionedbefore,suchconditionsareassociatedwith reducedecologicalvalidity,which isa keyadvantageofsample collectioninparticipants’homesettings.
4.2.2. Increasingstatisticalpower
Researchersmaywonderwhetherproblemsofsamplinginac- curacycannotbeovercomebysimplyincreasingstatisticalpower, e.g., in large-scaleepidemiological research.Indeed, this would bethecaseifinaccuracyoccurredrandomly,i.e.,notsystemati- callyrelatedtorelevantparticipantcharacteristics(cross-sectional research)orsituationalfactors(intra-individualresearch).Under suchcircumstances,inaccuratesamplingwould merelyincrease theerror ofCAR estimates, whichcould betackledbyincreas- ingthenumberofobservations.However,extensivedataindicate that non-adherence, a factor which is likely to strongly affect samplinginaccuracy,doesnotoccurrandomlybutcovarieswith relevantpsychologicalfactors.Forexample,researchhasshown thatadherencetomedicaltreatmentsregimensisinfluencedby individual differences in depressiveness and/or social support (meta-analyses:DiMatteo,2004;DiMatteoetal.,2000).Research focusingontheCARalsoconfirmedaninverserelationshipbetween perceivedsocialsupportandsamplinginaccuracy(Kudielkaetal., 2007b).In a largemulti-ethnic study,inaccurate samplingof a diurnal cortisolprofile(including post-awakeningsamples) was relatedtolowerincome,educationlevels,and(marginally)eth- nicity (Golden et al., 2014). Furthermore, effects of sampling accuracyhavebeenfoundtointeractwithhealthstatus,i.e.,female fibromyalgiapatientswerelessinfluencedbybeinginformedabout theuseofobjectivemonitoringstrategiesthanhealthy controls (Brodericketal.,2004).Thisindicatesthatinaccuratesamplingis likelytoco-varywithparametersthatareofcentralinteresttoPNE inquiry,i.e., psychologicalorhealth-related factors.Under such circumstances,failingtocontrolforsamplingaccuracyposesthe eminentthreatthattruerelationships maybeobscuredorfalse relationshipsmaybeaccepted.Inthiscase,“(i)ncreasingtheNor thenumberofsamplescollectedwillyieldthesamelevelanddirection oferror.Infact,increasingstatisticalpowerwouldonlyincreasethe researcher’sconfidenceinafalseresult.”(Brodericketal.,2004;p.
648).
Thereislessevidenceonstatecorrelatesofsamplinginaccu- racy fromintra-individualCAR research.Brodericket al.(2004) observednodifferencesinsamplingaccuracybetweenweekdays and weekends, a factor frequentlyassociated with altered CAR profiles(Kunz-Ebrechtetal.,2004b;Schlotzetal.,2004).Still,it isclearlyconceivablethatinaccuratesamplingmayco-varywith statepsychologicalfactorsrelevant for PNEresearch(e.g.,state arousalorstress,prospectivememoryload,sleepcharacteristics;
Lawetal.,2013).Hence,itcannotbeexcludedthatfailuretoobjec-
tivelycontrolforinaccuratesamplingconfoundsintra-individual CARdata.Again,thisproblemcannotbealleviatedbyincreasing statisticalpower.
4.2.3. ExclusionofCARnon-responders
Inaccuratesamplinghasoftenbeenassociatedwithflattened orevennegativeCARprofiles(e.g.,Brodericketal.,2004;Dockray etal.,2008;Kupperetal.,2005).Thishasledtotheproposition thatissuesofsamplinginaccuracymaybeaddressedbyexclud- ingparticipantswhofailtoshowa cortisolincreasefromS1to latersamplesastheseare‘suspectednon-adherents’(Thornetal., 2006).However,thisisunlikelytobeasufficientapproach.The effectsofinaccuratesamplingonCARestimatesarelikelytobenon- linearandcontinuous(seeFig.3b–d):comparedtofullyaccurate sampling,smalldelaysafterawakeningmayfirstresultinanover- estimationoftheCARwhichthenturnsintothewell-documented underestimationoftheCARwithlongerdelays(>15–20min).The complete absenceofa post-awakeningincrease (i.e.,anegative CAR)islikelytooccuronlyifthedelaybetweenawakeningand theinitiationofsamplingexceedsthepeakoftheunderlyingCAR (between30 and45min,seeFig.3d).Theexclusionofnegative CARprofileswouldthuseliminateonlyextremecasesofinaccu- ratesamplingbutnotmildormoderatecases,whichalreadyhave thepotentialtosubstantiallybiasresults.
Inaddition,theexclusionofCARnon-respondersmaybyitself inducebias.Itisstillunresolvedwhether,givenfullyaccuratesam- pling,therearegenuineoccurrencesofparticipantsnotshowinga positiveCAR.Preliminaryevidencesuggeststhatthismayindeed bethecase:inaccuratelysampleddatabasedonobjectivemoni- toring,noincreaseoronlyaminorCAR(<2.5nmol/L,Wüstetal., 2000b;or<1.5nmol/L;Milleretal.,2013a)emergedon13.1%of studydays ininfants(Stalder etal., 2013), on18.0% ofdays in toddlersandyoungchildren(Bäumleretal.,2013), inadultson 14.7%(Dockrayetal.,2008),andon19.7%ofdaysinhealthypar- ticipants(Smythetal.,2013).Ofnote,patientswithbrainlesions, particularlyinthehippocampalformation(Buchananetal.,2004;
Wolfetal.,2005),appeartoshowmoregenerallyattenuatedor evenabsentCARs(seealsoSection6.3).Overall,thesedataindicate thatabsentCARsmayrepresentgenuinephenomenathatsimply formthelowerendofadistributionofCARmagnitudes.Following thisassumption,absentCARsshouldoccurmorefrequentlywithin groupsthatasawholeexhibitareducedCARprofile.CARresearch ofteninvestigatesgroupdifferences(e.g.,betweenclinicalpatients andcontrolsubjects),thusbydefinitiontryingtoprovethatone grouphasalowerCARthananother.Hence,aCARnon-responder exclusionstrategywouldbiasdatabysystematicallyexcludinga greaterpercentageoflowvaluesfromonegroupthanfromanother.
4.2.4. Informationalstrategies
Inresearchexaminingdiurnalcortisollevels(includingpost- awakeningsampling),participantswhowereinformedabouttheir samplingaccuracybeingverifiedbyelectronicmonitoringsystems showeda76%reductionincumulatedsamplingdeviationscom- paredto‘non-informed’participants(Kudielka etal.,2003)and moredayswithaccuratesampling(informed:90%,non-informed:
71%;Brodericketal.,2004).Importantly,besidesimprovingthe actual rates of sampling accuracy, ‘informed’ participants also tendedtocorrectlyself-reporttheirsamplingtimes,evenifthey hadnotsampledaccurately(Brodericketal.,2004;Kudielkaetal., 2003).Subsequentresearchwithparticipantsbeinginformedabout objective monitoring mostly confirmed high accuracy of post- awakeningsamplingintheseindividuals (Griefahnand Robens, 2011;Smythetal.,2013)althoughthiswasnotthecaseinarecent studyonalargemulti-ethnicsample(seeTable1b;Goldenetal., 2014).
Table2
Strategiesformaximizingparticipantadherencewiththesamplingschedulea. Face-to-facestrategies&materials
Increasingparticipantmotivation:
- Engageparticipantswiththeresearchgoalsandclearlyexplaintheimportanceofstrictadherencetothem - SpecificallyemphasizetheimportanceofcollectingS1immediatelyuponawakening
- Informparticipantsaboutthefactthattheirsamplingaccuracyisobjectivelymonitored - Establishappropriatesamplingdatesthatareconsideredas‘convenientandtypical’
- Allowparticipantstoaskquestions
Increasingtheclarityoftheprocedure:
- Gothroughtheprotocolwithparticipantsindetail - Ifpossible,practicerelevantpartswithparticipants
- Provideadditionaltake-homeinstructions(writtenformorasinstructionalvideo/DVD) - Adviseplacingsamplekit(withapen)besidebedthenightbeforesamplecollection - Defineandexplainwhatismeantbythe‘themomentofawakening’
- Makecollectionkituserfriendlyandpopulationappropriate(e.g.,largeprintinelderly) - Helpparticipantstoorganizethecollection(e.g.,usecolorcodedsamplingdevices)
Additionalstrategies
- Calloremailtheeveningbeforesamplingtohighlightinstructionsandremindaboutstartingthenextmorning - Useofautomatedremindertoolsthatareactivatedbyparticipantsonawakening
- Reminderphonecallsortextmessagingatthesamplingtimesthroughresearchpersonnel(onlyappropriateforpopulationswithwell-definedawakening times)
aStrategieswerepartlyadoptedfromAdamandKumari(2009).
Still,theabovedatamayindicatethepossibilitythatmerely informingparticipantsabouttheuseofobjectivecontrolmethods couldprovideastrategyforobtainingreliabledataonsampling accuracy(i.e.,throughself-reports).Inthiscontext,‘mock’strate- giescouldbeconsidered,withparticipantsbeingtoldthatobjective monitoringstrategiesarebeingusedwithoutthisactuallybeing thecase. Besidesethicalconsiderations, it isimportant tonote that theefficacy of such an approach hasnot been tested yet.
Thisisnottrivial astheeffectivenessofa mockcompared toa real‘informed’strategymaybereducedbyseveralroutes,suchas non-verbaltransmissionoflowerexpectationsfromexperimenters toparticipants (givenexperimenters’ awareness that noobjec- tivemonitoringisbeingused)orpassingonofinformationabout thenon-functionalityofobjectivemonitoringbetweenparticipants (e.g.,instudentpopulations).Apotentialsolutionmaybean‘open’
strategyaspartofwhichobjectivemonitoringisemployedinaran- domsubgroup,whileallparticipantsaretoldthatthereisachance ofbeingmonitored(AdamandKumari,2009).Whilethisavoids deceivingparticipants,itisunclearwhetherinformationaboutthe merechanceofbeingmonitoredisequallyeffectiveascertainty aboutthisfact.Hence,withoutfirmevidenceshowingtheeffec- tivenessofsuch anapproach,it isrecommendedthat objective monitoringisemployedacrossallparticipantsandisnotsubsti- tutedwithaninformationalstrategy.Notwithstanding,evidence clearlysuggeststhatparticipantsshouldalwaysbeinformedabout theuseofobjectivemonitoringstrategies.
5. Maximizingadherence
Irrespectiveof objectivemonitoring, it is expedient towork towards maximizing participant adherence. Such strategies are cost-efficientastheypreventdatalossthroughtheexclusionof inaccurately sampled data and increase data quality (i.e., fully adherentdataaresuperiortostatisticallyinferred/correcteddata).
Table2listsstrategiesformaximizingadherenceinCARresearch.
Severalof thesestrategiesarederived fromtheauthors’collec- tiveresearchexperience,withoutformalpublishedevidenceon effectivenesstesting.
Animportantopportunityforincreasingparticipantadherence isprovidedbytheinitialface-to-facemeeting.Strategiesemployed duringthis meetingmaybothraise participants’motivation for beingadherentandhelptoincreasetheclarityofthestudypro- cedure.Animportantwaytoraisemotivationistryingtoengage participantswiththeresearchgoals.Besidesconveyingthegen- eralpurposeofthestudy,thisinvolvesexplainingtheimportance ofbeingadherentinCARresearchandtheconsequencesofnon- adherence.Toensurethatparticipantsfullyunderstandthewhole studyprocedure, itis consideredimportantthat researchersgo throughthe protocolin detail withthem andpractice relevant components(e.g.,thesalivasamplingprocedure).Aspartofthis, itshouldbeexplained preciselywhatismeantbythe‘moment ofawakening’ inorder tostandardize thiscritical aspectacross participants(Adam&Kumari,2009).Werecommendthatsucha definitionshouldfocusontheregainingofconsciousnessasthe centralcharacteristicoftheawakeningmoment(e.g.,“Whenyou areawake, i.e.,you areconscious: youknow whoandwhere you are;youareinastatethatisclearlydifferentfromwhenyouwere sleepingeventhoughyoumaystillfeeltired.”).Inaddition,itshould bemadeclearthatparticipantsshouldnotinitiatesamplingafter prematurenightlyawakenings(e.g.,“Ifyouwakeduringthemid- dleofthenightandplantogobacktosleep,donotbeginsampling;
pleaseonlybeginwhenyouareawakeforthefinaltimebeforeyou plan to getup for the day.”)and that theyshould refrainfrom dozingorsnoozing duringtheCAR samplingperiod(e.g.,“Dur- ingthisstudy,please donot fallback tosleepor‘doze’afteryour initialawakening.Youcanstayinbedorgetoutofbedbutplease stayawake(evenifyouarenotfullyalert)duringandafterthesaliva samplingperiod.”).Besidesusingtheinitialface-to-facemeeting toclarifysuch criticalquestions aboutsample timing,it isalso importantthatappropriatesamplingdatesarenegotiatedbythe researcher and participantand agreed as‘convenientand typi- cal’.
Besidesface-to-facecontact,take-homeinstructionsinwritten formshouldbeprovided.Forsomepopulations,theadditionaluse ofinstructionalDVDshasprovenuseful(e.g.,Stalderetal.,2013).
Overall,itisimportanttomakeinstructionsasexplicitandpractice- orientatedaspossible,e.g.,participantsmaybetoldtoplacethe
samplekitandapenbesidethebedbeforegoingtosleeptoavoid post-awakeningdelaysthroughhavingtosearchforthematerial (AdamandKumari,2009).Strategiesthatmakethecollectionkit moreuserfriendlyandhelpparticipantsorganizethecollection (e.g.,colorcodingofmaterial)arealsodeemedhelpful.Researchers havefurtherhadpositiveexperienceswithusingreminderphone calls,emails,ortextmessagesontheeveningpriortosampling (e.g.,Smythetal.,2013).Besidesremindingparticipantsofimpor- tant procedures (e.g., towear actigraphydevices tobed), such measuresalsosignalanextraeffortmadebytheresearchteam, thusagainhighlightingtheimportanceofaccuratelyfollowingthe studyproceduretoparticipants.Inaddition,recentstudieshave employedmethodsforremindingparticipantsabouttimesofpost- awakeningsamplecollection.Theseincludeautomatedstrategies, e.g.,reminderwatches(Franzetal.,2013),remindingthroughpar- ticipants’mobilephones(Garcia-Bandaetal.,2014)orelectronic reminders,e.g.,timersthatareactivatedbyparticipantswhentak- ingS1andthenbeep/flashatthelatersamplingtimes(e.g.,Doane andAdam,2010;GriefahnandRobens,2011).Inaddition,reminder phonecallsortextmessagingatparticipants’individuallypredicted samplingtimeshasbeenemployed(e.g.,Oskisetal.,2009).While suchstrategiesmayincreaseadherencebypreventingagainstthe forgettingof sampling,theycannot providecertaintyaboutthe accuracyofsampling.Hence,theyshouldonlybeviewedascom- plementaryapproachesbutcannotreplaceobjectivemonitoring strategies.
6. Dealingwithcovariates
Researchershavetodealwiththefactthathormonesecretionis relatedtoalargenumberofstateandtraitfactors(Schlotz,2011).
Dependingontheresearchcontext,thesecovaryingfactorsmaybe consideredconfounders,mediators,moderatorsordirectvariables ofinterest(AdamandKumari,2009;Kudielkaetal.,2012;Schlotz, 2011).Ifacovariateisnotofmaininterest,themostcriticalques- tioniswhetheritconfoundsobservedassociations(Schlotz,2011).
ConfoundingisgivenwhenacovariateisrelatedtoboththeCAR andthevariable(s)ofinterest,thuscreatingaspuriousrelationship betweenthem,andneedstobeaddressed.However,evenifafac- torisonlyrelatedtotheCARbutnottoothervariablesofinterest, thismayincreasetheerrorvarianceofthemodelandthusreduce statisticalpowerfordetectingassociationswiththeCAR(Schlotz, 2011).Strategiesforpreventingunwantedinfluencesofcovariates inambulatoryPNEresearchcanbegroupedintoinstructional,sta- tisticaladjustmentandexclusionstrategies(Kudielkaetal.,2012;
Schlotz,2011).
6.1. Instructionsaboutpost-awakeningbehavior
Besidesinformingparticipantsaboutthenecessity tocollect samplesincloseaccordancewiththespecifiedsamplingtimes(Sec- tions3–5),furtherinstructionsmayaddressparticipantbehavior overthepost-awakeningperiod.Table3providesanoverviewof factorstobeconsideredinthiscontext.Themostcommoninstruc- tionshavebeenforparticipantstotakenilbymouthotherthan water,refrainfromsmokingandomitcleaningtheirteeth(toavoid abrasionandvascularleakageintosaliva)untilafterthefinalsam- pling(Clowetal.,2004).Thereissupportforaninfluenceofthefirst twofactors:Cortisolsecretionisknowntobeacutelyinfluenced bycaffeineandnicotineintake(review:Kudielkaetal.,2009),food consumption(particularlyhighproteinfoods;e.g.,Gibsonetal., 1999;Rosmondetal.,2000)andbloodglucoselevels(Rohleder andKirschbaum,2007).Thissuggeststhatbreakfasting(incl.caf- feinatedorsugareddrinksand/orprotein-richfoods,e.g.,eggs)or smoking duringthepost-awakening phase mayaffect theCAR.
Table3
Guidelinesforinstructingparticipantsontheirbehavioroverthemorningperiod.
Factor: Recommendation:
Specificmodeofawakening(e.g.,onlybyalarmclock) Norestriction Untilhavingfinishedthemorningsamplingroutine:
Drinkingandeating:
- Caffeinateddrinks Disallowa
- Sugareddrinks Disallowa
- Food/breakfast Disallowa
Smoking Disallowa
Toothbrushing/dentalhygiene Norestriction
Physicalactivity:
- Remainsupine Norestriction
- Restrictmovement Norestriction
- Physicalexercise Disallowa
Important:participantsshouldalwaysbeencouragedtoreportonengagingin anyofthesebehaviors.
aDependingontheresearchcontext,itmayalsobejustifiabletoallowengage- mentinthesebehaviorsandtostatisticallyadjustforthisaccordingly.
By contrast,toothbrushingis atleastunlikely tobeassociated withstronggroupeffectsassalivarycortisollevelswerefoundto beunaffectedbynormaldentalhygiene(Gröschletal.,2001)or evenvigoroustoothbrushing,despitethelatterleadingtoblood leakageintosaliva(Kivlighanetal.,2004).Researchinchildren, usingsalivarytransferrinlevelsasamarkerofbloodcontamination, alsoconcurredwiththegeneralnotionthatbloodcontamination throughdentalhygieneisunlikelytohaveastrongeffectonsalivary cortisollevels(Grangeretal.,2007).
Similarly,currentfindingsspeakagainstaninfluenceofphysi- calbehavior/activitylevelsinthenormalrangeontheCAR,which hasbeenfoundtobeunaffectedbyposturalchanges(i.e.,remain- ing supine vs.standing/behaving normally; Hucklebridgeet al., 2002;Wilhelmetal.,2007)orthelevelofmotilityoverthepost- awakeningperiod(Stalderet al.,2009).However,thisdoesnot applytophysical exercising,which isknowntoinducecortisol reactivitywhenperformedaboveacertainintensitylevel(e.g.,Hill etal.,2008;KirschbaumandHellhammer,1994).Finally,partici- pants’modeofawakening(spontaneousvs.alarmclock)hasbeen foundunrelatedtoexpressionoftheCAR(Stalderetal.,2009;Wüst etal.,2000b).Still,astheabovestudiesdidnotcontrolforthebeta error,itcannotbeexcludedthatsmalleffectsexistbutwerenot detectedintherespectivestudysamples.
Together, an effect onthe CAR is particularly suggested for eating,drinking(caffeinated orsugared beverages), smokingor engaginginphysicalexerciseduringthepost-awakeningperiod.
Concerningthesebehaviors,inmostresearchcontextsitisrec- ommended that researchers (i) instruct participants to abstain fromthesebehaviorsuntilaftertheyhavefinishedpost-awakening sampling. Alternatively, (ii) in case researchers feel that these restrictionsimposeatoosevereburdenonparticipants’ normal routines(i.e.,reducingwillingnesstoparticipateand/orecological validity),participantsmaybeallowedtoengageinthesebehaviors butshouldthenbestronglyencouragedtoreportthissystemati- cally(e.g.,throughthediarylogsystem).Thisiscriticaltofacilitate subsequent statisticaladjustment for such potential influences.
Thelatterpointalsoappliestothosebehaviorswithoutaproven influenceontheCAR(modeofawakening,dentalhygiene,moder- atephysicalactivity),forwhichitisstillrecommendedtoobtain self-reportdata.Furthermore,ininstanceswhenparticipantsare allowed toeat, drinkand/or brush their teeth,they shouldbe instructedtorinse theirmouthafterwards andtoabstainfrom engagingin thesebehaviorsin theimmediateperiod (1–2min) beforesampling.