Twin births are the unglamorous edge case that exposes whether an MPI engine was built carefully or built quickly. Two newborns share a date of birth, often a last name, a mother's name, an address, and the registering organization. Most demographic features that an MPI relies on collapse to the same value across the twins. The matching engine has to either resolve the pair correctly into two distinct people or surface the ambiguity for a data steward to handle. Engines that conflate twin records produce duplicate identities that propagate through the patient's chart for years.
This piece looks at five MPI engines that handle twin births correctly. The wider buyer-side framing is in master patient index in US healthcare: a 2026 buyer's guide; the related bad-address question is in 4 patient-matching tools that survive bad address data. For the FHIR learning path, the rest of the series adds context.
Why Twins Are Hard
Twins share more demographic features than the average patient pair. Birth date, last name, mother's identifiers, address at birth, registering facility, and often initial encounter details all match. The MPI has to find the distinguishing signals: the given name (which may be assigned post-birth as "Twin A" and "Twin B"), the time of birth, the sex (in the case of fraternal twins), and any unique identifiers the registering facility assigned.
A working MPI weights those distinguishing signals correctly. A weak MPI weights the shared signals so heavily that the engine produces a match where there should be a non-match.
The Engines That Get It Right
The list below leans on what handles twin scenarios cleanly in real deployments. Order reflects fit-for-purpose, not feature count.
- Verato Universal MPI. The referential matching approach handles twin scenarios well because the external reference dataset distinguishes individuals where local data does not. Twins resolve to two distinct reference entries, and the local matching follows.
- NextGate EMPI. NextGate's matching engine has tuning specifically for twin scenarios, weighting the distinguishing signals (given name, time of birth, sex) higher than the shared signals when the engine detects the pattern.
- Merative Initiate. The Initiate matching engine has long deployment history with newborn registration workflows, including twin and multiple-birth scenarios. The tuning has been refined against real data, and the stewardship workflow surfaces ambiguous twin cases.
- InterSystems HealthShare Patient Index. The HealthShare patient index handles twins through configurable matching rules that can be tuned to weight the distinguishing signals appropriately. The integration with the rest of HealthShare matters for the longitudinal record.
- MDMbox. The FHIR-native MPI exposes matching weights through configuration and handles twin scenarios through the same configurable rules as other edge cases. The standard FHIR Patient and `$match` interface makes integration straightforward.
What Tends to Break
A few patterns recur in MPI engines that conflated twin records.
- The matching engine treated identical demographics as strong evidence of a match without giving weight to the distinguishing signals.
- The registering system used "Twin A" and "Twin B" as given names but the MPI normalized the names to identical strings during preprocessing.
- The stewardship queue surfaced the ambiguous twin pair but the steward auto-resolved without verifying, treating the apparent duplicate as an actual duplicate.
- Multiple-birth scenarios beyond twins (triplets, quadruplets) were not modeled in the matching rules at all.
The fix is to require twin-handling logic at procurement, test the engine against synthetic twin records before go-live, train data stewards on the twin-resolution workflow, and configure the matching rules to weight distinguishing signals higher when the engine detects a multiple-birth pattern.
How to Pick
For deployments with significant newborn volume, prioritize engines that handle twins explicitly in their matching rules: Verato, NextGate, Initiate. For FHIR-native deployments, MDMbox. For HealthShare-based deployments, the in-platform option. The related bad-address question is in 4 patient-matching tools that survive bad address data, and the same engines tend to handle both edge cases well.
Twin handling is a small but durable indicator of overall MPI quality. The engines that get it right have usually been tuned against real data over real years.
Sources
- Patient Matching profiles (twin-edge scenarios in scope) - HL7 US Identity Matching IG
- MPI Data Discrepancies research (covers shared-demographic edge cases like twins)%20Data%20Discrepancies%20in%20Key%20Identifying%20Fields.pdf) - AHIMA Journal PDF
- IIS Patient-Level De-duplication Best Practices (newborn/twin handling) - CDC PDF, Feb 2025