Biometrics, algorithms help boost hospitals’ patient-matching rates

When a patient walks into a hospital for an appointment, there’s usually an unspoken expectation: They’ll share their name, date of birth and maybe a few other demographic details with a registrar. That helps staff pull the correct medical record, so the patient will be able to see their physician, now armed with their complete medical history and health data.

It seems like a simple process, but the back end is much more complicated, filled with a mix of technological elements like data standards and algorithms, not to mention room for human error.

A few seemingly small, but consequential, problems could emerge if a name or date of birth is entered with a typo, if a patient has recently moved to a new address, if there are inconsistencies in the way addresses are written, or if patients with similar information are confused with one another.

Similar demographics resulted in confusion at a Camden, N.J., hospital in November, after medical staff gave a kidney designated for one patient to another transplant recipient with the same first name, last name and name suffix.

The mix-up took place when a transplant coordinator at the hospital, Virtua Our Lady of Lourdes Hospital, spotted the patient’s name on a transplant list, not realizing there were two patients with the same name. She didn’t review the patient’s date of birth or Social Security number, which would have revealed the mix-up, according to Dr. Reginald Blaber, executive vice president and chief clinical officer at Virtua Health, the hospital’s parent organization.

After a review of the incident, “we realized that we did not have a process in place to safeguard against human error,” Blaber said. “As an institution, we should have had a fail-safe, where we forced her to confirm that she had the right patient—going through multiple means of identification.”

Virtua now uses four identifiers for all transplant patients—name, date of birth, last five digits of the Social Security number and the match number from the United Network for Organ Sharing—and there are multiple points at which a second staff member is required to confirm those details.

“We knew this was not our best day,” Blaber said. “We knew we had to do better.”

The patient who should have received the transplant got a kidney about a week later, according to the hospital.

A classic example of how data can be misleading came after a Houston-based health system reviewed its patient population, and found it had 2,488 patients named Maria Garcia, 231 of whom had the same date of birth.

If any of those Maria Garcias aren’t matched with their correct record, a physician may not be fully informed of their medical history—a particular concern if, say, someone has “a disease or ailment that requires certain care,” said Jose Serna, senior manager for patient access management at the system, Harris Health System.

It’s even more challenging to match patient records between hospitals, where match rates can be as low as 50% to 60%. That’s a foundational challenge for interoperability, noted Ben Moscovitch, the Pew Charitable Trusts’ project director for health information technology. If a hospital can’t identify a shared patient at a nearby facility, it won’t be able to get their records. “Bottom line is that effective patient matching is critical to achieving interoperability,” he said.

That makes seamless patient matching a particular point of concern as companies are rushing to develop vaccines, antibody tests and other methods to support immunity to COVID-19. Generally, immunization registries have been plagued by many of the same matching issues as the rest of the healthcare ecosystem. There could also be challenges integrating COVID-19 or antibody tests into medical records, as many patients are visiting third-party labs or drive-through sites.

To ease patient-matching struggles, some countries have issued a national patient identification number. The U.S. is not one of them—Congress for decades has prohibited HHS from using funds to develop a unique patient identifier, citing issues related to privacy and security.

The House of Representatives in June 2019 voted to overturn the ban. The Senate did not follow suit.

“For 20 years we’ve not been able to discuss” a national patient identifier, said Julie Pursley, director of health information management practice excellence at the American Health Information Management Association. While most agree a national patient identifier wouldn’t be the ultimate solution, it could be a helpful piece of the puzzle.

Many hospitals today have seen success rolling out new technologies and best practices to aid with patient matching, such as biometrics, algorithms and staff training.

At Harris Health System, Maria Garcia wasn’t the only problem. About 249,000 of the system’s patients shared a first and last name with at least one other patient, according to an analysis the system completed years ago—which made patient matching a time-consuming task for registration staff, who manually searched for the correct record by asking for a patient’s name and date of birth.

Harris Health, which treats a large immigrant population, also faces challenges since “a lot of our patients don’t have Social Security numbers,” a data element many healthcare organizations use to bolster matching rates, said Amber Wingo, the system’s administrative director of patient access and revenue cycle systems. “Combined with the common names and the (other) commonalities, that was one of our big issues,” she said.

Besides patient-care concerns, poor patient matching has steep costs. Inaccurate patient identification accounts for roughly $1,950 in duplicative medical care costs per inpatient and $1.5 million in denied claims per hospital each year, according to a survey by Black Book Market Research.

So to get around those challenges, Harris Health implemented a system that identifies patients based on the pattern of veins in their hands; it has been a mainstay of the system’s patient-matching strategy since 2011. Patients who come visit a hospital or clinic place their hand on a device—“give us a low-five,” Wingo says—that scans their palm.

That scan, along with a patient’s date of birth, are the two data elements used to identify patients.

Harris Health’s duplicate record rate is now 7%, down from 10% before installing the palm-vein scanners. And because of more accurate matching, the system now employs just one full-time staffer to evaluate and merge possible duplicate records, instead of a full team.

The vast majority of patients have enrolled in the palm-scanning system, according to Wingo. There was apprehension among some patients at the beginning, system leaders acknowledged. Much of the concern related to whether the biometric information would be shared with law enforcement. The palm-vein data is only used internally and not shared outside the organization, Serna said.

To make sure patients understood that, Harris Health rolled out the palm-vein scanners to its hospitals and clinics in a phased approach. For each of the facilities in which the technology was live, staff hung posters in English and Spanish explaining the new process and would direct patients with additional questions to patient advocates.

But matching records to palm scans doesn’t work for everything. Harris Health still relies on searching for patients based on their standard demographic data elements, like names and dates of birth, when matching patients with their records from outside the health system—including from a local health information exchange.

While programs that recognize patients based on palm, fingerprint and iris scans have proved useful for matching patients within the same organization, getting good matches between different facilities is more challenging.

In focus groups conducted by Pew, biometrics proved to be one of the most popular matching solutions patients cited interest in. But without the entire healthcare system adopting a single, standardized form of biometrics, it won’t be useful for matching between facilities.

University of Utah Health gets daily reports from a statewide health information exchange, UHIN. The reports include information on which of the system’s primary-care patients visited an emergency department and were admitted to or discharged from another organization.

The process to match patient information from other organizations is set up in the back end, so it doesn’t require manual intervention from the care management team at University of Utah Health, said Shelly Medley, care management supervisor at the system.

UHIN’s reports typically include a patient’s medical record number, some demographic data, the hospital they went to, and their admission and discharge date. Armed with that information, University of Utah Health’s care management team can reach out to patients to see if they need a follow-up appointment, as well as to ensure they understand any care plans or new medications.

Medley said the care management team will always look up the patient by their medical record number—provided in UHIN’s reports—to double-check it’s the correct patient, but there are rarely any problems.

University of Utah Health has been working with UHIN for years. That has included a patient-matching improvement project that UHIN, University of Utah Health and Intermountain Healthcare collaborated on from 2015 to 2017, which involved developing protocols for Utah’s poison-control center to collect demographic information from callers.

“The key to any HIE is matching patients across different systems and different organizations,” said Cody Johansen, UHIN’s director of operations.

To match patients across facilities, UHIN uses probabilistic matching algorithms that compare multiple demographics, including name, date of birth, gender, address, phone number and Social Security number, to determine the likelihood of two records being from the same patient. If enough of those items match closely enough, the software will link the two records.

If some of the demographics match, but the system is unsure, it’ll flag the records for additional review.

UHIN runs those possible matches through software that uses external, third-party data—such as information from public records—to determine whether to merge two records. This method, called referential matching, works by creating a more comprehensive view of a patient with information like previous names and addresses.

Regularly reviewing those possible matches is important to avoid a long queue of records needing confirmation.

Still, Johansen stressed that while matching algorithms are important, health systems and HIEs need to ensure they’re working with high quality and consistent data.

“You can do so much with an algorithm, but if the source data isn’t clean to start out with then there are limitations on what you can do,” Johansen said.

To get clean source data, it’s not just about the IT underpinnings—there’s also a human component that demands special training. That’s been key to helping Children’s Health in Dallas maintain a low duplicate record rate. The system boasts a duplicate record rate of just 0.1% or 0.2%, according to Katherine Lusk, chief health information management and exchange officer at Children’s Health.

“We really focus on data integrity,” Lusk said, which means the system gives immediate feedback to registrars when a mistake they’ve made results in a duplicate or erroneously merged patient record.

To create a new medical record, registrars at Children’s Health are prompted to input complete legal name and address in the format used by the U.S. Postal Service for patients, as well as date of birth, gender, race, phone number and even email address. That information—along with previous names and addresses—is the same data used to match patients that visit the system in the future.

Email addresses aren’t typically used for matching, although most hospitals do collect the information from patients, according to Pew’s Moscovitch. That’s despite the fact email addresses are one of a few “important data elements” recent research has suggested could help to “facilitate greater match rates,” Moscovitch said.

And, according to a study published in the Journal of the American Medical Informatics Association last year, standardizing last names and addresses—specifically to the format used by the U.S. Postal Service—has proved helpful for improving matching sensitivity.

After a new medical record is created at Children’s Health, data integrity specialists manually review each one to determine if it’s a potential duplicate that needs to be merged with another file, or if it really is a new patient.

If a data integrity specialist unearths a problem—a record that’s been erroneously created or merged—“immediately, the registrar knows that they’ve made a mistake,” Lusk said, as the data integrity specialist sends the registrar feedback on what policies they might have missed.

Children’s Health also shares quarterly reports with a health information management committee, showing duplicate record rates by department.

Hospitals need to set up these feedback loops, so registrars are alerted when they’ve pulled the wrong patient record or input a patient’s demographic information incorrectly. That should include training registrars not only on the technical aspects of their responsibilities, but also on “the importance of what they’re doing,” such as the possible clinical ramifications if something goes awry, AHIMA’s Pursley said.

“Show them how to correct it in the future,” Pursley said of acknowledging mistakes. But beyond that, explain how “identifying the patient upfront is a core function to everything that we do.”

Another component of Children’s Health’s patient- matching strategy has involved figuring out when standard demographic data—even if input perfectly—can still trip up some algorithms.

In the case of a newborn, for example, a hospital will need to develop protocols for how to create a medical record for an infant who hasn’t been named yet. There’s also challenges that arise with young twins or triplets, since many will share the same date of birth, address and phone number. In those cases, a data integrity specialist might need to delve in to the clinical data for assurance that they’re looking at the right patient.

Twins and triplets proved challenging when working with regional HIEs, Lusk said. HIEs would often try to erroneously merge young siblings with one another, since their patient-matching algorithms and practices were developed for adults—so Children’s Health learned it needed to keep a close eye on those cases and flag patients who are part of multiple births.

“In our very fragile population, oftentimes we have to look just a little bit deeper,” Lusk said.