By Kevin Venner, Jennifer Ross
What is Randomisation in a Clinical Trial?
Randomisation is the process for how subjects are assigned to Treatment (groups, arms, etc.) in a clinical trial with introducing a deliberate element of chance. If randomisation is not utilized or utilized inappropriately, then assumptions may be made on upcoming Treatment assignments. Knowledge of what Treatment is being assigned next can consciously or unconsciously influence decisions on whether to enroll a subject or which subject to enroll next. This influence is called Selection Bias. When properly implemented, Randomisation protects against Selection Bias to ensure the observed Treatment effect is due to the Treatment itself and not due this bias.
Further, randomisation helps achieve the required number of subjects per Treatment, which is also known Treatment Balance. With the random allocation of subjects throughout the trial, it is expected that each Treatment will have similar subjects for evaluation.
Interactive Response Technology (IRT) enables the global execution of randomisation and medication management across multiple sites. This removes the need for site-specific code-envelopes and other burdensome manual randomisation processes. Instead of relying on an unblinded contact to track randomisation, the IRT maintains an auditable dataset including details of each subject’s randomisation transaction.
Determining how to design randomisation in the IRT begins by reviewing the clinical trial’s protocol. Typically, the protocol provides randomisation details such as the Treatments, allocation ratio, number of subjects enrolled (sample size) and other information as applicable (e.g., stratification factors, cohorts, etc.).
This article focuses on the most common randomisation methodologies implemented in IRT:
- Central Randomisation
- Stratified Randomisation with Blocks Pre-Allocated to Strata
- Stratified Randomisation with Blocks Allocated On-Demand to Sites
How is Central Randomisation Methodology Implemented in IRT?
Central Randomisation is when all subjects are randomised within the same scheme regardless of any subject characteristics or demographics. Clinical trials use a Central Randomisation design when Treatment safety and efficacy is evaluated across all randomised subjects, and no sub-group analysis is planned. This evaluation requires Study-Level Treatment balance.
For an example, assume a protocol’s randomisation design specifies:
- Treatments: 2 (Active vs. Placebo)
- Treatment Allocation Ratio: 1:1
- Sample Size: 20
- Stratification: None – N/A
To obtain the Study-Level Treatment balance given the sample size (N=20) and ratio (1:1), the IRT will need to randomly allocate 10 subjects to Active and 10 subjects to Placebo. This is achieved via a Blocked Randomisation List, sometimes referred as Randomisation Schedule. The Randomisation List is generated with a specified Block Size that includes randomly ordered Treatment assignments within each block. For example, if the Block Size is 4, then for every 4 records, 2 Active records and 2 Placebo records would appear in random order. This blocking technique is the basis for each type of Randomisation List utilized across the 3 common Randomisation Methodologies. See below for an illustration of a Blocked Randomisation List for Central Randomisation.
Example 1: Central Randomisation List
Sequence Number | Randomisation Number | Treatment Code | Treatment Description | Block Number | SubjectID |
10001 | 10001 | A | Active | 1001 | 1 |
10002 | 10002 | A | Active | 1001 | 2 |
10003 | 10003 | B | Placebo | 1001 | 3 |
10004 | 10004 | B | Placebo | 1001 | |
10005 | 10005 | A | Active | 1002 | |
10006 | 10006 | B | Placebo | 1002 | |
10007 | 10007 | A | Active | 1002 | |
10008 | 10008 | B | Placebo | 1002 | |
10009 | 10009 | B | Placebo | 1003 | |
10010 | 10010 | B | Placebo | 1003 | |
10011 | 10011 | A | Active | 1003 | |
10012 | 10012 | A | Active | 1003 | |
10013 | 10013 | B | Placebo | 1004 | |
10014 | 10014 | A | Active | 1004 | |
10015 | 10015 | B | Placebo | 1004 | |
10016 | 10016 | A | Active | 1004 | |
10017 | 10017 | B | Placebo | 1005 | |
10018 | 10018 | A | Active | 1005 | |
10019 | 10019 | A | Active | 1005 | |
10020 | 10020 | B | Placebo | 1005 |
The assignment of subjects to the randomisation records in the IRT is simple! At randomisation, the IRT identifies the next available record (based Sequence Number order) and assigns to the subject. The subject’s ID is permanently linked to their assigned Randomisation Number and Treatment.
Looking at the Central Randomisation design in the Example 1 list, the 1st, 2nd, 3rd subjects are assigned to Treatments A, A, and B, respectively. The 4th subject will be assigned to Treatment B, which will complete the block. When all 20 subjects are assigned, 5 blocks will be completed and Treatment allocation ratio of 1:1 will be maintained at the overall Study-Level.
What if a Clinical Trial Requires Sub-Group Level Treatment Arm Balance?
If Treatment effect differences are expected across certain subject sub-groups, then the randomisation design may need to maintain the Treatment Balance within specified subject sub-groups. The clinical trial’s protocol defines these sub-groups as Stratification Factors. This sub-group level balance can be achieved within a Stratified Blocked Randomisation List.
For an example, assume a protocol’s randomisation design specifies:
- Treatments: 2 (Active vs. Placebo)
- Treatment Allocation Ratio: 2:1
- Sample Size: 90
- Stratified by:
- Prior Treatment (Yes vs. No)
- Symptom Score: (1 vs. 2 vs. 3)
Based on the above details, approximately 60 subjects will need to be assigned to Active, and 30 subjects assigned to Placebo. Within each cross-combination of the protocol’s Stratification Factor Levels (defined as Stratum), the blocks should maintain the 2:1 Treatment allocation ratio. The Randomisation List for this study can be designed with a Block Size of 6 (with 4 Active and 2 Placebo records) and the below Stratum definitions:
Stratum | Prior Treatment | Symptom Score | Stratum Description |
1 | Yes | 1 | Prior Treatment: Yes; Symptom Score: 1 |
2 | Yes | 2 | Prior Treatment: Yes; Symptom Score: 2 |
3 | Yes | 3 | Prior Treatment: Yes; Symptom Score: 3 |
4 | No | 1 | Prior Treatment: No; Symptom Score: 1 |
5 | No | 2 | Prior Treatment: No; Symptom Score: 2 |
6 | No | 3 | Prior Treatment: No; Symptom Score: 3 |
Each Stratum is pre-allocated its own set of randomised blocks within the Randomisation List, which essentially creates a sub-list for each Stratum. Example 2 below shows the 1st block in the Stratum 1 sub-list and the 1st block in the Stratum 6 sub-list.
Example 2: Stratified Randomisation List with Blocks Pre-Allocated to Stratum
Sequence Number | Randomisation Number | Stratum | Stratum Description | Treatment Code | Treatment Description | Block Number | SubjectID |
10001 | 10001 | 1 | Prior Treatment: Yes; Symptom Score: 1 | A | Active | 1001 | 1 |
10002 | 10002 | 1 | Prior Treatment: Yes; Symptom Score: 1 | A | Active | 1001 | |
10003 | 10003 | 1 | Prior Treatment: Yes; Symptom Score: 1 | B | Placebo | 1001 | |
10004 | 10004 | 1 | Prior Treatment: Yes; Symptom Score: 1 | B | Placebo | 1001 | |
10005 | 10005 | 1 | Prior Treatment: Yes; Symptom Score: 1 | A | Active | 1001 | |
10006 | 10006 | 1 | Prior Treatment: Yes; Symptom Score: 1 | A | Active | 1001 | |
… | … | … | … | … | … | … | … |
60001 | 60001 | 6 | Prior Treatment: No; Symptom Score: 3 | B | Placebo | 6001 | 2 |
60002 | 60002 | 6 | Prior Treatment: No; Symptom Score: 3 | A | Active | 6001 | |
60003 | 60003 | 6 | Prior Treatment: No; Symptom Score: 3 | A | Active | 6001 | |
60004 | 60004 | 6 | Prior Treatment: No; Symptom Score: 3 | A | Active | 6001 | |
60005 | 60005 | 6 | Prior Treatment: No; Symptom Score: 3 | A | Active | 6001 | |
60006 | 60006 | 6 | Prior Treatment: No; Symptom Score: 3 | B | Placebo | 6001 | |
… | … | … | … | … | … | … | … |
To randomise subjects, the IRT first determines the subject’s Stratum, then identifies that Stratum’s sub-list and assigns the next sequential record. If the 1st subject has Prior Treatment = Yes and Symptom Score = 1, then they are assigned to Randomisation Number 10001 and Treatment A. If the 2nd subject has Prior Treatment = No and Symptom Score = 3, then they are assigned to Randomisation Number 60001 and Treatment B. As subjects are randomised within each Stratum, the blocks assignments are completed and the ratio of 2:1 is sustained.
Is this the only way IRT can randomise subjects within Stratification Levels?
NOPE!
The pre-allocation of blocks to stratum demonstrated above is the most common method for stratified randomisation, but there are situations where pre-allocation is not the best fit.
When Site is a Stratification Factor, it is highly possible that more Sites are added mid-study. If pre-allocating blocks to Site (creating a sub-list for each Site), then each time a Site is added in the IRT, a new list would also be needed. This incurs unnecessary downtime and subsequent costs.
To avoid this headache, utilize On-Demand Allocation of Blocks to Sites! The list is generated in the same way as the Central Randomisation List in Example 1, with no blocks pre-allocated to any specific Site. Then the IRT allocates blocks to each Site On-Demand.
As an example, assume a protocol’s randomisation design specifies:
- Treatments: 2 (Active vs. Placebo)
- Treatment Allocation Ratio: 1:1
- Sample Size: 100
- Stratified by: Site
A Blocked Randomisation List with a Block Size of 4 is generated without any blocks pre-allocated to Sites. At Randomisation, the IRT first checks if any blocks with available records exist for the subject’s Site. If no, then the IRT identifies / assigns the next set of available Block(s) to the subject’s Site and assigns the 1st record to the subject. If yes, then the subject is assigned to the next available record within the block(s) allocated to their Site.
The number of blocks to assign to each Site can be set based on each study’s preference. For ease, the example will demonstrate allocating just 1 block at a time.
Assume the 1st 3 subjects are at the following Sites:
- SubjectID=1, Site 1234
- SubjectID=2, Site 3232
- SubjectID=3, Site 1234
Since the 1st subject is at Site 1234, the 1st Block (1001) is assigned to Site 1234. SubjectID = 1 is assigned to the 1st record in that Block (Randomisation Number 10012, Treatment A). The 2nd subject is at Site 3232, the IRT assigns the 2nd Block (1002) to the Site and 1st record in that Block (Randomisation Number 10006, Treatment B) to SubjectID = 2. The 3rd subject is at Site 1234, which has records available for assignment, thus the 2nd record in Block 1001 is assigned (Randomisation Number 10004, Treatment B) to SubjectID = 3.
Example 3: Stratified Randomisation with Blocks Allocated On-Demand to Stratum (Site)
Sequence Number | Randomisation Number | Treatment Code | Treatment Description | Block Number | Site | SubjectID |
100001 | 10012 | A | Active | 1001 | 1234 | 1 |
100002 | 10004 | B | Placebo | 1001 | 1234 | 3 |
100003 | 10002 | B | Placebo | 1001 | 1234 | |
100004 | 10001 | A | Active | 1001 | 1234 | |
100005 | 10006 | B | Placebo | 1002 | 3232 | 2 |
100006 | 10011 | B | Placebo | 1002 | 3232 | |
100007 | 10009 | A | Active | 1002 | 3232 | |
100008 | 10007 | A | Active | 1002 | 3232 | |
100009 | 10005 | B | Placebo | 1003 | ||
100010 | 10008 | A | Active | 1003 | ||
100011 | 10003 | A | Active | 1003 | ||
100013 | 10010 | B | Placebo | 1003 |
As shown above, this approach sets up the Site stratification within the Blocked Randomisation List On-Demand. In this design, the IRT can allocate blocks to new Sites (or even Stratum!), without having to generate subsequent Randomisation Lists.
How Should Randomisation Numbers be Ordered in the Randomisation List?:
For simplicity, in the Examples 1 and 2, the Randomisation Number is ordered sequentially and equal to the Sequence Number. However, it may be necessary for Randomisation Numbers to appear in random (Scrambled) order for blinding purposes. Example 3 (On-Demand Allocation of Blocks to Sites) warrants Scrambled Randomisation Numbers since it involves assigning a single block at a time. Scrambling the Randomisation Numbers prevents anyone from identifying the Block Size. In the example, if the Randomisation Numbers were ordered sequentially, study personnel may be able to figure out that Randomisation Numbers are assigned in sets of 4, which is equal to the block size. To scramble or not to scramble should be agreed upon by sponsor’s Biostatistical Representative and the IRT List Generators.
Important: The Block Size is an Unblinding Parameter that should only be known to study personnel involved in the design and implementation of the Randomisation List. Knowledge of the block size can lead to potential Selection Bias!
Final Thoughts:
The IRT randomisation list approaches discussed in this article are just the surface of randomisation possibilities! More complex methods for randomisation are achievable through IRT (e.g., Covariate Adaptive Randomisation (minimization), Target Adjusted Algorithms, Hierarchal Algorithms, Adaptive Designs, Master Protocols, etc.).