Healthcare technology continues to advance with ever greater sophistication and functionality. Yet one key component in the health IT toolkit has not kept pace with innovation. Patient matching technologies, including conventional master patient indexes (MPIs), hit a wall 10 years ago, and it’s time to admit it: conventional MPIs are a dying breed.
The reason is that patient matching needs have grown exponentially at exactly the same time the underlying technology for patient matching solutions reached its limits. The challenge lies in the limitations of the algorithms used for matching, the underpinning of today’s MPIs.
Case in point: patient matching errors cost the average healthcare facility $17.4 million a year, according to the 2016 National Patient Misidentification Report, a recent survey of healthcare executives. The same report cites that 86% of respondents say they have witnessed or know of a medical error that was the result of patient misidentification.
Conventional MPIs have Reached Their Mathematical Limits
An MPI’s job is to connect patients to their medical records—and to ensure that disparate medical records belong to the same person. MPIs accomplish this by comparing a patient’s demographic profile (name, address, birthdate, Social Security number, phone, email, etc.) to every patient profile in the master patient database to find a match. If two profiles meet a statistical threshold to be considered a match, the MPI declares they are the same person, and that patient’s records are linked together under a common medical record number.
This process works perfectly if patient demographic data contained in medical records are perfect. But that’s rarely the case. Instead, patient records hold inconsistent dirty data due to data entry errors, system defaults for birthdays or Social Security numbers, name or address changes over time, nicknames or aliases, junior/senior ambiguity, and cultural name variations, to mention a few.
Since data quality is the problem, why not tune the algorithms to accommodate for this? Because probabilistic algorithms have a mathematical limit to their matching abilities, which was reached a long time ago. And no amount of algorithmic tuning can fix this problem—no amount of tuning can enable an MPI to match one record with a maiden name and old address to another record containing a married name and new address. Or to match two records containing sparse and different data elements, like one record with just a name and birthdate and another with just a name and phone number.
To compensate for these shortcomings, healthcare organizations invest an enormous amount of time, effort, and money to correcting patient data and keeping data up-to-date. They form data governance committees to enforce strict data standards; hire specialized IT people to tune the MPI’s algorithms; perform data cleansing exercises to periodically clean up the MPI; and mobilize data stewardship teams to manually review and resolve matches that the MPI cannot make on its own. Even still, match rates aren’t nearly what they should be. And with the proliferation of patient data, it’s only getting worse.
It’s Time for a New Approach
Healthcare organizations desperately need an alternative to conventional MPIs. Two current industry-sponsored competitions are searching for new ideas for patient matching. The Office of the National Coordinator for Health IT (ONC) is sponsoring one and the other is a $1 million prize from the College of Healthcare Information Management Executives (CHIME). Additionally, the debate around a national patient ID is resurfacing, and HHS may soon begin researching the impact of requiring hospitals to improve patient matching in order to participate in Medicare.
Providers are scrambling to find point solutions that might help, including investments in biometric readers, ID cards, and even smartphone apps. But these are all temporary patches for a problem that runs much deeper. And they fail to address the billions of legacy medical records that already exist that don’t contain biometric stamps, ID numbers, or smartphone QR codes.
The healthcare industry would be wise to look at other industries to learn how they match records to their customers. For example, many retail and banking organizations leverage a “referential matching” approach to identify people and match records with extremely high accuracy and precision. These referential matching approaches compare an organization’s customer records to a reference database that contains a vast and comprehensive array of commercially available demographic data about each person in the U.S.—data that include name changes and variations due to marriage, current and old addresses, and common nicknames and aliases. By matching two records to this “answer key” of demographic data, referential matching approaches can determine if two records belong to the same person even if those records have old, incomplete, or very different demographic data. In other words, referential matching can make matches that algorithmic approaches can never make.
Because of their extremely high accuracy and precision, it is absolutely critical that referential matching solutions begin making their way into healthcare, where accurate matching can dramatically reduce costs and improve patient safety.
I propose that the first step is to build a new generation of cloud-based MPI solutions that incorporate referential matching approaches. Because these next-generation MPIs are cloud-based and leverage reference databases spanning the entire U.S., they will be universally accessible for any healthcare organization to simply “plug into.” And they will be able to address an entirely new universe of patient identitification and patient matching challenges that conventional MPIs can never address because of their limited matching capabilities and lengthy implementation times. This new generation of “universal” MPIs will enable today’s interoperability and population health initiatives; they will enable the rapid mergers of hospital systems and EHRs; and they will enable tomorrow’s innovative health IT vendors to quickly gain advanced matching capabilities for their platforms simply by “plugging in” to a universal MPI.
This is where the healthcare industry is headed. Conventional MPIs have reached their breaking point and a dramatically different, totally new approach is necessary.
The Future of Patient Matching
ONC has recommended a new goal for organizations to reduce their duplicate rates from the current 8-12% national average down to 0.5% by 2020. That’s a great goal, but we know that duplicate rates are only going to rise in the coming years as data quality degrades and the number of connected IT systems multiplies.
Conventional MPIs have been essential components of the healthcare IT ecosystem since the 1990s. Today, they are no longer effective and need to be laid to rest. Fortunately, the industry is starting to pay attention and is opting for greater accuracy and improved results in the form of next-generation universal MPIs.
Mark LaRow is Chief Executive Officer of Verato, a cloud-based platform that resolves and matches identities faster, better, and cheaper than conventional conventional master patient index (MPI) and master data management (MDM) technologies.
