To ensure accurate matching, the system has to have personnel review about 1% or 2% of its records, Oriol said. But with a health information exchange, “we're up to 9% with these things that require people to intervene and match. … It's not ideal.”
Moreover, sometimes by the time the matching question is resolved, the physician has already seen the patient and lost the opportunity to incorporate the external information into their decisionmaking. “We'd like to have that information flowing like when you open a tap at your house,” he said.
Katherine Lusk, chief health information management and exchange officer at Children's Health, Dallas, said her hospital has had a quality assurance and data governance program for patient matching for many years. Children's Health is one of 11 organizations participating in a pilot by the American Health Information Management Association to create a national information governance model, which will include guidance on patient matching.
The model, scheduled for public release this year, will include a tool to identify gaps in an organization's patient-matching approach, such as inadequate training, Lusk said.
Cost is also another reason to fix the matching problem, said Marc Probst, vice president and CIO at Intermountain Healthcare, Salt Lake City. The integrated delivery system studied it a few years ago and estimated it could save between $4 million to $5 million a year simply by doing a better job of matching records.
Intermountain recently announced results of a patient-matching improvement project with its statewide HIE, the Utah Health Information Network, which includes University of Utah Health Care, its chief rival. “The University of Utah came up with a pretty good algorithm that got us closer,” Probst said. But much of the improvements came by virtue of participants reaching consensus on doing basic things the same way, such as agreeing on a larger than normal number of data fields to use for matching.
It included not only the first and last names, sex and dates of birth, but also Social Security numbers, home phone numbers, race and home addresses. The project standardized how those attributes are recorded, which improved the rate of automatic matching across the exchange to 95% from 10%.
Dan Chavez, executive director of San Diego Health Connect, that Southern California community's local health information exchange, is scheduled to talk at HIMSS about patient matching and how his HIE overcame “the largest obstacle to EHR exchange.”
Chavez estimates that 30% of EHRs have basic data on patient identities such as names, addresses or Social Security numbers that are old, incomplete or incorrect and therefore can't be matched across providers without manual intervention. The San Diego HIE cut its manual effort by 75% and doubled its matching accuracy using third-party data.
Earlier this month, CHIME kicked off a National Patient ID Challenge with HeroX, an online platform to promote challenges linked to tech innovations. They're offering $1 million in prize money—to be raised by CHIME and other sponsors—to encourage developers to address patient matching on a national scale with 100% accuracy.
Physician informaticist Dr. Barry Hieb said his not-for-profit organization, Global Patient Identifiers, based in Tucson, Ariz., will be a CHIME prize competitor. The GPI approach is to generate and store unique patient identifiers, but not store patient medical records themselves. A GPI database will keep track of those providers that have records for each stored identifier. Users of the system will be charged a few cents each time they query it for the whereabouts of a patient's records.
Improving patient matching will take far more than a technology upgrade, important as that may be, said
Dr. Charles Jaffe, CEO of Health Level 7, a healthcare standards development organization.
“However good patient-matching algorithms are, I would never argue that you have the perfect algorithm,” Jaffe said. “It's really contingent on the quality of the demographic data you've collected and that tends to be error-filled. Even if we do get a patient identifier, we'll still have problem with data entry in the identifier.”