The do's and don'ts of randomization in clinical trials
Jolee Graham, biostatistics director, Nicki Payne, director, IRT, and Katy Tam, director, biostatistics, discuss the components of successful randomization in research.
There are various elements and characteristics that define a clinical trial to be adequate and well-controlled per Code of Federal Regulations (CFR) Title 21 by the Food and Drug Administration (FDA). Inclusion of an appropriate comparator treatment and random assignment of treatment to subjects to minimize bias are among the characteristics listed in the regulations. To further reduce the chance of bias, “blinding” studies in the trials is important. Blinding ensures all data are being collected and all analyses are being completed without any bias entering the collection because those collecting the data are unaware of what treatment each subject is taking.
Below we discuss key elements for randomization, blinding and avoiding bias in clinical trials:
- Do use an interactive response technology (IRT) system. While employing traditional blinding methods—such as manual randomizations, blinding information via envelopes, etc.—can be less costly, using an IRT system is critical to prevent unblinding and bias in a clinical trial. IRT systems eliminate the possibility of human error and offer a wide range of advantages to traditional, manual methods including:
- Allowing separate randomization and material schedule. When using a kit-based randomization (packaging material off of the randomization schedule), damaged or expired drug kits impact treatment balances. A separate material schedule maintained by an IRT system assures that any damage in material has no impact on the randomization schedule as the IRT would link the assigned randomization record from the randomization schedule to available material from the material schedule.
- An audit trail. IRT systems have the ability to keep a record of all users who enter and/or update data in the system including, but not limited to, details of any code breaks. For example, the system could track the time a code was broken and who broke it.
- Data integrity. Because an IRT system is fully-validated, risks associated with data quality and integrity are much lower than with traditional manual randomization methods. The IRT system can ensure and prove randomization records are assigned in an order that cannot be accomplished with manual randomization methods. By nature of the system, an IRT can be proven to be unbiased, while a manual method cannot, opening manual methods to more regulatory scrutiny than IRT systems.
- The ability to randomize or perform emergency unblinding 24 hours a day, seven days a week, 365 days a year. The flexibility of an IRT system allows randomization to occur at any day or time, as well as enforcing randomization limits to prevent over-enrolling, and is always available in the case of an adverse event or subject’s visit to the emergency room that requires unblinding or other emergency event.
- More complex algorithms. For example, the use of Dynamic Minimization Randomization using the Pocock and Simon method (1975) or trial-specific adaptive randomizations is possible through an IRT system.
- Do assign appropriate block size in the randomization schedule to ensure well-balanced treatment assignment and avoid the risk of unblinding. Too small of a block can increase the unblinding risk; however, too large of a block increases treatment imbalance, particularly in stratified schedules.
- Do maintain detailed audit trails to avoid approval delays. Maintaining accurate audit trails is essential to ensuring regulatory compliance. IRT systems allow these trails to be tracked electronically, providing a record of randomizations and when they occurred, any code that was broken and when it occurred, and any unblinding that may have occurred and which user was associated with that unblinding. IRT systems also assign permissions to access blinded data based on user credentials, greatly reducing the probability of accidental unblindings. Additionally, it is critical to internally separate team members into blinded and unblinded groups to avoid crossing of data and to maintain records of those who are newly added to either group with the appropriate sign off.
- Don’t have too many stratification factors and associated levels compared to the number of subjects to be randomized in the clinical trial. Having too many stratification factors and levels not only increases the risk of treatment imbalance but also increases the risk of not having enough subjects in each stratification factor level to be able to determine statistical significance.
As randomization is a key foundation of an adequate and well-controlled clinical trial, it is important that it is performed in a manner that ensures no accidental, intentional or perceived bias and maintains the scientific integrity of the trial. PPD continues to prioritize randomization in clinical trials to ensure accurate results in our research through these methods and others.