Contents
Introduction
TransCelerate's member companies collectively invest $125 billion a year in pharmaceutical R&D [1]. Separately, TransCelerate estimates that roughly 90% of the clinical trial protocols produced across the industry still exist as unstructured Word or PDF documents [1]. That is the starting point for understanding the Unified Study Definitions Model, or USDM. USDM is not a new form to fill out. It is a different way of representing what a protocol actually is: not prose describing a study, but a structured, machine-readable dataset that other systems can read directly, where the integrations and data mappings exist to support it.
The distinction matters more than it might sound. A traditional protocol document communicates a study design to a human reader, who then re-keys pieces of it into an electronic data capture (EDC) system, a clinical trial management system (CTMS), a randomization platform, and eventually into SDTM datasets for regulatory submission. USDM asks a different question: what if the design were captured once, in a form that downstream systems could consume directly, wherever the integrations and data mappings exist to support it, instead of a person retyping it into each one? This article walks through what USDM is, what problem it was built to solve, where it stands today, and what changes, and what does not change, when a sponsor moves from documents to structured data.
USDM vs traditional protocol comparison table
| Comparison | Traditional Protocol Document | USDM-Structured Protocol |
|---|---|---|
| Source of truth | Narrative Word/PDF, interpreted separately by each team | Single structured dataset referenced by downstream systems [2][3] |
| Downstream reuse | Manual re-entry into EDC, CTMS, IRT, and SDTM builds | Direct consumption where integrations exist; can seed SDTM Trial Design domains [2] |
| Amendment handling | Each change re-typed and re-checked across documents and systems | A changed element can be traced to every place it is referenced, though reconciliation still requires review [5] |
| Regulatory alignment | Sponsor-specific templates and formats | Built to align with ICH M11's CeSHarP exchange specification [2][6] |
| Implementation burden | Familiar tooling; low upfront investment | Requires governance, controlled terminology, and validated system integrations [2][7] |
| Current maturity | Industry-standard today; roughly 90% of protocols [1] | Emerging; USDM v4.0 released June 2025, adoption still voluntary [1][2] |
Source of truth
- Traditional Protocol
- Narrative Word/PDF, interpreted separately by each team
Downstream reuse
- Traditional Protocol
- Manual re-entry into EDC, CTMS, IRT, and SDTM builds
- USDM-Structured Protocol
- Direct consumption where integrations exist; can seed SDTM Trial Design domains [2]
Amendment handling
- Traditional Protocol
- Each change re-typed and re-checked across documents and systems
- USDM-Structured Protocol
- A changed element can be traced to every place it is referenced, though reconciliation still requires review [5]
Regulatory alignment
- Traditional Protocol
- Sponsor-specific templates and formats
Implementation burden
- Traditional Protocol
- Familiar tooling; low upfront investment
Why This Topic Matters in Clinical Trials
Protocol amendments are the clearest evidence that document-based protocol authoring has a cost. A Tufts Center for the Study of Drug Development analysis of 836 protocols across phases I through IV found that 57% had at least one substantial amendment, and researchers judged 45% of those changes avoidable [5]. Phase II protocols averaged 2.2 global amendments and Phase III protocols averaged 2.3, with a median direct cost of $141,000 per Phase II amendment and $535,000 per Phase III amendment [5]. Protocols with substantial amendments also showed longer recruitment periods and fewer patients enrolled against original targets [5].
Some of that downstream burden is amplified by how protocol information is stored and reused. When eligibility criteria, the schedule of activities, and endpoint definitions live only inside narrative paragraphs, every downstream team, from data management to site contracting, has to interpret that language independently and translate it into their own system's format. TransCelerate describes the result plainly: protocols lack "a common, machine-readable format that can be easily exchanged among disparate systems," producing what it calls significant manual duplication of effort and an average four-month gap between protocol approval and study startup [4]. USDM is the technical standard CDISC and TransCelerate have advanced to close that gap, though industry adoption of it is still emerging rather than settled [2]. A related checklist for clinical trial startup workflow walks through where that four-month gap tends to originate today.
The operational burden behind document-based protocols
Clinical trial protocols estimated by TransCelerate to still exist as unstructured documents [1].
Average gap between protocol approval and study startup described by TransCelerate [4].
Median direct cost of a substantial Phase III protocol amendment in the cited Tufts CSDD analysis [5].
Current Evidence and Research Landscape
USDM is the flagship data standard to come out of TransCelerate's Digital Data Flow (DDF) initiative, developed jointly with CDISC [1][2]. Rather than replacing existing CDISC foundational standards such as CDASH and SDTM, USDM sits upstream of them, providing a structured starting point that those standards can consume more efficiently [2]. Development has moved through five phases since July 2021: an initial reference architecture (Phase 1), biomedical concepts and time-point modeling (Phase 2), alignment with the ICH M11 protocol template and support for complex designs (Phase 3), continued refinement (Phase 4), and a scoping phase now underway (Phase 5) [2]. USDM version 4.0 was published on June 3, 2025 [1][2].
Structurally, a USDM study is organized around a small number of nested objects. CDISC's own materials describe the model's core classes as Study, StudyVersion, and StudyDesign, where StudyVersion carries titles, identifiers, and amendment history, and StudyDesign holds the arms, epochs, encounters, timelines, activities, and eligibility criteria that make up the trial itself [2][9]. The model separates "shape" elements, such as arms and epochs, from timeline elements such as visits and scheduled activities, and it expresses the whole thing as JSON so that each fact is defined once and referenced elsewhere by identifier rather than copied [3][9]. USDMPlay, an educational resource built around the standard, describes the underlying philosophy as "write once, read many": capture the design a single time during planning, and let downstream systems draw from that same source instead of re-deriving it [3].
The reference implementation work has moved in parallel. TransCelerate, Microsoft, Accenture, and CDISC began building a Study Definitions Repository Reference Implementation in 2021, releasing a minimum viable product in June 2022 as an open-source model of a repository that stores and exchanges USDM-formatted study data [7]. CDISC's more recent Implementation Handbook, released in 2026, walks through generating SDTM Trial Design domains such as TA, TE, TV, TI, and TS directly from a USDM-structured study definition, removing a step that has historically been done by hand [2]. To illustrate the logic: a single Encounter element defining, say, a Week 4 visit window can in principle drive the EDC's visit structure, the corresponding SDTM TV domain record, and the site-facing visit calendar from one definition, rather than that window being keyed in three times by three different teams. A related project applies the same logic to the schedule of activities table specifically, aiming to turn what CDISC calls "static, ambiguous schedule tables" into structured, computable data [2].
How USDM represents a clinical trial as structured data
Each study fact is defined once and referenced elsewhere by identifier rather than repeatedly copied into separate documents and systems.
Where validated integrations and mappings exist
Operational Impact for Sponsors, CROs, and Sites
For a sponsor, the practical difference shows up first in study startup. Under the document model, a clinical operations team writes the protocol, and separate teams then reinterpret it to build the EDC, configure the CTMS, define randomization rules, and set up the interactive response technology. Each translation step is an opportunity for the kind of small inconsistency, a visit window defined slightly differently in two systems, an inclusion criterion phrased ambiguously, that eventually surfaces as a query or a protocol deviation. TransCelerate's own framing of the problem points to the four-month average delay between protocol finalization and site activation as a direct consequence of this repeated manual translation [4].
Amendments expose the same weakness from a different angle. When a change to eligibility criteria or a dosing schedule is made in a Word document, propagating that change correctly to the EDC, the CTMS, site training materials, and the informed consent form requires someone to track every place the original language was referenced and update each one. A structured model narrows that problem considerably, since a change to a single USDM element can, in principle, be traced to every system that consumes it. That traceability does not eliminate the scientific and regulatory work an amendment requires. It targets the manual reconciliation across systems, updating the EDC, the CTMS, site training materials, and consent language separately, that plausibly contributes to the cost and delay Tufts CSDD documented, though that study measured overall amendment prevalence, cost, and performance impact rather than breaking out how much of the $141,000 to $535,000 per amendment traces specifically to reconciliation work [5]. For more on how a single amendment ripples through a trial's other documents, see Kitsa's breakdown of downstream amendment impact. Sites benefit too: a structured schedule of activities can drive visit calendars and budget grids directly, rather than requiring a coordinator to transcribe a table from page 47 of a PDF.
From repeated transcription to structured reuse
Traditional Protocol
Separate updates across each consuming system
USDM-Structured Protocol
Trace affected downstream references
USDM enables structured reuse and traceability, but downstream systems still require appropriate integrations, mappings, validation, and human review.
Regulatory and Documentation Considerations
Regulators have moved to formalize this shift rather than leave it to individual sponsors and vendors. The ICH M11 guideline on the Clinical Electronic Structured Harmonised Protocol, known as CeSHarP, introduces both a harmonized protocol template and a technical specification for exchanging protocol content electronically across ICH regions [6]. In the European Union, ICH M11 reached Step 5 adoption on 15 December 2025, with the stated purpose of ensuring protocols are "prepared in a consistent fashion" and enabling "the interoperable electronic exchange of protocol content" on an open, non-proprietary basis [6]. USDM's Phase 3 and Phase 4 development work was explicitly aligned to ICH M11, which is why the two efforts are frequently discussed together rather than as competing standards [2]. It is worth being precise about what that alignment does and does not require: ICH M11 introduces the CeSHarP template and specifies a technical format for electronic exchange; it does not name the CDISC USDM standard as a requirement. USDM is one of the clearest implementation paths aligned with that specification, but a sponsor's implementation obligation runs to the exchange format itself, where that format has been adopted into applicable regulatory expectations, not to a specific vendor's version of USDM. Like all FDA guidance documents, the U.S. version explicitly states that it "does not establish legally enforceable responsibilities" and instead describes the agency's current thinking and recommendations, with sponsors free to use an alternative approach that meets the underlying statutory and regulatory requirements [10]. Adoption timelines also differ by region: the EU reached Step 5 on 15 December 2025, and FDA finalized its own guidance, template, and technical specification for CeSHarP in May 2026, after having circulated them as drafts in December 2022 and June 2025 [6][10].
This is a meaningful departure from how sponsors have historically treated protocol format, largely as a matter of internal template preference. Robert DiCicco, a TransCelerate vice president, has argued that digital protocols create "opportunity for higher-quality data and a more connected ecosystem where insights can be shared," and Michelle Rohrer of Roche has framed early adoption as a way for sponsors to be ready for ICH M11 compliance while also capturing efficiency gains before the standard becomes a baseline expectation rather than an option [1]. Regulatory readiness, in other words, is becoming a second reason to adopt structured protocols alongside the operational efficiency case, not just a distant compliance deadline.
AI and Automation Perspective
Structured protocol data is also what makes credible AI use in clinical development feasible in the first place. A large language model asked to draft or review a section of an unstructured PDF protocol is working from a document with no formal semantics, no reliable way to distinguish an inclusion criterion from a note in a table footnote. A model working against a USDM-structured study design has clean, typed inputs: defined arms, defined epochs, defined eligibility elements, each tagged with what it is rather than merely how it reads.
That does not mean automation should run unsupervised, and it does not mean every use of AI in protocol drafting falls under the same regulatory framework. FDA's January 2025 draft guidance on artificial intelligence in regulatory decision-making is explicit about its own boundaries: it does not cover AI used in drug discovery, and it de-scopes AI "used for operational efficiencies (e.g., internal workflows, resource allocation, drafting/writing a regulatory submission) that do not impact patient safety, drug quality, or the reliability of results from a nonclinical or clinical study" [8]. Where AI is used to help draft protocol text without touching the substance of eligibility criteria, endpoints, or study design elements that affect result reliability, it likely sits in that excluded category. Where an AI system is generating or checking the structured design elements themselves, such as eligibility criteria, endpoints, or schedule of activities logic that other systems will consume, the guidance's risk-based credibility framework is the more relevant standard to apply. This guidance remains in draft as of this writing, labeled "not for implementation" pending finalization, so its recommendations should be read as FDA's current thinking rather than a settled rule [8].
That framework sets out seven steps: define the question the model addresses, define its context of use, assess model risk based on influence and consequence, build a credibility assessment plan proportionate to that risk, execute it, document results and deviations, and determine whether the model is adequate for its intended use [8]. The guidance is explicit that where a human reviews AI output before it is acted on, evaluation should assess "the performance of the human-AI team, rather than just the model in isolation" [8]. It also calls for development and test data to be independent of each other and for ongoing monitoring of model performance for drift after deployment [8]. Applied to protocol work, this points toward AI systems that draft against a structured model and flag inconsistencies for a qualified reviewer, rather than systems that generate final study-design content without that review.
How Kitsa Fits Into This Problem
Kitsa's KScribe works on regulatory and clinical documents, including protocols, informed consent forms, investigator brochures, and clinical study reports, with an emphasis on keeping shared study details consistent across those documents rather than treating each one as a standalone drafting exercise. That general direction, defining a study's design elements in one place and carrying them consistently into the documents and systems that depend on them, is aligned with what USDM and ICH M11 are pushing the industry toward, even where the underlying tooling differs from a full USDM implementation.
What USDM Does Not Solve
Structured data does not replace governance or scientific judgment
Structured data changes what is possible, but it does not remove the work of getting there. Standing up a USDM-based workflow means agreeing on governance for who can change a study definition and when, adopting CDISC's controlled terminology consistently across teams that may have used their own naming conventions for years, and building or buying validated integrations between the repository and each downstream system, EDC, CTMS, randomization, and safety reporting among them [2][7]. Those integrations also need their own change control: when a study definition changes mid-trial, each connected system has to be re-validated against the new version rather than assumed to still be in sync, which is its own operational discipline, not a byproduct of adopting the model. None of that is optional: a structured model that only some systems can actually read delivers little of the benefit and adds a new maintenance burden on top of the old one.
USDM also does not, by itself, resolve scientific or medical judgment questions. A poorly designed eligibility criterion is still poorly designed whether it lives in a paragraph or a structured field, and a structured schedule of activities still requires clinical and operational review to confirm the visit windows and assessments make sense. The model changes how information is stored and shared. It does not substitute for the expertise that decides what that information should say.
Key Takeaways
- USDM represents a clinical trial's design, including objectives, population, arms, epochs, and the schedule of activities, as structured, machine-readable data rather than narrative text [2][3].
- TransCelerate estimates that roughly 90% of today's clinical trial protocols are created and managed as unstructured documents, which is the gap USDM and the Digital Data Flow initiative are meant to close [1].
- A Tufts CSDD study of 836 protocols found 57% had a substantial amendment, with median direct costs of $141,000 in Phase II and $535,000 in Phase III, and judged 45% of amendments avoidable [5].
- ICH M11 (CeSHarP) reached Step 5 adoption in the EU on 15 December 2025, and FDA finalized its own M11 guidance, template, and technical specification in May 2026; both set out the exchange standard itself, not USDM by name, and neither creates a legally binding requirement to use USDM specifically [2][6][10].
- Adopting USDM is not just a data-modeling exercise: it requires governance, consistent use of controlled terminology, and validated integrations with every downstream system before the benefits show up [2][7].
- USDM version 4.0, released June 3, 2025, and the open-source Study Definitions Repository built by TransCelerate, Microsoft, Accenture, and CDISC give sponsors concrete tooling to implement structured protocols today [1][2][7].
- FDA's draft AI guidance frames credible automation as risk-proportionate and human-supervised, which matters directly for AI tools drafting or checking protocol content built on structured data [8].
FAQ
Is USDM a replacement for the protocol document itself?
No. USDM captures study design as structured data that generates or informs the protocol document, the EDC build, and other downstream systems from a single source, but it includes provisions for the narrative content that still appears in the protocol document that regulators and investigators read [2][3].
Does USDM replace SDTM or other CDISC standards?
No. CDISC describes USDM as complementary to standards like CDASH and SDTM, sitting earlier in the process to provide the structured design inputs those standards then use, rather than replacing them [2].
Is adopting USDM mandatory?
No, not by name. ICH M11 introduces the CeSHarP template and specifies a technical format for electronic exchange. That specification reached Step 5 adoption in the EU in December 2025, and FDA finalized its own M11 guidance, template, and technical specification in the United States in May 2026. USDM was built in alignment with that specification and is one of the clearest paths to meeting it, but FDA guidance documents, including this one, explicitly state that they do not establish legally enforceable responsibilities and describe the agency's current recommendations rather than a binding mandate; a sponsor could in principle use an alternative approach that still satisfies the underlying regulatory requirements [2][6][10].
Is ICH M11 final in the United States now?
Yes. After circulating draft versions in December 2022 and June 2025, FDA published final M11 guidance, template, and technical specification documents in May 2026, following endorsement by ICH regulatory members in November 2025. That brings the U.S. in line with the EU's Step 5 adoption from December 2025, though both remain guidance rather than statute, so exact implementation expectations can still vary by region, by submission pathway, and by how mature a given sponsor's own systems and governance already are for consuming structured protocol data [6][10].
What is the Digital Data Flow initiative's relationship to USDM?
USDM is the core data standard produced by TransCelerate's Digital Data Flow initiative in partnership with CDISC. The initiative also includes reference tooling, such as the Study Definitions Repository built with Microsoft and Accenture, that demonstrates how USDM-formatted data can move between systems [1][2][7].
Does structured protocol data eliminate the cost of protocol amendments?
It does not eliminate the underlying scientific or regulatory work an amendment requires. What it targets is the manual reconciliation across systems, updating the EDC, CTMS, training materials, and consent language separately, that plausibly contributes to the cost and delay Tufts CSDD documented, though that research measured overall amendment prevalence, cost, and performance impact rather than isolating reconciliation as a specific cause [5].
Can AI tools safely draft protocol content from USDM data?
It depends on what the AI is touching. FDA's January 2025 draft guidance excludes AI used for operational efficiencies, such as drafting a submission, that do not affect patient safety, drug quality, or study-result reliability. Where AI is instead generating or checking structured design elements that other systems will rely on, FDA's guidance points toward a risk-based credibility assessment and a qualified human reviewing the output, rather than treating AI-generated content as final without review [8].
References
- [1]TransCelerate BioPharma Inc. "TransCelerate Advances Industrywide Shift To Digital Protocols." PR Newswire, October 8, 2025. https://www.prnewswire.com/news-releases/transcelerate-advances-industrywide-shift-to-digital-protocols-302577913.html
- [2]CDISC. "Digital Data Flow (DDF) for Clinical Trial Protocols." CDISC. https://www.cdisc.org/ddf
- [3]USDMPlay. "What is USDM? Complete Beginner Guide." USDMPlay. https://usdmplay.com/learn/what-is-usdm
- [4]TransCelerate BioPharma Inc. "Digital Data Flow Asset - Clinical Trial Process." TransCelerate BioPharma Inc. https://www.transceleratebiopharmainc.com/initiatives/digital-data-flow/
- [5]Getz KA, et al. "The Impact of Protocol Amendments on Clinical Trial Performance and Cost." Therapeutic Innovation & Regulatory Science, 2016. https://link.springer.com/article/10.1177/2168479016632271
- [6]European Medicines Agency. "ICH M11 Guideline on Clinical Study Protocol Template and Technical Specifications." EMA. https://www.ema.europa.eu/en/ich-m11-guideline-clinical-study-protocol-template-technical-specifications-scientific-guideline
- [7]TransCelerate BioPharma Inc. "How TransCelerate, Microsoft, Accenture, and CDISC are Working to Revolutionize Digital Data Flow." TransCelerate BioPharma Inc., 2022. https://www.transceleratebiopharmainc.com/how-transcelerate-microsoft-accenture-and-cdisc-are-working-to-revolutionize-digital-data-flow/
- [8]U.S. Food and Drug Administration. "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products." FDA, Draft Guidance, January 2025. https://www.fda.gov/media/184830/download
- [9]CDISC. "usdm" (official USDM reference repository, README). GitHub. https://github.com/cdisc-org/usdm/blob/main/README.md
- [10]U.S. Food and Drug Administration. "M11 Clinical Electronic Structured Harmonised Protocol (CeSHarP); International Council for Harmonisation; Guidance for Industry; Availability." Federal Register, May 22, 2026 (final guidance, superseding the June 6, 2025 draft technical specification and template notice). https://www.federalregister.gov/documents/2026/05/22/2026-10295/m11-clinical-electronic-structured-harmonised-protocol-cesharp-international-council-for