Why Document-Centric Clinical Operations Are Failing

Introduction

Phase III clinical trials now generate an average of 5.96 million data points per protocol, according to a 2025 collaborative initiative by TransCelerate BioPharma and the Tufts Center for the Study of Drug Development ^[1]. This figure is based on submitted research from the TransCelerate Optimizing Data Collection initiative and reported through an industry announcement. A decade earlier, that volume was approximately 1.2 million, based on Tufts CSDD's 2021 analysis, which found that 3.6 million data points represented roughly three times the volume collected by late-stage trials ten years prior ^[2]. The operational systems managing those trials have not kept pace. In most sponsor and CRO organizations, the primary unit of clinical work is still the document: a static file produced, reviewed, revised, stored, and hoped to remain consistent with every other document it references. The gap between the volume and velocity of modern trial data and the limitations of document-based operations is generating measurable costs in avoidable amendments, startup delays, and growing difficulty satisfying a regulatory framework that increasingly prioritizes data integrity over document completeness.

This article examines what document-centric clinical operations look like in practice, where they contribute to delays and quality risk, what the evidence shows about the costs, and why the regulatory and operational consensus is shifting toward a different model.

What Document-Centric Clinical Operations Actually Means

The term sounds abstract, but the reality is familiar to anyone who has worked in clinical research. Document-centric operations treat each trial artifact as a discrete, largely self-contained file: the protocol, the investigator's brochure (IB), the informed consent form (ICF), the statistical analysis plan (SAP), the DSUR, the CSR. These documents are authored separately, reviewed in isolation, version-controlled through naming conventions or TMF folder structures, and reconciled manually when changes in one create obligations to update another.

The logic behind this approach made sense decades ago, when trials were smaller, regulatory submissions were physically delivered, and the principal challenge was generating paper records that held up to audit. The paper-based case report form, the three-ring binder of site files, the NDA delivered on disc: these were artifacts of an era in which producing any consistent documentation at all was a genuine achievement.

The problem is that document-centric thinking has persisted long past its useful life. The procedures are now electronic, but the operating logic is the same. Teams author a protocol in Word, distribute it via email or SharePoint, and attempt to cascade updates manually to consent forms, site training materials, and the IB when eligibility criteria change. Monitoring relies on checking document completeness rather than reading signals in trial data. Regulatory decisions are made by reviewing binders rather than querying structured records. As an operational observation: the document is still the system.

Why This Topic Matters Now

Three converging pressures have made document-centric operations increasingly costly to sustain.

Protocol complexity has outpaced manual management. Tufts CSDD reported in 2021 that Phase II and III protocols now involve 263 distinct procedures per patient, a 44% increase since 2009 ^[2]. TransCelerate BioPharma's collaborative research with Tufts CSDD documented that over the past decade, Phase III pivotal trials saw a 283% increase in data points collected and a 40% increase in total procedures ^[3]. When a protocol contains that many interdependent specifications, maintaining consistency across all downstream documents through manual review is a significant and unreliable burden at scale.

Amendment rates show that the document-authoring model is under strain. A 2024 study published in Therapeutic Innovation and Regulatory Science, drawing on Tufts CSDD data from 950 protocols and 2,188 amendments, found that the prevalence of protocols requiring at least one amendment increased from 57% in 2015 to 76% ^[4]. The mean number of amendments per protocol rose 60%, from 2.1 to 3.3, over that period ^[4]. Earlier Tufts CSDD research documented that 45% of amendments were avoidable, driven by protocol design deficiencies including design flaws, errors and inconsistencies in the protocol narrative, and infeasible execution instructions ^[5]. The median direct cost to implement a substantial amendment was $141,000 for a Phase II protocol and $535,000 for a Phase III protocol, and the authors of that study estimated the actual cost including indirect factors is likely three to four times larger for Phase III programs ^[5].

The time cost is equally significant. The 2024 Tufts CSDD study found that the average time from identifying the need for an amendment to last oversight approval is now 260 days, and that investigative sites operate with different versions of the clinical trial protocol for an average of 215 days during that window; a duration that has nearly tripled over the past decade ^[4]. During those 215 days, a site that implemented the amendment two months ago and a site still awaiting ethics committee approval are running the same trial under different operational specifications.

ICH E6(R3) has formally redirected the regulatory framework. The revised Good Clinical Practice guideline, adopted by the ICH at Step 4 on January 6, 2025, represents the most significant reorientation of GCP since E6(R2) in 2016 ^[6]. The EMA implemented E6(R3) on July 23, 2025 ^[7]. The FDA issued a corresponding guidance in September 2025, which, consistent with standard FDA guidance practice, represents the agency's current thinking and is not legally binding ^[8]. The guideline moves away from prescriptive documentation requirements toward quality-by-design (QbD) principles, risk-proportionate oversight, and emphasis on data reliability over document completeness. This article uses "data-centric" as an operational interpretation of the direction E6(R3)'s data governance provisions point toward, not as a regulatory phrase from the guideline itself.

Where Document-Centric Operations Contribute to Risk

The Cross-Document Consistency Problem

A clinical trial protocol is not a standalone artifact. It is the source document for a cluster of interdependent records: the IB informs the risk profile described in the ICF; the eligibility criteria in the protocol define the patient populations the SAP must account for; the endpoint definitions determine what the CSR must report. When these documents are generated independently, by different teams, using different templates, at different times, inconsistencies accumulate without any structured mechanism to detect them.

Tufts CSDD research (2016) identified "errors and inconsistencies in the protocol narrative" and "infeasible execution instructions and eligibility criteria" as among the leading causes of avoidable amendments ^[5]. These are authoring failures, not scientific uncertainties. They arise during the drafting phase, before formal QMS review, when protocol teams working in parallel on different sections do not have a structured way to check that what one team wrote is consistent with what another committed to.

To make the cascade concrete: a sponsor discovers during enrollment that an eligibility criterion excluding prior immunotherapy is too restrictive and requires amendment. That single change does not stay in the protocol. It must propagate to the ICF, which informs participants of inclusion criteria; to the SAP, which may have pre-specified analyses tied to the original population; to the eDC system, which screens patients against protocol criteria; to site training materials; and to any country-specific protocol variants. Each update requires someone to open a document, identify the relevant passage, revise, route for review and approval, and confirm that the updated version replaced the prior one at every site. The Getz 2024 data, 215 days of sites operating under different protocol versions, illustrates how persistent those version gaps can be during an active amendment cycle ^[4].

The ACRP noted in a February 2026 analysis that documentation delays in clinical trial startup frequently trace to "varying workflows or manual processes" that fail to synchronize reviewing cycles, track document versions, and preserve audit trails across the site network ^[9].

Startup Delays and the Documentation Bottleneck

Study startup is where document-centric operations impose their most measurable cost. An ICON industry survey conducted in June 2025, covering more than 100 principal investigators and senior site personnel, found that 55% of sites reported the time from site selection to full activation takes five months or longer ^[10]. The same survey found that 66% of sites experience contract and budget delays "often" or "always," and 92% cited sponsor and CRO documentation support as the area most in need of improvement ^[10].

Site pre-selection decline rates, where sites refuse participation before formal selection, rose from 35% in 2021 to 47% in 2023 according to ICON's longitudinal data ^[10]. That increase reflects a compounding dynamic: as trial complexity grows, the administrative burden at site activation grows with it, and sites weigh the overall onboarding cost when deciding whether to participate. Documentation burden is one component of that calculation; contractual, financial, and competitive factors contribute as well. The ACRP analysis confirms that documentation specifically is a documented bottleneck: startup delays commonly trace to incomplete regulatory packets, missing forms, outdated training records, and version conflicts in the documents sites receive from sponsors ^[9].

The Problem With Monitoring That Reads Documents Instead of Data

Traditional source data verification requires a clinical research associate to visit a site, pull records, and compare them against data entered into the eDC system. When the clinical question is whether a trial is performing as designed, document comparison is a lagging indicator, not a real-time signal. By the time an on-site review identifies a protocol deviation or eligibility error, it was created days or weeks earlier.

Kelly, Spreafico, and Siu (2020), writing in the British Journal of Cancer, identified the administrative burden of maintaining numerous static site files over prolonged follow-up periods as one specific operational inefficiency that compounds site workload in oncology trials ^[11]. They recommended transitioning to digitally scanned electronic formats and consolidating long-term follow-up patients across multiple open studies into shared protocols; practical steps that reduce per-site document burden without requiring advanced AI or full platform transformation ^[11].

ICH E6(R3) Section 3.11.4 introduces a risk-proportionate monitoring framework that explicitly supports centralized, data-driven oversight as an alternative or complement to traditional on-site monitoring, with monitoring strategies designed to detect risks and data quality issues in a timely and proportionate manner ^[6]. ACRP's summary of FDA's E6(R3) adoption highlights "audit trails, metadata, traceability, and secure system validation" as the practical compliance implications for sponsors ^[12]. Risk-based quality management under E6(R3) is more naturally supported by structured, timely data access than by static document archives alone.

The Amendment Cascade as a System Warning Signal

Protocol amendments function as a diagnostic on the document-authoring system. When an amendment is filed, it means that the protocol was insufficient to govern the trial as originally designed. The amendment then requires a cascade of secondary updates across the ICF, IRB/IEC submissions, site-level training materials, country-specific regulatory filings, and the eDC system; each managed through the same manual, document-by-document process that produced the original error.

With 76% of protocols now requiring at least one amendment and the mean count at 3.3 per protocol, this cascade repeats multiple times across the average study ^[4]. In nearly half of historical cases, the amendment could have been avoided if protocol design deficiencies had been caught during drafting ^[5]. That those deficiencies were not caught reflects the absence of automated cross-document validation in the protocol authoring workflow, not individual reviewer failure.

Regulatory and Documentation Considerations

ICH E6(R3), adopted at Step 4 on January 6, 2025, articulates the operating principles that document-centric organizations must now align with across ICH regions ^[6]. Four provisions are particularly relevant.

Quality by design. E6(R3)'s Principles section and Section 3.10 establish that sponsors must identify factors critical to trial quality during protocol development and build proportionate controls around them, rather than relying on post-hoc document inspection to catch quality failures ^[6]. A protocol authoring process that generates Word documents without structured validation of cross-document dependencies does not align naturally with this expectation.

Risk-proportionate monitoring. Section 3.11.4 of E6(R3) establishes that sponsors must implement a monitoring approach, including centralized data review, proportionate to the risks identified in the trial quality risk assessment ^[6]. The FDA described the revision as embracing "innovations in trial design, conduct, and technology" and explicitly supports remote and centralized monitoring approaches ^[8]. These approaches are better supported by structured, timely data access than by static document archives alone.

Data governance and the lifecycle of records. E6(R3) introduced a dedicated Data Governance section (Section 4) covering the full data lifecycle: capture, relevant metadata including audit trails, data corrections, transfer and migration, retention, and destruction ^[6]. Section 3.16 covers data handling and record keeping for sponsors, including requirements for computerised system validation, user accountability, access controls, and traceability. ACRP's analysis of FDA's adoption highlights these provisions as placing heightened expectations on sponsors for data integrity throughout the trial lifecycle ^[12].

ICH E6(R3) Section 4.2 establishes data lifecycle requirements for attributability, legibility, contemporaneity, accuracy, completeness, security, and reliability of clinical data ^[6]. Industry practitioners commonly discuss these requirements using the ALCOA+ framework, which extends the original ALCOA principles with completeness, consistency, endurance, and availability ^[13]. That taxonomy is useful shorthand, but the formal ICH requirements are in Section 4 directly, and sponsors should align their systems to those provisions.

Essential records, not essential documents. E6(R3) replaces the older "essential documents" concept with "essential records," encompassing all formats of evidence whether paper or electronic, structured or unstructured, and requiring that these records be maintained with version control and directly accessible for regulatory inspection ^[6]. E6(R3) Section 4.2.2 specifically addresses the role of metadata and audit trails in supporting record integrity ^[6]. This shift reflects the guideline's intent that quality assurance be built into the data governance framework of a trial, not demonstrated retrospectively through a complete records repository.

Sponsors operating in the European Union should note that the EMA implemented E6(R3) on July 23, 2025, with Annex 1 covering interventional trials now in force ^[7]. FDA adoption, communicated in September 2025, represents the agency's current expectations for U.S. trials ^[8]. Annex 2, covering decentralized and non-traditional trial designs, remains under consultation ^[7].

Document-Centric vs. Data-Centric Operations: A Practical Comparison

"Data-centric" is used here as an operational description of how trial workflows are reorganized when structured data, rather than authored files, governs the consistency of clinical records. It is not a regulatory phrase from ICH E6(R3), though the guideline's emphasis on data lifecycle governance, essential records, and quality by design points in this direction.

This comparison does not imply that documents disappear under a data-centric model. TMF/essential-record repositories remain required under GCP, and essential records must be maintained and directly accessible throughout the trial and retention period, as established in ICH E6(R3) Sections 3.16 and the Annex 1 essential records provisions ^[6]. What changes is the authoring logic: documents become outputs of structured processes rather than the primary operational system.

The Evidence on Data-Driven Clinical Operations

The SCOPE 2026 meeting produced a documented consensus among clinical operations leaders: the industry is "rapidly moving from fragmented, document-driven processes toward connected, data-centric ecosystems," and experts identified digital protocols as "the backbone of the next era of automation, efficiency, and intelligence in clinical research" ^[14]. The meeting discussions emphasized transforming the protocol from "a static document into a dynamic digital asset" that can orchestrate downstream workflows and enable "adaptive, data-driven decision-making across the trial life cycle" ^[14].

Kelly, Spreafico, and Siu (2020) identified operationally grounded efficiency improvements including transitioning from paper-based site document storage to digitally scanned electronic formats and consolidating follow-up patients from multiple open protocols into a single shared protocol; both of which reduce the per-document labor cost of running a trial without requiring advanced AI or full platform transformation ^[11].

Babaeipour, Charest, and Wright (2026), in a preprint evaluation of 23 publicly available clinical trial protocols, found that a clinical-research-specific retrieval-augmented generation (RAG) system achieved an average extraction accuracy of 69.6% across six protocol content categories, compared to 62.6% for standalone LLM prompting ^[15]. In a controlled experiment with simulated clinical research coordinator workflows, AI assistance reduced task completion time and improved accuracy, with the largest gains in content scattered across multiple protocol sections, including adverse events, interventions, and site requirements ^[15]. This is a preprint on a small simulated sample and has not yet undergone peer review; results should not be generalized to regulatory submission contexts without further validation.

AI and Automation Perspective

AI applied to regulatory document generation and cross-document validation addresses a specific, bounded problem: the cost and error rate of producing documents that are internally consistent, protocol-aligned, and compliant with current regulatory guidance. The appropriate frame is not replacement of medical writing expertise but reduction of the authoring and reconciliation burden that currently consumes significant clinical operations capacity.

An AI system trained on regulatory document structures can flag when eligibility criteria in a draft ICF diverge from the protocol's inclusion/exclusion specifications, identify when an IB section describing a safety signal has not been reflected in the corresponding DSUR, or generate a draft CSR section that references the endpoints defined in the SAP, reducing the manual transcription step that introduces errors. What AI cannot do is verify the scientific accuracy of a clinical judgment, validate a regulatory strategy, or replace the review of a qualified medical writer or regulatory affairs professional.

ICH E6(R3) Section 4.3 requires that computerised systems used for data capture and record generation be fit for purpose, validated, and subject to appropriate controls including audit trails and access management ^[6]. AI-generated regulatory documents are not exempt from this requirement; they require expert review and validation before submission. The contribution of AI in this space is best understood as a reduction in authoring cycle time and a more consistent starting point for human review, not an autonomous submission capability.

The more significant long-term shift may be in converting static protocols into structured data assets. When a protocol is authored as machine-readable fields, defined eligibility criteria, endpoint specifications, version-controlled outputs, the downstream consistency problem becomes amenable to software-level validation rather than dependent on human comparison at each document boundary. That is the transition the SCOPE 2026 panelists were describing, and the direction that E6(R3)'s data governance provisions implicitly support.

How Kitsa Fits Into This Problem

According to Kitsa's product documentation, KScribe supports AI-assisted regulatory document generation for protocols, CSRs, SAPs, ICFs, and IBs, with agents designed to enforce regulatory guideline adherence and cross-document consistency. Kitsa also identifies DSUR generation among its regulatory document generation capabilities ^[16]. In the context of this article's argument, those capabilities are directly relevant to the manual authoring and reconciliation burden associated with avoidable amendments and startup delays. Human review remains the final step; the tool is designed to improve the quality and consistency of the starting point for that review, not to replace it. For organizations managing multiple concurrent programs or high-amendment therapeutic areas like oncology, reducing the authoring cycle time per document type compounds across a portfolio.

Key Takeaways

• Phase III trials now average 5.96 million data points per protocol, roughly five times the volume from a decade ago, per a 2025 TransCelerate BioPharma and Tufts CSDD initiative announcement ^[1].
• Protocol amendment prevalence rose from 57% to 76% between 2015 and 2022. The mean number of amendments per protocol increased 60% to 3.3. Implementation now takes an average of 260 days from need-to-amend identification to last oversight approval, during which sites operate with different protocol versions for an average of 215 days ^[4]. Tufts CSDD found 45% of amendments were avoidable, and Phase III amendments carry a median direct cost of $535,000 each ^[5].
• 55% of investigative sites report that activation from site selection takes five months or longer. 92% cite documentation support as the area most needing improvement from sponsors and CROs ^[10].
• ICH E6(R3), adopted January 6, 2025, introduced a dedicated Data Governance section (Section 4) covering the full data lifecycle, replaced "essential documents" with "essential records," and established quality-by-design and risk-proportionate monitoring as foundational requirements. The EMA implemented it July 23, 2025; the FDA issued corresponding guidance in September 2025 ^[6],[7],[8].
• Cross-document inconsistencies in consent forms, protocols, IBs, and SAPs frequently trace to the drafting phase, before formal QMS review, and are difficult to detect reliably through manual comparison alone ^[4],[5].
• The SCOPE 2026 industry consensus identified the transition from document-driven processes to connected, data-governed ecosystems as the defining operational shift of the current era in clinical research ^[14].
• The "data-centric" model described in this article is an operational interpretation of where E6(R3)'s data governance requirements point, not a regulatory phrase from the guideline itself.

FAQ

References

1. [1]TransCelerate BioPharma and Tufts Center for the Study of Drug Development. "TransCelerate and Tufts CSDD Uncover Opportunities to Rethink Data Collection and Optimize Protocol Design." Drug Discovery News, September 15, 2025. https://www.drugdiscoverynews.com/transcelerate-and-tufts-csdd-uncover-opportunities-to-rethink-data-collection-and-optimize-protocol-design-16646
2. [2]Tufts Center for the Study of Drug Development. "Rising Protocol Design Complexity Is Driving Rapid Growth in Clinical Trial Data Volume." GlobeNewswire, January 12, 2021. https://www.globenewswire.com/news-release/2021/01/12/2157143/0/en/Rising-Protocol-Design-Complexity-Is-Driving-Rapid-Growth-in-Clinical-Trial-Data-Volume-According-to-Tufts-Center-for-the-Study-of-Drug-Development.html
3. [3]TransCelerate BioPharma. "Optimizing Data Collection Initiative." TransCelerate BioPharma Inc., September 2025. https://www.transceleratebiopharmainc.com/initiatives/optimizing-data-collection/
4. [4]Getz K, Smith Z, Botto E, Murphy E, Dauchy A. "New Benchmarks on Protocol Amendment Practices, Trends and Their Impact on Clinical Trial Performance." Therapeutic Innovation and Regulatory Science, 2024;58(3):539-548. doi:10.1007/s43441-024-00622-9. PMID: 38438658. https://link.springer.com/article/10.1007/s43441-024-00622-9
5. [5]Getz KA, Stergiopoulos S, Short M, Surgeon L, Krauss R, Pretorius S, Desmond J, Dunn D. "The Impact of Protocol Amendments on Clinical Trial Performance and Cost." Therapeutic Innovation and Regulatory Science, 2016;50(4):436-441. doi:10.1177/2168479016632271. https://link.springer.com/article/10.1177/2168479016632271
6. [6]International Council for Harmonisation. "Guideline for Good Clinical Practice E6(R3)." ICH Step 4 Final Guideline, January 6, 2025. https://database.ich.org/sites/default/files/ICH_E6%28R3%29_Step4_FinalGuideline_2025_0106.pdf
7. [7]European Medicines Agency. "ICH E6 Good Clinical Practice; Scientific Guideline." EMA, effective July 23, 2025. https://www.ema.europa.eu/en/ich-e6-good-clinical-practice-scientific-guideline
8. [8]U.S. Food and Drug Administration. "E6(R3) Good Clinical Practice (GCP)." FDA Guidance Document, Docket FDA-2023-D-1955, September 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/e6r3-good-clinical-practice-gcp
9. [9]ACRP. "Administrative Weaknesses That Slow Down Clinical Trial Operations and How to Fix Them." February 17, 2026. https://acrpnet.org/2026/02/17/administrative-weaknesses-that-slow-down-clinical-trial-operations-and-how-to-fix-them
10. [10]ICON plc. "ICON Survey Reveals Increasing Clinical Trial Startup Delays, Underscoring Need for Human-Centred Site Activation Solutions." Press Release, December 2, 2025. https://www.iconplc.com/news-events/press-releases/icon-survey-reveals-increasing-clinical-trial-startup-delays
11. [11]Kelly D, Spreafico A, Siu LL. "Increasing Operational and Scientific Efficiency in Clinical Trials." British Journal of Cancer, 2020;123(8):1207-1208. doi:10.1038/s41416-020-0990-8. PMID: 32690866. https://www.nature.com/articles/s41416-020-0990-8
12. [12]ACRP. "FDA Publishes ICH E6(R3): What It Means for U.S. Clinical Trials." September 23, 2025. https://acrpnet.org/2025/09/16/fda-publishes-ich-e6r3-what-it-means-for-u-s-clinical-trials
13. [13]Medicover MICS. "ICH GCP E6(R2) vs E6(R3): Key Differences and Practical Implications." November 6, 2025. https://medicover-mics.com/ich-gcp-e6r2-vs-e6r3-key-differences/
14. [14]Applied Clinical Trials Online. "Beyond the Buzzwords: What SCOPE 2026 Revealed About Clinical Trial Planning and Operations." May 2026. https://www.appliedclinicaltrialsonline.com/view/scope-2026-clinical-trial-planning-operations
15. [15]Babaeipour R, Charest F, Wright M. "AI-Assisted Protocol Information Extraction for Improved Accuracy and Efficiency in Clinical Trial Workflows." arXiv:2602.00052v2 [cs.IR], April 16, 2026. Banting Health AI, Toronto. https://arxiv.org/html/2602.00052
16. [16]Kitsa. "KScribe: AI Regulatory Document Generation." Kitsa.ai product documentation. https://kitsa.ai/regulatory-document-generation

Suggested internal links: AI-powered regulatory document generation