Interview with WorldCare Clinical Associate Director of Imaging Services

Q: When should a Blinded Independent Central Review (BICR) be used?

A: BICRs are most prevalent in imaging studies and are also done for histology, safety reviews and endpoint assessment committees. When a subjective assessment has a lot of variability and is critical to a trial’s success, sponsors find that incorporating the requirement for a BICR of imaging and clinical data eliminates bias and improves the homogeneity of the research population, which should result in a decrease in trial sample size.  Given the significant investment in late-stage clinical development, a BICR can provide value by decreasing the data variability and potentially result in a cost savings.

Q: Why is site bias a concern?

A: Knowing the referral population can artificially improve performance of a site review in a diagnosis trial. This can be a major issue in open labeled trials, which require a BICR to determine on-study response to treatment. This issue can often be totally avoided when independent experts are used for both screening and on study reads. Recent research shows an increased response rate (RR) for the site reads when compared to a BICR.

Q: What are the different types of BICRs and when should each type be used?

The most common are single reads or double reads with adjudication. In general, single reads are used when cost is a major consideration, a faster turnaround time is needed or when it’s not the primary endpoint of the trial.  In the single read paradigm, concerns regarding precision are less of an issue than concerns of site bias. A double read with adjudication will add precision compared to a single read design and provides an indication of inter-reader variability. An adjudication read is usually only performed as a tie breaker (i.e., if the primary readers disagree) and is often used when the data is part of the primary trial endpoint, and/or when data interpretation is complicated. There is also a mixed model double read with adjudication, which has the added advantage of incorporating a site assessment and results in a cost savings due to less inter-reader variability. However, it can’t be accomplished in all scenarios as it requires the same questions be asked of both the site and BICR.

Q: Why not just do a double blind?

A: Double blinding is not always possible because treatments with respect to regimen/dosing, can be much different. Investigators can figure out the blinding schema. However, if it’s determined that the blinding was discernible, it can invalidate the trial results.

Q: What should be adjudicated?

You should only adjudicate those parameters that are directly related to the endpoint, otherwise your adjudication rate can be very large. High adjudication rates are sometimes an indication of inadequate reader training. With regard to the BICR, high accuracy usually infers good precision, but over time precision can be decreased in trials due to definitional drift.  Programs for ongoing training and testing help to minimize variability.  Too often variability results are obtained before the start of reads and upon completion of the trial, eliminating any opportunity to improve the quality of assessments.  Commonly used tests of inter- and intra-observer variability require larger sample sizes, the selection of cases at random, and provide poorly understood, if not useless, information unless corrected for chance.  This practice is not only costly, but does not achieve the desired effect of controlling quality.

Q: How can variability be better controlled?

Assessing heterogeneity across BICRs statistically provides information on reader accuracy while minimizing costs associated with recirculating large sample sizes of cases for traditional inter- or intra-reader variability.  This is most helpful in trials that don’t have multiple reads of the same images.  A variability plan focused on ongoing assessments of difficult cases, not cases selected at random, can provide a more confident assessment of reader variability.  By analyzing reader performance during the conduct of the study, sponsors benefit from retraining of readers resulting in homogenous datasets.  These concepts are applicable to BICR of radiology, pathology, dermatology, or other clinical information.  If good inter-reader results are obtained, intra-reader testing may not be needed.