A decade ago, the Study Data Tabulation Model (SDTM) was perhaps an adequate format for clinical trial data submissions. It was effective in standardizing data submitted from study sponsors, ensuring consistency in format and semantics. Since then, however, the way we use data has changed. Although SDTM has evolved to accommodate more and more disease areas and study designs, each new version feels like a patch on the prior version rather than a for-ward-looking standard. This brings clinical researchers to the current state of affairs—a patch-work system that fails to meet today’s needs or prepare for tomorrow’s. Where do we go from here?

The design flaws
As the opportunity to use new types of data and endpoints has increased, so do the challenges associated with SDTM. At its core, SDTM is a categorization—or interpretation—of collected data by the sponsor. But since source data is not submitted to the FDA, they never see the orig-inal database. Tables are merely a single “view” on relational data, so the long-term value of the data is lost without explicit relations. In order to make collected data as useful as possible, the raw data must remain findable, shareable and accessible under the right framework and governance.

Another major limitation of SDTM is the “T”, which stands for ‘tabular’.  Although researchers are accustomed to analyzing and reporting data in this rectangular format, relations between variables are actually much more complex than what can be easily expressed in a 2-dimensional tabular format. Also, SDTM splits up all findings into arbitrary domains. This was done to group relatable variables together, but in practice it doesn’t always make sense. For example, “oxygen saturation” measured from a blood sample falls into the Laboratory Test Re-sults domain, while the same thing measured using pulse-oximeter is submitted under Subject Visits. Similarly, microbiology tests —which were previously in LB—have now been relegated to the new MB domain (perhaps due to file sizes). For those not accustomed to the SDTM standard, these arbitrary domains make things needlessly complex. They also add constraints when combining data from non-SDTM sources and SDTM sources—which is happening more frequently as clinical research embraces real world data (RWD).

Lately, today’s SDTM is strongly based on SAS-XPT format, which stipulates US-ASCII and a maximum character count of 8 for variable names and all test codes, 40 for labels, and 200 for data. And because CDISC doesn’t accept modern coding systems (e.g., SNOMED-CT, LOINC, UCUM), sponsors must map everything to CDISC-mandated terminology, which are “lists with-out relations”. For example, in coding systems such as SNOMED-CT every code/concept is de-scribed using relations (e.g. a Systolic Blood Pressure measurement is a measurement of the Pressure in the Systolic Phase), whereas in CDISC terminology Systolic Blood Pressure is ‘only’ a type of Vital Signs (VS). A lot of valuable information is lost by ‘flattening’ these terminologies while also hampering true interoperability.

Possible solutions
In my opinion, we need a data-centric, linked solution where the relations between variables, patients, and visits are modeled out. Ultimately, we need a patient-centric solution modeled around how the data is actually captured in a clinical trial. There are many challenges to doing so—but technology is actually far from the biggest barrier.

An improved version of SDTM could potentially be used for RWD trials as a standard publica-tion and storage format. The ‘physical’ format could be something like XML or JSON-LD, which are common data exchange formats, thereby making it easy to build software solutions around them. Importantly, there are ways to provide a schema for XML/JSON-LD that checks if the generated file follows a given specification (e.g., a “new” SDTM format versus XPT). Of course, some industry folks find XML antiquated and JSON-LD an unsatisfactory compromise. Perhaps the answer lies in another format—but what exactly that looks like remains uncertain, RDF perhaps?

Another part of the solution lies in a more modern data exchange format. HL7 FHIR is becom-ing the standard of data exchange within healthcare, at least in the US, and will only gain im-portance given the rise in RWD and eSource. At the same time, clinical research must fully embrace the FAIR data principles. This would allow clinical trials of the future to be coupled to an ML/AI algorithm capable of perfectly defining the context and automatically generating SDTM metadata. Once rich linked data is common-place, rendering SDTM datasets and metadata becomes a trivial automated step between col-lection and publication.

Can the future of SDTM lie in RDF?
Recently, Tim Williams and Marc Andersen demonstrated how they could separate analysis results from the tabular ADaM structure using RDF Data Cubes and then recreated the analysis displays (TFLs). This thinking further evolved in a later project with Armando Oliva, M.D. in the “Clinical Trials Data as RDF” project where they deconstructed SDTM and modeled the data using the clinical trial process, then materialized the data into the tabular SDTM format.

The courage to change
Currently, there doesn’t seem to be any initiative at CDISC to start from scratch or move away from tabular formatting. It seems unlikely that the FDA will impose a new technology on the industry. After all, the FDA will likely always want to analyze large datasets in tabulated for-mat, although graph-based data sets have also proven to be very useful to find patients similar to each other, predict diagnoses and find drug-drug interactions. Also, many sponsors have legacy systems—and new ones under development—based on the SDTM standard.

One possible game changer is if U.S. legislation mandated another standard for submissions— especially one that allows for the expression of relationships between variables and study events. Congress has already mandated the use of HL7 FHIR for healthcare data interoperabil-ity between providers, payers, and patients so we may see similar legislation affect clinical re-search. Still, it seems most probable that the FDA will not lead the change. SDTM represents a large investment for both the FDA and trial sponsors and pay-to-play standards organizations have a vested interest in maintaining the status quo. And, although flawed, SDTM has brought benefits, such as decreased approval times. Therefore, SDTM won’t be abandoned without a viable replacement.

SDTM may remain for some time, but there needs to be much evolution over the next 5 to 10 years to keep up with the new types of data and new dimensions of data that could potentially be leveraged. The current framework can—and should—morph into new structures. We’ll likely see the earliest innovation come from agile companies pursuing graph-based FAIR approaches for their own data integration and data quality purposes. One thing is certain: a patient-centric and data-rich future calls for a fresh approach.

---

Derk Arts MD, PhD is CEO & Founder at Castor. He has more than 15 years of experience in medicine, research and technology. He founded Castor to solve the biggest issues in clinical research: a lack of inclusivity, patient focus and impact of data. Castor enables sponsors worldwide to run patient-centric trials on a unified platform that helps them maximize the im-pact of research data on patient lives. Dr. Arts believes the key to achieving lasting change in the industry is through scalability and standardization.