MATCH IT Targets the Symptom, Not the Matching Engine
Congress returned to the MATCH IT Act in March 2025 with a familiar promise: reduce the patient matching failures that continue to disrupt care, billing, and privacy protections across the U.S. healthcare system. Backed by AHIMA, HIMSS, CHIME, AMIA, and the American College of Physicians, the bill would direct HHS to define “patient match rate,” require ONC to adopt a minimum demographic data set for certified health IT, and create a voluntary Medicare bonus for providers who meet an accuracy threshold once one exists.
The problem is not whether patient matching matters. It clearly does. Inaccurate matching can delay treatment, trigger incorrect medication or procedure decisions, expose protected health information, create duplicate records, and drive claim denials. Black Book Research has estimated that one in three healthcare claim denials stems from inaccurate patient matching, with an average hospital losing $2.5 million annually and the industry absorbing an estimated $6.7 billion a year.
One sentence captures the stakes: measuring a flawed process more consistently does not fix the process itself.
Why Standardizing a Metric May Leave Errors Intact
The bill’s strongest feature is also its narrowest. By standardizing how patient matching performance is measured, federal policy could create a common yardstick for vendors, health systems, and regulators. But a uniform metric does not tell organizations which identity-resolution method they should use, and that omission leaves the underlying failure mode untouched.
Most electronic health record systems rely on probabilistic scoring across demographic fields such as name, date of birth, sex, address, phone number, and sometimes Social Security number. Each health system then sets its own confidence threshold for what counts as a match, while “possible” matches are routed to manual review. This approach can work reasonably well inside a single enterprise, but it becomes fragile across organizations because thresholds differ. The same pair of records may be considered a match in one system and a mismatch in another, even when the demographic inputs are identical.
That variability is not a cosmetic issue. It is the central reason patient identity errors persist in interoperability settings. When the operating rules vary, the result varies with them.
Hybrid Deterministic-Probabilistic Matching Follows a Different Logic
A better approach combines probabilistic scoring with deterministic rules. In practical terms, that means an algorithm does not rely only on soft demographic similarity. It also checks hard identity anchors such as legal name, date of death, Social Security number, place-of-death county code, gender, record ownership status, abandonment status, and upload status. A hybrid model can capture cases that a narrow probabilistic system would push into manual review or miss altogether.
This matters because patient matching is not merely about reducing clerical noise. Identity resolution is a safety function. When a record is misassigned, downstream clinicians may make decisions using incomplete or incorrect information, and health plans may process the wrong claim under the wrong member file. The clinical harm can be immediate when a medication history is attached to the wrong person or a critical result disappears into a duplicate chart.
Hybrid logic does not eliminate human oversight. It reduces how often humans are forced to clean up system-level ambiguity.
Vital Records Systems Show What High-Stakes Matching Requires
Vital records systems offer a useful model because they operate under far less tolerance for error. State systems register roughly 2.8 million deaths each year and feed identity data downstream to Social Security, Medicare, insurers, and pension boards. In that environment, a false positive can interrupt benefits for a living person, while a false negative can allow a fraudulent claim to continue unnoticed. The consequences are financial, legal, and deeply personal.
Recent real-world mistakes show how damaging mismatches can be. In 2025, the Social Security Administration mistakenly declared two living men dead, and both had their benefits cut off over demographic mismatches that took months to resolve. Those cases underscore a basic reality: identity errors are not hypothetical edge cases. They can produce direct harm when a system treats incomplete demographic similarity as certainty.
State electronic death registration systems have responded by using hybrid deterministic-probabilistic methods. Combining a deterministic match on Social Security number and date of death with a probabilistic check on name variants can resolve cases that a name-and-birth-date-only engine would send to manual review. In high-stakes identity workflows, that broader logic is not exotic. It is operationally sensible.
Configuration Still Matters, but the Method Has to Be Right First
The Office of the National Coordinator for Health IT has noted that hybrid systems can require complex configuration, governance, and monitoring. That is a real implementation burden. Hybrid matching is not a silver bullet, and different organizations can still set different thresholds, which means variation can persist even when the core logic improves.
Yet the existence of trade-offs does not make the method interchangeable with a simpler one. A probabilistic-only model leaves more room for inconsistent classification because it depends heavily on local policy choices. A hybrid model narrows that room by anchoring the decision to both statistical similarity and deterministic identity checks. In other words, it raises the floor before organizations start customizing thresholds.
For hospitals and health systems, that difference is operationally important. A vendor that supports only probabilistic matching may be perpetuating avoidable duplicates, denials, and manual reconciliation work. A hybrid engine can improve results without waiting for Congress to define a new reporting framework.
Health Systems Should Ask a More Precise Question
The legislative debate has drifted toward whether Congress can establish a cleaner national standard for patient matching. A more useful question is whether the current matching architecture is capable of producing reliable identity decisions in the first place.
For health IT leaders, the next step is concrete: ask vendors whether the matching workflow uses deterministic rules alongside probabilistic scoring, or only probabilistic scoring. If it is the latter, the system may be relying too heavily on local thresholds and manual review to compensate for an incomplete method. In a field where a wrong merge or a missed match can alter treatment, privacy, and payment, method selection is not an implementation detail. It is the core of the safety problem.
Congress may still pass a reporting fix. The harder work is inside the engine room of the record-matching system, where better logic can prevent the next duplicate chart before it reaches the bedside or the claims department.