A master patient index in a clinical research stack does work that looks invisible until it goes wrong. Subjects enrolled at three sites end up with three different identifiers; cross-site visits become hard to reconcile; the sponsor's data team spends weekends reviewing dedup reports that should have been preventable. A FHIR MPI is the layer that takes that work upstream, into the moment the identifier is first issued. In 2026 it is one of the quieter foundations a research platform stands on.
This guide walks through what an MPI does in a research context, the capabilities that matter most in 2026, and how to think about sponsor-hosted versus federated when the trial network spans CROs, hospital systems, and decentralized subjects on personal devices.
For more on the broader stack, the clinical data exchange hub on the homepage points to the related explainers.
What an MPI Is in Research
An MPI maintains a single canonical identity for a subject across every system that touches the trial: the EDC at each site, the sponsor's safety database, the ePRO platform, and the lab vendor's portal. In FHIR-native deployments the canonical record sits in a Patient resource, with cross-reference identifiers from each contributing system held as identifiers within that record.
In research the MPI carries weight beyond what hospital MPIs do. It feeds randomization stratification, drives the per-subject audit trail, and underpins the sponsor's view of enrollment at any point in the study.
Capabilities That Matter Most in 2026
Three capabilities separate a usable research MPI from a generic patient-matching tool:
- Honest probabilistic matching with documented thresholds and a reviewer interface for borderline cases, not a black-box score.
- Source-aware identifier handling, so a subject's EHR identifier from one site does not collide with another site's local identifier.
- A clean reconciliation workflow for the moment a subject visits a second site or moves between CROs in a multi-CRO trial.
Most MPIs handle the first capability adequately. A smaller set handles the second. A still smaller set handles the third without forcing a manual review queue that grows faster than the data team can process.
Sponsor-Hosted or Federated in a Multi-CRO World
In ordinary healthcare the MPI typically lives in one place: the hospital or the health network. In research, the MPI question is shaped by the multi-CRO reality of most large trials, where no single party owns the identity layer.
A sponsor-hosted MPI puts the canonical identity in the sponsor's environment. Sites and CROs push identifiers into the sponsor MPI; everyone resolves through it. This is the cleaner pattern for sponsors that want full control.
A federated eMPI distributes the identity layer, with cross-references maintained at each contributing site and reconciled through a shared protocol. This is the cleaner pattern for trial networks where the sites are not willing to push identifiers to a sponsor-owned store.
The Sponsor-Hosted MPI vs Federated EMPI comparison walks through the trade-offs in detail.
Common Pitfalls You Should Know About
A handful of things bite teams in their first research MPI deployment. Sites push different identifier types under the same FHIR identifier slot, and the MPI cannot tell which is which. Sponsor reconciliation runs find too many borderline matches because the threshold was tuned on synthetic data. Mid-study amendments add new sites, and the identifier-namespace plan does not account for them.
The fix is the same in each case: pick an MPI that has been run on a real multi-CRO trial with the same complexity, not on a single-site demo.
Where to Go From Here
Once you understand what a research-grade MPI needs to do, the natural next reads are the product listicles. The top 5 FHIR-native MPI tools for multi-site trial networks covers the products specifically, and the top 7 patient matching algorithms for investigator-site reconciliation takes the algorithm angle.
Picking the right one is less about benchmark match rates and more about whether the reconciliation workflow fits the way your trial actually runs. That is the question worth holding onto before any procurement call.