The eTMF quality check, explained.

What a trial master file is, and why inspectors open it before anything else

One trial document, from a signature at a site to an inspector's screen

1. A document is createdAt Site 114 the investigator, the doctor who leads the trial there, signs Form FDA 1572 on June 12. A monitor writes a visit report on September 23.
2. It travels to the fileBy site portal, by a vendor's file drop or by email from the monitor. Here it can wait for weeks.
3. It is filed and checked, and the list is checked tooWhat is this document, where does it go, are its date and signature there, and what does it make due? This is where a missing delegation log, the site's record of who may do what, is caught or missed. This guide is about this step.
4. The team chases what is missingThe monitor who covers the site asks for the document, with the reason and a date.
5. An inspector reads the fileSometimes at short notice, and sometimes years later if the file supports a marketing application, the request to sell the medicine. Whatever the file shows that day is the evidence.

The people, sites and dates in this example are fictional.

Why a trial master file is hard to keep: four facts that surprise newcomers

Essential record, TMF zone, delegation log, expected document list: the words to know

Fourteen terms are defined in the one-page summary.

What goes in a trial master file: essential records, the investigator site file and the eleven TMF Reference Model zones

In shortOne trial has one TMF, in two parts: the sponsor's and the investigator's. ICH E6(R3) now speaks of essential records, which covers data as well as documents. Most teams file to the TMF Reference Model, a shared index of eleven zones that CDISC, a clinical data standards body, now calls the TMF Standard Model. Version 3.3.1 is current, and its successor was anticipated for early 2027.

EMA's guideline is plain about the purpose. The documentation "should be sufficient to adequately reconstruct the activities undertaken in conducting the trial, along with decisions and justifications", and should allow an inspector to confirm compliance "without the need for additional explanation" from staff.2 A reader who needs someone to explain the file has found a problem with the file.

The sponsor's file and the investigator's file hold different things on purpose

Ten of the eleven TMF Reference Model zones, and which of them differ by site

Zone names as used in this guide's example; CDISC publishes the terminology.5 The model has eleven zones.6 This guide and its example cover the ten above and leave Zone 07 out of scope. "Filed" is typical practice, and your own TMF plan decides. Each zone divides into sections and artifacts, the model's word for a document type.

EU Regulation 536/2014, ICH E6(R3), EMA's TMF guideline and Part 11: what binds an eTMF, and since when

In shortEuropean law requires the file outright. US rules require the records and leave the shape to you. ICH E6(R3) sets shared practice, and EMA's guideline is the most detailed regulator's text on a TMF we found. The UK's rules changed on April 28, 2026. None of them names a filing deadline in days.

21 CFR is Title 21 of the US Code of Federal Regulations, where FDA's rules sit.

The eight controls EMA lists for a primary eTMF

EMA is the only regulator we found that lists controls for an eTMF one by one.2

EMA adds that the system "should be validated to demonstrate that the functionality is fit for purpose". Metadata should be formally defined, with a predefined document date, and a file name should never stand in for a visible version number.2

How long to keep a trial master file: 25 years in the EU, 15 in Canada, two after approval in the US

In shortRetention rules differ more than most people assume, and they recently moved in both directions. The UK raised its period from 5 years to 25 for new applications, and Canada cut its period from 25 to 15. A global trial usually keeps to the longest rule that applies. For the record, 25 years outlasts most software, so plan for readable formats and verified migration.

This table covers these five only, and other countries set their own periods. Your legal team decides which apply to your trial.

One new sub-investigator, four documents: the delegation log, Form FDA 1572 and financial disclosure

In shortA new doctor on a site's team is a routine event in a trial, and it creates several records at once. Some rest on rules and some on your plan. Keeping them apart stops you from chasing the wrong deadline. Under FDA's own FAQ, a new sub-investigator does not need a new Form FDA 1572.

Which deadline is the law's and which is your own

FDA's final FAQ on Form FDA 1572 is dated May 2010. A May 2021 draft revision, still labeled a draft on FDA's page, proposes changes on signature waivers for sites outside the US (questions 10, 11 and 13) and a new section on them; FDA's notice lists no change to the answer used here.27 Who belongs in Section 6 of the form depends on whether the person performs significant trial-related duties.24

EMA's report on 2024 inspections lists the failure in plain terms: "tasks performed outside of the delegation period" and "discrepancies between documented delegation of trial related tasks and actual activities performed".28 Dr. Hofstra's case in the example is the first of those, and it can be resolved only by the signed log and a clear record of what happened.

How to calculate TMF completeness, and why the expected document list decides the answer

In shortTMF completeness is documents filed divided by documents that should be there by today. The top number is easy to count. The bottom number is an expected document list worked out from each site's milestones, and it is where most of the error lives. One site can score 100%, 86%, 67% or 48% on the same morning, depending on the bottom number chosen.

The same 12 filed documents at Site 114, divided four ways

A TMF heatmap shows these scores as a grid, with zones on one axis and sites on the other and each cell colored by its score. Read the chart above as four managers looking at the same cell on the same morning. The first sees green and does nothing. The second sees amber and plans to look next quarter. The third sees red and asks why. The fourth sees red on a site that opened only months ago, and soon learns to ignore red.

Rules that turn milestones into a dated list of documents

An expected document list is a calculation with four inputs: the site's milestones, your plan's rules, events that appear only inside documents, and exclusions a person has approved. A rule has three parts: when this happens, these documents become due, by this date. A well-kept rule adds a fourth: where the event will be read from. "A staff change" means little until you say whether it comes from a staff directory, the delegation log or the text of visit reports.

E6(R3) supports this approach in a single line: a structured content list "may be used to prospectively identify essential records".3 The Reference Model's implementation guide asks the question behind most rules: is the artifact necessary to achieve a milestone, such as site activation?6

Three states a cell can have besides a score

How to add site scores into a study figure without misleading yourself

Add the top numbers, add the bottom numbers, then divide. Averaging site percentages gives a site with 4 expected documents the same weight as a site with 400. Suppose one site has 4 of 4 and another 360 of 400 (illustrative). The average of the two percentages is 95%. The true figure is 364 of 404, or 90.1%.

On the fictional study in this guide, 36,412 filed of 38,655 due gives 94.2%. That single figure suits a board. The number that changes what anyone does tomorrow is the smallest one with an owner: this zone, at this site, today.

No regulator sets a target figure. A 2015 cross-company working party found "no detailed regulatory guidance" on tolerance and proposed limits of its own, such as under 1% missing for product delivery records, which across 300 such documents allows two.29 That is a published proposal with no regulatory status. If you adopt a tolerance, your plan must say what happens when you exceed it.

TMF timeliness metrics: which clock to start, what late means and why a pile before inspection looks bad

In shortTimeliness is the share of filed documents that reached the file within your target. Decide which date starts the clock before anything is loaded, because you cannot re-score a file that mixed them. A complete file can still be late, and a late one is evidence that nobody could rely on it while the trial ran.

EMA asks you to consider which dates you use for filing: receipt, filing into the eTMF, approval or expiry.2 Pick one for the clock and write it in the plan. The example uses the date on the document, with a 30-day target. Some teams use the date of finalization. Either works if the plan says so and the system stores it as metadata.

The site was activated on June 30 and screened its first patient on July 14, all while the sponsor's file lacked the form signed on June 12. E6(R3) says some essential records "should generally be in place prior to the start of the trial".3 A note to file, signed by the investigator, is a common way to explain such a delay after the fact. It records what happened and leaves the delay in the record.

TMF inspection readiness: the same score on inspection day, two different histories

A review that happens once a quarter cannot show this curve. An audit trail can, and inspectors can read one. A look at the file just before an inspection shows what the inspector will see, and by then the dates cannot change. So measure timeliness while filings can still be made on time.

TMF QC review: EMA's six areas, risk by document type and what a sample can and cannot catch

In shortEMA asks for risk-based checks in six areas, and only some can be sampled. One working party found that nearly half of required document types are critical if missing. Sampling finds misfiled documents at a known rate. It finds no missing documents at all, which is why completeness and quality need separate checks.

The six areas EMA says a TMF quality review should consider

Which missing documents do the most harm: one working party's ratings

In 2015, Hecht and colleagues, quality specialists from several companies, assessed the 148 core document types required in Version 2 of the Reference Model. They asked what the absence of each would mean for patients' rights and safety and for data integrity.29 Their conclusion deserves a place in any QC plan: it is important "to assess the importance of the individual documents rather than to simply consider the overall number of documents filed".29

What a QC sample of 96 can miss, and what no sample can find

Say 2% of automatically filed documents are in the wrong place (an assumption). Each document you sample has a 98% chance of being fine. The chance a sample of n sees no error at all is 0.98 to the power of n. A sample of 96 misses every error about 14% of the time. So a weekly sample does tell you the rate, as long as it is big enough.

Now say the delegation log never arrived. No sample of filed documents contains it, at any size. Only a comparison with the expected document list can find it. Keep both checks in your plan and report them apart, so that a high quality score never reassures anyone about what is absent.

ICH defines quality control as the "operational techniques and activities undertaken to verify that the requirements for quality of the trial-related activities have been fulfilled".3 Verifying needs a written requirement. So your plan should state what right looks like for each document type: its location, mandatory metadata and signature. EMA's report on 2024 inspections lists "no formally documented process for TMF quality control" among its findings.28

TMF inspection findings: essential documents lead EMA's table, and four in five UK sponsor inspections had findings

In shortPublished inspection data point the same way. In EMA's 2024 report, essential documents is the largest single sub-category of critical and major findings. In the UK's figures for 2012 to 2015, more than four in five commercial sponsor inspections found TMF problems. These are inspection findings on selected trials, so they show what goes wrong more clearly than how often it does across the industry.

What one critical finding looked like, in the MHRA's words

The MHRA's metrics report for April 2018 to March 2019 gives an unusually detailed account of a critical finding on record keeping.4 The finding had been major at the organization's previous inspection and was raised to critical. Its main points read like a checklist of what to prevent:

Which TMF check catches which failure: list, dates, sample, audit trail or read-through

The US picture, from FDA's bioresearch monitoring data

FDA uses different words. It publishes no TMF category, because US rules do not use the term. In its fiscal 2024 metrics, the most common observations at clinical investigators include failure to comply with Form FDA 1572 requirements, and inadequate or inaccurate case history records and study records.32

What a weak TMF risks: extra inspector days, a critical finding, and at worst a rejected application

In shortMost FDA inspections end without action, and a finding usually means corrective work. The serious end of the scale is real: the MHRA wrote in 2015 that problems providing the file could lead to rejection of a marketing application. In between, the cost is staff time and the credibility of everything else the inspector reads.

How the damage grows, step by step

A weak file also hurts participants, in a less visible way. EMA's guideline explains that the file exists so that staff can manage the trial and monitors can check it.2 A site running with a team member who is not on the log, or a laboratory whose normal ranges changed without anyone filing the new ones, is a trial being run on assumptions. The file is where those assumptions get caught.

Who uses eTMF systems: the last public survey, and where the Reference Model and E6(R3) push the work next

In shortGood public data is scarce. The largest survey we could read is from 2014, run by a vendor, and paper and file shares dominated then. No source we could check publishes market size, documents per trial or typical completeness scores. The direction is clearer: E6(R3) takes in data as well as documents, and the next TMF standard is built around data flow.

Three shifts a TMF team should plan for: records, data standards and a running measure

AI on a trial master file: what FDA, EMA and the EU AI Act say, and which rules do not reach it

In shortNo regulator has written a rule for AI on a TMF. FDA's draft sets operational efficiency aside when it does not touch patient safety, drug quality or study reliability. EMA names eTMFs in its systems guideline and leaves AI for a possible later annex. So the working rules are the ordinary ones: validate, keep an audit trail, sign with a meaning and name who decided.

Which TMF tasks suit software, and which stay with a person

What AI filing has shown so far, and what we could not find

AI can classify documents and read fields, and vendors publish high accuracy figures in case studies. We found no peer-reviewed study of AI filing accuracy on a TMF, so treat those figures as claims until you test on your own documents. eTMF products advertise expected document lists tied to milestones, and service firms sell heatmap reviews as a project. We found no public figure for how many teams keep the list current from events found inside the documents themselves.

Handle a confidence score with care. It is the model's estimate of its own certainty. Treat it as a routing setting, and measure the real error rate by sampling what passed it. The ALCOA++ principles in EMA's guideline apply to any step that writes to the record: attributable, legible, contemporaneous, original, accurate and the rest.21 Whoever or whatever classifies a document, the trail has to say so, and Part 11 requires that a change "shall not obscure previously recorded information".18

Who keeps a trial master file today: the TMF lead, QC reviewers, monitors and a spreadsheet of expected documents

In shortA TMF lead owns the file, but monitors, sites, vendors and every functional team feed it. Documents wait in queues, and the list of what is due is often kept by hand. No public time study exists, so the split below is one estimate for a fictional sample. On that estimate, more than half of the days are waiting.

The people a document passes through before it counts as filed

Where 38 days go, in one estimate for a by-hand file

In house, outsourced, or with software: four ways to keep the file

From a red cell on the TMF heatmap to four gaps with an owner: one morning at Site 114

In shortThis is the example from the top of the page, inside the product. Lena Ortiz, TMF lead and QC reviewer on VLT-302, opens the heatmap at the start of her day. Site 114 is red in Zone 05. She opens the cell, sees why 18 documents are due, reads the passage that revealed the staff change, and assigns four gaps to the site's monitor.

1

Ten TMF zones and 17 sites on one heatmap, scored at 6:00 am

TMF Reference Model zones run down the side, and sites across. Each cell counts only the documents that site should have by today. The study reads 94.2% complete, 33 of 46 sites are ready in every zone shown and 37 gaps are open.

Trial-level zones show a dash, because they are filed once for the study. A red dot marks late filings. Selecting Site 114's Zone 05 cell opens the panel on the right: 67% complete, 12 of 18, and 75% on time.

2

Why 18 documents are expected, and what is wrong with them

The milestone strip explains the 18: selected March 9, activated June 30, first patient July 14, a sub-investigator added September 15, Visit 3 on September 22. Seven more documents are not expected yet, because Visit 4 is on November 17.

Below it: five missing, three filed late, one waiting for QC. The panel offers two actions: assign the four gaps, and ask for a note to file on the late Form FDA 1572.

3

The same heatmap, switched to TMF timeliness

One switch redraws every cell as the share filed within the plan's 30 days. Site 114's Zone 05 reads 75%. A site's two scores can differ: Site 122's Zone 05 is 86% complete and 78% on time.

Form FDA 1572 tops the panel's late list: signed June 12, filed September 14, 94 days after signing and 64 past the target.

4

A staff change found inside a visit report

No spreadsheet knew Dr. Hofstra had joined. The completeness checker, one of the solution's four AI agents, read it on page 3 of the Visit 3 report, set it against delegation log v1, which has no entry for him, and against section 6.3 of the sample's TMF plan. It then opened gap G-1052 with each passage one click away.

The checklist beside it shows the whole zone by section, each document filed, late, missing or expiring. The study nurse's GCP certificate is flagged as expiring on October 31.

5

Four gaps go to the site's monitor in one step

The Zone 05 owner rule suggests Omar Haddad, the CRA for Sites 114 and 116. Lena assigns the delegation log with the three gaps that go with it, due October 13, the date the follow-up letter asked for. The plan's own 14 days for the log ended on September 29, so this one is already late, and the record says so.

Nothing goes to the site until a person presses Send.

6

A drafted request for the late Form FDA 1572

The gap follow-up agent drafts a request to Dr. Varga from the filing record and the TMF plan: signed June 12, filed September 14, 94 days, the plan asks for 30. It asks her for a short note to file, so an inspector can see the delay was known and handled.

Lena can edit it. It leaves only when she presses Send.

7

The gap register, with every owner and date

The four Site 114 gaps now sit with Omar. The register lists the 14 most urgent of 37 open: Site 107's local lab normal ranges and certificate, 28 days open and overdue since October 2, chased by Dana Okafor; Site 119's August temperature log with Ewa Nowak.

Select any screen to open it larger. Dates and times are shown in US form.

Batch 0412: 240 documents read and matched to the TMF Reference Model, and one letter that needed a person

In shortDocuments arrive from a site portal, a vendor file drop and CRA email. Each is matched to an artifact and its details are read. At 90% confidence or more, with date and signature checks passing, it files itself. Below that, a QC reviewer decides and signs. Every reason on screen is tied to a passage.

1

Batch 0412, step by step: three channels, 612 pages, 31 sites re-scored

The batch arrived at 8:12 am: 162 from the site upload portal, 51 from the vendor file drop and 27 from CRA email, 612 pages in all. The classifier matched each to a Reference Model v3.3.1 artifact. The metadata extractor read site, date, version, signatures and people named.

Then the completeness checker re-scored the 31 sites the batch touched. Site 114 stays red, because delegation log v2 was not in the batch.

2

Six filings held for a QC reviewer

QC reviewers see only what the agents doubted, six of the 240. The queue shows each held document with its proposed artifact and confidence.

One of them, a protocol signature page at 96%, is held for a different reason: its date is missing, and the sample's rule files nothing automatically without one.

3

A letter that looks like a report, at 64% confidence

The Visit 3 follow-up letter came in titled "Site Visit Summary", a title visit reports often use, so 29% went to Monitoring Visit Report. But it is a letter to the principal investigator with agreed actions, and it says the full report is held by the sponsor.

The classifier proposes Monitoring Visit Follow-up Letter, highlights each passage it relied on, states the reason that pulled the other way, and holds it. Every detail read carries its own confidence and its source.

4

Lena confirms the filing with her signature

Lena files the letter with the meaning of her signature: "Reviewed: filing and details are correct". The screen records name, time and meaning, the three things 21 CFR 11.50 looks for in a signed record.19

Site 114's Zone 05 moves from 67% to 72%, 13 of 18. The delegation log is still with Omar.

5

The cell moves, and stays red

One confirmed filing lifts the cell by a single document. A confirmed letter does nothing for the delegation log. The cell measures what is due and filed, so it can only turn green when the missing documents arrive and pass QC.

Expected document rules, a site not yet due, the inspection copy, the TMF dashboard and your settings

In shortAround the heatmap sit the screens a TMF team uses all year. The rules that make documents due are visible and editable. A new site's hatched cells mean nothing is due yet. An inspection copy goes to a named approver first. The dashboard tracks the backlog and settings hold your plan.

Four TMF agents, and the people who confirm filings, approve exports and press Send

In shortFour specialist agents file, read, score and draft. None can sign a filing, remove an expected document or send anything to a site. A QC reviewer signs doubtful and critical filings. Only a person marks a document not required, with a reason and an approval, and the clinical trial manager approves what leaves for an inspector.

What each TMF agent does, and where it stops

Who confirms a filing, who removes an obligation and who releases the file

The people in the VLT-302 sample, in their own words

From trial documents and your TMF plan to a scored file with owned gaps

Systems a TMF quality check connects to

Audit trail, signatures and credentials a clinical quality team will ask about

Each agent step, filing, assignment, reminder and signature is recorded with who, what and when. A signed confirmation carries name, time and meaning.19 The record is designed to stay available for the 25 years the EU requires.1 These are controls that support your compliance with Part 11, and validating your configured installation stays with you, as EMA's guideline expects.2,18

Botminds' information security is managed to ISO 27001 and audited under SOC 2 Type II. No product can be certified for good clinical practice or Part 11, and we make no such claim. The solution runs at enterprise scale across many studies, with roles, maker-checker approvals (one person prepares, another approves) and an exportable audit trail.

What a filing backlog, a stale expected list and a late-found gap cost a sponsor

In shortThree kinds of cost sit here: the license, the hours, and the risk at inspection. Only the first is on a quote. Published figures frame the scale of the hours and the price of a day. No source publishes TMF hours per study or per document, so your own numbers are the ones to trust, and you can find most of them in a week.

Six TMF costs you can measure in a week, starting with 50 documents' dates

What a late delegation log does that no spreadsheet prices

What automated filing and scoring return: specialist hours, a list that keeps itself current and fewer surprises

In shortPut in your own numbers. The calculator values specialist hours released from filing and QC, and hours released from keeping the list. It deliberately leaves inspection outcomes out, because nobody can honestly price them. Every default is illustrative or a sample figure. Replace them.

Filing, QC, the expected list and inspection days: where TMF automation saves

TMF automation calculator: documents a year, minutes per document and hours on the list

How ready your TMF team is: plan, index, milestones, sign-off and a score you can defend

In shortPick the answer closest to your team today. Eight questions score 0 to 2, and the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

Twelve questions for any eTMF or TMF quality software vendor, us included

In shortAlmost every tool demonstrates well on clean documents. These questions test what a demonstration hides: where the expected list comes from, whether a filing can be traced, what happens to a low-confidence document, and who can make a gap disappear.

Testing TMF automation in ten weeks: a closed study, planted staff changes and one live study beside your usual process

In shortStart with a study you have closed, because you already know what its file looks like. Write the rules, re-score it, plant errors to see whether the checks fail when they should, then run one live study in parallel. Decide on five numbers.

Days to file, planted gaps found and false alarms on your own study

Trial master file questions answered: completeness, timeliness, retention, delegation logs and AI

In shortQuestions TMF teams and buyers commonly search for, each answered on its own.