Duplicate Charts Start at Registration, Not in Cleanup

Every management services organization eventually funds a duplicate chart cleanup. Somebody runs the potential duplicates report, a team works through it for a quarter, and the number goes down. Then it climbs back, because nothing changed at the place duplicate charts are actually created, which is a person on the phone deciding whether the Robert Martinez in front of them is the Bob Martinez already in the system.

Cleanup is treating the symptom. The generation rate is the disease, and it is set by what happens in the first ninety seconds of an inbound call.

For a management services organization the problem has a second layer. Each affiliated practice registers patients its own way, on its own configuration, and the same person can exist several times across entities without anything flagging it. Every acquisition adds another population of records that were built to somebody else’s standard.

The work that fixes this is not a data project. It is a change to how the front office searches before it creates.

Duplicates are a registration behavior, not a database property

The evidence points at the point of capture rather than at the matching software.

ONC’s patient identification and matching report describes an error rate below eight percent as the industry-recognized standard for matching while noting many organizations exceed it, and cites a study of 112 master patient indexes that found a mean duplication rate of eight percent, with a quarter of those indexes above ten percent. The same report notes that a 2008 study traced the majority of patient identification errors in one emergency department back to registration, where information was entered incorrectly.

That is the whole argument in two findings. The rates are meaningful, and the origin is the intake conversation.

The report is also clear that manual review is a permanent cost rather than a temporary one. Matching thresholds are deliberately set to produce duplicates rather than false positive overlays, as a patient safety choice, which means a human review queue is a structural feature of running an accurate index and not a sign that something is broken.

An MSO has a structural version of the problem

A single practice creates duplicates. An MSO creates duplicates and overlaps, and the second kind is harder to see.

The structures vary enormously and nobody should assume theirs is typical. Some organizations run as a single MSO-level EHR account with separate tablespaces underneath, each requiring its own authorization and each potentially carrying different appointment types. Others run every practice off one enterprise with completely different configurations per practice. Others again combine their practices with provider-level department groups distinguished by a name prefix. There is no single answer to what an organization of this kind looks like.

Underneath that, locations are sometimes distinct billing entities inside the same system, which changes what a basic question like “when was the patient last seen” even means.

ONC’s terminology is useful here. A duplicate is more than one record for the same person in one system. An overlap is more than one identifier for the same person across two or more facilities in the enterprise, and it commonly arises after institutional merging. Every acquisition an MSO makes is an overlap generating event, and the front office at the acquired practice has no way to know it.

Search harder before you create

The behavior that prevents duplicates is boring: search thoroughly, and search more than one way, before creating anything.

Under time pressure a person searches once, on the spelling the caller gave, does not find it, and creates a chart. The variations that defeat that single search are the same ones every time. Nicknames, Robert and Bob and Bobby. Changed surnames. Hyphenated names entered with and without the hyphen. Transposed birth dates. A phone number that moved to a new person.

An automated intake does not get tired at 4pm, and that is most of the advantage. It can run enhanced matching against multiple attribute combinations, check phone and date of birth independently of name spelling, and surface near matches for confirmation rather than silently creating a second record. The right default is to make creating a new chart the harder path and finding the existing one the easier path, which is the opposite of how most workflows feel to a rushed person.

This is also the recommended practice rather than an invention. ONC’s SAFER Guide on patient identification sets out the practices that work. Registration staff should be trained to look for an existing record before creating a new one. The registrar should be prompted to consider potential matches at the moment a new record is created. Records with a high degree of similarity that fail to match because of missing demographic data should be flagged for manual review.

The confirmation step matters as much as the search. When the system finds a likely match it should verify with the caller using something they can answer, then attach to the existing chart. A near match acted on without confirmation is an overlay, where one patient’s information lands in another patient’s record, and that is a worse outcome than a duplicate.

Where a person must decide

Merging records is not an automated decision, and it should not become one. The AI’s job ends at surfacing a high confidence potential match with the evidence attached. A person decides.

Any near match the patient cannot confirm goes to staff. So does every case where the demographic data conflicts in a way the caller cannot resolve, and every suspected overlay, which needs urgent human attention because clinical information may already be in the wrong chart.

Family relationships are the recurring trap. Parents and children sharing a name, spouses sharing an address and phone, twins sharing a birth date and a surname. These look like duplicates to any matching algorithm and are not. Linking family charts correctly is administrative work with real consequences, and it belongs to a person who can ask a direct question.

Measure the creation rate, not just the backlog

The two numbers to watch are the duplicate creation rate, meaning new duplicates against registration volume, and the database duplicate rate. Most organizations only ever look at the second one, which is a backlog measure that moves when a cleanup runs and tells you nothing about whether the process improved.

Watch creation weekly by location and by the staff workflow that produced the record. It will not be evenly distributed. It concentrates in specific sites, specific shifts, and specific intake paths, and that is actionable in a way an enterprise-wide percentage never is.

For an MSO the same report should be readable per entity, because a newly acquired practice generating duplicates at three times the rate of the others is a training and configuration problem you want to find in month one rather than in next year’s cleanup.

Key Takeaways

Stop funding cleanups without changing intake. Duplicates are created at registration and a cleanup that does not change the creation rate buys a temporary number.
Duplicate rates are meaningful and the origin is the intake conversation. One study of 112 master patient indexes found a mean duplication rate of eight percent, with a quarter above ten percent.
Expect a permanent manual review queue. Matching thresholds are deliberately tuned to produce duplicates rather than overlays, so human review is a structural cost of an accurate index.
Know your own structure. MSO configurations vary widely, from separate tablespaces per practice to one enterprise with different configurations per site, and locations may be distinct billing entities.
Treat every acquisition as an overlap generating event. More than one identifier for the same person across facilities is the MSO-specific failure and it arrives with the deal.
Search multiple ways before creating. Nicknames, changed surnames, hyphenation, transposed dates, reassigned phone numbers. Make creating a new chart the harder path.
Never merge automatically and never attach on an unconfirmed near match. An overlay is worse than a duplicate.
Track the creation rate weekly by location and workflow, not just the total backlog. It concentrates, and that makes it fixable.

An MSO that measures only its duplicate backlog is measuring the output of a process it is not watching. The charts get made one at a time, by people doing their best on a call with someone who is spelling their name for the third time. Give that moment a better search, a confirmation step, and permission to hand the hard ones to a person, and the cleanup you fund next year is much smaller than the one you funded this year.

Sources