Contributed Commentary by Stefan Dürr

June 8, 2018 | Based on the ICH Guideline on Statistical Principles for Clinical Trials (E9), the most important design techniques for avoiding bias in clinical trials are blinding and randomization.

Randomized controlled trials (RCTs) are considered the gold standard for conducting clinical trials. Randomization helps to avoid selection bias by introducing a deliberate element of chance into the assignment of treatments to patients. This ensures that the investigator or site staff cannot know or easily predict what the treatment will be for the next patient randomized, and the decision to enroll a patient in the clinical trial is not biased by the knowledge of treatment the patient will receive.

The most common way to randomize patients is by using a randomization schedule in which the element of chance is restricted through the use of blocks. This allows sites to maintain a better balance compared to unrestricted randomization. This method also helps to increase comparability, but at the same time increases the risk of selection bias towards the end of a block sequence. It is therefore of the utmost importance to keep the site and other relevant staff blinded to the block size, especially in open label trials. Using multiple block sizes can be another way to reduce the predictability of future treatments.

The knowledge of the patient’s treatment could potentially lead to conscious or unconscious bias in the way the site staff recruits the patients (selection bias), how they are treated (performance bias), the assessment of endpoints (detection bias), the handling of withdrawals, and how data is excluded from analysis.

The patient attitude to the treatment if known can lead to a difference in reporting symptoms (response bias) or withdrawal from the study (attrition bias). To reduce the chance of bias to the data, clinical trials should be conducted in a double-blind design whenever possible.

The trial blinding must be maintained until all data sets are locked. Until then investigators, site staff, sponsors and most other participants must not be aware of the treatment patients are receiving. However, there are certain functions like clinical supply or safety that require access to unblinded data. Having detailed standard operating procedures (SOPs) in place is imperative. This ensures a standardized, documented approach on access to unblinding information is appropriately restricted to authorized individuals throughout the lifecycle of the study.

Since IRT (Interactive Response Technology) systems are used to randomize patients and keep studies blinded, it is also imperative to discuss unblinding risks as part of the specification process. The sponsor must define:

1. Which system responses contain unblinding information and provide appropriate access
2. Which reports contain unblinding information and provide appropriate access
3. A process for user access for unblinded user groups

It is equally important to be aware that data points that are not unblinding in one study design can become unblinding in a different study design.

Unblinding incidents that happen during the study are critical events and need to be reported and explained. While it is obvious when an unblinding incident occurs, it is much more tricky to identify and prevent partial unblinding. We define partial unblinding as the knowledge that two or more patients are on the same treatment or that two patients are on different treatments.

A major source of partial unblinding risks is in the setup of the clinical supply strategy. One important aspect is the use of sequence numbers and sorting as it relates to shipping and dispensing of clinical supplies. For example, if we have a study with two treatments (active and placebo), the study would ship and dispense the drug kits based on the lowest available kit number.

 The site has the following site stock:

• Kit 301 (Active)
• Kit 598 (Placebo)
• Kit 612 (Placebo)
• Kit 912 (Active)

The site randomizes four patients in the below order:

1. Patient 1 gets kit 301
2. Patient 2 gets kit 912
3. Patient 3 gets kit 598
4. Patient 4 gets kit 612

If the site staff is aware that the selection is by kit number, then based on this example it would be clear that Patient 2 is in a different treatment than patient 3 and 4, as the patient got a kit number that skipped over the two kit numbers received by patient 3 and 4. This would then also mean that patient 3 and 4 are in the same group.

To avoid this scenario, the kits need to be dispensed either randomly or based on a scrambled sequence that is not known to blinded staff.

• Kit 912 (Active, Sequence 1)
• Kit 301 (Active, Sequence 2)
• Kit 598 (Placebo, Sequence 101)
• Kit 612 (Placebo, Sequence 102)

A similar issue with partial unblinding can happen if we review how kits from different shipments are dispensed, so it is advised to use different sequence numbers for shipping and dispensing.

However, we can introduce new partial unblinding issues through the use of sequencing when we sort kits in reports or responses. In the above example, the sequence ranges are separate for Active and Placebo, so these must not be used for sorting kits in reports. The only safe way to sort kits is to sort them by the kit number.

When designing the study and planning the clinical supply, sponsors must ensure that all unblinding risks are identified, analyzed and appropriately mitigated. Maintaining the trial blind is a critical part of maintaining integrity and validity of the clinical trial data.

Stefan Dürr is Senior Director of Client Delivery and Head of Drug Supply Center of Excellence at Cenduit LLC. He can be reached at stefan.duerr@cenduit.com.