Determining Sample Size - EANS · PDF fileFeasibility studies ... Summary – determining sample size ... traditional soap and water bed baths without proper evaluation of

10-07-2016

1

Determining Sample Size considera3ons for interven3on studies & surveys

Dr.BetsievanGaal

Introduc3on Sample Sizes

• Whyisitimportanttoconsidertherequiredsamplesizeforastudy?

EuropeanAcademyofNursingScience–2016 2

10-07-2016

2

Introduc3on Sample Sizes

• Whyisitimportanttoconsidertherequiredsamplesizeforastudy?

EuropeanAcademyofNursingScience 2hHp://freshspectrum.com

Objec3ves of this lecture

• UnderstandingoftheprinciplesofsamplesizesintermsofevaluaLngcomplexintervenLons

• UnderstandingofdifferentmethodsofcalculaLngsamplesizesappropriateforevaluaLngcomplexintervenLons

• UnderstandingthecontribuLonofclustersincalculaLngsamplesizes

EuropeanAcademyofNursingScience 3

10-07-2016

3

What are important concepts to know

• ResearchpopulaLon• Asample•  Samplescheme• Design

EuropeanAcademyofNursingScience 4

Researchproblem

Researchques0on Methodology

Importance of good sample sizes

• RepresentaLveness:doesyoursamplerepresentthepopulaLon?

•  SelecLonbias:over/underrepresenLngofsomesubgroups

• GeneralisaLon:canyougeneralisetheresultsofthesampletothepopulaLon?

EuropeanAcademyofNursingScience 4

PopulaLons&SamplesThefocusisonapopulaLonthoughwecanonlyexaminesamples

10-07-2016

4

Sample size methods

1.  Probabilitysampling–aselect/arandomsample

2.  Non-probabilitysampling–select/anonrandomsample??

EuropeanAcademyofNursingScience 4

Sample size: 1. probability sampling

When?•  ListofthewholepopulaLon• GeneralisaLonisimportant

• OYenquanLtaLvestudies

EuropeanAcademyofNursingScience 4

10-07-2016

5

Sample size: 1. probability sampling

1.  Asimplerandomsampling2.  StraLfiedrandomsampling3.  Clusterrandomsampling4.  SystemaLcrandomsampling

EuropeanAcademyofNursingScience 4

Sample size: 1. probability sampling 1 a simple random sampling

• AllpeopleinthepopulaLonhavethesamechangetobeselected•  Forexample:

•  PaLents•  Students

EuropeanAcademyofNursingScience 4

Polit,D.F.&Beck,C.T.(2012,p.208).Nursingresearch.Genera/ngandassessingevidencefornursingprac/ce(ninthediLon).Philadelphia:LippincoHWilliams&

Wilkins.

Small table of random digits

10-07-2016

6

Sample size: 1. probability sampling 2 Stra3fied random sampling

• PopulaLonsdividedintosubpopulaLons:strata•  Forexample:socialeconomicstatus/school

EuropeanAcademyofNursingScience 4

Burns,A.C.&Bush,R.F.(2011).Principesvanmarktonderzoek.ToepassingenmetSPSS.Amsterdam:PearsonBeneluxB.V.

Sample size: 1. probability sampling 3 cluster random sampling

• PopulaLonsdevidedintosubpopulaLons:strata•  Forexample:socialeconomicstatus/school

EuropeanAcademyofNursingScience 4

CenterforstaLsLcsUniversityofHasselt,Belgium

10-07-2016

7

Sample size: 1. probability sampling 4 Systema3c random sampling

• WholepopulaLons

EuropeanAcademyofNursingScience 4

CenterforstaLsLcsUniversityofHasselt,Belgium

Summary: 1. probability sampling

• GeneralizetothepopulaLon•  LesschangeofselecLonbias,butnoguaranty•  LesspracLcal–Costly&Lmeconsuming

EuropeanAcademyofNursingScience 4

10-07-2016

8

Sample size: 2. non probability sampling

When?• GeneralisaLontothewholepopulaLonisnotimportant• Notenoughresources:money,Lme•  TracingthepopulaLonisdifficult???

• UsedforquanLtaLve&qualitaLvestudies

EuropeanAcademyofNursingScience 4

Sample size: 2. non probability sampling

1.  Conveniencesampling2.  Snowballsampling3.  Quotasampling4.  Purposefulsampling

EuropeanAcademyofNursingScience 4

10-07-2016

9

Sample size: 2. non probability sampling 1 Convenience sampling

•  Alsocalledaccidentalsampling•  Surveyamongstudentsofanuniversityyouworkwith•  Surveyamongnursingofahospitalyouknow•  Surveyamongpeoplewhoareinoneshoppingarea

EuropeanAcademyofNursingScience 4

Sample size: 2. non probability sampling 4 Snowball sampling

•  Alsocalledchainsampling

EuropeanAcademyofNursingScience 4

hHp://www.wordpress.com

10-07-2016

10

Sample size: 2. non probability sampling 3 Quota sampling

•  Aconveniencesamplingwithinastrata•  StrataisdefinedbydataofthepopulaLon

•  Gender•  EducaLonlevel•  Medicaldisease

EuropeanAcademyofNursingScience 4

Strata

Conveniencesampling

Sample size: 2. non probability sampling 3 Quota sampling

EuropeanAcademyofNursingScience 4

Strata Popula0on Conveniencesampling Quotasampling

Men 100(20%) 5(5%) 20(20%)

Women 400(80%) 95(95%) 80(80%)

Total 500(100%) 100(100%) 100(100%)

10-07-2016

11

Sample size: 2. non probability sampling 2 Purposeful sampling

•  When:•  Selectspecificpeopleinyourstudy•  ForexampletesLnganewintervenLon

EuropeanAcademyofNursingScience 4hHp://www.memrise.com

Summary: 2. non probability sampling

• ArepracLcal• BeawareofselecLonbias• ComparethesamplewiththepopulaLon

EuropeanAcademyofNursingScience 4

10-07-2016

12

Calcula3ng sample sizes

Why?• PracLcalconsideraLons:limitedbudget/subjects•  Toensureameaningfulresultofthestudy• Avoidsstudiesthatare

•  toosmall(underpowered)•  solargethattheydetecteffectsthatarenotrelevant(overpowered)

EuropeanAcademyofNursingScience 4

Calcula3ng sample sizes

Because:•  ThefocusisoYenonapopulaLonthoughtypicallywecanonlyexaminesamples

•  ThetaskofstaLsLcsisto:•  EsLmatetheproperLesofapopulaLonfromthoseofthesample•  EsLmatetheeffectofanintervenLoninapopulaLonfromtheeffectinasample

EuropeanAcademyofNursingScience 4

TalkwithyourstaLsLcianaboutcalculaLngthesamplesize

10-07-2016

13

Calcula3ng sample sizes

Factorsthatinfluencethepowerofastudy• Desiredpowerlevel–ErrorsintesLnghypothesis:TypeI/II•  EsLmatedeffectsize• Designofthestudy• Measurementlevel•  Theoutcome• Analysistechniques

EuropeanAcademyofNursingScience 4

Calcula3ng sample sizes

EuropeanAcademyofNursingScience 4

Decisionbasedonsta0s0caltestUnknownifhypothesisistrue Notreatmenteffect

Treatmenteffectexists

Notreatmenteffect Correctdecisionp=1–α

TypeIerrorp=α

Treatmenteffectexists

TypeIIerrorp=β

Correctdecisionp=1–β=Power

TheoutcomeofastaLsLcaltestandtheirprobabiliLes

PowerCorrectlyrejectafalsehypothesis

10-07-2016

14

Example: Washing Without Water study

EuropeanAcademyofNursingScience 4

Aims •  To compare skin damage •  To compare costs

Interven

0on

Tradi0onalBathing

WashingwithdisposableWashinggloves

Control

Work in groups

ReadthesummaryoftheWashingWithoutWaterstudyandanswerthenextquesLonswithyourgroup.1.  Whatsamplesizemethodshasbeenusedinthatstudy?2.  WhatinformaLondoyouneedtocalculatethesamplesizebefore

thestartofthisstudy?

EuropeanAcademyofNursingScience 4

10-07-2016

15

Work in groups

1.  Samplesizemethod:

2.  NeededinformaLontocalculatethesamplesize:

EuropeanAcademyofNursingScience 4

Sample size & the MRC framework

EuropeanAcademyofNursingScience 4

DevelopmentIdenLfyingtheevidencebaseIdenLfyingordevelopingtheoryModellingprocessandoutcomes

Feasibilityandpilo0ngTesLngproceduresEsLmaLngrecruitmentandretenLonDeterminingsamplesize

Evalua0onAssessingeffecLvenessUnderstandingchangeprocessAssessingcosteffecLveness

Implementa0onDisseminaLonSurveillanceandmonitoringLongtermfollow-up(CraigPetal.2008)

10-07-2016

16

Sample size & the MRC frame work Development

•  IdenLfytheimportanteffect–theoutcome–andhowtomeasures• MakingplausibleesLmateoflikelyeffect•  IdenLfyingexisLngstudiesfromwhichesLmatesoflikelyeffectandsamplingvariabilitycanbedrawn

•  EsLmaLngtheprevalenceofaproblem•  Findingstudysubjects•  Impact/returnoninvestment

EuropeanAcademyofNursingScience 4

Sample size & the MRC frame work Feasibility studies

•  InsufficientpriorinformaLontocalculatesamplesizerequired…• Gatheringdataaboutvariabilityininstrumentsetc.fordefiniLveevaluaLon:

•  Assessingfeasibilityofrecruitmentanddropoutsthatwilleffectsamplesize•  EsLmatesofvariability(SD)ofthepopulaLon•  IdenLfyingbaselinerates

•  TesLngstudyprocedures

EuropeanAcademyofNursingScience 4

10-07-2016

17

Sample size & the MRC frame work Defini3ve evalua3on

• PowercalculaLonrequiredinadvancefordefiniLvestudy• Mustdefinemainendpointandtheprecisionrequired(smallesteffectthatisimportant):

•  ‘Posthoc’powercalculaLons(whatthepowerwas,aYerthestudy)aremeaningless

EuropeanAcademyofNursingScience 4

Some specific considera3ons

• Cluster(randomised)studies•  StandardtechniquesdonottakeintoaccountcorrelaLonsbetweenindividualswithinclusters

•  Largersamplesarerequired

• CosteffecLveness•  VariaLonoccurson2parameters:costandeffect•  EffectonsamplesizevariesdependingonthestrengthanddirecLonofthecorrelaLon

EuropeanAcademyofNursingScience 4

10-07-2016

18

Summary – determining sample size

•  Inthislectureshowedthe• DifferentmethodstosampleyourstudypopulaLonfromthepopulaLons

•  TheimportanceofapropersamplesizecalculaLon•  TheimportancetomeetyourstaLsLciananddecidewhatthebestsamplesizeisforyourstudyas

•  Onwsizedoesnoetfitall,aseelding

EuropeanAcademyofNursingScience 4

LectureSamplesize: WorkinginGroups 12–07–2016

WashingWithoutWater–study

Readthesummaryofthisstudyandanswerthenextquestions:

1. Whatsamplesizemethodshadbeenusedinthisstudy?

2. Whatinformationdoyouneedtocalculatethesamplesizebeforethestartofthisstudy?

Background:No-rinsedisposablewashglovesareincreasinglyimplementedinhealthcaretoreplace

traditionalsoapandwaterbedbathswithoutproperevaluationof(cost)effectiveness.

Objectives:Tocomparebedbathsforeffectsonskinintegrityandresistanceagainstbathingandcosts.

Design:Clusterrandomizedtrial.Randomizationwasperformedpriortobaselinedatacollectionandat

thelevelofnursinghomewards.Residentswithinthesamewardwereconsideredtobeacluster

Setting&Participants:Fivehundredadultcare-dependentresidentsfromnursinghomewards.

Methods:Theexperimentalcondition‘washingwithoutwater’consistsofabedbathwithdisposable

washglovesmadeofnon-wovenwaffledfibers,saturatedwithano-rinse,quicklyvaporizingskincleaning

andcaringlotion.Thecontrolconditionisatraditionalbedbathusingsoap,water,washclothsand

towels.Bothconditionswerecontinuedfor6weeks.

Theprimaryoutcomeforthisstudywasprevalenceofskindamage,definedasclinicalsymptomsof

intertrigo,(contact)dermatitis,orcandidiasis.Wedistinguishedtwolevelsofseverityofskindamage:

anyskinabnormality/lesionandsignificantskinlesions:1)Anyskinabnormalities/lesionswereconsidered

presentwhenatleastoneofthefollowingsymptomswereobserved:brightreddiscoloration;erythema;

white,greenoryellowdiscolorationofthewoundbed;atrophicandshinyskin;satellitelesionsandkissing

lesions,fissures,erosionsorulcerations.Thesesymptomsshouldbepresentonthebuttocksoroneofthe

following10skinfolds:eyes,neck,armpits,elbows,sub-mammaryregion,umbilicus,abdomen,groins,

analcleft,andbetweentoes.2)Significantskinlesionswereconsideredpresentiftheskinwasnotintact

onthebuttocksoroneofthefollowingfourskinfolds:sub-mammaryregion,abdomen,groins,andanal

cleft.

Additionaloutcomes:resistanceduringbedbaths,costs.

Results:Anyskinabnormalities/lesionsovertimedecreasedslightlyintheexperimentalgroup,and

increasedslightlyinthecontrolgroup,resultingin72.7%vs77.6%ofresidentshavinganyskin

abnormalities/lesionsafter6weeks,respectively(p=0.04).Therewerenodifferencesinsignificantskin

lesionsorresistanceafter6weeks.