2015-08-16

Analyses Across Multiple Studies

One of the main reasons for FDA building the Janus Clinical Trials Repository (CTR) is to facilitate analyses across multiple studies. This is particularly useful for safety analyses of rare events where pooling across studies is desirable to increase power.

I was at the FDA during the time when Vioxx went off the market. Although I was not directly involved in FDA's analysis of the available data for adverse myocardial ischemic events, I do remember the review team's struggles associated with both finding/accessing the data and getting them into an analyzable format. If the CTR had been available, the analyses could have been performed much more quickly.

As the CTR was being built to support this use case, unexpected challenges emerged. I would best characterize them as an inadequate degree of semantic interoperability with the SDTM; i.e. ambiguities in the way certain SDTM concepts are defined. The result is one cannot automate an "apples to apples" comparison of data across studies using existing SDTM variables.

One example is the Reference Start Date (RFSTDTC). The definition is (in SDTM v. 1.4):

Reference Start Date/time for the subject in ISO 8601 character format. Usually equivalent to date/time when subject was first exposed to study treatment. Required for all randomized subjects; will be null for all subjects who did not meet the milestone the date requires, such as screen failures or unassigned subjects.

The problem here is the word "usually." The net result is that RFSTDTC is sponsor-defined and can mean something different from study to study. Because study day (--DY) is derived from the RFSTDTC, it is also problematic. One cannot be sure that analyses that rely on the RFSTDTC mean the same thing across multiple studies without first doing a manual review of the study and verifying that RFSTDTC has a constant definition across the studies. Given that these studies sometimes span different applications and sponsors, it is not a trivial task.

What we really need is a standard definition for RFSTDTC that is constant across all studies. One approach is to take the possible different ways that sponsors are defining RFSTDTC and identify each as a separate concept, with a different name and concept identifier (e.g. date of randomization; start date of the first treatment epoch; date of first exposure to investigational drug). For some studies, these may all be the same, but not for others.

No comments:

Post a Comment