How AI-enhanced identity matching is solving patient ID woes in healthcare systems

Accurately matching patient records across different systems is an eternal challenge in healthcare data management. When a patient named “Elizabeth Smith” arrives at the hospital, is she the same person as “Beth Smith” who visited last year and lives in the same town but has a different phone number? What about the “Liz Smith” from two years ago with the same address but a slightly different birthdate?

Patient-by-patient, matching records is simple, if tedious. But determining which records belong to which patients at scale is a remarkably complex project, with profound implications for patient care, administrative efficiency, and healthcare costs. Patient misidentification costs the average hospital $1.5 million annually, with the U.S. healthcare system losing over $6 billion from denied claims related to inaccurate patient identification, per research from Black Book.

“[Patient matching] sounds like an easy problem to solve, but it turns out to be ridiculously difficult,” said Drew Ivan, Chief Architect at Rhapsody, during his session Bringing the Right AI to the Job at ViVE 2025. “If you get two records and they have the same name, address, and birth date, it’s probably the same person. But if they have the same name and address and a different birth date, is it the same person or not? It could be a son with the same name as his father, or it could be a mis-entered birthdate.”

EMPI systems, pesky gray areas, and scaling challenges

At the heart of patient identification is the Enterprise Master Person Index (EMPI), which serves as the authoritative repository for patient identities across an organization. When functioning correctly, an EMPI maintains a single, comprehensive view of each patient despite data existing in multiple systems. Healthcare organizations typically match patients by comparing demographic information like names, birth dates, and addresses across records, assigning weights to different elements and applying probabilistic matching rules.

Traditional EMPIs calculate similarity scores between records, automatically matching those with high scores and creating new patient entries for those with low scores. However, this approach leaves a problematic “gray area” of records with moderate similarity scores that could potentially be matches or non-matches. 

As healthcare organizations grow and data volumes increase, the traditional approach faces significant scaling challenges. These potentially duplicate cases enter a queue for manual review by data stewards, trained specialists who investigate and resolve ambiguous matches. This creates a substantial bottleneck, particularly when large volumes of new data enter the system through mergers, acquisitions, or system migrations.

“As you get bigger and bigger populations that you’re trying to manage, you need more and more data stewards,” Ivan explained. “The number of data stewards you need scales as the size of your problem.” 

Replicating human judgment with AI

To overcome these limitations, Rhapsody is turning to artificial intelligence (AI), specifically machine learning and neural networks, to enhance the traditional matching process. Rather than replacing existing algorithms, these AI systems augment them by learning from patterns in human decision-making.

Ivan notes that despite adding increasingly sophisticated features like nickname support customized for different regions and countries (i.e., recognizing that “Mike Johnson” and “Michael Johnson” may be the same person in the U.S.), Rhapsody “hit a wall where no amount of algorithmic effort was able to give us another increment of improvement.”

The machine learning approach breaks through this ceiling by learning from actual human decisions rather than relying solely on pre-programmed rules. This allows the system to recognize subtle patterns that would be nearly impossible to codify otherwise. The result is a far more nuanced matching capability that significantly reduces the “gray area” requiring human intervention.

A patient ID solution with a long list of benefits

The AI-supported approach directly improves data quality by reducing both the volume of records requiring human review and the time these potential duplicates spend in queues. This translates to lower risk in the downstream clinical and administrative systems relying on accurate patient identification and increased operational efficiency and patient safety.

By combining the precision of traditional algorithms with the learning capabilities of artificial intelligence, a system like Rhapsody’s can address one of healthcare’s most persistent challenges. As this technology matures and becomes more widely adopted, it promises to improve the integrity of patient data across the healthcare ecosystem, saving billions while directly benefiting patients.