Reference Data for Clinical Trials: Sites, Investigators, and Countries in One Clean List

One Site, Many Records: How Trial Master Data Drifts Apart

Start with a single site in a single Phase 3 study and follow it through the systems. The study manager creates it in the CTMS during feasibility, typing the hospital name from an email signature. The EDC vendor builds the study and creates the same site from a spreadsheet the CRO sent, with a site number and an abbreviated name. The IRT team loads the site from a different extract so the site can receive drug. The eTMF gets a site folder named after the principal investigator, because that is how the TMF team searches. The safety database records the reporter’s institution from the case form, which the site coordinator filled in by hand. Finance creates a payee record from the site’s invoice letterhead, which shows the hospital’s research foundation, a separate legal entity with its own tax number.

None of these steps is wrong on its own. Each team needed a record, and each created one in the system it controls. But six months later the sponsor has six records for one site, and none of them agree on the name, the address, or which legal entity is involved. When the principal investigator moves to a new hospital and the site is transferred, some systems are updated, some are not, and one system overwrites the old address, erasing where the study was conducted for the first two years.

The Systems That Hold Site and Investigator Data

The table below lists the systems that typically hold a copy of site, investigator, or country data, what each one uses it for, and how the copy usually gets created. The point is not that any of these systems is badly designed. The point is that each one creates its own record unless someone tells it not to.

The Problem Is Measurable in Public Data

Sponsors sometimes assume that name variation is a minor cleanup issue. Public registries show how large it gets when free text is allowed. A 2019 analysis in Trials examined interventional drug and biologic trials registered on ClinicalTrials.gov that started between 2005 and 2014. In 12% of 35,121 trials, the investigator’s name or role was missing. Among 71,359 name and role pairs, 17% of the “names” were not names of real people at all. The authors identified 19 categories of name variants and found that 13% of the names had variants that a program could not resolve.¹

A companion study of the Clinical Trials Registry of India by an overlapping research group found the same kinds of problems, including variations in names and incomplete details for ethics committees. Its recommendations are the same ones a sponsor should apply internally: fewer free-text fields, more drop-down selection, and pre-registration of the names of people and organizations so that each trial record selects from an existing list rather than typing a new entry.² That last recommendation is, in plain terms, a master list with controlled selection. Registries reached that conclusion after years of messy data. Sponsors can apply it from the start.

How This Differs From Coding Dictionaries and Reconciliation

Two related problems are worth setting aside so the scope is clear. Coding vocabularies such as MedDRA, WHODrug, and CDISC Controlled Terminology are external reference sets with release schedules, and the governance question there is version control. Reconciling participant-level data between EDC, safety, and IRT is a different job again: making sure the same adverse event or randomization appears consistently in each system. This article is about neither. It is about the operational master data that every one of those systems depends on: which site is which, who the investigator is, where the site is, and which country it belongs to. If those records are wrong, reconciliation compares records that are not about the same site to begin with.

What Bad Site Data Does at Inspection and at Payment Time

Leaders tend to fund master data work when they can see what it prevents. For site and investigator data, the clearest consequences arrive in three places: FDA inspection planning, financial disclosure and screening, and payment transparency reporting.

FDA Uses Your Site List to Plan and Direct Inspections

For marketing applications, FDA asks applicants to provide a comprehensive table of all clinical sites in each major study, a table of entities that took on contracted study activities, by-site participant data listings, and a summary-level clinical site dataset named clinsite. The CBER draft guidance published in June 2024 explains why. FDA relies on the site information to select sites for inspection, and accurate contact information lets FDA’s field investigators contact clinical investigators to schedule inspections and make sure the inspection goes to the correct location, meaning where the records are available for review. The same draft notes that applicants have often provided this information in variable formats that do not support timely inspection planning.³ The CBER draft states that it will contain the same requirements as CDER’s draft guidance on the same topic, first published in February 2018, and both remained drafts as of the most recent technical conformance guide.^3,4

The format FDA asks for is the useful part for anyone designing a master. The Bioresearch Monitoring Technical Conformance Guide (version 3.1, September 2024) lays out the clinical site table with a site identifier, the investigator’s name with any prior clinical investigators in parentheses, the site address at the time of the study with the updated address where one exists, and contact information both at the time of the study and as updated. Its example flags a site where the clinical investigator changed at the sponsor’s request before study completion.⁴

Form FDA 1572 Changes Have to Reach the IND

Under 21 CFR 312.53(c), before an investigator begins participating, the sponsor must obtain a signed Form FDA 1572 with the investigator’s name and address, the research facilities where the study will be conducted, the clinical laboratories, the responsible IRB, and the list of subinvestigators.⁵ FDA’s information sheet on the 1572 adds two details that matter for data design. The investigator’s name in Section 1 should be the full legal name, and when information on a signed 1572 changes (an IRB address, a new subinvestigator, a new clinical laboratory), the investigator documents the change and informs the sponsor so that the sponsor can update the IND, for example through an information amendment.⁶ A sponsor whose CTMS shows a new subinvestigator that never made it into an IND amendment has a gap that an inspector can find by comparing two documents the sponsor produced.

Financial Disclosure Depends on a Complete Investigator List

Under 21 CFR 54.4, an applicant must submit a list of all clinical investigators who conducted covered clinical studies and must certify or disclose financial interests for each one.⁷ The definitions in 54.2 make this a data problem as much as a compliance one. “Clinical investigator” includes any listed or identified investigator or subinvestigator directly involved in treating or evaluating research subjects. “Significant payments of other sorts” are payments from the sponsor to the investigator or the institution, to support the investigator’s activities, worth more than $25,000 during the study and for one year after it ends, excluding the expense of running the clinical studies themselves.⁸

Two failures follow from duplicate investigator records. First, the same person under two spellings can be certified under one and missed under the other, leaving the application’s list incomplete. Second, payments that should be totaled for one investigator get split across two records, so each record falls under the $25,000 threshold while the true total is over it. Neither failure is visible to the people who prepare the disclosure unless the investigator records are already resolved.

Screening Against Debarment Lists Uses Names

FDA’s debarment list for drug product applications identifies debarred persons by last name and first and middle names, with the effective date and term of each debarment.⁹ There is no identifier to match on. Screening investigators, subinvestigators, and key vendor staff against it therefore depends on having the person’s full legal name, which is exactly the attribute that drifts most across systems. Screening a nickname from the CTMS against a list published under a legal name is a weak control.

Open Payments Research Reports Carry Investigator Identity

For applicable manufacturers, research payments under a written agreement or protocol are reported to CMS separately from other payments. Under 42 CFR 403.904(f), the report includes the receiving institution or individual, the total research payment, the study name, related products, and, for each covered recipient who is a principal investigator, the name as listed in NPPES, the National Provider Identifier, a state license number, specialty, and primary business address. The ClinicalTrials.gov identifier is optional.¹⁰ Under 42 CFR 403.912, failing to report timely, accurately, or completely carries a civil money penalty of $1,000 to $10,000 per payment, with an annual cap of $150,000, rising to $10,000 to $100,000 per payment and a $1,000,000 cap for knowing failures. These base amounts are adjusted annually for inflation.¹¹

The research payment report is built by joining site payments to principal investigator identity. If the PI on the study-site record is stale because the investigator moved and nobody end-dated the old link, the payment gets attributed to the wrong person or the wrong institution. Commercial teams face a version of this problem with healthcare professional master data, but research reporting has its own twist: the PI attribution comes from clinical operations records, not from the commercial customer master, and the two are rarely reconciled.

In the EU, CTIS Draws on Organization Master Data

The EU picture runs through EMA’s Organisation Management Service (OMS), which stores validated organization names and location addresses, each with a unique Organisation_ID and Location_ID. OMS does not assign roles: the same organization can be a marketing authorization holder for one product and a sponsor for another.¹² In the Clinical Trials Information System, organization search retrieves OMS data by default. For specific areas, including Part II trial sites, the site involved in a serious breach notification, and inspected sites, users who cannot find an organization in OMS can search for organizations registered locally in CTIS, and then add a new local organization if needed.¹³ A sponsor that already knows the OMS identifiers for its EU sites can select the right record every time. A sponsor that does not will create local entries, one submission at a time.

Separate the Person, the Place, and the Study Site

Most of the duplication described above comes from one design mistake: treating “the site” as a single thing. In practice, the word “site” means at least three different things depending on who is speaking, and a single record cannot hold all three without breaking.

When a CRA says “site 0104,” they mean a study-specific assignment that exists for one protocol and has an activation date and a closure date. When a feasibility lead says “the site,” they mean the hospital or research center, which exists independently of any study. When a monitor says “the site has a new PI,” they are talking about a person who has a career, credentials, and a history of moving between institutions. These three things change at different rates and for different reasons. A useful master separates them.

Why the Study Site Must Stay Separate

The study site is the entity people most often try to merge away, because two study sites at the same hospital look like duplicates. They are not. Site 0104 in protocol A and site 2210 in protocol B may share a hospital and a principal investigator, but they have different IRB approvals, different activation dates, different site numbers in the EDC, and different contracts. Merging them would corrupt every participant record attached to either site number. The right model links both study sites to the same person and the same location, which is what lets you answer “which studies has this investigator run at this hospital” without collapsing the records that the participant data depends on.

The study site is also where history lives. When a principal investigator leaves and a new one takes over, the study site keeps its number and its participants, but the PI link is end-dated and a new link starts. When the investigator moves to another hospital and brings the study along, the regulators and the EDC usually see a new site with a new number, while the person record stays the same and gains a new affiliation. Both cases are routine. Both break a model in which the investigator’s name is a text field on the site record.

Why the Place Needs Two Levels

The organization and location split is not just a technical preference. It solves a specific payment and inspection problem. The legal entity that signs the clinical trial agreement and receives payment is often not the same as the physical place where participants are seen. A university hospital may contract through its research foundation, see participants at two clinics across town, and send samples to a laboratory it owns at a third address. Section 3 of the 1572 lists research facilities, and FDA’s information sheet says that an investigator who sees participants at more than one site should list all of them.⁶ A master that records one address per “site” cannot represent this. A master with an organization, its locations, and parent and child relationships can.

Identifiers in Use and What Each One Cannot Do

A master list needs its own internal identifiers for every person, organization, location, and study site, because no external identifier covers all of them. External identifiers are still worth storing, because they make matching far more reliable and they let the master match the identifiers that regulators and registries use. What matters is knowing what each identifier was designed for and where it stops being useful.

FDA’s BMIS Is a Lookup, Not a Master

FDA’s Bioresearch Monitoring Information System publishes clinical investigators, CROs, and IRBs named on a Form FDA 1572 or 1571 submitted to CDER since October 1, 2008, updated quarterly. FDA is explicit about two limits. BMIS is not intended to be a comprehensive list of investigators, because sponsors are not required to submit the 1572 to FDA and studies outside the US not run under an IND do not generate one. And the file contains a separate entry each time an investigator appears in a new submission, so an investigator named in 10 INDs appears 10 times. The published field layout also caps the first name at 14 characters and the last name at 24.¹⁴ BMIS is a useful cross-check. It cannot serve as your investigator master, and loading it as one would import duplication by design.

ROR Shows How to Handle Renames and Mergers

The Research Organization Registry is designed for research affiliations, and its data model includes organization types such as healthcare, education, and company, a status of active, inactive, or withdrawn, and relationships of parent, child, related, predecessor, and successor. Two rules in its documentation are worth copying directly into a site master. When an organization only changes its name, ROR adds the new name and keeps the previous one on the same record, with no successor relationship. When an organization stops operating, its record becomes inactive and can point to a successor, and ROR suggests that users replace inactive or withdrawn records with the successor where one exists. For DOI metadata, ROR recommends keeping the name and ID historically correct even after an organization is merged.¹⁵ That is the same principle the BIMO site table applies to addresses: keep what was true at the time, and link it to what is true now.

Shared Investigator Registries Reduce Re-Collection

Industry efforts to share investigator profiles across sponsors have aimed at the same problem from the outside. TransCelerate’s Investigator Registry initiative, now marked complete, was built to enhance the Shared Investigator Platform and create a shared repository of business contact details and study participation information for consenting investigators. Its stated benefits include less need for investigators to keep updating multiple records and less duplication of site qualification work.¹⁶ Where a sponsor participates, storing the registry identifier on the person record is a strong matching key. Where an investigator has not consented to sharing, the sponsor still needs its own record.

Country and Region Codes: ISO 3166 Is the Start, Not the Whole Answer

Country looks like the easy part. In practice it is where many sponsors find the most variation per record, because every system that has a country field accepts something different: two-letter codes, three-letter codes, English names, local names, and the occasional city typed into the wrong box. Fixing countries first is usually the fastest early result in a site master program, but it helps to know the standards well enough to avoid creating a new problem.

What ISO 3166 Provides

ISO 3166 has three parts. Part 1 covers countries, dependencies, and other areas, with a two-letter code (alpha-2) that ISO recommends as the general-purpose code, a three-letter code (alpha-3), and a three-digit numeric code. Part 2 covers principal subdivisions such as states and provinces, written as the country’s alpha-2 code followed by up to three characters; ISO’s own examples are ID-RI for the Riau province of Indonesia and NG-RI for the Rivers province of Nigeria. Part 3 covers formerly used country names, with four-letter codes. ISO notes that the standard does not itself define country names, which come from United Nations sources, and that the codes are free to use. The current codes are on ISO’s Online Browsing Platform, where users can subscribe to notifications of changes.¹⁷

Two practical points follow. First, choose one code form as the stored value (alpha-2 is the usual choice) and derive the others, rather than letting each system pick. Second, store the code, never the name. Names change even when codes do not. ISO 3166-1 changed the English short name for TR from “Turkey” to “Türkiye” effective July 11, 2022, while the code stayed TR.¹⁸ A system that stored the code needed no change. A system that stored the name now holds two spellings for one country, and any report that groups by country name splits it in two.

Why FDA Outputs Need GENC

The US government does not use ISO 3166 directly. The Geopolitical Entities, Names, and Codes (GENC) Standard is the US government profile of ISO 3166. It exists because US law requires federal agencies to use geographic names approved by the US Board on Geographic Names, and because US recognition policy does not always match the entities in ISO 3166.¹⁹ FDA announced in 2019 that it would adopt GENC on December 17, 2020, noting that GENC uses ISO 3166 names and code elements wherever possible, with changes only where US law or recognition policy requires them.²⁰

This shows up directly in inspection planning outputs. The BIMO Technical Conformance Guide changed its instructions from ISO codes to the GENC code list in version 3.0 (August 2022), and in the clinsite dataset specification the COUNTRY variable takes the GENC code for the country where the site is located, while STATE takes the GENC unabbreviated preferred name of the subdivision.⁴ For most countries the GENC and ISO codes are the same, which is exactly why the differences get missed. The safe design is to keep ISO 3166 as the stored code, maintain a governed crosswalk to GENC, and generate FDA outputs from the crosswalk rather than hand-editing the submission dataset.

Regions Belong to the Company

Operational regions (for example, “North America,” “Western Europe,” “Asia Pacific”) are not part of any standard. They are company decisions that reflect how clinical operations is organized, and they change when the organization changes. One company might group Israel with Europe for clinical operations and with the Middle East for commercial reporting; another might move Mexico between North America and Latin America after a reorganization. Neither choice is wrong. What goes wrong is embedding the region into the country field, or letting each system keep its own region mapping. Keep regions as a separate, dated mapping from country codes to region names, owned by clinical operations leadership, so that a reorganization changes one table instead of six systems.

Matching and De-Duplication Rules for Sites and Investigators

Building the master starts with finding duplicates in the records you already have, and it continues for as long as new records arrive. The record linkage literature is mature, and the core ideas apply directly. A tutorial in the International Journal of Epidemiology describes the main options: deterministic linkage, which links records only when chosen fields agree exactly; probabilistic linkage, which scores agreement across several fields and allows partial agreement such as small spelling differences; cleaning and standardizing fields before comparison; limiting comparisons to blocks such as the same region so the work is feasible, at the risk of missing matches across blocks; and setting two thresholds so that pairs fall into links, non-links, and an uncertain middle group that needs clerical review. The authors note that trying to catch more true matches means more clerical review.²¹

What differs for trial master data is not the method but the rules about what may be merged. The three entities need different rules, because a wrong merge does different damage to each.

Standardize Before You Compare

Most apparent duplicates disappear once fields are standardized consistently. For organizations and locations, that means expanding or removing common abbreviations (St., Saint, Med Ctr, Medical Center), removing department suffixes into a separate field, normalizing punctuation and case, parsing addresses into street, city, subdivision code, postal code, and country code, and validating postal codes against the country. For people, it means separating titles and degrees from names, recording the legal name from the signed 1572 as its own attribute, holding preferred names and former names as aliases, and normalizing transliteration for names written in non-Latin scripts. Keep the original value alongside the standardized one. Stewards need to see what was entered, and auditors may ask.

Rules for People

• Match on identifiers first. Two records with the same verified NPI, the same national license number in the same jurisdiction, or the same registry identifier are strong candidates. Treat an identifier conflict (same name, different NPI) as a reason to keep records apart until a steward checks.
• Never merge on name alone. Common surnames repeat within a single hospital. Require at least one confirming attribute beyond the name: an identifier, a date of birth where you are permitted to hold it, a CV, or a confirmed history of affiliations.
• Do not use email or phone as the deciding attribute. Both change when an investigator moves institutions, which is the exact moment duplicates are created.
• Treat a move as a new affiliation, not an edit. When an investigator changes institutions, end-date the old affiliation, add the new one, and leave every existing study-site link pointing at the location where that study took place.
• Send uncertain pairs to a person. The clerical review queue is not a failure of automation. It is where the hard cases belong, and its size tells you whether the matching thresholds are set sensibly.

Rules for Places

• Match the organization and the location separately. Two records may be the same organization at different locations, or different organizations (a hospital and its research foundation) at the same address. Both are common and both are legitimate.
• Use block keys that respect geography. Comparing within the same country code and postal code region keeps the work manageable. Review a sample of cross-block pairs periodically so that a mistyped postal code does not hide a duplicate forever.
• Prefer external organization identifiers where they exist. An OMS Location_ID or a ROR ID on both records is strong evidence. Absence of an identifier is not evidence of a different organization.
• Model departments as children, not duplicates. “Department of Oncology, University Hospital” is a child location or unit of the hospital, not a different hospital. Keep department detail in your own master: EMA has told CTIS users that OMS does not capture department or unit details for hospitals, universities, and similar institutions.²²

Rules for Study Sites

• Do not de-duplicate across protocols. Two study sites with the same PI and hospital in different protocols are two records by design. Link them; do not merge them.
• Do de-duplicate within a protocol. Two records for the same protocol and site number in different systems are the same study site and should carry the same master ID.
• Do not reuse site numbers. A closed site’s number should stay retired within the protocol, so that participant records and the TMF never point to two different sites under one number.

Survivorship and Merge History

When two records are confirmed as the same entity, the master needs a rule for which value survives in each attribute. Rules work best when they are set per attribute based on the source of authority, not by ranking whole systems. The legal name comes from the signed 1572. The organization’s legal name and registered address come from the organization’s own documents or OMS. The contracting and payee entity comes from finance’s vendor onboarding. The site’s phone number comes from the site’s most recent confirmation. A system that is authoritative for one attribute is often unreliable for another.

A merge should never delete a record. The record that does not survive should be marked as merged, with a pointer to the surviving record, and every downstream system that held the old ID should receive the merge so it can update its cross-reference. EMA has described OMS to CTIS users as a master data system that does not delete records: a record that is no longer relevant or was created in error is inactivated, historical names and identifiers are retained, and when two records are merged, all of their identifiers stay valid and map to the surviving record. ROR handles erroneous records the same way, by withdrawing them and pointing to the correct one.^15,22 Even a regulator’s master can contain duplicates: one of the questions EMA answered in that session came from an organization registered twice in OMS under two organization IDs.²² The same approach keeps a sponsor’s audit trail intact: anyone can trace a historical report back to the record it used, even after that record was merged.

Ownership, Stewardship, and How Downstream Systems Subscribe

A clean list goes out of date quickly without owners. The reason site master data drifts in the first place is that every team owns its own copy and nobody owns the shared one. Fixing that does not require a large new function. It requires assigning accountability per attribute group, giving a small number of stewards the authority to make changes, and making the master the only place where new sites and investigators are created.

Who Owns What

The table below shows an ownership split that works for many mid-size sponsors. The names of the roles will differ; the principle is that the team closest to the source of authority for an attribute is accountable for it, and stewards in a central team carry out the changes.

Note that bank details belong to finance and should stay in the finance system, linked to the organization by master ID. Putting banking information into a clinical master widens access to sensitive data without improving anything the clinical side does.

The Change Request Is the Control

Every change to the master should arrive as a request with a reason and supporting evidence, the same way EMA asks for supporting documents with OMS change requests.¹² For a new investigator, that is the 1572 and CV. For a site address change, it is the site’s written confirmation. For an organization merger, it is the legal notice. The steward approves or rejects, the master records who changed what and when, and the change is published to subscribers. This also gives quality assurance something to review: a log of changes with evidence, rather than six systems edited by whoever had access.

ICH E6(R3), which reached Step 4 at ICH in January 2025, places responsibility for selecting qualified investigators and institutions with adequate resources and facilities on the sponsor, and lists verification of investigator and site staff qualifications and site resources as monitoring activities.²³ Those verifications produce exactly the evidence that a change request should carry. Linking the two means that the act of qualifying a site also updates the master, rather than creating a document in the TMF and leaving the CTMS as it was.

How Downstream Systems Subscribe

Subscription means each downstream system stores the master ID for every site, person, and location it holds, receives changes from the master, and stops creating its own records. The mechanics vary with each system’s capabilities, but the pattern is consistent.

When a CRO Runs the Systems

Many sponsors run some or all of a study inside a CRO’s CTMS, EDC, and payment systems. The sponsor still owns the regulatory outputs: the BIMO site table, the financial disclosure list, and the research payment reports. That makes master IDs a contract term. The CRO agreement should require the CRO to accept the sponsor’s master IDs at site creation, return them in every site-level extract, report new or changed sites to the sponsor’s stewards within a defined window, and include them in the transfer at study close. Without that clause, the sponsor gets a clean list for its in-house studies and a separate, unmatched list for every outsourced one.

Where to Start: A Six-Step Build

A site and investigator master does not need a multi-year program or a new platform to start producing value. The sequence below works with existing systems, starts with the smallest and most widely used list, and ends with a regulatory output that proves the work.

Conclusion

Site, investigator, and country records are some of the most repeated data in clinical development, and some of the least governed. Every system needs them, every team creates them, and nobody is asked to keep them consistent. The consequences show up far from where the records were created: in an FDA site table that cannot show where a study was conducted, an investigator list that misses a subinvestigator, a payment total split across two spellings of one name, or a research payment report attributed to an investigator who left the institution a year earlier. The fix is not a new system. It is a model that separates the person, the place, and the study site, a governed country list, matching rules that know what may and may not be merged, owners for each attribute, and downstream systems that subscribe instead of copying.

Sakara Digital works with pharma and biotech organizations on the data foundations that clinical operations, quality, and compliance teams rely on, including operational master data for trials. If you are looking at a set of site and investigator records that no longer agree with each other and want an independent view of where to start, we are happy to have that conversation.

For Further Reading