Utilizing IRT

Scientists have long recognized the value of "blinding" certain participants in a study to reduce or eliminate biased results. In clinical trials, blinding refers to keeping trial participants, investigators, and/or assessors unaware of the assigned intervention, so that they will not be influenced by that knowledge. According to international standards set forth in ICH E9, the most important techniques to avoid bias in clinical trials are randomization and blinding.

Randomization addresses selection bias. Randomization schemes are put in place because it's vital that there be no way of knowing which treatment group the next subject will be placed in to avoid bias on that basis.

Blinding, on the other hand, is intended to guard against performance bias (unequal care of patients); detection bias (unequal assessment of results); and attrition bias (biased evaluation of protocol deviations or dropouts from the study).

Interactive response technology (IRT) has had a huge impact on patient randomization and drug allocation strategies because it allows powerful computer algorithms to be employed. However, even with this support, partial unblinding—and therefore bias—can occur.

Direct and partial unblinding

Direct unblinding is well understood and happens most often when documentation that holds the treatment information falls into the hands of staff involved with the study. This can include shipping documents or release documents that can contain kit numbers that can be linked to a specific material type.

In IRT, blinded and unblinded reports are issued with the intent that they only be seen by the appropriate group. But mistakes happen, and it's important to guard against, for example, an unblinded pharmacy that receives information about a treatment group or material types. Faxes pose an especially big risk of unblinding, because you never know who's going to pick up a fax that contains unblinding information that could directly affect the study.

Unlike direct unblinding, partial unblinding occurs at the moment when there is full certainty that two patients belong to the same treatment group, or to different groups, without exactly knowing what the treatment group is. One way partial unblinding can occur is through the packaging. If you send a kit to a site containing four vials of the same treatment and give one to patient A and one to patient B, it's now clear that patients A and B are in the same treatment group.

The danger of partial unblinding

Partial unblinding is a serious issue in clinical trials because it opens the door to more and more accurate unblinding as the study progresses. Based on their own clinical judgment, investigators or other site staff can often guess the treatment of a patient with higher probability than if based on chance alone.

The difficulty is that an investigator with some unblinding clues may, consciously or unconsciously, begin to look for corroborating evidence, and over time the accuracy of his or her guesses improve. Once you know two or three subjects are on the same treatment the chances increase that an investigator can find out what that treatment is because the data can be pooled. Further, it's not inconceivable that an entire site could be unblinded in this way, correctly guessing patients and treatments and thereby introducing a significant data bias into the study results.

Sources of clues

Any document that lists multiple kit numbers can lead to partial unblinding if it is not sorted by the kit number. This would include shipment requests, shipment documentation, shipment packing, site stock reports and IRT notifications.

In addition, shipping or dispensing different lot numbers at the same time will lead to partial unblinding, especially if dispensing is done sequentially. Particularly when drugs are being resupplied to a site or supplied in response to a specific event, dispensing based on the same sequence as the kit list can easily give investigators clues that can potentially lead to biased behavior.

In IRT, randomization and dispensing are split into two separate steps: a randomization list that holds the treatment groups, and a kit list that holds the medication kit numbers. In this case, when a patient gets randomized, he or she first gets assigned the treatment group, and then gets assigned a kit number of the site stock at the site of the appropriate type. Because the randomization is now centralized, the next subject at a different site can be randomized to the same block with a kit number of the appropriate medication assigned accordingly. The randomization and material lists are split.

Using IRT, designers should create appropriately randomized material kit lists that disassociate treatments and cohorts and ship and dispense them following a random sequence. Whenever possible, avoid shipments that are based on single dispensing events and be certain to blind shipments by sending at least two material types; in fact, study organizers should stop using lot numbers for shipping and dispensing if one of the material types is no longer available.

Conclusion

The computing power in a modern IRT system when combined with thoughtful planning can ensure clinical study designers a blinded study.

IRT can also help minimize allocation bias that may occur when covariates that affect the outcome are not equally distributed between treatment groups and the treatment effect can be confounded by the covariate. Although this balance would be hard to maintain manually, a well-built IRT system makes the process straightforward and reliable. Additionally, IRT-based randomization can be utilized to track unequal group sizes such as those found in escalating dose studies, or when regulatory agencies may require a minimum number of patients exposed to treatment.

Partial unblinding can be a serious problem if it's not addressed from the beginning of the study. However, by using interactive response technology and a thoughtful selection and allocation plan the potential effects can be virtually eliminated.

Editor's note: For futher information on the topic please visit http://bit.ly/Y7mBcm to listen to an on-demand podcast.

A modern IRT system when combined with thoughtful planning can ensure clinical study designers a blinded study.

Stefan Duerr is Associate Director of Project Management at Cenduit LLC, www.cenduit.com/.