RANDOMISATION, BIAS AND BLINDING IN CLINICAL TRIALS

Dipesh Mistry & Jessica SmithSeptember 2010

OVERVIEWOVERVIEW

• Randomisation-What, why, how and when

• Allocation Concealment

• Bias

• Blinding

WHAT IS RANDOMISATION?WHAT IS RANDOMISATION?

DEFINITIONRandomisation is a process by which each participant has

the same chance of being randomly assigned to one of two or more groups e.g. Group A & Group B.

WHY RANDOMISE?WHY RANDOMISE?

Possible explanations for a difference would include: 1.the intervention exhibits a real effect;

2.the outcome difference is solely due to chance3.there is a systematic difference (bias) between the groups due to factors other than the intervention

Eligible patients

Group A

Group B

Allocate Patients

Analysis

Follow up

WHAT IS ALLOCATION BIAS?WHAT IS ALLOCATION BIAS?

LBPPatients

Less SeverePatients

More SeverePatients

WHAT IS ALLOCATION BIAS?WHAT IS ALLOCATION BIAS?

Is there a difference because:

- the treatment actually works- of the way he has allocated patients

RANDOMISATION DEPENDS ON:

1.the generation of an unpredictable allocation sequence

2. the concealment of that sequence

ALLOCATION RATIO’SALLOCATION RATIO’S1. Equal Allocation

Participants have the same chance of being assigned to trial arms

• Easily accommodates multiple arm studies• Generally the most efficient design• Easy to implement

2. Unequal Allocation• Assigns more participants to intervention• e.g. BEST – 2:1 allocation

TYPES OF RANDOMISATIONTYPES OF RANDOMISATION

• Simple Randomisation

• Permuted Block Randomisation

• Stratified Block Randomisation

• Minimisation Method

SIMPLE RANDOMISATIONSIMPLE RANDOMISATIONDefinitionMost basic form of randomisation where each treatment

assignment is ”memory less” - made without regard to previous assignments

Some examples:• Unbiased coin toss for each trial participant• Roll an unbiased die• Sequence of Random Numbers from statistical textbooks• Computer generated random sequence

ILLUSTRATIONILLUSTRATION

• Two Groups (criteria: {2,4,6,8,0}=A, {1,3,5,7,9}=B):

A computer generated random sequence:

4,8,3,2,7,2,6,6,3,4,2,1,6,2,0,…….

4 8 3 2 7 2 6 6 3 4 2 1 6 2 0A A B A B A A A B A A B A A B

ILLUSTRATIONILLUSTRATIONA computer generated random sequence:

4,8,3,2,7,2,6,6,3,4,2,1,6,2,0,…….

Two Groups: different randomisation ratios (e.g. 1:2)(criteria:{1,2,3}=A, {4,5,6,7,8,9}=B; ignore 0’s)

4 8 3 2 7 2 6 6 3 4 2 1 6 2 0B B A A B A B B A B A A B A -

ILLUSTRATIONILLUSTRATIONA computer generated random sequence:

4,8,3,2,7,2,6,6,3,4,2,1,6,2,0,…….

Three Groups: (criteria:{1,2,3}=A, {4,5,6}=B, {7,8,9}=C; ignore 0’s)

4 8 3 2 7 2 6 6 3 4 2 1 6 2 0B C A A C A B B A B A A B A -

SIMPLE RANDOMISATION SIMPLE RANDOMISATION SUMMARYSUMMARY

Pros and Cons+ Simplistic implementation+ Allocation is random and unpredictable− Can produce unbalanced allocation− Can lead to analysis complications e.g. interim

analyses

Imbalance SolutionReplace with another allocation list (criteria: >10)Restrict the randomisation

BLOCK/PERMUTED BLOCK BLOCK/PERMUTED BLOCK RANDOMISATIONRANDOMISATION

Definition• Allocation list is comprised of “blocks”• Each block contains one possible combination

(permutation) of possible treatment allocations• Allocation is balanced at end of each block

Basic Implementation

1 Block size is an multiple of treatment arms i.e.2 arms → b = {2, 4, 6, . . .}, 3 arms → b = {3, 6, 9, . . .}

2 List all possible combinations e.g.For block size of 2: AB or BA

3 Choose a criteria for each block of patientsDigits 0-4 = AB Digits 5-9 = BA

4 8 3 2 7 2 6 6 3 4 2 1 6 2 0AB BA AB AB BA AB BA BA AB AB AB AB BA AB AB

4,8,3,2,7,2,6,6,3,4,2,1,6,2,0,…….

ILLUSTRATION

1 Lets say: - want to compare 2 groups e.g. Grp A & Grp B- choose block size of 4 patients

2 List allocation combinations for block size of 4AABB BAAB ABBA ABAB BABA BBAA

3 Choose criteria for each block of patients1 - AABB 2 - BAAB 3 - ABBA4 - ABAB 5 - BABA 6 – BBAA ignore 0,7,8,9

4 8 3 2 7 2 6 6 …- -

4,8,3,2,7,2,6,6,3,4,2,1,6,2,0,…….

ABAB ABBA BAAB BAAB BBAA BBAA …

BLOCK RANDOMISATION SUMMARYBLOCK RANDOMISATION SUMMARY

Pros and Cons+ Balance between arms is guaranteed by block’s end+ Interim analysis can still have balance− Bias can occur if allocation sequence is determined

RecommendationsDo NOT use block size of 2Use reasonably large blocks to avoid predictabilityNot too large if interim analysis intended

STRATIFIED BLOCK RANDOMISATION

Definition

Balancing treatment groups with respect to prognostic factors which may be related with participant response in order to prospectively achieve treatment group comparability

WHAT DOES THIS MEAN?WHAT DOES THIS MEAN?

REMEMBER: randomisation is to ensure treatment group comparability

HOWEVER: certain prognostic factors may be predictors of response e.g. age, sex,

THEREFORE: we want to make certain that there is a balance in both groups for these factors

HOW? – have a separate block randomisation scheme for each category

ILLUSTRATIONILLUSTRATION

• Single Strata: Age, (block size 4)

AGE <50 BABA AABB ABBA BBAA BAAB …AGE ≥50 BAAB ABBA BBAA ABAB BABA …

GROUP A GROUP B

AGE <50 50% 50%

AGE ≥50 50% 50%

ILLUSTRATIONILLUSTRATION• Two Strata: Gender & Age, (block size 4)

Male AGE <50 BABA AABB ABBA BBAA BAAB …Male AGE ≥50 BABA AABB ABBA BBAA BAAB …

Female AGE <50 BAAB ABBA BBAA ABAB BABA …

Female AGE ≥50 BAAB ABBA BBAA ABAB BABA …

GROUP A GROUP BMale AGE <50 50% 50%Male AGE ≥50 50% 50%

Female AGE <50 50% 50%Female AGE ≥50 50% 50%

STRATIFIED STRATIFIED RANDOMISATION SUMMARYRANDOMISATION SUMMARY

• Pros and Cons

• + Balance important variables between arms• + Improves power by reducing variance• − Complicates allocation process• − Too many strata can lead to sparse data

MINIMISATION METHODMINIMISATION METHOD

DefinitionMinimisation (Adaptive randomisation) is an accepted

statistical method to limit imbalance in randomised clinical trials in conditions with known important prognostic factors.

Its called minimisation because imbalance in the prognostic factors is minimised

ILLUSTRATIONILLUSTRATION- - Lets say we have 40 patients alreadyLets say we have 40 patients already- Each factor (gender & age) should consist of 40 patients- Each factor (gender & age) should consist of 40 patients- - Next patient (41Next patient (41stst patient) is Female and <50. patient) is Female and <50.

- For each treatment, add number of patients in the - For each treatment, add number of patients in the corresponding 2 rows:corresponding 2 rows:

Group A = 9 + 8 = 17 Group B = 8 + 7 = 15Group A = 9 + 8 = 17 Group B = 8 + 7 = 15

Factor Level GROUP A GROUP B Gender Male 11 12 Female 9 8Age AGE <50 8 7 AGE ≥50 12 13

NEXT ASSIGNMENT TO GROUP B

MINIMISATION METHOD SUMMARYMINIMISATION METHOD SUMMARY

Pros and Cons+ Cannot determine next allocation+ Maintains balance across groups+ Advantage over stratified block randomisation, as

randomisation does not occur within strata− Can be technically challenging to implement

ALLOCATION CONCEALMENTALLOCATION CONCEALMENT- different to blinding- different to blinding- avoids selection bias- avoids selection bias

Trial Type MechanismSingle-centre trial -Independent person responsible for patients

registration and randomization- Use randomisation list or sealed envelopes

Multi-centre Trial After gaining consent from eligible patient:-Central randomisation by telephone-Interactive voice response system-Fax or internet

Central registration office not feasible/desirable

-Sealed envelopes containing treatment allocation inside

PRACTICAL CONSIDERATIONSPRACTICAL CONSIDERATIONS

STUDY TYPE RANDOMISATIONSmall studies BlockLarge studies BlockLarge, Multi-centre studies Stratified Block / MinimisationLarge, prognostics factors Minimisation

RECAPRECAP• Randomisation

- Reduces allocation bias- Ensures comparability of Groups

• Allocation Concealment- Reduces selection bias- Again, ensures comparability of Groups

• OTHER FORMS OF BIAS AFTER RANDOMISATION….

TYPES OF BIASTYPES OF BIAS

• Patient bias

• Care Provider bias

• Assessor bias

• Analysis and Interpretation bias

PATIENT BIASPATIENT BIAS

DEFINITIONThe patients knowledge of the treatment being

received may affect the outcome of the study.

• patient’s knowledge that they are receiving a “new” treatment may substantially affect the patient’s subjective assessment.

CARE PROVIDER BIASCARE PROVIDER BIAS

DEFINITIONThe care provider’s knowledge of which treatment a

patient is receiving may affect the way the provider• Deals with the patient• Treats the patient• May give patient information about the treatment the

patient is receiving

ASSESSOR BIASASSESSOR BIAS

DEFINITIONAssessor’s knowledge of which treatment the patient is

receiving may affect the way the assessor assesses outcome

• Affect validity of conclusion of the study• If assessment conducted whilst patient still receiving

treatment may indirectly provide patient with information about the treatment they are receiving.

ANALYSIS AND INTERPERTATION BIASANALYSIS AND INTERPERTATION BIAS

Knowledge of the treatment arm may affect resultant analysis of the data as

• Seeking explanation of an “anomalous” finding when one is found contrary to the study hypothesis.

• Accept a “positive” finding without fully exploring the data

Knowledge of the treatment arm may affect decisions made by external monitors

• Terminate the study due to adverse events• Terminate the study for superiority of treatment

BLINDINGBLINDINGBlinding = masking treatmentAll of these potential biases can be avoided if everyone

involved within a study is blinded to the treatment being given to the patient.

HIERARCHY OF BLINDING• Un-blinded • Single Blind• Double Blind• Complete Blind

SINGLE BLINDSINGLE BLIND

The patient (or sometimes the clinician) is blinded to the treatment given.

Often used when double blinding is impractical for reason such as

• Need to adjust medication• Potential side effects

DOUBLE BLINDDOUBLE BLIND

Neither the participant nor the physician conducting the study know which treatment is being given to the participant

Minimises both potential patient biases and potential assessor biases

Should be used whenever possible

DOUBLE BLINDING TECHNIQUESDOUBLE BLINDING TECHNIQUES

Placebo for each possible treatment• Tablets identical in physical appearance• Tablets with similar taste and smell• Same carrier used for IV infusions

Other treatments “shammed” as far as possible• Sham surgery

DOUBLE BLINDING- ALWAYS FEASIBLE?DOUBLE BLINDING- ALWAYS FEASIBLE?

Double blinding may not be possible:

• May not be ethically permissible to blind patient

• Study using an exercise intervention

PROBLEMSPROBLEMS WITH WITH DOUBLEDOUBLE BLINDING BLINDING

SIDE EFFECTSIf drug produces side effects = difficult to blind patient and

ethically wrong to produce a placebo that induces side effects.

EFFICACYIf treatment truly effective may become clear as to which

treatment the patient is receiving – rare!

COMPLETE BLINDINGCOMPLETE BLINDING

Effectively it is the best approach BUT

• Requires two groups of people for data processing, one group to encode data and one group to perform the analysis.

• Therefore not economically feasible. Used sometimes by major drug companies

COMPLETE BLINDING COMPLETE BLINDING TECHNIQUESTECHNIQUES

• Analysis uses coded treatment groups

• Analysis uses coded side effects

• Analysis uses coded laboratory tests

SUMMARYSUMMARY• BIAS different types of bias exist in all clinical trials• Randomisation deals with • Allocation concealment deals with • Blinding deals with other types of bias

Allocation bias

Selection bias

THANK YOU FOR LISTENING

ANY QUESTIONS?

Next week…Next week…

BY: Janet Dunn & Louise HillerDATE: Thursday 30th SeptemberTIME: 12.20- 1.45 pm

ROOM: T0.08/09

ANALYSISANALYSISSimple and to the point, defining termsSimple and to the point, defining terms