Choosing a Master Patient Index for a FHIR-native EHR is fewer feature checkboxes and more about operational tradeoffs. The seven questions below are what production teams end up asking after the RFP is signed and integration starts.
1. Deterministic-only or hybrid probabilistic? Deterministic MPI works cleanly under 500k patients. Beyond that, probabilistic scoring becomes necessary. The US Core Patient profile supports both signals, but not all MPIs use them equally.
2. Real-time vs. batch matching latency. Real-time (<200ms) matching supports point-of-care workflows. Batch (hourly or nightly) supports analytics but not intake. Verify both SLAs.
3. Merge tooling — automated or HIM-review? Auto-merge is fast but risks bad merges. HIM review is safer but adds queue depth. Best-in-class MPIs support both, with confidence thresholds gating auto vs manual.
4. Death registry integration. State death registries update quarterly. MPIs that don't reconcile flag deceased patients as active for months.
5. Duplicate detection cadence. Some MPIs run duplicate detection nightly, some weekly. Weekly leaves duplicates in the wild too long for high-volume intake environments.
6. Split resolution. Patient charts can be split (one person, two records). Split resolution needs a merge, but linking merged charts to downstream FHIR resources (Patient.link) requires cascade update tooling.
7. HIPAA de-identification for reporting. MPI systems that emit patient counts for reporting must handle small-cell suppression. Regulatory requirement varies by state.
Vendor scorecard, mid-2026
| Vendor | Deterministic | Probabilistic | Auto-merge | Death registry |
|---|---|---|---|---|
| Verato | Full | Full | Configurable | Native |
| NextGate | Full | Full | Configurable | Add-on |
| Aidbox MDM | Full | Basic | Manual | Manual |
| MITRE FRIL (open) | Basic | Full | Manual | External |
MPI is a five-year decision. Get the seven questions above answered before signature, not after.