Okay, first – some context…
Without the FDA requiring SEND datasets, we would not have seen the industry-wide adoption and implementation of the standard. The change made by the industry, continues to fascinate me, in terms of both speed and scale.
This drive for submission provides us with a well-defined standard, and one that is well suited to single study review. However, it is becoming increasingly more apparent that there are some shortcomings for cross study analysis and data mining. The reason for this is that, while SEND allows for accurate representation of a study’s results in electronic, machine readable form, it also allows for a significant variability from study to study.
Now that I have set the scene, I’d like to discuss a recent paper by the Japan Pharmaceutical Manufacturers Association (JPMA) SEND Taskforce Team. This describes their analysis of multiple SEND packages from a variety of suppliers. It details key areas of variability in how data are represented from study to study. If data mining and cross study analysis gets you as giddy-as-a-kiddy on a Christmas morning, then I’d highly recommend that you take a deep dive into the paper for yourself. It contains very detailed results, calling out things like the specific variables that are most prone to variation between providers.
One of the key areas that the paper discusses is the scope and application of SEND Controlled Terminology (CT). It will not come as a surprise to anyone who is routinely working with SEND Datasets, that many key variables do not have CT defined for them. They allow for a free text description. The paper calls out many examples, including Clinical Signs where even variables like the test name, and the severity are not controlled.
Stepping away from Clinical Signs, the discussion on CT reminded me of work being conducted by PHUSE regarding the lack of CT for the Vehicle being used on the study. While, for single study analysis, a free text description is perfectly adequate, when it comes to data mining, the lack of CT proves problematic. For this specific issue, PHUSE are recommending a particular structure, format and nomenclature be used to describe the vehicle.
Such recommendations, to enforce supplementary rules and standardization – essentially, further CT in addition to the regular SEND CT – adds complexity to the creation of SEND datasets. That complexity will then increase the time and cost to produce SEND Datasets. That discussion will open up another debate, which I’ll leave for a different day.
Suffice to say, that SEND provides an accurate representation of a study’s results in electronic form, well-suited to single study review. However, there are shortcomings relating to multi-study usage, but these can be overcome. The JPMA paper does a very good job of calling out these issues to address.
As usual, drop me a note if you like to discuss this further.
Till next time,
Marc
