Duplicate MPI Pairs After a Six-Site EHR Go-Live
Harbor Valley Health System went live on a new EHR eight months ago across six registration sites and a 410-bed hospital. The EMPI now holds 412,000 records, and a vendor audit identified 3,710 potential duplicate pairs (0.9%). Of those, 214 pairs carry conflicting blood types, allergy entries, or different dates of birth, and 46 of the 214 already have active medication orders filed on both records. Registration turnover is 34%, and the downtime procedure still creates a new medical record number at the laboratory interface. The system signed an HIE participation agreement that begins in 60 days. The CMIO wants the duplicate list worked top to bottom by two analysts who hold merge rights in the EHR and no written merge policy.
1. How should the 3,710 pairs be sequenced, and what must be in place before any merge is performed?
Read the worked answer
Sequence by patient-safety consequence, not by list order or count.
- Tier 1 (do first): the 46 pairs with active orders on both records. Two live medication profiles can produce duplicate dosing, duplicate anticoagulation, or conflicting allergies at the point of care. These need same-day clinical review, not a clerical merge.
- Tier 2: the remaining 168 pairs inside the 214 with conflicting demographics, blood type, or allergies. Verify against a source document (photo ID, prior chart, lab report) before merging.
- Tier 3: cosmetic duplicates where demographics agree and no clinical data conflicts - safe to batch.
Before any merge:
- A written merge/unmerge policy with named roles, a two-person verification step, and a defined unmerge/audit-trail capability. Merges are effectively irreversible in many systems; a mass merge without unmerge is a one-way door.
- A pre-merge image of both records retained per record-retention policy, since the legal record must remain reconstructable.
- Merge rights limited to trained staff with documented competency - the two analysts should not hold unrestricted rights merely because they were available.
- A hold on HIE go-live linkage until Tier 1 and Tier 2 are cleared, because outbound exchange propagates a bad link to every trading partner.
2. Which root causes should you investigate first, and who should own the fix?
Read the worked answer
Test the causes that explain volume, not the ones that are easiest to fix.
- The laboratory downtime interface is the highest-yield suspect: it produces a fresh MRN every downtime event and does so silently. Pull the count of MRNs created by the interface during the last 12 months and match them against the duplicate list. If a disproportionate share traces there, the fix is a staged-record process and interface change control, owned by IT with HIM sign-off.
- Turnover-correlated registration error - compare duplicate origination rates by site and by registrar tenure. Six sites with 34% turnover usually show one or two outliers; that points to training and search-discipline problems, not to the system.
- Search behavior - whether staff search by name only when DOB is unknown at the point of registration.
Ownership is a governance question. The HIM/data governance function owns the standard (search rules, verification elements, merge policy); registration leadership owns compliance; IT owns interface and EMPI tool configuration. Assigning all three to one analyst guarantees the policy never gets written. The data governance committee should be chartering this, not the CMIO informally.
3. What single measure should the data governance committee track monthly, and why that one?
Read the worked answer
Track new duplicate pairs created per 1,000 registrations, by site, with the 214 high-risk pairs reported separately as an open-item aging count.
Reasoning:
- The raw duplicate count is a stock, and a stock will rise forever in a growing system even if the process is improving. It tells you nothing about whether the fix worked.
- Duplicates per 1,000 registrations is a flow and is comparable across sites of different volume.
- The high-risk subset must be separate because it measures unresolved patient-safety exposure and needs an aging clock (how many older than 7 days), not a rate.
- A data quality measure reported without an owner and a threshold is decoration. Attach a target (for example, under 2 per 1,000) and require the site with the worst rate to report a corrective action each month.