The following is a guest blog post by Wes Rishel in partnership with Verato.
Henry Ford famously said, “If I’d asked them what they wanted, they would have said ‘faster horses.'” When it comes to patient matching – the cornerstone of health information interoperability – we seem to be asking for faster horses. But what we need is a totally new approach.
The “horse” here is probabilistic patient matching. Probabilistic algorithms match two patient records by comparing them directly to each other and determining the probability that the two records belong to the same patient. Basically, if the demographic data (like name, address, and birthdate) looks very similar across the two records, then a match is made.
These algorithms have been the preferred approach to resolving and matching patient identities since the 1980s. But today’s healthcare landscape is very different from that of the 1980s. Healthcare organizations are no longer simply providers or payers – there are now Health Information Exchanges (HIEs), Accountable Care Organizations (ACOs), care management companies, and even health systems with their own insurance plans.
There is now a larger push to share and exchange patient data between all of these organizations and with state and federal agencies and do analytics on a massive scale for research and population health. And we can anticipate an explosion of patient data coming from many new sources, including patient portals, patient engagement applications, telemedicine applications, personal health records, and Internet of Things (IoT) medical devices.
All of these factors make today’s patient matching challenges much more difficult than those of the 1980s, and yet we’re applying the same matching approach we used three decades ago. The consequences are drastic: up to one in five patient records are not accurately matched within the same health care system according to CHIME, and as many as half of patient records are mismatched when data is transferred between health care systems according to the ONC.
As a technology adviser over the years, I frequently advised governmental and private entities with over a million patient records that, as flawed as it was, probabilistic matching was their only choice. But probabilistic matching has clearly reached its limits. Even large and expensive efforts within healthcare organizations to improve and tune probabilistic algorithms achieve only incrementally better results. It is time to move on from the “horse.” We need a totally new approach.
A completely new approach in healthcare is a familiar approach elsewhere
It is time to emulate Henry Ford and find a completely new approach to patient matching. But it is also important to recognize that Ford didn’t actually invent the car. He didn’t even invent mass production, which had already been applied in other industries. His contribution was the vision that applying mass production to automobiles would open up a whole market, the gumption to gather the investment and execute, and the stubbornness to ignore naysayers.
So it is with patient matching. We simply need innovators that have the vision to apply proven identity matching approaches to the healthcare industry – as well as the gumption and stubbornness necessary to thrive in a crowded and often slow-moving healthcare IT market.
Many industries – including retail and financial services – already have viable and proven solutions to match and link their customer records, and these are the solutions we should look to as an industry to solve our own patient matching challenges.
Most proven solutions hinge on cross-correlating the demographic data from customer records with demographic data from third-party sources, including public records, credit agencies, or telephone companies. Importantly, this third-party demographic data includes not just current and correct attributes for a person, but also out-of-date and incorrect attributes – like previous addresses, maiden names, and common typing errors for birthdates or phone numbers.
By referencing these comprehensive sets of third-party demographic data during the matching process, these “Referential Matching” approaches can significantly outperform probabilistic matching algorithms. For example, Referential Matching can match one record that contains a maiden name, old address, and birthdate with another that contains a married name, new address, and phone number. Both of these records match to the same person in the third-party reference database, which has the entire set of demographic attributes for that person. In essence, this third-party reference database acts as an “answer key” for demographic data.
Results from this approach were recently published in Journal of AHIMA 88, “Applying Innovation to the Patient Identification Challenge” by Lorraine Fernandes, RHIA, Jim Burke, and Michele O’Connor, MPA, RHIA, FAHIMA. This article reviewed how Healthix, the largest public health information exchange (HIE) in the nation, used a vendor built on referential matching architecture to resolve 54.1 million MRNs down to 21.9 million unique identities. These 21.9 million unique individual records are now clear and available to meet key clinical and operational needs.
Referential Matching needs to make its way to the healthcare industry, and luckily it is already being used by many of the largest health systems, payers, and HIEs. But this is not enough. The costs of poor patient matching are too dramatic to keep pushing for faster horses: inaccurate matching decreases quality of care, has drastic implications for patient safety and privacy, costs millions of dollars of lost revenue each year to denied claims, and increases costs to our healthcare system due to systemic inefficiencies, redundant tests and procedures, and unnecessary IT and labor expenditures.
The healthcare industry should take a lesson from Henry Ford. The winning disruptive patient matching solution need not be created, but only adapted from other industries. As another wise man said, “discovery consists of seeing what everybody has seen, and thinking what nobody has thought.”