1,853
Documents
11
Zones
96.8%
Complete
25y
Retention
Contents
1. If It Is Not in the TMF, It Did Not Happen
The Trial Master File is the evidence that this trial was conducted the way the protocol, the regulations and the sponsor's own procedures say it was. It is not a filing cabinet and it is not an archive. It is the artifact an inspector reads to decide whether to believe the submission.
| Value | |
|---|---|
| Documents expected across all zones | 1,853 |
| Zones, per the DIA TMF Reference Model | 11 |
| Completeness at the status date | 96.8% |
| Timeliness — filed within the required window | 91.2% |
| Documents currently overdue | 143 |
| Retention after the last approval | 25 years |
The sentence that governs everything below: a decision that was made well and recorded badly is,
at inspection, a decision that was made badly.
This is uncomfortable and it is not a technicality. An inspector cannot observe how carefully a monitoring visit was conducted; they can only read the report. They cannot know whether an ethics approval preceded the first participant; they can only compare two dates. The file is not a description of the trial's quality — for every purpose that matters after the fact, it is the trial's quality.
Which is why the TMF is a program-management responsibility rather than an administrative one. It is the only deliverable in this program that every function contributes to and no single function owns.
This is uncomfortable and it is not a technicality. An inspector cannot observe how carefully a monitoring visit was conducted; they can only read the report. They cannot know whether an ethics approval preceded the first participant; they can only compare two dates. The file is not a description of the trial's quality — for every purpose that matters after the fact, it is the trial's quality.
Which is why the TMF is a program-management responsibility rather than an administrative one. It is the only deliverable in this program that every function contributes to and no single function owns.
2. The Eleven Zones
Structured against the DIA TMF Reference Model rather than an internal invention, so that an inspector, a partner or an acquirer finds the file laid out the way they expect.
| Zone | Name | Owner | Custody | Documents | Share | What it evidences |
|---|---|---|---|---|---|---|
| 1 | Trial Management | C. Tyrrell | Sponsor | 38 | 2% | Plans, minutes, oversight records. ⚠ The zone that proves the sponsor exercised the oversight it did not transfer. |
| 2 | Central Trial Documents | Dr. S. Aldridge | Both | 24 | 1% | Protocol and all amendments, investigator brochure editions, sample case report form. The definitional documents everything else is measured against. |
| 3 | Regulatory | Dr. P. Raghunathan | Sponsor | 19 | 1% | IND submissions, safety reports to the agency, annual reports. Distinct from Zone 4 — regulators are not ethics committees. |
| 4 | IRB or IEC and other Approvals | Dr. R. Molyneux | Both | 260 | 14% | ⚠ One approval package per site, each on its own institutional calendar. The largest document count in the file and the one that gates activation. |
| 5 | Site Management | Dr. R. Molyneux | Both | 1,040 | 56% | Feasibility, selection, initiation, monitoring visit reports, close-out. Four documents per site on average, and the zone an inspector samples first. |
| 6 | Investigational Product | Dr. K. Oyelaran | Both | 96 | 5% | Shipping, accountability, temperature excursion records, destruction certificates. Reconciles to the 41,600-kit supply record. |
| 7 | Safety Reporting | Dr. N. Halloran | Sponsor | 258 | 14% | Expedited reports, DSURs, safety notifications to investigators. Must reconcile exactly to the safety database — two systems, one truth. |
| 8 | Central and Local Testing | Dr. F. Achterberg | Both | 47 | 3% | Laboratory certifications, method validation, reference ranges, normal ranges per site. |
| 9 | Third Parties | G. Petrossian | Sponsor | 31 | 2% | Vendor agreements, transfer of obligations, audit reports. ⚠ Where the sponsor proves what it transferred and what it kept. |
| 10 | Data Management | Dr. F. Achterberg | Sponsor | 26 | 1% | Data management plan, edit checks, coding dictionary versions, lock certificates. |
| 11 | Statistics | Dr. F. Achterberg | Sponsor | 14 | 1% | Statistical analysis plan and its versions, randomization records, analysis datasets documentation. ⚠ The SAP's DATE in this zone is what proves it preceded the lock. |
| Total | 1,853 | 100% |
Three zones carry 84% of the documents, and all
three scale with sites rather than with the program.
Site Management (1,040), IRB approvals (260) and Safety Reporting (258) between them account for most of the file. A program with the same budget, the same duration and forty sites instead of 260 would have a TMF a fraction of this size.
TMF effort scales with the number of relationships, not with the size of the program — which is the same shape as the internal labor envelope, and for the same underlying reason.
Site Management (1,040), IRB approvals (260) and Safety Reporting (258) between them account for most of the file. A program with the same budget, the same duration and forty sites instead of 260 would have a TMF a fraction of this size.
TMF effort scales with the number of relationships, not with the size of the program — which is the same shape as the internal labor envelope, and for the same underlying reason.
The custody column matters more than it looks. Documents marked “Both” exist in two places — the sponsor TMF and the investigator site file — and both copies have to be complete. An inspection at a site examines the site file; an inspection at the sponsor examines the sponsor file; and a discrepancy between them is itself a finding.
3. Contemporaneous, or Not a TMF at All
| Principle | What it requires |
|---|---|
| Contemporaneous | Filed as work happens, not assembled before an inspection. ⚠ A TMF built retrospectively is detectable — creation dates, version gaps and suspiciously uniform filing dates all show it, and an inspector who finds one stops trusting the rest of the file. |
| Complete | Every expected document present, and every absence explained. ⚠ A documented reason for a missing document is acceptable; a silent gap is not. |
| Attributable | Every document traceable to who created it and when. This is the same ALCOA+ standard applied to the record of the trial rather than to the data. |
| Legible and available | Retrievable by an inspector within the session. A document that exists and cannot be found in twenty minutes is, for inspection purposes, missing. |
A TMF assembled before an inspection is detectable, and it destroys the credibility of
everything else in the file.
The tells are structural rather than subtle: creation dates clustered in a narrow window, version histories with gaps, filing dates that are uniform where the underlying work was not. An inspector who finds a retrospectively assembled file does not conclude that the documents are wrong — they conclude that they cannot tell whether the documents are wrong, which is worse and which cannot be remedied during the inspection.
This is the ALCOA+ contemporaneous requirement, the most frequently cited data-integrity deficiency in the industry, applied to the record of the trial rather than to the trial's data.
The tells are structural rather than subtle: creation dates clustered in a narrow window, version histories with gaps, filing dates that are uniform where the underlying work was not. An inspector who finds a retrospectively assembled file does not conclude that the documents are wrong — they conclude that they cannot tell whether the documents are wrong, which is worse and which cannot be remedied during the inspection.
This is the ALCOA+ contemporaneous requirement, the most frequently cited data-integrity deficiency in the industry, applied to the record of the trial rather than to the trial's data.
The practical consequence is that the TMF has a filing clock, not a filing deadline. The program does not aim to have the file complete by Gate 5; it aims to have every document filed within days of existing, continuously, for 27 months. Those are different disciplines and only one of them survives an inspection.
4. Filing Windows
| Document type | Must be filed within |
|---|---|
| Executed contract or agreement | 5 business days of full execution |
| IRB or IEC approval | 5 business days of receipt |
| Monitoring visit report | 15 business days of the visit |
| Expedited safety report to the agency | Same day as submission |
| Protocol amendment | 5 business days of approval |
| Anything else | 30 calendar days of creation or receipt |
Timeliness is measured separately from completeness, and it is always the worse number.
The file is 96.8% complete and 91.2% timely, and that gap is not an inconsistency — it is the honest shape of the measure. Documents get filed eventually; the question a filing window asks is whether they were filed when they should have been, and a document filed three months late is permanently late no matter how complete the file eventually becomes.
A TMF metric that reports only completeness will read green for the entire program and tell nobody anything. The fact base asserts that timeliness must be the lower of the two, because a file where they are equal has almost certainly stopped measuring one of them.
The file is 96.8% complete and 91.2% timely, and that gap is not an inconsistency — it is the honest shape of the measure. Documents get filed eventually; the question a filing window asks is whether they were filed when they should have been, and a document filed three months late is permanently late no matter how complete the file eventually becomes.
A TMF metric that reports only completeness will read green for the entire program and tell nobody anything. The fact base asserts that timeliness must be the lower of the two, because a file where they are equal has almost certainly stopped measuring one of them.
143 documents are currently overdue. That number is reported rather than smoothed, and the reason it is not zero is worth stating plainly: with 231 sites active and monitoring visits running continuously, a zero would mean the measure had been redefined rather than the work perfected.
5. How Completeness Is Known
Completeness is a claim, and a claim needs a method. Three mechanisms produce the number in §1, and none of them is somebody's opinion.
| Mechanism | How it works | What it catches |
|---|---|---|
| Expected-document list | Every zone has a defined set of documents that must exist for each study, site and milestone. Completeness is the count present divided by the count expected — a denominator fixed in advance rather than derived from what happens to be filed. | ⚠ Documents nobody created. A file measured against itself is always 100% complete. |
| Quality control review | A 15% sample of filed documents is reviewed against a checklist: right document, right version, legible, correctly indexed, correctly dated. Current error rate 2.3%. | Documents that are present but wrong — superseded versions, unsigned copies, misfiled zones. |
| Reconciliation against source systems | Zone 6 reconciles to the supply record, Zone 7 to the safety database, Zone 11 to the randomization system. The file must agree with the systems it describes. | ⚠ The class of error where the file is internally consistent and disagrees with reality. |
The expected-document list is the whole control, and it has to be written before the documents
exist.
A TMF measured against the documents it happens to contain is complete by definition, always, and the metric is worthless. The list is fixed at study start, extended when a protocol amendment creates new expectations, and it is the denominator that makes 96.8% mean something.
This is the same principle as the gate criteria being fixed at the previous gate, and the variance thresholds being set at baseline. A standard set after the fact is set by someone who already knows the answer.
A TMF measured against the documents it happens to contain is complete by definition, always, and the metric is worthless. The list is fixed at study start, extended when a protocol amendment creates new expectations, and it is the denominator that makes 96.8% mean something.
This is the same principle as the gate criteria being fixed at the previous gate, and the variance thresholds being set at baseline. A standard set after the fact is set by someone who already knows the answer.
6. Adjacent Records, and Who Owns Them
| Not the TMF | Where it lives | Why the distinction matters |
|---|---|---|
| The clinical database | EDC, and the analysis datasets derived from it | The TMF holds documents about the data — the plan, the edit checks, the lock certificate. It does not hold the data. |
| The investigator site file | At each of the 260 sites | ⚠ A separate, parallel file the site owns. The sponsor is responsible for confirming it exists and is complete, and cannot maintain it. |
| The submission | The eCTD dossier | The NDA is assembled from evidence the TMF holds, but the two are structured differently and serve different readers. A complete TMF does not produce a submission. |
| The corporate archive | Records management, after closure | The TMF transfers to archive at closure and is retained 25 years. ⚠ Archived is not the same as accessible — retrieval within an inspection window is still required. |
Findings come from the second of those, more than from any other.
The sponsor cannot maintain 260 investigator site files, does not own them, and is nonetheless answerable for their completeness. That obligation is discharged through monitoring — every site visit includes a site file review, and the finding rate from those reviews is a leading indicator for what a site inspection would produce.
It is the same structure as everything else the program transferred: execution sits with the site, the obligation stays with the sponsor.
The sponsor cannot maintain 260 investigator site files, does not own them, and is nonetheless answerable for their completeness. That obligation is discharged through monitoring — every site visit includes a site file review, and the finding rate from those reviews is a leading indicator for what a site inspection would produce.
It is the same structure as everything else the program transferred: execution sits with the site, the obligation stays with the sponsor.
7. Inspection Readiness, and Closure
The TMF is the first thing requested in almost every inspection, and the request is usually for a specific document within a specific zone rather than for the file as a whole.
| Requirement | Standard | Why |
|---|---|---|
| Retrieval time | Any document produced within 20 minutes of request | ⚠ A document that exists and cannot be found is, for inspection purposes, missing. This is the single most common way a complete file fails an inspection. |
| Index accuracy | Zone and subzone correct for every document | An inspector navigating by the Reference Model expects to find things where the model says they are. Misfiling is indistinguishable from absence. |
| Absence rationale | Every expected-but-missing document has a recorded reason | A documented absence is a controlled state. A silent gap is a finding, and invites the question of what else is missing. |
| Access control | Part 11 compliant, with audit trail | The eTMF is a regulated system. Who filed what, and when, must itself be evidenced. |
The file outlives the program by more than twenty years.
This program closes 30 June 2030. The TMF is retained for 25 years after the last approval, which means it must remain complete, indexed and retrievable long after every person who built it has moved on and every system it was built in has been replaced.
That is the real argument for using a published reference model rather than an internal structure. The people who will need to navigate this file have not been hired yet, and the only thing that will make it navigable to them is that it is laid out the way the industry lays out every other one.
This program closes 30 June 2030. The TMF is retained for 25 years after the last approval, which means it must remain complete, indexed and retrievable long after every person who built it has moved on and every system it was built in has been replaced.
That is the real argument for using a published reference model rather than an internal structure. The people who will need to navigate this file have not been hired yet, and the only thing that will make it navigable to them is that it is laid out the way the industry lays out every other one.
Transfer to archive is a closure deliverable with a named owner and an acceptance criterion, not an administrative afterthought. The closure report records it alongside the post-marketing obligations, and for the same reason: both are things the program hands to somebody who will still be responsible for them when the program no longer exists.