Comment from Ian Marschner
AnonymousSupportAcademic
Summary: The commenter, likely an academic or researcher given the citation of a peer-reviewed medical journal, supports the guidance but suggests refining the definition of "concurrent control" to be broader for platform trials. They also advocate for including specific requirements regarding "Participant Randomization Schemes" to ensure valid data interpretation in data-sharing arrangements.
Section III<br/><br/>The guidance to include only concurrent controls in the primary analysis is welcome (Section B). The fourth paragraph of Section B, beginning at line 195, clarifies that any controls that could not have received a comparator drug should not be included in the concurrent controls analysis of that drug. This is a welcome and important clarification. However, here and elsewhere in the document, “concurrent control” is implicitly defined as a control participant that was randomized contemporaneously. It would be better to embed the requirement that a control should have been able to receive the comparator drug into the definition of a concurrent control. That is, rather than a paragraph beginning with “In addition to including only concurrent control data …”, it would be better for this paragraph to be framed such as “For some platform trials the definition of concurrent control needs to be broader than just the requirement to have been randomized contemporaneously.” The paragraph could then present the same clarifications and examples using a broader definition of concurrent control. A recent paper dealt with precisely this issue and introduced the concept of a “Concurrently Randomized Cohort” which captures this broader definition of concurrent controls; see Marschner et al. BMJ Medicine 2023;2:e000497. doi:10.1136/ bmjmed-2023-000497<br/><br/>In addition, in Section B it would be useful to note that, as well as drug-specific eligibility, another reason why some controls may not have been able to receive a comparator drug is because some platform trials allow sites to offer some, but not all, of the comparator drugs, meaning that controls at those sites could not receive all comparator drugs, even if they were eligible. For example, there may be region-specific restrictions on drugs that can be offered.<br/><br/>Section IV<br/><br/>The final two paragraphs of this section deal with data sharing and data access. Platform trials and other master protocol designs present specific challenges for data sharing. Valid interpretation requires not just the treatment that each participant was allocated to, but also knowledge of the treatments that could have been allocated to each participant, and also the randomization probabilities and other characteristics of the design (e.g. stratification factors). All of these aspects can differ between participants in a platform trial and they need to be fully documented in a “Participant Randomization Scheme” that forms an integral component of any data sharing arrangements, as defined and discussed in the above publication (Marschner et al, 2023). The guidance should note that data sharing for master protocol designs necessitates the collection and sharing of a participant randomization scheme, without which valid interpretation may be compromised.<br/><br/>