What the World of Health Data Has to Learn from a Turbulent Time in U.S. Financial History
In 1809, the Farmers Exchange Bank of Chepachet, Rhode Island found itself having issued $560,000 worth of bank notes with only $86.46 in assets on hand. The ensuing financial collapse conferred upon the Farmers Exchange Bank a dubious distinction: it was the first bank to fail in the United States.
Prior to the National Banking Acts of the 1860s, there was no national currency in the United States, and in states without state-chartered banks, unscrupulous bankers in rural regions realized that bank notes issued by their banks were less likely to be redeemed for silver or gold (or other assets) the farther away the notes were used. Hence, rural bankers in New England happily flooded the city of Boston with their notes. In financial history, this time is known as the era of “Wildcat Banking,” a period marked by volatility and frequent financial upheaval.
The world of health data in its current state is roughly analogous to the free-for-all of the wildcat banking era. Different hospitals, health systems, nursing homes, insurance companies, home health agencies, and tech companies all store our health data in various forms without standard “denominations.” For example: If you were to request some specific data from a healthcare entity, such as the laboratory values from a serum sodium test, you might actually get the values for a urine sodium test in response. By the same token, one clinic’s categories of race and ethnicity may likely differ from another’s. It is as if all the players in health care are issuing their own “currency”—in this case, healthcare data—without the standardization or “exchange rates” that make easy and secure transactions possible.
In an effort to correct this era of “Wildcat Health Data,” Congress enacted one of the most bipartisan pieces of legislation of this factional era, the 21st Century Cures Act (2016). The Act contains robust language on data “interoperability” and sanctions for “data blocking.” In March of last year, as the COVID-19 pandemic hit, the first regulations enforcing interoperability were entered into the Federal Register.
If you’re wondering why this is important, try for a moment to imagine how difficult life might be without national currencies, rules governing financial transactions, and the ability to perform accounting?
In healthcare, we hear much rhetoric about our profession being data-driven, yet our wildcat approach to collecting, storing, managing, and exchanging data makes it exceedingly hard to actually be so. All our data are being collected in such heterogeneous and not readily transferable ways that the health care economists, business analysts, operations analysts, clinicians and data scientists who are tasked with looking at and analyzing health data do so only with Herculean effort—one perhaps comparable that of an early 19th-century Bostonian trekking to Rhode Island to redeem a bank note.
So many potentially beneficial insights—how our healthcare system works or doesn’t; new therapeutic possibilities; new efficiencies in providing care—are left on the table because of how hard it is to work with Wildcat health data. Even within healthcare institutions, because electronic health records are provided by vendors whose motivations and incentives are not aligned with those of clinicians, it is often difficult to interpret our own data. And when we do figure it out, this knowledge is not easily transferred to other places—we’re just dealing with our own proprietary bank notes.
Our Wildcat era of health data leads to fragmentation, data arbitrage, unfairness, and mistakes. It’s hard to do accounting when one person uses doubloons and another uses Bitcoin.
Such fragmentation is more than inconvenient: it both causes and perpetuates serious inequities. When it becomes difficult to perform true comparative analyses across different clinical, research, and operational areas, inequities—whether structural, intentional, or unintentional—will inevitably arise. Therefore, a necessary prerequisite for fostering an environment that can truly identify and respond to inequity is interoperability. Such interoperability also makes it possible to identify and correct issues related to clinical quality in an agile fashion. After the Civil War, when we created an interoperable national currency, this not only facilitated business transactions, but also gave rise to the ability to audit and create best accounting practices around financial transactions.
Catching FHIR: Building Meaningful Interoperability
Blue FHIR (pronounced “fire”) represents an organizational effort to create transparent and interoperable data transactions and accountability at Duke. While there is now federal statutory language mandating interoperability, the fleshing out the specifics still will still require the development of community standards and best practices. In order to meaningfully participate in a robust national dialog about what these practices should look like, we must figure out what looks right to ourselves.
So what is “FHIR”? It is the Fast Healthcare Interoperability Resources (FHIR) standard promulgated by the Health Level 7 (HL7) organization as the open-data standard that the 21st Century Cures Act embraces for data transactions across all of healthcare. Some things to note:
- FHIR cannot simply be “turned on” by a vendor. Although Epic (our electronic health record platform vendor) is gradually implementing FHIR, it has taken an ambivalent stance regarding such standards (Epic’s CEO wrote a strongly worded letter against interoperability just prior to the federal rule-making in early 2020). As such, our Duke Crucible engineers have identified serious deficiencies in Epic’s implementations. Therefore it will be incumbent on internal effort to meet regulatory goals.
- FHIR is the most fundamental data approach to achieving our quality and equity aims. FHIR confers the ability to create consistent, computable clinical quality or equity outcome measures that can be openly interpretable to all stakeholders at Duke and are also reproducible & consistent (so that a given metric for a Performance Services report will be exactly the same as a metric for a payer or CMS)
- In order to build FHIR Definitions of the things we care about at Duke, a nexus of clinical leadership, technical architecture, and operational leadership are required. Blue FHIR is a key activity to coordinate these interests to identify and prioritize how we implement FHIR. Again, we should be driving this prioritization, not vendors.
- The interoperability provided by FHIR is a rising tide that lifts all boats. This interoperability layer makes a panoply of complementary initiatives easier: remote patient monitoring, digital health, population health and value-based care, de-identification, and data sharing.
- The creation of a national consensus on the electronic definition of conditions and treatments will require a national dialog in which all players in healthcare must participate. For Duke to lead in these conversations, Duke should be completely facile and knowledgeable about FHIR and how we would want to use it internally to conduct the conversations about how it should be used nationally.
Some Parting Thoughts
This blog post will be my last as Director of Duke Forge. I have been given a tremendous opportunity to work in a new and growing space where I can help develop novel approaches to how we help people maintain or improve their health. Stay tuned. Meanwhile, I am glad to see that the teams I have built have the momentum and robust operational management needed to continue on missions such as Blue FHIR.
I recently came across a wonderful blog post by data scientist Vicki Boykis titled “Neural nets are just people all the way down.” In her post, she notes that the most famous applications of machine learning and artificial intelligence only exist because people—not machines!—have done all the brute-force work of annotating images with labels like “pizza” or “traffic light.” What’s more, linguistic hyponyms (such as “pizza,” “pepperoni pie,” or “slice”) must also be curated by humans to ensure that the meanings of these labels make sense.
This human dimension of data science has always been at the heart of Forge’s mission, and this mission continues. Healthcare is not delivered by machines, and in our lifetimes cannot and should not be delivered by algorithms outside of the context of humans. Automation undoubtedly can smooth out burdensome red tape and ease repetitive workflows, but the profession—the calling— of maintaining, improving, and saving lives is “people all the way down.”