A Choir Without Harmony Is Just Noise: Accepting the Challenge of Data Harmonization in Transplantation

The novelty of big data in healthcare is subject to debate. For transplant, big data turns 40 this year. In 1984, the US Congress passed the National Organ Transplantation Act, which in part established the Organ Procurement and Transplantation Network to maintain a national registry for organ matching. Since that time, several notable advancements have been made in national transplant data quality and scope. What started as a registry for organ matching has become a vehicle for public reporting of patient outcomes and a powerful resource in a toolkit of national data sources used by researchers to drive innovation and progress.1,2 Despite its size, our national data system lacks vastness. We still fail to collect prelisting patient or process metrics, and definitions of many variables are elusive. Rejection, for example, may be defined by a bolus of solumedrol, a histological examination, or the presence of varying degrees of donor-specific antibodies in serum. The art of clinical practice remains prominent in transplantation, contributing enormous value alongside an ever-evolving science. This presents a challenge for a uniform model of data collection spanning multiple allografts, health systems, geographic boundaries, and allocation systems. Yet, the power of uniformity as it pertains to data collection and organization is undeniable, and progress sacrificed in its absence is also difficult to quantify. Analytic capabilities are evolving at lightning speed, and in the wake of reports from several national regulatory bodies, it is inevitable that the transplant national data system will be improved. But how? Hosseini et al3 laid out the essential next steps along with a well-referenced foundation in data concepts and terms that are commonly deployed yet inconsistently defined. In this month's issue of Transplantation, they make clear the case for establishing a common data model for transplant which ideally would incorporate pre-waitlist steps as well as robust, individual- and community-level social determinants of health. As the authors point out, a common data model serves as a framework to define data elements through structured vocabularies while specifying collection protocols and variables essential for linkage across datasets. A well-established common data model built through robust collaborations between clinicians, informaticists, and data scientists can translate the lexicon of the bedside into a discrete, organized system of measures that can translate clinical practice into actionable information.4 Common data models have been developed in a range of clinical disciplines, but the Observational Medical Outcomes Partnership (OMOP) is a prime candidate for the adoption of a common data model in transplant. Established in 2008 and housed at Columbia University, OMOP has supported a range of specialties in adopting a common data model to accelerate research to meet the demands of high-priority public health needs. Other examples outlined in the article, including PCORNET, i2b2, and All of Us, provide a foundation for integrated data sets that could allow our field to reach beyond the transplant center into the upstream disease states that feed the unfortunately high demand for a severely limited organ supply. The elusive "denominator" problem in transplant could be quantifiable through the adoption of such a model, allowing transplant research to incorporate the biologic and social mechanisms of disease that disproportionately affect the same populations too frequently eliminated from our waitlists. But are we ready? Transplant feels universal to those of us who practice it clinically, but the variability in health system operations, reimbursement policies, and a myriad of health delivery processes all affect the lift required to turn harmonized, unified transplant data from fantasy to reality; and that is within the United States. Across national borders, the effort becomes compounded by varying healthcare delivery systems, health insurance practices, and legal requirements for data privacy. Even in a hypothetical scenario where transplant processes were executed and reimbursed under a unified system, we are still left to grapple with several unique aspects of transplant that are hard to capture. These include changing health states over time, unclear boundaries between phases of transplant care, ambiguous diagnostic codes, and a mountain of unstructured data that contains invaluable information about motivation for transplant, caregiving network support, and adherence to care regimens. Yet, without discrete, organized, curated data, the capabilities of advanced analytics will remain out of reach for us and our science. More importantly, our patients will suffer as a result. So, what is the next step? Transplant clinicians will find this article an accessible primer to the foundational concepts of data harmonization. Those of us at the bedside need to work alongside clinical informaticists, biostatisticians, and data architects to ensure that our practice is effectively translated into discrete, organized measures. This effort starts locally, but a larger, international collaborative effort is urgently needed to adjudicate which patient, process, allograft, and outcome measures are most essential for large-scale studies. Those measures must be strictly defined through consensus and brought to a data science team, in cooperation with electronic health record servicers, who are poised to translate them into defined variables mapped to specific points in our health system data infrastructures. This is an effort that will fail if executed in silos, and professional transplant societies are well poised to lead collaborative efforts to bring these definitions and protocols to national and international regulatory bodies. It is difficult to say if the field of transplantation is ready to align with an established common data model, such as OMOP. We will not know until we try, and we must try together. Transplant is the home of the early adopter, and several have been calling for data harmonization efforts for years. The challenges seem overwhelming, but the rewards are without limits. Let us roll up our sleeves and get to work.