Randomization
On this page (14)
Most clinical studies, and in particular the canonical randomized controlled trial use randomization – i.e. randomly assigning study participants to one of at least two groups or study arms. While this can be done manually, it is much more convenient to have the software handle the randomization process
Our software supports this by accepting externally prepared randomization lists that define the available randomization slots. All randomization functionality is provided by the randomization module that can be booked for your study.
Randomization permissions
Randomizing a participant is possible for roles that have the randomize permission. Users in such roles can trigger randomization for study participants associated with their study center(s).
The View randomization permission, on the other hand, allows users to see which study arm (group) a particular participant has been randomized to. Users/roles with this permission will not be blinded with respect to randomization. Accordingly, great care should be taken in deciding who shall receive that permission.
Finally, someone needs to upload/manage randomization lists in the system. All users with the top-level permission Study general settings and go-live have the ability to do that. Randomization lists that are downloaded at some point when the study is running contain information about slots that have been assigned, already. While the list does not show participant IDs, it is not sufficient for unblinding, but it does contain information that conveys information about randomization status of existing participants. Accordingly, this permission should be given to a small, selected group of users, only.
Randomization lists
Our software uses pre-generated randomization lists that are provided by study admins and then used to assign a slot in the list and the respective group (= study arm / sub-cohort/ ...) to participants. In the following sections we will explore some common randomization schemes and how they can be set up.
Randomization setup is located at Setup > general > randomization
Randomization lists need to be in CSV format (using semicolons as field separators and commas for decimal separator). The first row defines the column names, all other rows are data. Each data row in the list represents one slot that is available to a participant. So if you want to recruit 50 participants, the randomization list must contain at least 50 slots (rows). Once all slots have been used/assigned, no additional participants can be randomized.
We recommend to start by downloading a randomization list template. The resulting CSV file shows the expected format, required columns as well as some examples for optional columns. The download link is located in the introductory text in the randomization section f the setup.
Upon randomization, the system will assign the participant to the first unused and matching slot. Therefore, row order in the list determines randomization order. Randomization lists ensure that you can have full control over the size of the total cohort as well as the sub-cohorts (groups).
Simple randomization
Simple randomization means that participants are randomly assigned to groups with a certain probability – much like tossing a coin or throwing dice. You can define as many groups as desired. The probability for each group is defined by the proportion of slots for the group in the list.
So if you want to recruit 100 participants and allocate them in a 60/40 ratio to the two groups Drug and Placebo, you need to create a list with 60 rows for group Drug and 40 rows with group Placebo.
For a study with 90 participants equally distributed over three groups A, B and C, the lists must have 90 data rows – 30 for each group.
The randomization list for simple randomization requires only one single column group. Creating a simple randomization list is easy:
- Create a csv file with one column header: group.
- Add as many rows for each group as you need for the desired ratio and cohort size.
- Randomly shuffle the order of the data rows.
Block randomization
Simple randomization ensures that at the end of recruitment, the study cohort has the desired group sizes and ratio. However, group ratios can vary substantially over the course of recruitment. Block randomization is a popular method to ensuring that group ratios remain close to the desired values over the entire course of recruitment.
Explaining the exact method for creating a block-randomization list is beyond the scope of this manual and can be found in the statistical literature. Briefly, the process is this:
- Decide on number and ratio of groups
- Decide on suitable block size k
- Create blocks of size k that contain permutations of your groups.
- The number of blocks is determined by the desired cohort size as well as block size.
Common statistical software systems such as SAS or R offer tools for generating block-randomization lists.
Randomizing by study site (center)
When a single randomization list is applied to a multi-center study, there is no control over the individual center's contribution to the study groups. If you want to ensure that group balance is also maintained within the individual centers, you need to randomize stratified by center. The easiest way to do that is to include a center column in your randomization list. Upon randomization, the system will then find the first free slot for the respective center.
Alternatively, you can upload separate randomization lists for each center.
The system will allow mixing of global and per-center lists. In that case, the system uses center-specific lists before falling back on the global list(s) in case not free slot is found in the former. This may be useful in certain situations but should not be used in most normal settings.
Stratified randomization
Sometimes, it is desirable to keep a certain balance of known covariates within each group. Common examples include sex or age group. This can be accomplished easily:
First, you need to make sure that the relevant variables are included in a form and that this form is included early in the study so that the data is available at the time of randomization. Stratified randomization only works with categorical variables – typically implemented as single-choice questions like "Sex: male/female".
In order to allow stratification by these questions, you need to assign an
alias to the respective questions (e.g. $sex) and include the alias as a
column in your randomization list. The resulting list then looks something like
this:
group $sex
A m
B f
A f
B m
[...]
You can include multiple aliases for stratification. E.g.:
group $sex $age_group
A m 18_30
B f 18_30
A f 18_30
B m 18_30
A m 31_45
B f 31_45
A f 31_45
B m 31_45
[...]
The stratification aliases should be categorical – i.e. single-choice questions. If you know what you are doing, computed fields also work, but make sure you fully understand what the possible return values are.
Crossover designs
Some studies implement a crossover design. E.g. participants who were randomized to group A are treated with the drug under test over the course of the first visits while group B receives a placebo. After finishing a certain visit, the participants are then switched to the opposite group so that all participants receive both treatments at some point during the trial.
Although our software has no specific feature to switch group assignment, it is easy to implement crossover designs by using slightly adapted group definitions. Instead of defining Group A as "Drug" and B as "Placebo", you should define A as "Drug then Placebo" and B as "Placebo then drug".
Appending to the randomization list
Sometimes things don't go according to plan. For example, you may need to recruit more participants than planned, initially. So you will have to extend the randomization list to do that in a running study. While you cannot directly append rows to an existing list, you can easily upload another one of the same format. The system will treat additional lists as extensions and will move to the next list once the previous one is exhausted.
Downloading the list
Randomization lists that are already in use by the system can be downloaded by users with setup permission. The downloaded list has gained additional columns that were appended by the system:
| column | type | Description |
|---|---|---|
used |
true/false | true if the slot has been allotted to a participant, false otherwise |
created_at |
timestamp | Creation time of the slot |
created_by |
user who uploaded the list | |
deleted_at |
timestamp | time of deletion, empty otherwise |
deleted_by |
timestamp | user who deleted the slot |
Deleting lists
It is possible to delete lists from the system. However, lists will only be fully deleted if none of their slots has been used at the time of deletion. Any slots that are already in use cannot be deleted once the study is live.
Randomization workflow
In the participant view, a Randomization widget will be displayed. Clicking Randomize will initiate the randomization process and after confirmation, the system will search for an available slot that matches all stratification criteria, starting from the top of the list.
User with view randomization permission will be able to see the group/arm a participant has been randomized to.
Both triggering randomization and looking up the result are recorded in the audit trail. Please note, that randomizing a participant is an irreversible event. I.e. you cannot undo the randomization once it has been done.
Emergency unblinding
Especially in drug trials, it may become necessary to break blinding of the site personnel with respect to the study arm a participant is in, so that appropriate treatment can be administered in an emergency.
This can be achieved by making sure, that accounts with view randomization permissions are available at the sites but not used during normal work. Access data for these accounts may be kept in sealed envelopes or kept by previously designated people.
Special Alias
You can access the randomization result in computed fields and visibility rules via
the special alias $_randomization_group.
Caution! This feature has the potential to break blinding! Only use after careful consideration.
FAQ
Q: I want to control the group probability but not the exact size of my sub-cohorts. Can that be done?
A: Yes. Just make sure you randomization list contains many more slots (rows) than you will need. That way, you can set the desired group ratio but nor exact group size.
Q: Can I use the result of a computed field for stratified randomization?
A: Yes, you can. Just make sure there is an alias for the computed field and that it returns the desired categories.
Questions about Magana Trial Manager?
Try setting up your own eCRF in a free test account, or reach out to our team in Regensburg — we usually reply within one business day.