Why patient matching needs a dual-track approach

Patient matching with records affects all of us, most of the time without our knowledge.

Whether it is the receptionist at a doctor’s office asking for your date of birth or someone on a care team pulling records from a Health Information Exchange, one of two things is happening behind the scenes. An identifier is establishing your identity across different silos of information, or some kind of statistical matching is being performed.

The identifier is a simple solution, but historically has not been viable. Despite this, our current technology landscape offers two tools to make a universal patient identifier (UPI) practical—the advent of big data and the ubiquity of smartphones.

A 2008 RAND report provides a comprehensive overview of why one UPI per patient at a national, state or regional level is a clear win for patient identity matching. The conclusion was straightforward:

“The foregoing analysis indicates that a healthcare system in which every patient has a unique, non-disclosing patient identifier is clearly desirable for reducing errors, simplifying interoperability, promoting NHIN architectural flexibility, and protecting patient privacy,” the report states.

Barriers to UPIs start at the federal level, where an essential moratorium on studying a UPI was instituted by Congress. For this reason, many have suggested that a hybrid model will be the only feasible path to UPIs. In a hybrid model, statistical matching is used to build a network of UPI-enabled patients. When the network reaches a critical mass, we will start to see the benefits.

There are two possible models for a UPI gaining traction and eventually reaching critical mass. One is the advent of tools driving the current trends in machine learning—large datasets, abundant computing power and the network effects inherent on the Internet. The second is a distributed model where patients leverage the ubiquity of smartphones and the power of a consumer-facing player with a household name, like Apple or Google, to connect to their health records.

The current hype cycle in machine learning is being driven by the abundance of computing power available in the cloud, as well as tools to manage extremely large amounts of data in a distributed fashion. If all HIEs in America suddenly joined forces to share a patient matching infrastructure, we have the technology to hold and analyze all their data. More importantly, the wealth of data could lead to breakthroughs in how we match patients.

For example, most statistical matching algorithms currently rely on some level of human intervention, such as the thresholds at which matches are considered accurate. However, this is a job much better left to the machines because they are able to simultaneously see the entirety of the data, and process feedback instantly. A similar strategy driven by people would take years of experimenting.

Decentralization of the UPI can be imagined through Apple’s recent launch of electronic health record (EHR) integration. Patients manually link their iPhone to their patient portal account, and can then carry their data with them. In the future, data would hypothetically be readable/shareable wherever the phone goes. Even more amazing, future enhancements would include data like demographics, emergency contacts and even insurance information syncing over to the EHR with a fingerprint, much like ApplePay.

UPI via a smartphone must work across the fragmentation of the smartphone ecosystem, as well as the fragmentation of the EHR ecosystem. Apple’s medical records solution is limited to a handful of sites among three leading EHR vendors, and the expansion trajectory is unclear. A similar solution does not exist for Android, which takes up the majority of the market. Health systems will also need to deal with the obvious security, fraud and privacy concerns. Despite this, the presence of personal devices in our lives is here to stay, and may well be the vector for UPI.

Patient matching boils down to two opposite strategies—either you know any two given patients are the same person through a unique identifier or you rely on statistical matching. A unique identifier is a clear win, but is impractical to do overnight for a number of reasons. Machine learning and smartphones have already transformed a number of industries. Time will tell whether or not one of the oldest problems in healthcare can see new light under the lens of these new paradigms.