By David Kronfeld, Head of Real World Data Innovation, Medidata Solutions
Medtech is just starting to access the rich well of real-world data (RWD), which offers some incredible opportunities to build more successful and useful devices. RWD technology can be employed to improve patient recruitment, efficiently deploy a sales team, identify unmet needs in regions, and to tell a value story.
But there is a problem with medical device companies’ expectations of RWD. Data aren’t magic—they are a digital representation of what actually happens in a real care setting—and there is no such thing as complete data.
Although many organizations are working to improve how data are collected, processed, and managed, significant challenges of applying RWD remain. While we wait for technology to catch up, here are some tips on how to think about RWD:
Problem #1: Data sources are fragmented
Similar to clinical trials, CROs and sponsors face a variety of data sources from which to pull for medical device trials. Electronic medical records (EMRs), claims data, and registries are commonly used sources of healthcare- and patient-related data for medtech. Each of these may or may not be useful, depending on the business case we are trying to solve, and each has its limitations.
- Electronic medical records
EMRs are messy. EMR data lives in silos—many, many silos—and as of yet, there are no universal standards that allow users to map different datasets together, nor are there any scaled data-sharing initiatives. Efforts are underway to improve the problem, but not at the pace device makers want.
- Claims data
Claims data are extremely powerful. These databases are used to drive improvements in population health and address issues related to cost, quality, and outcomes. They include information at the patient encounter level regarding diagnoses, treatments, and billed and paid amounts. They complement EHRs and combine to offer a broad view of patient interactions. The data comes de-identified for HIPAA compliance, but retains the ‘chain of custody’ and can be used to find patients through their unique provider identifiers. Although this type of data is also not perfect, it can be valuable depending on the use—especially for commercial applications.
Registries typically focus on patients who share a common reason for needing care, allowing physicians to see what treatments are available and how patients with different characteristics respond to certain therapy. They can yield great insights. However, registries are retrospective in nature and are only as good as the information that goes in, which leads us to our second problem.
Problem #2: Data are not as accurate or complete as you might expect
The most frustrating idea about data is that it doesn’t yet offer us the ability to connect insights to individual patients. This can be wildly unsatisfying, especially when it comes to patient recruitment. Often, the best data can do is tell us how hard recruitment will be.
EMR systems are simply digital versions of old-style patient records where doctors would manually write notes and file them away. Although they contain a lot of information, they are not designed to report insights for individual patients, nor across patient populations. Further, the data are only as good as the information provided by the physician. Data recording is a required task for billing and reimbursement, yet if a physician doesn’t use all the checkboxes and pull-down menus properly, the results are less valuable.
Healthcare providers tend to prefer free text fields because they match how caregivers have historically recorded care. Those fields often contain accurate, higher value information. Natural language processing will be able to create insights from that treasure trove (but isn’t quite there yet), and extra care and security has to be applied considering that this field often includes protected health information.
Problem #3: Patient records are fragmented
Individual patients often have multiple records, which further complicates an already fractured system. People often think that if they’ve seen the same doctor for years, all their data are together. But that might not be true. For instance, the records might be different if they’ve visited the emergency room or been referred to a specialist.
Claims data can often be highly accurate and reflect a continuum of care, but only for the time a patient was under that coverage. A single patient might change insurance, and change care providers several times in their lives. Each will have a record of every system they’ve ever touched, resulting in a wide medical data footprint. In a use case, we ask, how much of that data footprint is needed to yield insights?
How data are getting better
When sponsors ask, “How complete is the data?” often the answer is, “It is as complete as it can be.” This may or may not be complete enough for an FDA submission. The inherent challenges of data need to be respected, but companies should also not be deterred from pursuing a RWD strategy.
Technology companies are getting better at linking datasets to create unified footprints. For example, in April, Datavant acquired UPK (Universal Patent Key). UPK provides HIPAA-compliant de-identification services for healthcare data, while Datavant provides services to safely link their data to improve medical research and patient care. The acquisition could help streamline both services. Along a similar vein, HealthVerity has tokenized technology that provides HIPAA-compliant methods to link records. Over time, these technologies will help.
Other companies are looking to unify data on the level of single EHR. Pharmacy cooperatives, such as altScripts, have rich data from their EMR systems. However, we should understand that these systems also have gaps. For example, physician ordering, and internal or external labs during a hospital stay might not be part of the EMR or EHR—they have their own systems. So, a complete medical record needs to be one step above the level of an EMR and accept a variety of data types.
Such health-system level companies are worth watching. These organizations include Guardian Research Network (GRN) and Syapse, both of which focus on oncology patients. They are looking at how to marry inpatient activity, outpatient activity, medical records, and other clinical systems within the healthcare setting to create a holistic view of the patient journey.
Lastly, we are approaching the point where interested parties will soon move away from “wanting the data” to only wanting the insights it brings. There’s an enormous amount of data out there, and managing and applying data science to it all is an expensive process. When it comes to RWD, every opportunity has a consequence. This is why it is important to first understand the business problem you’re trying to solve. When you’re clear on that, RWD is a tool that can yield innovative advancements for medical product developers.