Skip to main content
Bureau of AI
Category: Model Lifecycle & MLOps

Substantial Modification

Also known as:
Simply put

A substantial modification is a change to a clinical trial made after authorities have already issued their decision on the trial. In the context of the EU framework, such a change is treated as substantial when it is likely to significantly affect the safety or rights of participants, or the reliability of the trial's data. When a sponsor wants to make such a change, they generally must submit it for review rather than implementing it freely.

Formal definition

Under the EU Clinical Trials Regulation (Regulation (EU) 536/2014), as reflected in national competent authority and EMA guidance, a Substantial Modification (SM) is defined as any change to any aspect of a clinical trial made after notification of a decision on a previously submitted application. A modification is characterized as substantial where it is likely to have a significant impact on the safety or rights of the trial subjects or on the reliability and robustness of the data generated. Procedurally, an SM is submitted as an application—typically through the Clinical Trials Information System (CTIS)—accompanied by documentation such as a cover letter and a modification description summarizing the changes. Note: the specific criteria distinguishing substantial from non-substantial modifications, and the exact submission requirements, are jurisdiction- and framework-dependent; the evidence here reflects the EU CTR context and should not be assumed to apply to other regulatory regimes.

Why it matters

The concept of a Substantial Modification establishes a control point in clinical trial oversight: once a competent authority has issued its decision on a trial application, the sponsor cannot unilaterally change core aspects of the study that could affect participant safety, participant rights, or the reliability of the trial data. In the EU framework under Regulation (EU) 536/2014, changes meeting the substantial threshold must generally be submitted for review rather than implemented at the sponsor's discretion. This preserves the integrity of the authorization on which the original approval was based and ensures that regulators retain visibility into material changes over the life of a trial.

The practical significance lies in the threshold judgment. A change is treated as substantial when it is likely to have a significant impact on subject safety or rights, or on the reliability and robustness of the data generated. Misclassifying a substantial change as non-substantial can mean implementing an unauthorized change; conversely, over-classifying routine changes can create unnecessary submission burden and delay. Because the criteria distinguishing substantial from non-substantial modifications are jurisdiction- and framework-dependent, sponsors operating across regulatory regimes cannot assume that the EU CTR characterization transfers to other jurisdictions.

Who it's relevant to

Clinical Trial Sponsors
Sponsors bear responsibility for determining whether an intended change is substantial and, if so, for submitting it for review through the appropriate channel (typically CTIS in the EU CTR context) before implementation. They must prepare the accompanying documentation, such as a cover letter and a modification description, and cannot assume the same classification rules apply across different regulatory regimes.
Regulatory Affairs and Compliance Professionals
These professionals apply the substantial-versus-non-substantial threshold in practice, deciding which changes require formal submission and which do not. Because the distinguishing criteria are jurisdiction-dependent, they must track the specific requirements of the applicable framework and national competent authority guidance rather than relying on a single universal standard.
Competent Authorities and Reviewers
Authorities evaluate submitted Substantial Modifications to confirm that changes made after their original decision do not adversely affect participant safety, participant rights, or data reliability. EMA training materials, for example, provide step-by-step guidance for evaluating a substantial modification application submitted through CTIS.
Investigators and Trial Site Staff
Because a Substantial Modification affects aspects of an ongoing trial after authorization, investigators and site staff need to understand that certain changes cannot be freely implemented and are subject to review, ensuring that trial conduct at the site remains aligned with the authorized and modified protocol.

Inside SM

Change to Intended Purpose
A modification that alters what the system is designed to do, or the conditions or context in which it is intended to be used, is commonly treated as a factor in assessing whether a change is substantial. In several AI-governance frameworks, an unforeseen change to intended purpose can be a decisive element.
Impact on Compliance and Risk Profile
A change that affects a system's continued conformity with applicable requirements, or that materially alters its risk profile, is typically a core component of the substantial-modification concept. The threshold question is generally whether the change is significant enough to warrant renewed assessment rather than treated as routine maintenance.
Distinction from Pre-Planned Changes
Changes that were anticipated and specified in advance—for example, within a predetermined change or learning plan documented at the time of initial assessment—are frequently excluded from what counts as a substantial modification, whereas unplanned or out-of-scope changes are more likely to qualify.
Trigger for Re-Assessment or Re-Evaluation
A defining functional feature is that a substantial modification, once identified, typically triggers some form of renewed review, revalidation, or re-conformity activity, rather than being absorbed into ongoing monitoring alone.

Common questions

Answers to the questions practitioners most commonly ask about SM.

Does 'substantial modification' in AI regulation mean the same thing as it does in clinical trials or medical device contexts?
No. Although the phrase appears in several regulatory domains, the term should not be treated as interchangeable across them. In AI governance discussions, particularly those referencing the EU AI Act, 'substantial modification' is used in the context of changes to an AI system that may affect its compliance status. Clinical-trial or medical-device usages of similar phrasing belong to different regulatory regimes with their own definitions, thresholds, and issuing authorities. When applying the concept, always confirm which framework and jurisdiction you are working within, because the operative definition and its legal consequences differ by regime.
Does any change to a deployed model count as a substantial modification?
Not necessarily. A common error is treating every retraining, parameter update, or configuration change as a substantial modification. In many frameworks the concept is intended to capture changes significant enough to alter the system's risk profile, intended purpose, or compliance posture, rather than routine maintenance. The precise threshold depends on the applicable framework and how it defines the term, and some routine updates may be explicitly excluded. Because definitions and boundaries are still evolving and can be contested, organizations should not assume a single universal test applies.
How can an organization determine whether a planned change crosses the threshold into a substantial modification?
Typically organizations establish a documented assessment process that evaluates whether a change affects the system's intended purpose, its risk classification, or the assumptions underlying prior compliance and validation work. Factors commonly considered include changes to intended use, materially different data inputs, altered performance boundaries, or changes affecting affected populations. The specific criteria should be mapped to the applicable framework's definition rather than a generic checklist, and the assessment should be recorded so the rationale is auditable. Where the threshold is ambiguous, seeking legal or compliance input is prudent.
Which lines of defense are typically involved in evaluating a potential substantial modification?
In organizations using a three-lines model, the first line (model developers or owners) usually identifies and proposes the change and performs an initial impact assessment. The second line (independent risk or compliance functions) typically reviews whether the change alters the risk profile or triggers regulatory obligations. The third line (internal audit) may later assess whether the process for identifying and handling such modifications operated effectively. These roles should remain distinct so that those making changes are not the sole judges of their significance.
What documentation should accompany a determination about a substantial modification?
Documentation commonly includes a description of the change, the assessment of its impact on intended purpose and risk classification, the criteria applied, the individuals or functions involved, and the conclusion reached. If a change is judged not to be substantial, recording the reasoning is often as important as documenting one that is, because it demonstrates that the determination was deliberate. Retaining this record supports auditability and may be relevant if the compliance status of the system is later questioned. Specific documentation requirements depend on the governing framework.
How does a substantial modification relate to model revalidation activities?
A substantial modification often serves as a trigger for revisiting validation, since changes significant enough to alter a system's behavior or risk profile can invalidate assumptions from prior validation work. However, the two concepts are distinct: revalidation is an activity assessing whether a model remains fit for its intended use, while a substantial modification is a characterization of a change that may prompt such activity. Not every revalidation follows a substantial modification, and organizations may revalidate on schedule or in response to performance monitoring independent of any modification determination.

Common misconceptions

Any update or retraining of a model automatically constitutes a substantial modification.
In many frameworks, routine updates, planned retraining, or changes anticipated within a documented change plan are not treated as substantial modifications. Whether a change qualifies typically depends on its effect on intended purpose, compliance, or risk profile, not on the mere fact that a change occurred.
"Substantial modification" carries a single, universally fixed definition across all regulatory and standards contexts.
The term's precise meaning and thresholds are context- and instrument-specific and can differ across jurisdictions and sectors. Practitioners should rely on the definition applicable to their specific framework rather than assuming interchangeability.
Determining a substantial modification is purely a technical model-performance question.
The determination is generally a governance and compliance judgment about intended purpose, conformity, and risk—not solely a measurement of model performance change. Performance shifts may be one input, but the substantiality assessment is broader than performance metrics alone.

Best practices

Establish and document, in advance, the criteria your organization uses to distinguish a substantial modification from routine maintenance, tied to the specific framework(s) that apply to your system.
Where the applicable framework permits, define anticipated changes within a documented change or update plan at the time of initial assessment, so that pre-planned changes are clearly distinguishable from unplanned ones.
Assess each proposed change against its effect on intended purpose, continued compliance, and risk profile, and record the rationale for classifying it as substantial or non-substantial.
Route classification decisions through an appropriate governance and oversight process rather than leaving them to individual developers, so that determinations are consistent and auditable.
Maintain a change log and version history that captures what changed, when, why, and how it was classified, to support internal review and any external examination.
Revisit classification criteria periodically, since the treatment of substantial modification is evolving and may differ by jurisdiction and sector; do not assume a fixed threshold applies indefinitely.