Trial Records and TMF Review: FDA Citation Counts, Retention and Five Quality Checks (2026)

A trial file becomes useful when an authorized reviewer can connect a decision to the evidence available at the time. That may require an approval letter, a laboratory report, a system audit trail and a documented clinical assessment held by different organizations. A large collection of documents can still leave that connection unclear. The central design question is which records establish what happened, who controls them and how they remain usable throughout the required retention period.

This report combines a reproducible analysis of FDA inspection-observation frequencies with current US, EU and UK record requirements and a proposed five-gate quality framework. Its scope is medicinal-product trials. The numerical analysis identifies selected record obligations in the published data. Estimating TMF defect prevalence or predicting inspection outcomes would require different data and a validated method.

TL;DR

• Use the correct denominator. The selected FDA BIMO workbook rows contain 1,715 citation-frequency occurrences across FY2020–FY2025. They cover the BIMO program, rather than a cohort consisting solely of clinical-investigator inspections. [1] [2]
• Declare the records selection. Investigator case-history and IRB-minute references total 311 occurrences. Adding all observed sponsor-record subsections under 312.57 gives 324, or 18.9%. These obligations concern different record custodians.
• Separate content from retention. EU CTR Article 57 concerns the TMF's essential documents and accessibility; Article 58 addresses archiving, including the minimum 25-year period and traceable alterations. Participant medical files follow national law. [3]
• Track the legal trigger. US sponsor, investigator and IRB retention rules differ. UK retention also depends on the April 28, 2026 transition and other applicable obligations.
• Review five dimensions. Establish applicability and ownership, verify identity and integrity, check timing and version, trace relationships and decisions, and demonstrate retrieval and preservation. These are proposed quality controls, with study-specific scope and cadence.

1. Recompute the citation record before drawing an operating conclusion

FDA publishes annual inspection-observation workbooks organized by program. Our analysis uses the BIMO entries for fiscal years 2020 through 2025 from the August 23, 2026 normalized snapshot. The selected input contains 479 aggregate rows. Summing the frequency field gives 1,715 occurrences. Counting rows would answer a different question because one row can represent multiple occurrences of the same citation in a fiscal year. The public supplement preserves the selected rows, input hash and calculation definitions. [1] [2]

The first correction concerns program scope. These BIMO sheets include references related to investigators, sponsors and IRBs, among others. Calling the total “1,715 clinical-investigator inspections” would change both the population and the unit. The workbook aggregate also lacks the site-level identifiers needed to determine how many unique inspections contributed to a particular selection. Multiple observations can arise during one inspection.

The second correction concerns regulatory references. The combined 312.60 frequency is 522, but it includes distinct citation identifiers: 481 for protocol compliance and 41 for informed consent. Case histories under 312.62(b) contribute another 271. The pair of regulatory references totals 793, or 46.2% of the selected frequencies, while the underlying descriptions cover more than one issue. This combined share provides no direct estimate of eligibility errors or TMF defects.

The third correction concerns the meaning of an observation. FDA explains that a Form 483 communicates observed conditions and is considered alongside the inspection report, collected evidence and responses. It is not a final agency determination of a violation. Published aggregate frequencies are useful for describing the cited references; they provide limited information about the facts and subsequent resolution of any individual finding. [4]

A practical consequence follows for research teams: use the aggregate to generate questions, then assess the actual process with trial-specific evidence. A frequent case-history reference can motivate a review of source availability and traceability. It cannot establish that a missing central filing step caused those findings, that one software configuration would prevent them, or that all observations concern the same underlying behavior.

2. Why the records total is 311, 324 or 440

We calculated three explicit selections to show how the inclusion rule changes the result. The narrowest combines 312.62(b), investigator case histories, with 56.115(a)(2), IRB meeting minutes. Their frequencies are 271 and 40, respectively, producing 311 occurrences and 18.1% of the 1,715 total. That is an exact description of two references, rather than a comprehensive definition of recordkeeping findings. [2]

The next selection adds every observed subsection under 312.57. These comprise eight occurrences under paragraph (a), one under (b), three under (c) and one under (d), for 13 sponsor-record occurrences. Searching only for an exact unsuffixed “312.57” string would miss them. The resulting total is 324, or 18.9%. A reproducible method should show the matching rule so another analyst can recognize this difference.

A broader sensitivity calculation includes all observed subsections in the three regulation families: 312.62, 312.57 and 56.115. Their totals are 348, 13 and 79, respectively, giving 440 occurrences, or 25.7%. This selection includes additional subjects such as investigator drug-disposition records and retention. It still excludes documentation-related requirements elsewhere in the regulations. None of these definitions is a complete census of every possible records concern.

The distinction is operationally important. Investigator case histories are generally associated with investigator or institutional source records. IRB minutes belong to the IRB's record obligations. Sponsor records have their own requirements and custody arrangements. Combining their frequencies creates an analytical category. Custody remains defined by the underlying obligations and arrangements, while the causes of individual observations require case-specific evidence.

The annual frequencies also need careful interpretation. Investigator case-history counts are 30, 48, 36, 49, 56 and 52 from FY2020 through FY2025. IRB-minute counts are seven, one, 13, seven, nine and three. Sponsor 312.57 subsection counts are four, one, zero, two, five and one. The three-part annual totals are therefore 41, 50, 49, 58, 70 and 56. [2]

A rise or fall in these frequencies can reflect changes in inspected populations, inspection activity, observed practices, citation selection or other factors. Separating those influences would require additional information about the inspected populations and activity. We therefore avoid attributing the 2020 count to one cause or treating later counts as evidence that electronic filing failed to improve quality. Estimating a defect rate would require a suitable denominator and a consistent case definition at the relevant inspection or organization level.

For a sponsor, the useful output is a bounded list of obligations to examine, with the data's limitations attached. For a records team, the more actionable evidence comes from its own controlled sample: which expected records are missing, which are unusable, what decisions they support and how the problems arose. Those local questions are the basis of the five-gate framework later in this report.

3. Define the essential record and its custodian

ICH E6(R3), including Appendix C, describes essential records and their management across the trial lifecycle. The purpose includes evaluating trial conduct and the reliability of results. FDA issued its E6(R3) guidance in September 2025. Use the applicable regional implementation and the trial's circumstances when establishing the record set; a reference taxonomy supplies an organizing structure that needs to be assessed against study-specific record needs. [5]

EMA's TMF guideline distinguishes sponsor and investigator responsibilities and discusses files held across organizations and systems. It addresses structure, access, quality, copies and archiving. A usable TMF can therefore involve a documented arrangement across repositories, with appropriate control and accessibility. The operational task is to make the relevant records and relationships identifiable without unnecessarily centralizing confidential source information. [6]

We propose a record-location index with the record category, evidentiary purpose, responsible party, authoritative location, access route, expected timing and retention basis. Add the relevant study, country, site or system scope. A location entry should be specific enough that a replacement team member can request or retrieve the record. “At vendor” leaves the custodian, system, access conditions and continuity arrangements unresolved.

For example, a central laboratory report may originate in a laboratory system, reach the investigator through a portal and populate selected fields in an EDC system. Define which record supplies the original result, where corrections are maintained, and how the investigator retains access to the necessary information. The EDC value and the laboratory report have related purposes but can contain different context. Their relationship should survive a corrected result and a vendor transition.

Participant medical information needs particular care. A sponsor may require evidence of monitoring and appropriate source access without holding a full identified copy of every participant's medical file. The approved arrangements should specify what may be accessed, by whom and for what purpose. The GDPR framework for relevant European processing adds requirements concerning lawful processing, special-category data, roles and safeguards. Record design should preserve the necessary evidence while respecting those boundaries. [7]

A record can also remain in an operational system rather than becoming a standalone PDF. For an electronic signature or corrected database entry, the meaningful evidence may include metadata and an audit trail. Decide what must remain available to explain the record. Flattening the visible screen can lose the relationship between the action, identity, timestamp and subsequent change, even when the exported page appears complete.

4. Retention is a rule with a trigger, not a single date field

EU CTR Article 57 requires the TMF to contain essential documents supporting verification of trial conduct and data quality and to be accessible to Member States on request. Article 58 establishes archiving provisions, including retention of TMF content for at least 25 years after trial end unless other Union law requires longer. It also addresses traceable alterations and ownership transfers. Medical files are retained according to national law. These are distinct obligations, rather than a single article prescribing a universal folder list. [3]

The US IND investigator rule in 312.62(c) uses a different trigger: two years after approval of a marketing application for the drug for the investigated indication, or the specified circumstances involving discontinuation and FDA notification where no application is filed or approved. The sponsor rule in 312.57(c) likewise links retention to approval or to discontinued investigational shipment and delivery with FDA notification. Site closeout alone is an inadequate basis for starting either clock. [8] [9]

IRB records have a separate US requirement. Section 56.115(b) establishes at least three years after completion of the research, together with access for FDA inspection and copying. Apply the investigator and sponsor obligations separately, even where all three record groups concern the same trial. [10]

The UK's April 2026 guidance distinguishes applications submitted before April 28, 2026 from those submitted on or after that date. The newer regime generally requires at least 25 years for TMF documents; transitional trials generally retain a five-year TMF minimum, with specified extensions. Participant medical files generally have a 25-year requirement, and marketing-authorization or other applicable rules can extend retention. The transition depends on the application submission date, rather than the date a filing team installs a new system. [11] [12]

We propose keeping the legal rule and the event evidence alongside the calculated earliest review date. Fields might include jurisdiction, record class, applicable provision, triggering event, evidence of that event, minimum period, additional obligations, legal hold and approving owner. If the trigger is unknown or has not occurred, the record should remain in a state that makes that uncertainty visible. Guessing a date to complete a dashboard can create a false destruction instruction.

Retention decisions also need coordination between organizations. An investigator may depend on sponsor communication to know that a relevant regulatory event has occurred. A storage provider may know the contract end date but lack the clinical and regulatory context to determine whether records can be destroyed. Specify who supplies the event evidence, who reviews the rule and who authorizes any disposition. Keep the resulting decision attributable and available.

Consider a trial that closes its final site while development continues for the same drug. An operational closeout milestone can support archiving activities while the legal event that starts a US retention period may still be in the future. The proposed retention register should make that distinction explicit. Similarly, a commercial agreement ending after a fixed number of years must provide a way to continue retention when the applicable obligation lasts longer.

5. Gate one: applicability and ownership

The first proposed gate asks whether the expected record set reflects the trial as it is actually conducted. Begin with applicable requirements, the approved protocol, the selected systems and the assigned activities. Use a taxonomy to organize the result, then document which categories apply, when records are expected and who is responsible. A category can be inapplicable, pending a future event, or due and missing; those states should remain distinguishable.

A protocol amendment illustrates the need for a dynamic expectation. It may create a new approval, revised participant materials, additional training and changed system specifications. The expected set should identify the affected sites and the relevant version. It should also preserve what was expected before the amendment. Replacing the old expectation with the new one can make historical completeness impossible to interpret.

Define ownership across record creation, checking, custody and issue resolution. Identify the person or function that creates the record, the party that checks it, the custodian and the person who resolves a missing or disputed item. A small trial may combine these roles while documenting each responsibility explicitly. Define a backup for a departure or extended absence, especially where access depends on a particular individual's account.

For an external provider, translate the responsibility into a deliverable that can be recognized. “Maintain trial documentation” is too broad to establish whether a corrected report, audit history or configuration approval will be supplied. Specify the records, timing, format, relevant metadata, access and exception handling. Align the agreement with the operational index so a user can move from a missing item to the party able to resolve it.

An applicability review should also address records held outside the main filing system. The index can point to controlled system-validation evidence or an institutional source environment where appropriate. The gate passes when the team has a defensible expectation and an effective access arrangement, rather than when every potential category contains an uploaded file.

6. Gate two: identity and integrity

The second gate examines whether a present record can serve its intended evidentiary purpose. Check that it belongs to the correct trial and scope, is readable, contains the expected pages or components, and preserves the information needed to understand who created or approved it. The review should distinguish a cosmetic imperfection from a defect that affects interpretation, participant protection or data reliability.

For a scanned document, inspect the full record rather than only its first page. A readable cover sheet can conceal a clipped signature, missing attachment or reversed page. Where an electronic copy replaces an original, apply the relevant certified-copy and transfer controls. EMA's TMF guideline discusses certification and digitization, while ICH E6(R3) defines a certified copy in terms of verification of the information, including relevant metadata where applicable. [6] [5]

Electronic records require review of more than appearance. FDA's October 2024 electronic-systems guidance and EMA's 2023 computerized-systems guideline address controls relevant to electronic data, records and their lifecycle. Determine which metadata, audit history and access controls matter for the record's intended use. A vendor validation package can contribute evidence, while the responsible party still needs to assess the actual trial configuration and processes. [13] [14]

Our proposed sample includes a signed record, a corrected record, an attachment and an export. Ask the reviewer to identify the record's origin, the person or system responsible for the action, the relevant time and any subsequent change. Then compare what is visible in the live system with the retained or exported form. A difference may be acceptable if the necessary information is preserved and interpretable; an unexplained loss needs investigation.

When a defect is found, correct it transparently. Preserve the history of what was present, what changed, why the change was made and who authorized it. A late explanatory note should accurately describe the evidence available and the limits of retrospective reconstruction. Backdating a record or creating an unsupported account of an earlier event undermines the purpose of the file. The appropriate response may include documenting an unresolved gap and assessing its impact.

7. Gate three: timing and version

The third gate connects the record to the period or event it is meant to support. A document can be authentic and complete while being the wrong version for the action under review. Record the relevant approval, effective date, distribution and use as appropriate to the document type. Preserve superseded versions when they remain necessary to explain earlier trial conduct.

Take a consent review as an example. The useful chain identifies the approved participant-facing version, the version used for the individual, the date of the process and any later requirement for updated consent. A folder containing only the newest blank form leaves the historical question unresolved. The participant record and approval record may sit in different controlled locations, but the relationship must be available to an authorized reviewer.

Training provides another example. Define what change required training, which roles were affected and what evidence shows completion before the relevant activity where required. A generic annual training date can be unrelated to a newly introduced procedure. Conversely, an unchanged routine activity may have a different training need from a protocol-specific assessment. The review should use the actual assigned task and applicable requirements.

Filing timeliness can be measured, but the clock needs a definition. Distinguish the date of the underlying event, completion of the record, receipt by the filing party, initial upload and acceptance after quality review. A long event-to-file interval could reflect delayed creation, delayed transfer or delayed filing. These are different processes with different owners. Report the components needed to diagnose the problem instead of assigning every delay to the uploader.

We propose examining the distribution of lag by record category and importance. A median alone can hide a small group of seriously overdue records. Pair it with an upper percentile or an aging list, and preserve the denominator and cutoff date. Set escalation thresholds through the trial's quality plan and contractual arrangements. This report proposes no universal number of days that makes a record timely under every circumstance.

8. Gate four: relationships and decisions

The fourth gate tests whether several valid records tell a coherent story. Choose a consequential decision and trace its inputs, authority, outcome and follow-up. The sampled chain might concern eligibility, a protocol amendment, a safety assessment, a monitoring issue or a system change. Each chain should reflect the records and responsibilities relevant to that decision.

For eligibility, an authorized reviewer may need the applicable criterion, source measurement, laboratory range or time calculation and the investigator's assessment. Our companion eligibility analysis examines this study-build problem in detail. The TMF review should establish the appropriate record locations and oversight evidence while respecting the investigator's custody of participant source records. Uploading a duplicate medical history into a central repository is not the default solution to a missing relationship. [15]

A monitoring issue offers a sponsor-side example. Trace the observation in the monitoring record to the assigned action, supporting response and assessment of whether the issue was resolved. If the same issue recurs, retain the connection to previous actions so the reviewer can see whether the response changed. A status labeled “closed” should have an interpretable basis rather than simply indicating that a ticket was removed from an active queue.

For safety information, trace the relevant awareness evidence, clinical assessment and reporting actions under the applicable procedures. Multiple organizations may learn about an event at different times. The record should distinguish those facts and preserve the evidence used to assess the reporting obligation. A single manually entered date can conceal important differences unless its definition and source are clear.

The public FDA frequencies cannot reveal an inspector's sampling plan for a specific trial. We propose these exercises as internal quality checks, informed by the importance of the decision and known process risks. FDA's risk-based monitoring guidance supports planning and communicating monitoring in relation to important trial risks; the five gates and selection of review intervals are proposed here as study-specific implementation choices. [16]

A useful finding describes the broken relationship precisely. “Missing documentation” can mean that a record was never created, that access failed, that the wrong version was supplied or that the reviewer could not connect two identifiers. Capture the mechanism and affected scope. That information makes it possible to choose a correction and assess whether similar records need review.

9. Gate five: retrieval and preservation

The fifth gate tests whether authorized users can still obtain and interpret the records. Include both routine access and a plausible disruption, such as a departed administrator, expired vendor contract or unavailable primary system. A successful backup job is evidence about copying data; a successful restore-and-review exercise adds evidence that the retained material remains usable.

Define the retrieval task before timing it. Identify the requester, authorization, record scope and information needed to answer the question. A simple approved protocol and a cross-system participant trace have different retrieval demands. The elapsed time is useful as a process measure when the task and conditions are comparable. There is no universal “minutes to pass inspection” threshold established by the data in this report.

For an archive export, examine the index, document identities, versions, attachments, signatures, timestamps and audit information relevant to the intended use. Test search and navigation with someone who did not prepare the export. If understanding the files requires undocumented knowledge from a former vendor employee, the export needs additional context or a different preservation arrangement.

Preservation also includes continued control. Specify how access is granted and removed, how changes are traced, how integrity is checked and how incidents are investigated. Keep an inventory of formats and dependencies that may become difficult to support. A scheduled review can identify the need for migration before the only readable copy depends on obsolete software or an unavailable license.

A proposed migration acceptance test starts with a defined source population and reconciles it to the destination. Compare record counts and identifiers, then assess content and necessary metadata. Investigate exclusions and transformations explicitly. Where files are transformed, a changed byte hash can be expected; the validation must assess preservation of the required information rather than treating unequal hashes as automatically unacceptable or equal file counts as sufficient proof.

A preservation review should also test the meaning of metadata after export. A timestamp without its timezone, a signature identifier without the associated identity record, or a code without its definition can make an otherwise readable record ambiguous. Identify the metadata needed for each important record class and include it in the acceptance criteria. Preserve the explanation of any transformation, such as a date normalized into a different display format, so later reviewers can distinguish a presentation change from a change in the underlying event.

Access testing should use realistic roles. Ask an authorized investigator user to retrieve the retained participant information relevant to their responsibilities and ask a blinded sponsor user to retrieve the permitted oversight records. Then verify the behavior of a role that should lack access. Record both successful and appropriately refused requests. This exercise checks whether the preservation arrangement retains the intended boundaries as well as the files themselves.

For a trial using several systems, keep a small set of cross-system reference cases as part of migration planning. Each case should identify the expected records, relationships and interpretation before transfer. Reconstruct those same cases afterward and investigate any changed outcome. Select cases because they exercise important complexity, such as a corrected result or a superseded approval, and label the exercise as targeted validation rather than a statistical estimate of the entire archive's quality.

Document who authorizes retirement of the old environment and on what evidence. The decision should consider any unresolved discrepancies, applicable retention requirements, legal holds and the tested ability to retrieve records from the new arrangement. Archiving is a controlled transition of access and responsibility, not simply the date a project stops paying for a production subscription.

10. What a 95% completeness score leaves unanswered

Consider a hypothetical catalog containing 1,000 potential records. After documented applicability review, 100 are inapplicable and 100 are not yet due. The current due-and-applicable denominator is 800. If 760 are present, completeness is 95.0%, with 40 missing. Counting presence against all 1,000 catalog entries would produce 76.0% and answer a different question. Both the exclusions and the cutoff date need to be visible.

Now sample 80 of the 760 present records using a documented risk-based selection. Suppose 72 meet the defined quality criteria and eight have a defect. The sampled pass rate is 90.0%. It describes the reviewed sample. If selection deliberately emphasizes higher-risk records, the result should not be presented as an unbiased estimate for the entire file. A valid population estimate would require an appropriate sampling design and uncertainty assessment.

Completeness and quality therefore belong side by side, with different denominators and meanings. Add the number and nature of unresolved critical issues where relevant, because a high average can coexist with one record that matters greatly to participant protection or a central result. Define severity through the trial's quality process instead of assigning it solely from a document category.

A practical dashboard might show due-and-applicable presence, sampled quality results, overdue important records, unresolved relationship failures and retrieval-test outcomes. Each measure should identify its owner, cutoff, scope and action threshold. The threshold should prompt a decision or investigation. A chart that changes color without assigning an action adds display complexity without improving control.

Avoid converting these metrics into a guarantee of inspection success. Neither 95% completeness nor a particular sampled pass rate establishes regulatory acceptability. The value lies in finding and resolving specific weaknesses before they compromise a decision or make trial conduct difficult to verify. The supporting issue records are as important as the dashboard because they show what the team actually did with the information.

11. Make sponsor, CRO and site responsibilities explicit

A filing agreement should describe the work in enough detail that both parties recognize the same deliverable. It should address record categories, timing, review, access, correction, subcontractors, transfers and end-of-service arrangements. Align it with the actual trial operating model, including records retained by investigators, institutions and other custodians.

The US regulatory distinction deserves precision. Under 21 CFR 312.52, a sponsor may transfer specified Part 312 obligations to a CRO in writing. Obligations outside the written transfer remain with the sponsor, and the CRO assumes the applicable regulatory duties and potential action for transferred obligations. A blanket statement that only work, never obligations, can be transferred would misstate this provision. The written allocation needs to be reviewed in the relevant regulatory context. [17]

Our proposed operating agreement distinguishes the regulatory allocation from the practical filing workflow. For a monitoring report, identify the creator, reviewer, custodian, expected availability and party resolving findings. For investigator qualifications, specify which evidence is required, who obtains it and how changes are communicated. For system records, establish how the responsible party can obtain validation and change evidence needed for the trial's intended use.

Use event-based deadlines where they clarify the work, but define the event. A deadline measured from receipt behaves differently from one measured from a visit or final approval. If a record arrives incomplete, specify whether the clock continues, how the defect is escalated and what evidence remains available. These details prevent a contractual metric from rewarding repeated transfers of unusable material.

A CRO transition should include an open-item inventory. Reconcile the expected record set, outstanding queries, unresolved quality findings, access permissions and retention responsibilities. Ask the incoming team to retrieve a representative sample using the transferred index. That demonstration tests whether the handover conveys usable knowledge rather than only the volume of files moved.

The same principle applies when a site closes or an investigator leaves. Identify the continuing institutional custodian and the access route for retained records, as required by the applicable arrangements. Update contact details and responsibility records. A former employee's email address should not remain the sole method of locating records that must be available years later.

12. A practical review cycle and its limits

Start the review cycle with the trial's important decisions and current changes. New countries, amendments, providers, data sources or systems can alter the expected record set. Our companion decentralized-elements report examines the additional handoffs created by remote activities and distinguishes routine local care from protocol-specific work. Use those activity definitions to identify the relevant records and arrangements. [18]

Select a review sample with an explicit purpose. It may target newly introduced processes, known delayed categories or records supporting a critical decision. Record the population from which the sample was chosen and why it was selected. A targeted sample is valuable for finding problems, while its results need to remain labeled as targeted rather than representative of every record.

For each finding, distinguish immediate correction from the assessment of underlying cause and wider scope. A missing attachment may be supplied promptly, while the underlying interface defect affects many other records. A wrong version may require examination of activities conducted using it. Determine the response through the relevant clinical, quality and regulatory processes rather than assuming every issue can be closed by adding a file.

Follow up on whether the chosen action worked. A repeat sample can examine the affected process after correction, using a defined question and time period. Preserve the evidence of the decision, including a rationale when no broader action is warranted. This creates a more informative history than repeated assertions that the file was reviewed and found satisfactory.

A small trial may implement these controls through a controlled index and documented review process. A larger program may use an eTMF workflow with automated expectations, reminders and permissions. The appropriate tool depends on complexity and risk. In either setting, configure the process so that a human reviewer can identify the applicable record, understand its provenance and follow the consequential decision it supports.

Methods and limits

The numerical lane uses the R2-pinned FDA inspection-observation dataset with an August 23, 2026 snapshot. We verified the normalized file against its manifest hash, selected FY2020–FY2025 BIMO rows and summed published frequencies by exact reference and declared reference-family rules. The supplement contains 479 selected aggregate rows, annual results and the three sensitivity definitions. An independent rerun can reproduce the totals without interpreting individual case narratives. [2]

The legal and guidance lane was checked on September 18, 2026. It distinguishes binding provisions, agency guidance and our proposed implementation controls. The retention table covers selected medicinal-product obligations and leaves room for longer or additional requirements. The five quality gates, worked dashboard and suggested review exercises are an editorial framework. They have not been validated as a predictive model of inspection outcomes or as evidence of EClinCloud product effectiveness.

The aggregate observation data lack site-level denominators, complete finding narratives and case-level outcomes. They cannot establish the cause of a specific record problem, the share of defective TMFs or the effect of a technology. Their contribution is a transparent description of selected cited obligations. The trial-specific record map, quality samples and retrieval demonstrations supply the operational evidence needed for an actual review decision.

Sources

1. FDA — Inspection Observations, annual workbooks.

2. EClinCloud — Public BIMO records selections, annual frequencies and source lineage.

3. EUR-Lex — Regulation (EU) No 536/2014, Articles 57 and 58.

4. FDA — Form 483 Frequently Asked Questions.

5. FDA — ICH E6(R3) Good Clinical Practice, September 2025, including Appendix C.

6. EMA — Guideline on the content, management and archiving of the clinical trial master file, 2018.

7. EUR-Lex — Regulation (EU) 2016/679, GDPR.

8. eCFR — 21 CFR 312.62, investigator recordkeeping and retention.

9. eCFR — 21 CFR 312.57, sponsor recordkeeping and retention.

10. eCFR — 21 CFR 56.115, IRB records.

11. MHRA — Archiving and retention of clinical trial records, updated April 28, 2026.

12. MHRA — Clinical Trials Regulations transitional arrangements.

13. FDA — Electronic Systems, Electronic Records, and Electronic Signatures in Clinical Investigations, October 2024.

14. EMA — Guideline on computerised systems and electronic data in clinical trials, 2023.

15. EClinCloud — Eligibility Criteria and Screening Burden: A Study-Build Evidence Review.

16. FDA — A Risk-Based Approach to Monitoring of Clinical Investigations: Questions and Answers, April 2023.

17. eCFR — 21 CFR 312.52, transfer of obligations to a CRO.

18. EClinCloud — Decentralized Trial Elements: Evidence, Oversight and Country Review.