The Importance of Blinding and Data Masking in Clinical Trials
Blinded or confirmatory open label trials are essential in gathering evidence on the efficacy and safety of new drugs or treatments. However, maintaining blinding during interim analyses can be challenging, particularly when independent analyses through regular DMC are required. In such trials, rules should be defined, such as SOPs and DMC Charter, to ensure that the trial’s blinding is not compromised, and bias remains controlled. One way of handling DMCs is to have two separate teams, a blinded versus un-blinded statistical study team, responsible for DMC (blinded) and final analysis (un-blinded), respectively.
An alternative solution is to have two separate teams sitting in different locations, where the blinded statistical team is responsible for both DMC and final analysis activities. In this model, all outputs are programmed and quality-controlled by the blinded statistical team using blinded data and dummy treatment schedules as applicable. The blinded statistical team then transfers the final validated programs to the un-blinded statistical team, which reruns the programs using blinded data on the un-blinded study folder. After a comparison between the provided blinded outputs and the blinded outputs run on the un-blinded platform, the un-blinded statistical team updates flags to use actual randomization assignments and creates un-blinded outputs.
However, using dummy treatment codes may not be sufficient to maintain blinding in some circumstances, such as in an open trial where one may collect the relationship to a drug specific to one arm, or in a blinded trial where the occurrence of a particular adverse event may lead to guessing the arm. To overcome this issue, Cytel has developed a process and a tool to blind the data by applying systematic ‘data masking’ criteria, such as shuffling patients, records, and/or variables, blinding free text and medical coding dictionary. In this model, data goes through an additional blinding or masking process, where the un-blinded statistician runs the blinding/data masking criteria before transferring the manipulated data to the blinded statistical team.
The data masking process involves applying systematic criteria to blind the data effectively and uniformly across the study’s data. This approach allows the independent un-blinded statistical team to work with the data without compromising the trial’s blinding. The blinded team develops and validates the outputs using dummy treatment codes, and then the un-blinded team transfers the outputs to a separate area where they are handled with real treatment codes. The process described before for statistical program development is then applied.
Data masking can be beneficial when developing mock-up outputs when the sponsor requires the use of test data or a partial set of data. Working with a ‘data masked’ version of real data exposes the programming team to more ‘real’ data scenarios than working with test data. This can improve the programming team’s ability to handle real data scenarios, leading to better-quality outputs. Furthermore, data masking reduces the risk of unblinding the data during interim analyses, which can lead to biased decisions about the trial’s safety and scientific merit.
In conclusion, maintaining blinding in clinical trials is critical to ensure unbiased decisions about the trial’s safety and scientific merit. Rules should be defined, such as SOPs and DMC Charter, and two separate teams should be assigned, a blinded versus un-blinded statistical study team, responsible for DMC (blinded) and final analysis (un-blinded), respectively. However, using dummy treatment codes may not be sufficient to maintain blinding in some circumstances. In such cases, data masking is an effective approach to maintain blinding during interim analyses. Cytel’s data masking process and tool allow systematic criteria to be applied to blind the data effectively and uniformly across the study’s data, reducing the risk of biased decisions about the trial’s safety and scientific merit.
