It has always been a mystery to me why the FDA hasn't more enthusiastically embraced and offered incentives for NDA submissions to conform to the CDISC standards (A bit of this mystery was lifted for me at the CDISC INTRAchange in March). The INTRAchange is a meeting of the core CDISC participants and Dr. Theresa Mullin, Director of Planning and Informatics at FDA/CDER, spoke about the data review process at the FDA.
The FDA uses their Janus software as a repository to store the data, but they allow the reviewers to use different end-user tools to generate their views or develop their own code for analysis purposes.
The clinical data are typically presented in SAS files, but represented within these files in many different ways. For example, one study might collect medical history written as a narrative in a large text field, another might have collected medical history using a ‘check all that apply' methodology and only list the history that was checked. Medical History could also be represented by asking the patient to list each event in their medical history with the onset date separated out line by line. Some companies may code the history so that it is listed by body system, organ class, high level term, low level term and preferred term. Since there is no mandate to use standards, the reviewer is stuck with whatever they are presented with.
Astoundingly, less than 15% of the studies submitted to the FDA today follow any type of CDISC standard, although most companies agree that submitting data to the FDA using the CDISC standards will provide reviewers with a more consistent deliverable.
Dr. Mullen brought up several points around the submission of data using standards. Here are two that I found most interesting:
- There is no way for a Reviewer to know whether the data they are looking at is presented in a CDISC standard. Other unknowns are:
- Did the sponsor follow the CDISC standard precisely, or make some modifications or assumptions?
- Do some datasets follow the standard and others don't?
- Are some fields in some datasets standard, but not all?
- Are the field names standard names, but the data contained in the fields not collected in a standardized way?
- The FDA reviewers are not required to attend training on the CDISC standards. Those familiar with them know that the relationships and instructions for defining the data are complex and reading the complete documentation may not be feasible for all reviewers. Part of the problem stems from the fact that there is limited funding for training.
Thinking about the issues that the FDA faces brings up some new questions:
- How can the FDA be expected to receive benefits from receiving data in a standard way if reviewers are not trained on the standard?
- How can a reviewer sort through data easily if it is submitted in standards that are only partially standard?
- Where should the funding come for providing tools and training on standards to the FDA?
- How can the industry benefit from CDISC standards without FDA leadership, or at least participation?
I invite your comments and responses!