Tokenize me, Scotty!

In the 1992 US presidential election, the Clinton campaign strategists wanted to make sure that their staff focused on the economy, along with two other messages: “Change versus more of the same” and of particular relevance “Don’t forget health care.”? With no intention whatsoever to mix it up in today’s volatile political scene, I happily borrow from and alter the campaign slogan to shed light on an issue that should underlie most discussions on Real World Evidence: it’s all about the data.? Where do we get it from?? What was it originally intended for?? How complete is it?? How much of it is there?? Who will accept it?? How can I get it?

Recent technological advances having facilitated unprecedented access to data from a variety of sources, for those of us involved in RWE on a daily basis, the question often comes down to how much data will address our needs (once those needs have been fully articulated).? For us non-technical folks, tokenization seems to be the go-to method for grabbing as much data as possible from a variety of sources.? But is it?? Is it a matter of a mile wide, but only an inch deep?? If, as I often do, want truly robust real-world clinical information to “jump start” a prospective observational study, will tokenization do the trick, or do I need to dive directly into electronic medical records (via patient consent)??

While I believe it’s all about the data, I am quite limited in my ability to understand many of the technical issues that will ultimately impact how I can use the data, what I can say, and a variety of other implications.? So I turn to one of the leading experts in the RWD field, Sam Roosz, CEO of Crescendo Health, to show me the way!

Jeff: Sam, you’ve seen the full gamut of data access and quality solutions in your career.? Can you shed some light on this issue?? Am I even asking the right question?

Sam: Jeff, thanks for teeing up this conversation.? As you know I’m quite bullish about the potential for RWE-driven transformation of life sciences research, but also advise a judicious approach for any sponsor looking to leverage these tools.? A tool that may be perfectly suited to one research task may fail to meet quality standards for another: it’s essential to first determine the fit-for-purpose of a tool before deploying it.??

You mentioned tokenization, which is a great case study of this subject.? Tokenization in clinical research brings a number of benefits: the ability to link in data about study participants from claims or EMR sources beyond the site; access to social determinants of health data; and the ability to link in data from beyond the realm of healthcare providers (e.g., OTC medications).? However, it also has real limitations that need to be taken into account.? The most impactful is completeness: only a fraction of all health data is available for commercial use, and much of that data has use restrictions (e.g., around linking to certain other datasets or use cases for publication).? What this means in practice is that most sponsors will really see only 30-40% coverage of measures of interest for a prospectively enrolled cohort.? There also is the potential for impaired recruitment yield as patients become uneasy with the notion of their health data being linked to or from mysterious sources that are not transparent to them.? Furthermore, the process of de-identification requires data elements to be potentially removed from both EDC outputs as well as RWD inputs to ensure compliance, limiting the utility and auditability of the resulting data set.? All told, what this means is that tokenization can be quite powerful for hypothesis generation / exploratory outcomes but may indeed be ill-suited to support assessments of primary or secondary outcomes for a study.? Thankfully for those purposes we have another tool for the job: interoperability tools that enable the patient to grant research access to their health data.

Jeff: Tell me more about the nuances of EMR access.? And is this just a US phenomenon?

Sam: A patient’s right to access their health data is encoded into law in many parts of the world (including the US and EU).? Recently, rules stemming from the 21st Century Cures Act have further strengthened this right by establishing an enforcement body to ensure access, as well as mandating that EMR vendors and most payers offer patients the ability to bring their data into a health application of their choice via API (a software data “pipeline” between two locations).? What this means is that we have the ability to get complete coverage of ALL health data for a prospectively enrolled cohort by requesting patients consent to use of the data through their own legal rights.? At Crescendo, for example, we make this process a simple 10-15 minute onboarding after which we can assemble all study-relevant data on a patient’s behalf and make that data available back to a patient in a single personal health application while also sharing consented data with the sponsor.? Not only does this offer the ability to track patient outcomes wherever they are being captured (at the site, a clinic down the street, or a health system across the country), it also offers the possibility of minimizing a research site’s burden by reducing data entry requirements and creating a centralized repository of data for subsequent queries.

Jeff: Reducing site burden is critical in observational research, as we want to minimize any — even administrative — activity that wouldn’t reflect actual practice conditions.? That said, for a prospective study, most research-based interactions with patients — specifically, the use of fixed interval questionnaires to capture Quality of Life and other patient-reported outcomes — still isn’t standard practice.? So you’re saying that, if I want as complete and naturalistic a picture as possible of the patient journey from key clinical, economic, and humanistic perspectives, EMR access optimally complements PROs and, as necessary, site data input?? But are we compromising patient privacy?

Sam: Quite the contrary: at Crescendo, our commitment to patients is that the data they assemble on our platform will only be shared with themselves (the patients) and for the research purposes delineated in informed consent.? Patients are in the driver’s seat and control where the data goes and who uses it, so this approach is, in fact, quite privacy-affirming.? With that consent we can screen the full patient journey (as told in payer claims) for study-relevant episodes of care and then collect relevant information from health records from the institutions providing that care.? In some cases, this creates the opportunity to reduce complexity of site operations (as well as lowering the activation energy for study-naive sites and investigators), and in others, it enables a path to conduct robust observational studies in a siteless manner.

Jeff: Earlier you noted that tokenization requires key data to be deleted for anonymity: wouldn’t that be the same issue for EMR access???

Sam: ?We’re starting to touch on some pretty wonkish aspects of the HIPAA Privacy Rule but I will do my best to characterize it at a high level and we can perhaps delve deeper another time.? “De-identification” of RWD is a term of art under HIPAA: it requires removing far more elements than just the direct identifiers of the patients and often involves purging records of rare diagnoses and unstructured data elements.? In clinical research, we don’t traditionally work with data that has been de-identified under HIPAA but rather pseudonymous limited datasets where patient identifiers have been replaced with a Subject ID.? The data we generate at Crescendo looks and feels identical to the data one might get from an eCRF at a study site: we can share the full richness of data without any HIPAA de-identification data loss while still ensuring direct patient identifiers don’t enter our customer’s GCP environments.

Jeff:? Yes, indeed: I think we’re definitely in the weeds now!? Thanks, Sam!

Until next time (but I’d welcome any related discussion from any reader).