What ICH M11 CeSHarP Changes for AI-Assisted Protocol Authoring

Introduction

On November 19, 2025, the International Council for Harmonisation finalized M11, the Clinical Electronic Structured Harmonised Protocol guideline, at Step 4 ^[1]. The Committee for Medicinal Products for Human Use adopted it at Step 5 on December 11, 2025, with a coming-into-effect date set for June 11, 2026 ^[3]. The FDA published its own final guidance implementing M11 on May 22, 2026 ^[4]. Three milestones across ICH, the EMA's Committee for Medicinal Products for Human Use, and the FDA, one underlying change: a clinical trial protocol now has an internationally agreed structure for exchanging its content electronically, not just an internationally agreed outline.

That distinction is worth stating precisely, because it is easy to overstate. M11 does not hand the industry a finished data model. Its technical specification is designed to "enable development of a data model and an open, non-proprietary exchange message standard" ^[1], which is different from being that model. CDISC's Unified Study Definitions Model, built separately under the Digital Data Flow initiative, is the logical data model that actually does that job, aligned to M11's structure rather than supplied by it ^[7]. Protocol templates have existed for years, TransCelerate's Common Protocol Template among them, and technology-enabled precursors, TransCelerate's electronic Common Protocol Template (eCPT) initiative in particular, already pursued machine-readable protocol content and downstream reuse before M11 reached Step 4 ^[6]. What M11 adds, and what neither those static templates nor the pre-M11 eCPT work had, is a single technical specification developed through ICH and adopted for regional implementation: one set of data elements, identifiers, and conformance rules, rather than a company-level or industry-consortium standard, even though each region still adopts it on its own timeline. That combination, M11's structure plus a conformant data model, is what lands at the moment the industry is testing whether AI can reliably help write regulatory documents. The two developments are related but not identical, and the rest of this article tries to keep that boundary visible rather than blur it.

Why This Topic Matters in Clinical Trials

Protocol amendments are frequent, and Tufts Center for the Study of Drug Development data collected over three separate studies show amendments becoming more common. An early analysis of protocols approved between 2006 and 2008, published in 2011, reported an average amendment implementation cost of $453,932 and, based on timing data from 59 amendments, a median implementation cycle of 65 days ^[9]. A follow-up peer-reviewed study, analyzing 836 protocols and deriving cost estimates from 52 of them, found that 57% of protocols required at least one substantial amendment and reported a median direct implementation cost of $141,000 for a Phase II amendment and $535,000 for a Phase III amendment, with about 45% of amendments judged avoidable ^[13].

A more recent 2024 analysis found the share of protocols requiring at least one amendment had risen further, to 76%, with a mean of 3.3 amendments per protocol; that study did not update the cost figures, but it found that 77% of amendments were considered unavoidable, with regulatory agency requests and changes to study strategy the most common reasons cited, and that sites spend a mean of 215 days operating under different versions of the same protocol between the first and last required ethics committee approval of an amendment ^[10]. Read together, the three studies are consistent on one point and silent on another: amendment prevalence and amendment counts have both risen, from roughly 57% of protocols amended in the 2016 dataset to 76% in the 2024 one. On cost, the 2016 study's phase-specific medians remain the most defensible peer-reviewed benchmark, and the 2024 study did not update them, so the data does not support a claim that per-amendment costs have risen or fallen since 2016; it only supports that amendments themselves are becoming more common.

Most of that burden traces back to why amendments happen in the first place, regulatory feedback and evolving study strategy, rather than to document format alone. Where format does matter is in how much extra rework an amendment generates once it is needed. A protocol authored as free text in Word, then re-keyed into an EDC build, a statistical analysis plan, and a trial registry entry, creates separate opportunities for the same eligibility criterion or endpoint definition to drift out of alignment during that re-keying. ICH M11's technical specification addresses part of that problem: it defines a mix of discrete, code-list-bound data elements and narrative text blocks, rather than treating the whole protocol as undifferentiated prose, and structured elements are associated with defined identifiers, data types, controlled terminology, cardinalities, or business rules where the specification requires them ^[2][12]. Whether a protocol built this way actually populates downstream systems automatically, instead of being copied by hand into each one, depends on whether an organization has built and validated the data model and mappings that connect the structured protocol to those systems. M11 makes that outcome possible. It does not make it automatic, and it does not change why amendments happen in the first place.

Current Evidence and Research Landscape

ICH M11 is not a single document. It comprises three parts released together: the guideline itself, which sets out the design principles; a protocol template specifying required and optional headings, structure, and common text; and a technical specification defining the data elements, their cardinality, and their conformance rules for electronic exchange ^[1][2]. In June 2025, FDA opened a comment period specifically on the revised technical specification. It published an updated template alongside it, but only as a reference document to aid reviewers, and said explicitly that it was not seeking comment on the template itself ^[5]. Step 4 finalization of all three parts followed in November 2025 ^[1].

Regional adoption has not moved in lockstep. (Regional implementation details below are current as of September 2026 and may change; sponsors should verify current status directly with each agency before finalizing submission plans.) CHMP adopted the EMA's Step 5 package, referenced EMA/CHMP/ICH/778799/2022, on December 11, 2025, with a coming-into-effect date of June 11, 2026 ^[3]. The FDA's guidance, covering the same three documents, followed in May 2026 and is explicitly non-binding: sponsors may use alternative approaches provided they still meet applicable regulatory requirements ^[4]. The gap between EMA's effective date and FDA's guidance publication is a preview of what sponsors running global programs should expect more broadly. Other ICH regions have not necessarily moved as far. As of September 2026, Japan's PMDA lists M11 at Step 4 on its official ICH implementation page, with no published Step 5 adoption date ^[11]; sponsors should check that page directly for updates rather than assume Japan's timeline mirrors the EU's or FDA's. A multi-region trial may need to reconcile more than one regional implementation schedule before a single M11-conformant protocol can be filed everywhere at once.

M11's interoperability ambitions depend on a separate layer of standards to be realized in practice. CDISC's Unified Study Definitions Model, developed under its Digital Data Flow initiative with TransCelerate, was built with explicit alignment to the draft M11 protocol template as a design goal across its development phases ^[7]. USDM represents a clinical study as a logical data model, complete with controlled terminology and a REST API specification, and CDISC's implementation guidance documents how a structured study definition can generate CDISC Trial Design domains directly ^[7]. Generating other downstream artifacts, an EDC casebook or a registry record, from that same definition is a documented use case rather than a guaranteed outcome of adopting M11: it depends on whether an organization has actually built the mapping between its study definition and each downstream system. TransCelerate has separately released CPT V11, a version of its Common Protocol Template realigned to the finalized M11 template and technical specification, together with a mapping reference intended to help organizations migrate legacy CPT-based protocols rather than rebuild them from scratch ^[6].

Operational Impact for Sponsors, CROs, and Sites

The practical shift for regulatory and clinical operations teams is that a structured protocol section is no longer just text under a heading. Under M11's technical specification, a discrete data element carries a defined identifier, an expected format, and conformance rules governing that element itself, such as its data type and its permitted values from a controlled code list ^[12]. On its own, the specification does not enforce that an eligibility criterion stated in the protocol matches the wording in the informed consent form or the statistical analysis plan; that kind of cross-document consistency check is application logic that an authoring platform has to build on top of M11, not a guarantee the specification provides. Where a platform has built that logic, a mismatch that might otherwise surface only during downstream build, statistical programming, or pre-lock data review and cleaning can be caught during authoring instead. Where it has not, M11 conformance can assess required elements, data types, cardinalities, business rules, and controlled terminology where those constraints apply, but it says nothing about whether the underlying eligibility criterion is scientifically sound, operationally feasible at a given site, or medically appropriate. Either way, that judgment still requires human review.

That distinction also shapes what protocol amendments look like once a study is structured this way. Where an organization has built a working data model and validated the mappings between it and each downstream system, an amendment can be represented as a defined change set, and it becomes possible to trace computationally which systems, an EDC build, a randomization schedule, a trial registry record, are affected by a specific change. Where that implementation work has not been done, an M11-formatted protocol is still, functionally, a document a human has to read and re-key into each system, just one with cleaner headings. This does not eliminate amendments either way; the majority stem from causes like regulatory agency requests and evolving study strategy that structured authoring cannot prevent ^[10]. But a validated structured pipeline can expose certain internal inconsistencies, a mismatched eligibility criterion, an endpoint definition out of sync with the SAP, earlier, and it can reduce the effort and propagation risk involved in implementing a change once one is approved. It does not, on its own, prevent the design flaws, feasibility problems, or regulatory and strategic decisions that actually cause most amendments. These are benefits that follow logically from the architecture, and sponsors should treat them as expected outcomes to validate against their own data rather than results already demonstrated at scale.

Sites still face the coordination problem described above in the meantime: the 215-day average gap between a protocol amendment's first and last required ethics committee approval predates widespread M11 adoption ^[10], and closing it depends on site-level and ethics-committee review workflows as much as on protocol format. That is a strong argument for why an explicit, machine-readable version identifier on the protocol matters, but the identifier alone will not shorten a regulatory or ethics review cycle.

Regulatory and Documentation Considerations

Sponsors should read M11's status carefully in each region rather than assume it is treated the same way everywhere. FDA's May 2026 guidance explicitly frames M11 as a harmonized standard, not a mandate, and states that sponsors may use alternative approaches provided they still satisfy applicable regulatory requirements ^[4]. The EMA's Step 5 documents are framed differently: they describe the template and technical specification as intended to ensure protocols are delivered in a harmonised, regulator-acceptable exchange format for interventional trials, without the same explicit alternative-approaches language FDA uses ^[3]. Neither framing amounts to an immediate universal format cutover, but sponsors should confirm the specific submission expectations for the relevant procedure and agency rather than assuming FDA's optionality applies identically in the EU. FDA explicitly permits alternative approaches that satisfy applicable requirements, so continued use of legacy formats for US submissions is directly supported by the guidance's own text. For EU procedures, sponsors should confirm current submission-format expectations with the relevant authority rather than infer them from FDA's position. Either way, a sponsor that falls back on a legacy submission format may forfeit some of the interoperability and automated-exchange benefits the standard was built to provide, though a sponsor that maintains a structured source internally while still submitting a legacy-format artifact keeps more of those benefits than the format of the submission alone would suggest.

The technical specification itself is worth distinguishing from the software that will implement it. M11 defines structure: the data elements, their identifiers, their conformance rules. It does not mandate a specific authoring platform or file format, and CDISC's USDM is one of the standards designed to operationalize that structure rather than a required component of M11 compliance ^[1][7]. Regulatory teams evaluating authoring tools should confirm that a given platform actually validates against the published technical specification, rather than simply reformatting a document to look like the new template while leaving the underlying content unstructured.

For teams building an implementation plan, that distinction points to a practical checklist rather than a single yes-or-no compliance question:

• Whether the authoring platform maps content to the M11 technical specification's actual data elements, not just to the template's headings and section order
• Which CDISC USDM version, if any, the platform exports to or imports from, since USDM carries its own version history independent of M11's
• How controlled terminology and code lists are governed and kept current as ICH and CDISC update them
• Whether validation checks structural conformance only, or also flags cross-document inconsistencies that still require a human judgment call
• How an amendment is represented as a change set, and which downstream systems that change set is actually configured to reach
• Whether mappings to each downstream system, EDC, randomization, registry, have been built and validated, rather than assumed
• What audit trail and human sign-off exists before a structured element, or an AI-drafted value, is treated as final
• The specific regional implementation timeline that applies, since conformance with FDA's May 2026 guidance does not automatically align with every other region's Step 5 schedule

None of these questions has a universal answer. They depend on what a specific authoring platform has actually built, not on what ICH M11 alone requires.

AI and Automation Perspective

Structured protocol data can provide a more traceable and constrained input for AI drafting tools than free text, and a more checkable one. A large language model asked to draft protocol language from an unstructured brief has to infer and reconcile an eligibility criterion against whatever else the document says about it in narrative form, with no fixed vocabulary to anchor that comparison. Where an M11 element has a defined valid-value list, surrounding tooling can constrain a model to that element's controlled terminology and validate the resulting structure against the technical specification's conformance rules before a human ever reviews it; narrative text elements, which the specification also includes, do not offer that same constraint ^[12]. That does not make the draft correct, and it says nothing about whether the eligibility criterion the model wrote is scientifically sound or operationally realistic. It only means a certain class of structural error, a wrong field type, an undefined term, a missing required element, becomes more reliably machine-detectable than it would be in unconstrained free-text drafting, where the same check depends on a system correctly parsing prose rather than reading a typed field.

Whether a given use case falls under FDA's January 2025 draft guidance on AI in drug and biological product regulatory decision-making depends on the model's specific context of use and how its output feeds into a regulatory decision, not simply on whether protocols are involved. The guidance is explicit that it does not cover AI used for operational efficiencies, including drafting or writing a regulatory submission, when that use does not affect patient safety, drug quality, or the reliability of results from a nonclinical or clinical study ^[8]. An AI tool used only for formatting a protocol into the M11 template would generally sit outside that scope, provided its use does not affect patient safety, drug quality, or the reliability of study results. A model used to generate or check the substance of eligibility criteria, dosing rules, or safety monitoring language may fall inside it, where that output supports a regulatory decision about safety, effectiveness, or quality. In that case, the draft guidance recommends applying its seven-step, risk-based credibility framework: defining the model's context of use and building an evidence case scaled to how much harm a wrong output could cause ^[8]. As draft, nonbinding guidance, it describes a recommended approach rather than a currently enforceable legal requirement, though FDA's expectations at final adoption may differ from the draft text. The guidance itself acknowledges the line is not always obvious, and it encourages sponsors to engage with FDA early when they are uncertain whether a particular AI use falls within its scope ^[8]. Sponsors are responsible for developing and documenting the credibility evidence, but the framework envisions interactive feedback from FDA on the model's risk assessment and the adequacy of the credibility assessment plan, and either the sponsor or FDA can determine that a model's credibility has not been sufficiently established for its context of use ^[8]. Separately, in the specific context of postmarketing safety surveillance, the guidance states that sponsors remain responsible for regulatory compliance regardless of the technology used ^[8], a narrower point than a blanket statement covering every document-authoring use. Structured standards like M11 make part of the relevant credibility evidence easier to assemble, since a model's structural output can be tested against defined conformance rules rather than judged on prose quality alone. On their own, structured protocol fields do not satisfy the credibility assessment, documentation, sponsor oversight, or other controls that may apply once an AI output is used to support a regulatory decision, and the specific controls that apply depend on the model's risk level and context of use rather than following a single fixed checklist.

How Kitsa Fits Into This Problem

According to Kitsa's own product documentation, KScribe generates protocols, informed consent forms, statistical analysis plans, DSURs, and other regulatory and clinical documents from a structured study schema the company describes as conformant to CDISC USDM v4.0 and the ICH M11 technical specification, rather than from independent drafts of each document. Per that documentation, when a field changes, a secondary endpoint definition, for instance, the change is written back to the shared schema rather than to a single paragraph, and every document that consumes the field is flagged for the corresponding update and routed for human approval before it is treated as final. This description reflects Kitsa's stated product design rather than an independently audited conformance report. As ICH M11 and CDISC's USDM push the industry toward protocols authored as structured data rather than free text, an approach like this becomes increasingly valuable: it can reduce the manual reconciliation work involved when multiple documents consume the same approved definitions, though a shared schema is one implementation approach among several, not the only way to keep downstream documents consistent with a protocol.

Key Takeaways

• ICH M11 reached Step 4 finalization on November 19, 2025; CHMP adopted it at Step 5 on December 11, 2025 with a coming-into-effect date of June 11, 2026; and FDA published final implementing guidance on May 22, 2026, so a globally reusable protocol representation still has to reconcile more than one regional timeline ^[1][3][4].
• M11's technical specification is designed to enable development of a data model and exchange standard, not to be one itself; CDISC's USDM, built with explicit alignment to M11, is the logical data model that operationalizes that structure ^[1][7].
• FDA explicitly treats M11 as non-binding guidance and states that sponsors may use alternative approaches that satisfy applicable requirements; EMA's Step 5 package does not use the same wording, so sponsors should confirm EU procedural expectations rather than assume identical optionality ^[3][4].
• A 2024 Tufts CSDD study (Getz et al., analyzing 2022 data) found that 76% of protocols now require at least one amendment, with a mean of 3.3 amendments each, and that 77% of amendments are considered unavoidable, driven mainly by regulatory agency requests and evolving study strategy rather than document format ^[10].
• The same 2024 study found sites operate under different protocol versions for an average of 215 days between the first and last required ethics committee approval of an amendment, a gap driven by review workflows as much as by protocol format ^[10]. The later peer-reviewed cost benchmark cited here is a 2016 Tufts CSDD study: a median $141,000 per Phase II substantial amendment and $535,000 per Phase III amendment, based on 52 protocols; a 2011 analysis had put the overall average at $453,932, but the two figures use different scopes and statistics and should not be read as a trend ^[9][13].
• FDA's draft AI guidance explicitly excludes AI used purely for drafting efficiency; it may apply where a model's output supports a regulatory decision about patient safety, drug quality, or the reliability of study results, which can cover substantive protocol content but not routine template formatting, and the guidance encourages early FDA engagement when that line is unclear ^[8].
• Sponsors evaluating protocol authoring tools should confirm actual conformance validation against the published M11 technical specification and a named USDM version, not just visual alignment with the new template headings.

FAQ

References

1. [1]International Council for Harmonisation. "M11: Clinical Electronic Structured Harmonised Protocol (CeSHarP), Final Guideline." Step 4, November 19, 2025. https://database.ich.org/sites/default/files/ICH_Step4_M11_Final_Guideline_2025_1119.pdf
2. [2]International Council for Harmonisation. "ICH M11 Explainer." May 2026. https://database.ich.org/sites/default/files/ICH%20M11%20Explainer_May2026.pdf
3. [3]European Medicines Agency / Committee for Medicinal Products for Human Use. "ICH M11 Guideline, Clinical Electronic Structured Harmonised Protocol (CeSHarP)." Step 5, EMA/CHMP/ICH/778799/2022. Adopted December 11, 2025; effective June 11, 2026. https://www.ema.europa.eu/en/documents/regulatory-procedural-guideline/ich-m11-guideline-clinical-electronic-structured-harmonised-protocol-cesharp_en.pdf
4. [4]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. https://www.federalregister.gov/documents/2026/05/22/2026-10295/m11-clinical-electronic-structured-harmonised-protocol-cesharp-international-council-for
5. [5]U.S. Food and Drug Administration. "M11 Technical Specification: Clinical Electronic Structured Harmonised Protocol; Draft Technical Specification and Template; Availability." Federal Register, June 6, 2025. https://www.federalregister.gov/documents/2025/06/06/2025-10359/m11-technical-specification-clinical-electronic-structured-harmonised-protocol-international-council
6. [6]TransCelerate BioPharma Inc. "Clinical Content & Reuse Assets: Common Protocol Template (CPT) V11, Aligned to ICH M11." Accessed September 2026. https://www.transceleratebiopharmainc.com/assets/clinical-content-reuse-solutions/
7. [7]CDISC. "Digital Data Flow." Accessed September 2026. https://www.cdisc.org/ddf
8. [8]U.S. Food and Drug Administration. "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products; Draft Guidance for Industry." January 2025. https://www.fda.gov/media/184830/download
9. [9]Tufts Center for the Study of Drug Development. Analysis of 3,410 protocols approved 2006-2008 across 17 pharmaceutical and biotechnology companies; the 65-day median cycle-time estimate is based on timing data from 59 amendments. As reported in "Protocol Amendments: A Costly Solution." Applied Clinical Trials, May 2011. https://www.appliedclinicaltrialsonline.com/view/protocol-amendments-costly-solution
10. [10]Getz KA, Smith Z, Botto E, Murphy E, Dauchy A. "New Benchmarks on Protocol Amendment Practices, Trends and their Impact on Clinical Trial Performance." Therapeutic Innovation & Regulatory Science. 2024;58(3):539-548. Data collected 2022. DOI: 10.1007/s43441-024-00622-9. https://link.springer.com/article/10.1007/s43441-024-00622-9
11. [11]Pharmaceuticals and Medical Devices Agency (Japan). "ICH M11: Clinical Electronic Structured Harmonised Protocol (CeSHarP)." Status page, accessed September 2026, listing M11 at Step 4. https://www.pmda.go.jp/int-activities/int-harmony/ich/0095.html
12. [12]International Council for Harmonisation / U.S. Food and Drug Administration. "M11 Technical Specification: Clinical Electronic Structured Harmonised Protocol." Final, Step 4, November 2025, hosted by FDA. https://www.fda.gov/media/192648/download
13. [13]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 & Regulatory Science. 2016;50(4):436-441. https://pubmed.ncbi.nlm.nih.gov/30227022/