Master Patient Index systems in FHIR-based healthcare use specific operational patterns. Five that ship in 2026.
Pattern 1: Deterministic + probabilistic hybrid. Deterministic on identifier + name + DOB catches 92-95%; probabilistic handles the rest.
Pattern 2: Real-time + batch matching. Real-time (<200ms) at point-of-care intake. Batch (nightly) for duplicate cleanup.
**Pattern 3: Patient.link for merge tracking.** type: replaced-by and replaces represent merges. Downstream follows the chain.
Pattern 4: Empirical weight calibration. Off-the-shelf weights don't reflect regional distributions. Calibrate against your labeled dataset.
Pattern 5: HIM review queue. Auto-merge above threshold (typically 95-98%); manual review between 60-95%; auto-non-match below 60%.
Operational metrics
| Metric | Target |
|---|---|
| Auto-merge accuracy | >97% |
| Manual review turnaround | <10 days |
| Weekly new duplicates | <2% of intake |
| Death registry lag | <30 days |
| Split resolution turnaround | <14 days |
Common MPI mistakes
1. Skipping weight calibration. 2. Auto-merge threshold too aggressive. 3. No death registry reconciliation. 4. Weekly duplicate detection insufficient. 5. Missing HIM staffing.
Vendor selection
| MPI | Auto-merge | Probabilistic | Weight calibration | Merge tooling |
|---|---|---|---|---|
| Verato | Full | Full | Automated | Native |
| NextGate | Full | Full | Manual | Native |
| Aidbox MDM | Full | Basic | Manual | Native |
| MITRE FRIL | Basic | Full | Manual | External |
Scale considerations
1. Under 500k: deterministic + basic probabilistic. 2. 500k-5M: full hybrid, HIM team. 3. 5M+: enterprise MPI service, dedicated ops.
MPI is an operational discipline more than a technology choice. Sites investing in the five patterns above ship cleaner charts than sites that don't.
