randomization methods
DESCRIPTION
clinical trail methodsTRANSCRIPT
-
RandomizationinClinicalTrialsVersion1.0May2011
1. SimpleRandomization2. Blockrandomization3. Minimizationmethod
Stratification
RELATEDISSUES
1. AccidentalBias2. SelectionBias3. PrognosticFactors4. Randomselection5. Randomallocation
-
TwoIndependentGroups
Research isoften conductedusing experimentation involving themanipulationof a least
onefactor.Commonlyusedexperimentaldesigns includetheexperimentaldesignsforthe
twogroupproblem forsuperiority (i.e. isonetreatmentsuperiortoanotheronaprimary
outcome?),fornoninferiority(isanewtreatmentnoworsethananexistingtreatmentona
particularoutcome?),orforequivalence(aretwo interventionsessentially identicaltoone
anotheronaprimaryoutcome?).Strongconclusionsmaybedrawnfromsuchendeavours
providing the design is sufficiently powered and providing there are no threats to the
validity of conclusions. The use of randomization in experimental work contributes to
establishingthevalidityofinferences.
The following notes relate to the experimental comparison of two treatments with
participants randomized tooneof two treatments;suchdesignscanviewedasbeingRCT
designs[RCT===RandomizedControlTrialifonetreatmentisacontrolormoregenerally
RCT===RandomizedClinicalTrial].
SimpleRandomSamplesandRandomAllocation
Simplerandomselectionandrandomallocationarenotthesame.Randomselectionisthe
processofdrawinga sample fromapopulationwhereby the sampleparticipantsarenot
known inadvanced.ASimpleRandomSampleofsizek isasampledeterminedbychance
wherebyeachindividualinthepopulationhasthesameequalprobabilityofbeingselected
andeachpossiblesubsetofsizekinthepopulationhasthesamechanceofbeingselected.
Itishopedthatasimplerandomsamplewillgiveasamplethatisarguablyrepresentativeof
the population, and in doing so helpwith the external validity or generalizability of the
results.
RandomAllocation
Random allocation is a procedure in which identified sample participants are randomly
assignedtoatreatmentandeachparticipanthasthesameprobabilityofbeingassignedto
anyparticulartreatment.IfthedesignisbasedonNparticipantsand aretobeassigned1n
1 Randomization
-
to Treatment 1 then all possible samples of size have the same probability of being
assignedtoTreatment1.
1n
Example Purely forsimplicityofexpositionsuppose thereareN=4participants [Angela,
Ben,Colin,Dee],twoofwhomaretobeassignedtoTreatment1andtwotoTreatment2.
ThepossiblegroupsthatcouldbeassignedtoTreatment1are;1.[Angela,Ben],2.[Angela,
Colin],3.[Angela,Dee],4.[Ben,Colin],5.[Ben,Dee],6.[Colin,Dee].Rollingafairsixsided
diewouldbeonewayofperformingtherandomallocation.Forinstance,ifthedielandson
thenumber3 thenAngelaandDeewouldbeassigned toTreatment1andBenandColin
wouldbeassignedtoTreatment2.
The above is theway amethodologistwould consider random allocation. However the
abovedoeshaveitspracticaldrawbacks.ForinstancesupposeweconsideracaseofN=60
participantsand30aretobeassignedtoTreatment1andtheother30toTreatment2.For
theseparametersthereare155,117,520possibledifferentsamplesofsize30whichcould
assigned toTreatment1. Who in their rightmindwouldwriteout the listofallpossible
155,117,520combinations? Findingadiewith155,117,520sidesmightbedifficulttoo![A
computercouldbeusedtorandomlygenerateanumberfromtheintegers1to155,117,520
toselectasample.]
As described, random allocation can have practical problems but logically equivalent
pragmatic solutions exist (e.g. names in a hat with the first drawn out allocated to
Treatment1andtheremaindertoTreatment2).
1n
WhyRandomlyAllocate
Suppose two treatments, Treatment A and Treatment B are to be compared. Further
suppose the sample forTreatmentAareallmenand the sample forTreatmentBareall
female.Ifatanalysisadifferenceintheprimaryoutcomebetweenthetwogroupsisfound,
couldwe thenemphaticallyattribute thisdifferenceasa treatmenteffect? Clearlyunder
thisdesigntheanswertothatquestionwouldbeNo.Underthisdesignitcouldbeargued
that the effectmight be due to Sex, or to Treatment, or in fact bothmight affect the
2 Randomization
-
outcomeandtheiruniqueeffectscannotbedetermined.Inthisdesign,SexandTreatment
arecompletelyconfoundedandtheirseparateeffectscannotbeidentified.Inthisdesignan
argument foracausal effectdue to treatmentwouldnot standclose scrutinybecausea
plausiblealternativeexplanationforanydifferenceisevident.
In any practical situation the participants will have some (identifiable or unidentifiable)
characteristics thatmay be related to the outcome under investigation. A clustering of
thesecharacteristicswithanyonetreatmentcouldcauseasystematiceffect(asystematic
bias)betweenthegroupswhich isquitedistinct fromanytreatmenteffect (i.e.wewould
have a confounding effect). In the long run, random allocation will equalise individual
differences between treatment groups and in doing sowill remove extraneous bias and
allow the treatment effect to be established uncontaminated by other potentially
competing explanations. In any one experiment it is hoped that random allocationwill
minimisetheeffectofpossibleconfounders,reducingextraneoussystematicbias,leadingto
afaircomparisonbetweentreatmentsbyreducingthepossibilityofpartialconfoundingand
hencehelpingtoruleoutotherpotentialcompetingcausalexplanations.
Thedatageneratedunderanexperimentaldesignwill,mostlikely,beassessedusingformal
statisticalmethods.Thetheoryunderpinningpermutationtestsandrandomisationtestsis
based on the assumption of random allocation. Accordingly valid and defendable data
analysisplansmaybedevisedifrandomallocationisused.
SimpleRandomization
Themostcommonlyencounteredsituationinpracticeisatwotreatmentcomparisonwitha
predeterminedoverallsamplesizeNwithapredeterminedsamplesizeof forTreatment
1andsize forTreatment2( + =N).AtotalofNopaqueenvelopes, containingan
identifierforTreatment1and containinganidentifierforTreatment2,maybeshuffled.
Theorderoftheshuffledenvelopesdeterminestheallocationofparticipantstotreatments.
This process is relatively simple to organise, preserves the predetermined design
parameters,andcanbereadilyextendedtosituationswheremultipletreatmentsaretobe
compared.
1n
1n2n 1n
2n
2n
3 Randomization
-
SimpleSequentialRandomization
Acommonlyencounteredsituation isatwogroupcomparisonwheresamplesizes and
are required to be equal or approximately equal. In a two group trial, the process is
analogous to the tossofa coin such thateachparticipanthasanequalprobability tobe
allocated to either of the treatments. When the sample size is relatively large, simple
randomization is expected to produce approximately equally sized treatment groups
however this isnotguaranteedand thegeneral recommendation is toonly consider this
approachwhereoverallsamplesizeis200orabove.
1n
2n
PossibleProblemswithSimpleRandomization
Simple randomization reduces bias by equalising some factors that have not been
accounted for in theexperimentaldesigne.g.agroupofpeoplewithahealth condition,
different from the disease under study,which is suspected to affect treatment efficacy.
Anotherexample is thata factor suchasbiological sex couldbean importantprognostic
factor. Chance imbalances or accidental bias, with respect to this factor may occur if
biologicalsexisnottakenintoaccountduringthetreatmentallocationprocess.Anexample
ofaperfect randomizationwith respect togenderasan importantprognostic factor isas
showninFigure1.Figure2depictsanexamplewherethereisaccidentalbiaswithrespect
tobiologicalsex.
4 Randomization
-
30%
20%
30%
20% Treatment/Males
Treatment/Females
Control/Females
Control/Males
Figure1:Noaccidentalbias:perfectbalancewithrespecttoSex
10%
40%
30%
20%
Treatment/Males
Treatment/Females
Control/Females
Control/Males
Figure2:Accidentalbias:imbalancewithrespecttoSex
5 Randomization
-
Itmaybearguedthatrandomization istoo importanttobe lefttochance! Inthesecases
some practitioners may argue for a blocked randomization scheme, or a stratified
randomizationscheme,oronewhichdeliberatelyminimisesdifferencesbetweengroupson
keypredeterminedprognosticfactors.
BlockRandomization
Blockrandomization iscommonlyused inthetwotreatmentsituationwheresamplesizes
for the two treatments are to be equal or approximately equal. The process involves
recruitingparticipantsinshortblocksandensuringthathalfoftheparticipantswithineach
blockareallocatedtotreatmentAandtheotherhalftoB.Withineachblock,however
theorderofpatientsisrandom.
Conceptually there are an infinite number of possible block sizes. Supposewe consider
blocks of size four. There are six different ways that four patients can be split evenly
betweentwotreatments:
1.AABB,2.ABAB3.ABBA,4.BAAB,5.BABA,6.BBAA
Thenextstepistoselectrandomlyamongstthesesixdifferentblocksforeachgroupoffour
participants thatare recruited.The random selection canbedoneusinga listof random
numbers generated using statistical software e.g. SPSS, Excel, Minitab, Stata, SAS. An
exampleofsucharandomnumbersequenceisasshown;
9795270571964604603256331708242973...
Since there areonly sixdifferentblocks, allnumbersoutside the rangeof1 to6 canbe
droppedtohave;
52516464632563312423...
6 Randomization
-
Blocks are selected according to the above sequence. For example the first eighteen
subjectswouldbeallocatedtotreatmentsasfollows:
5 2 5 1 6
BABA ABAB BABA AABB BB
In the example, one group has two more participants than the other; but this small
differencemaynotnecessarilybeof great consequence. Inblock randomization there is
almostperfectmatchingofthesizeofgroupswithoutdepartingtoofarfromtheprincipleof
purely random selection. Note however this procedure is not the same as simple
randomization e.g. the first fourparticipants cannotbe all allocated to TreatmentA and
henceallpossiblecombinationsofassignmentarenotpossible.Notethatsimplesequential
randomizationisthesameasblockrandomizationwithblocksofsize1.
Stratification
Astratificationfactorisacategorical(ordiscretizedcontinuous)covariatewhichdividesthe
patientpopulationaccordingtoitslevelse.g.
sex,2levels:Male,Female
age,3levels:
-
Minimization
Usingthismethod,thefirstpatientistrulyrandomlyallocated;foreachsubsequentpatient,
the treatmentallocation is identified,whichminimizes the imbalancebetween groups at
thattime.Forexample,considerasituationwherethereare3stratificationfactors;sex(2
levels),age (3 levels),anddiseasestage (3 levels).Supposethereare50patientsenrolled
andthe51stpatientismale,age63,andstageIIIdisease.
TreatmentA TreatmentB
Sex Male 16 14
Female 10 10
Age
-
Therearetwopossiblecriteria:
Countonlythedirection(sign)ofthedifferenceineachcategory.TreatmentAisahead
intwocategoriesoutofthree,soassignthenextpatienttoTreatmentB
Addthetotaloverallcategories(26Asvs24Bs).SinceTreatmentA isahead,assignthe
nextpatienttoTreatmentB
Thesetwocriteriawillusuallyagree,butnotalways
Bothcriteriawillleadtoreasonablebalance
Whenthereisatie,usesimplerandomization
Balancebymarginsdoesnotguaranteeoveralltreatmentbalance,orbalancewithin
stratumcells
ProblemsandAdditionalBenefitsofRandomization
With some methods of allocation an imbalance due to the foreknowledge of the next
treatment allocation between the treatment groups with respect to an important
prognostic factor may also occur i.e. when an investigator is able to predict the next
subjects group assignment by examining which group has been assigned the fewest
patients up to that point. This is known as selection bias and occurs very oftenwhen
randomization is poorly implemented e.g. Preprinted list of random numbers can be
consultedbyanexperimenter beforenextpatientcomes in,orenvelopescanbeopened
beforenextpatientcomesin,oranexperimentercanpredictthenextallocationwithstatic
methodse.g.thelasttreatmentineachblockinblockrandomisation.
A practical issue in experimentation is allocation concealment, which refers to the
precautionstakentoensurethatthegroupassignmentofpatients isnotrevealedpriorto
definitively allocating them to their respective groups. In other words, allocation
concealment shields thosewho admitparticipants to a trial from knowing theupcoming
9 Randomization
-
10 Randomization
assignmentsandcanbeachievedbycodingthetreatmentsandnotusingtheirrealnames.
Thisiscalledmaskingandisdifferentfromblinding:
Masking or allocation concealment seeks to prevent selection bias, protects assignment
sequencebeforeanduntilallocation,andcanalwaysbesuccessfullyimplemented.
Incontrast,blindingseekstopreventsamplingbias,protectssequenceafterallocation,and
cannot always be successfully implemented. For example, it is impossible to implement
blindinginasituationwherethetreatmentisasurgicalprocedure.
Ingeneralsimplerandomallocation,aswellashavingverydesirabletheoreticalproperties
from a statistical perspective, often permits the implementation of very desirable
methodologyincludingmaskingandcangreatlyhelpwithpreservingtheoverallconclusions
fromexperimentalresearch.