IntJAyurvedaRes.2010JanMar1(1):5557.

PMCID:PMC2876926
doi:10.4103/09747788.59946

Samplesizecalculation
PrashantKadamandSupriyaBhalerao1

DepartmentofClinicalPharmacology,SethGSMedicalCollegeandKEMHospital,Parel,Mumbai400012,India
1
DepartmentofClinicalPharmacology,TNMCandBYLNairHospital,MumbaiCentral,Mumbai400008,India
Addressforcorrespondence:Dr.SupriyaS.Bhalerao,DepartmentofClinicalPharmacology,TNMCandBYLNairHospital,MumbaiCentral,
Mumbai400001,India.Email:supriya.bhalerao@gmail.com

CopyrightInternationalJournalofAyurvedaResearch

ThisisanopenaccessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense,whichpermitsunrestricteduse,
distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited.

INTRODUCTION

Oneofthepivotalaspectsofplanningaclinicalstudyisthecalculationofthesamplesize.Itisnaturally
neitherpracticalnorfeasibletostudythewholepopulationinanystudy.Hence,asetofparticipantsis
selectedfromthepopulation,whichislessinnumber(size)butadequatelyrepresentsthepopulationfrom
whichitisdrawnsothattrueinferencesaboutthepopulationcanbemadefromtheresultsobtained.Thisset
ofindividualsisknownasthesample.

Inastatisticalcontext,thepopulationisdefinedasthecompletesetofpeople(e.g.,Indians),thetarget
populationisasubsetofindividualswithspecificclinicalanddemographiccharacteristicsinwhomyou
wanttostudyyourintervention(e.g.,males,betweenages45and60,withbloodpressurebetween140
mmHgsystolicand90mmHgdiastolic),andsampleisafurthersubsetofthetargetpopulationwhichwe
wouldliketoincludeinthestudy.Thusasampleisaportion,piece,orsegmentthatisrepresentativeofa
whole.

ATTRIBUTESOF ASAMPLE

Everyindividualinthechosenpopulationshouldhaveanequalchancetobeincludedinthesample.
Ideally,choiceofoneparticipantshouldnotaffectthechanceofanother'sselection(hencewetryto
selectthesamplerandomlythus,itisimportanttonotethatrandomsamplingdoesnotdescribethe
sampleoritssizeasmuchasitdescribeshowthesampleischosen).

Thesamplesize,thetopicofthisarticle,is,simplyput,thenumberofparticipantsinasample.Itisabasic
statisticalprinciplewithwhichwedefinethesamplesizebeforewestartaclinicalstudysoastoavoidbiasin
interpretingresults.Ifweincludeveryfewsubjectsinastudy,theresultscannotbegeneralizedtothe
populationasthissamplewillnotrepresentthesizeofthetargetpopulation.Further,thestudythenmaynot
beabletodetectthedifferencebetweentestgroups,makingthestudyunethical.

Ontheotherhand,ifwestudymoresubjectsthanrequired,weputmoreindividualstotheriskofthe
intervention,alsomakingthestudyunethical,andwastepreciousresources,includingtheresearcherstime.

Thecalculationofanadequatesamplesizethusbecomescrucialinanyclinicalstudyandistheprocessby
whichwecalculatetheoptimumnumberofparticipantsrequiredtobeabletoarriveatethicallyand
scientificallyvalidresults.Thisarticledescribestheprinciplesandmethodsusedtocalculatethesamplesize.

Generally,thesamplesizeforanystudydependsonthe:[1]

Acceptablelevelofsignificance
Powerofthestudy
 Expectedeffectsize
Underlyingeventrateinthepopulation
Standarddeviationinthepopulation.

Somemorefactorsthatcanbeconsideredwhilecalculatingthefinalsamplesizeincludetheexpecteddrop
outrate,anunequalallocationratio,andtheobjectiveanddesignofthestudy.[2]

LEVEL OF SIGNIFICANCE

Everyoneisfamiliarwiththepvalue.Thisisthelevelofsignificanceandpriortostartingastudyweset
anacceptablevalueforthisp.Whenwesay,forexample,wewillacceptap<0.05assignificant,wemean
thatwearereadytoacceptthattheprobabilitythattheresultisobservedduetochance(andNOTduetoour
intervention)is5%.Toputitindifferentwords,wearewillingtoacceptthedetectionofadifference5out
of100timeswhenactuallynodifferenceexists(i.e.,getafalsepositiveresult).Conventionally,thep
valueof5%(p=0.05)or1%(p=0.01),whichmeans5%(or1%)chanceoferroneouslyreportinga
significanteffectisaccepted.

POWER

Sometimes,andexactlyconversely,wemaycommitanothertypeoferrorwherewefailtodetecta
differencewhenactuallythereisadifference.ThisiscalledtheTypeIIerrorthatdetectsafalsenegative
difference,asagainsttheonementionedabovewherewedetectafalsepositivedifferencewhenno
differenceactuallyexistsortheTypeIerror.Wemustdecidewhatisthefalsenegativeratewearewillingto
accepttomakeourstudyadequatelypoweredtoacceptorrejectournullhypothesisaccurately.

Thisfalsenegativerateistheproportionofpositiveinstancesthatwereerroneouslyreportedasnegativeand
isreferredtoinstatisticsbytheletter.Thepowerofthestudythenisequalto(1)andisthe
probabilityoffailingtodetectadifferencewhenactuallythereisadifference.Thepowerofastudy
increasesasthechancesofcommittingaTypeIIerrordecrease.

Usuallymoststudiesacceptapowerof80%.Thismeansthatweareacceptingthatoneinfivetimes(thatis
20%)wewillmissarealdifference.Sometimesforpivotalorlargestudies,thepowerisoccasionallysetat
90%toreduceto10%thepossibilityofafalsenegativeresult.

EXPECTEDEFFECT SIZE

Wecanunderstandtheconceptofeffectsizefromdaytodayexamples.Iftheaverageweightloss
followingonedietprogramis20kgandfollowinganotheris10kg,theabsoluteeffectsizewouldbe10kg.
Similarly,onecanclaimthataspecificteachingactivitybringsabouta10%improvementinexamination
scores.Here10kgand10%areindicatorsoftheclaimedeffectsize.

Instatistics,thedifferencebetweenthevalueofthevariableinthecontrolgroupandthatinthetestdrug
groupisknownaseffectsize.Thisdifferencecanbeexpressedastheabsolutedifferenceortherelative
difference,e.g.,intheweightlossexampleabove,iftheweightlossinthecontrolgroupis10kgandinthe
testgroupitis20kg,theabsoluteeffectsizeis10kgandtherelativereductionwiththetestinterventionis
10/20,or50%.

Wecanestimatetheeffectsizebasedonpreviouslyreportedorpreclinicalstudies.Itisimportanttonotethat
iftheeffectsizeislargebetweenthestudygroupsthenthesamplesizerequiredforthestudyislessandifthe
effectsizebetweenthestudygroupsissmall,thesamplesizerequiredislarge.Inthecaseofobservational
studies,forexample,ifwewanttofindanassociationbetweensmokingandlungcancer,sinceearlier
studieshaveshownthatthereisalargeeffectsize,asmallersamplewouldbeneededtoprovethiseffect.If
ontheotherhandwewanttofindouttheassociationbetweensmokingandgettingbraintumor,wherethe
effectisunknownorsmall,thesamplesizerequiredtodetectanassociationwouldbelarger.

UNDERLYING EVENT RATEINTHEPOPULATION

Theunderlyingeventrateoftheconditionunderstudy(prevalencerate)inthepopulationisextremely
importantwhilecalculatingthesamplesize.Thisunlikethelevelofsignificanceandpowerisnotselectedby
convention.Rather,itisestimatedfrompreviouslyreportedstudies.Sometimesitsohappensthatafteratrial
isinitiated,theoveralleventrateprovestobeunexpectedlylowandthesamplesizemayhavetobeadjusted,
withallstatisticalprecautions.

STANDARDDEVIATION(SDOR )

Standarddeviationisthemeasureofdispersionorvariabilityinthedata.Whilecalculatingthesamplesizean
investigatorneedstoanticipatethevariationinthemeasuresthatarebeingstudied.Itiseasytounderstand
whywewouldrequireasmallersampleifthepopulationismorehomogenousandthereforehasasmaller
varianceorstandarddeviation.Supposewearestudyingtheeffectofaninterventionontheweightand
considerapopulationwithweightsrangingfrom45to100kg.Naturallythestandarddeviationinthisgroup
willbegreatandwewouldneedalargersamplesizetodetectadifferencebetweeninterventions,elsethe
differencebetweenthetwogroupswouldbemaskedbytheinherentdifferencebetweenthembecauseofthe
variance.Ifontheotherhand,weweretotakeasamplefromapopulationwithweightsbetween80and100
kgwewouldnaturallygetatighterandmorehomogenousgroup,thusreducingthestandarddeviationand
thereforethesamplesize.

SAMPLESIZECALCULATION

Thereareseveralmethodsusedtocalculatethesamplesizedependingonthetypeofdataorstudydesign.
Thesamplesizeiscalculatedusingthefollowingformula:

2 2,
2(Z a + Z 1 )
n =
2

wherenistherequiredsamplesize.For

Z,Zisaconstant(setbyconventionaccordingtotheacceptederrorandwhetheritisaonesidedortwo
sidedeffect)asshownbelow:

ForZ1,,Zisaconstantsetbyconventionaccordingtopowerofthestudyasshownbelow:

Intheabovementionedformulaisthestandarddeviation(estimated)andthedifferenceineffectoftwo
interventionswhichisrequired(estimatedeffectsize).

Thisgivesthenumberofsampleperarminacontrolledclinicaltrial.

EXAMPLE

ThisissueoftheJournalhasanarticledescribingthebenefitsofayurvedictreatmentAyTPinpatientsof
migraineinanopenuncontrolledtrialdesign.[3]Ifanyonewishestoconfirmtheseresultsusinga
randomizedcontrolledtrialdesignwheretheeffectoftheayurvedicinterventionwillbecomparedto
standardofcareinheadacheasmeasuredbyVAShowwouldweplanthesamplesize?

Asseenabove,weneedthefollowingvalues:Z,Z1,,standarddeviation(estimated),and,the
differenceineffectoftwointerventions.Letusassumewewillacceptap<0.05asacceptableandastudy
with80%powerusingtheabovetables,wegetthefollowingvalues:Z,is1.96(inthiscasewewillbe
usingatwotailedtestbecausetheresultscouldbebidirectional).Z1,is0.8416.Thestandarddeviation
(basedonthedatainthepublishedpaper)wouldbeapproximately0.7.For,thepaperdescribesthatthe
ayurvedictherapyhasgivena35%effect.Previouslyithasbeenreportedthatsumatriptanat50mg
improvesheadacheby50%.[4]Thus,theeffectsizewouldbe15%(i.e.,0.15).

Thesamplesizeforthenewstudywillbe

2 2
 2 2
2(1. 96 + 0. 8416) (0. 72)
n =
2
(0. 15)

=362perarm.

Calculatingfora10%dropoutrateonewouldneedtocompleteapproximately400patientsperarmtobe
abletosaywithanydegreeofconfidencewhetheradifferenceexistsbetweenthetwotreatments.

LIMITATIONSOF THECALCULATEDSAMPLESIZE

Thesamplesizecalculatedusingtheaboveformulaisbasedonsomeconventions(TypeIandIIerrors)and
fewassumptions(effectsizeandstandardvariation).

ThesamplesizeALWAYShastobecalculatedbeforeinitiatingastudyandasfaraspossibleshouldnotbe
changedduringthestudycourse.

Thesamplesizecalculationisalsotheninfluencedbyafewpracticalissues,e.g.,administrativeissuesand
costs.

Footnotes
SourceofSupport:Nil

ConflictofInterest:Nonedecleared

REFERENCES

1.KirbyA,GebskiV,KeechAC.Determiningthesamplesizeinaclinicaltrial.MedJAust.
2002177:2567.[PubMed:12197821]

2.LarsenS,OsnesM,EidsaunetW,SandvikL.Factorsinfluencingthesamplesize,exemplifiedbystudies
ongastroduodenaltolerabilityofdrugs.ScandJGastroenterol.198520:395400.[PubMed:3875139]

3.PrakeshB,BabuSR,SureshkumarK.ResponseofAyurvedictherapyinthetreatmentofmigraine
withoutaura.IntJAyurvedaResearch.20101:2935.

4.CadyRK,SheftellF,LiptonRB,O'QuinnS,JonesM,PutnamG,etal.Effectofearlyinterventionwith
sumatriptanonmigrainepain:Retrospectiveanalysesofdatafromthreeclinicaltrials.ClinTher.
200022:103548.[PubMed:11048903]

FiguresandTables

error 5% 1% 0.1%
2sided 1.96 2.5758 3.2905
1sided 1.65 2.33

Power 80% 85% 90% 95%

ArticlesfromInternationalJournalofAyurvedaResearchareprovidedherecourtesyofMedknow
Publications