ederated AI could help oncology draw insights from data that cannot easily be brought together. Realizing that potential, however, will require more than connecting the data.

That fragmentation matters because many of the questions oncology now needs to answer cannot be addressed by a single data set or institution. Rare cancers and increasingly specific molecular subtypes may leave individual centers with too few patients. Relevant clinical, genomic, imaging, and other data may sit in different places. As cancer research becomes more specialized, the need to learn across those boundaries is growing.

“For years, innovation has not been constrained by lack of scientific insight, but rather by the F word: fragmentation,” said Alberto Grignolo, Editor-in-Chief of DIA Global Forum, opening a multidisciplinary discussion at the 2026 DIA Global Annual Meeting.

Federated artificial intelligence (AI) could help bridge some of those divides. Instead of bringing patient-level data together in one place, federated approaches bring the model to the data. Participating institutions train or refine the model locally and share model updates rather than the underlying records.

This can make it possible to learn across data sets that might otherwise remain separate. But it does not solve the harder questions around data quality, interoperability, governance, trust, incentives, and clinical relevance.

Ultimately, the value of federated AI will depend not on whether institutions can connect their data, but on whether doing so produces knowledge that improves research and patient care.

Oncology’s Fragmentation Problem

The problem is particularly acute in oncology. As Sean Khozin, CEO of the CEO Roundtable on Cancer and Project Data Sphere, noted, cancer biology has become increasingly complex. At the same time, advances in the field have driven greater specialization in medicine and research. The result is a paradox: oncology is generating increasingly sophisticated information while distributing that information across increasingly specialized domains and institutional silos.

The opportunity lies in learning across those data sets without necessarily pooling them. That is the basic premise of federated AI.

In a simplified federated system, a model is distributed to participating institutions and trained locally on their data. The underlying patient records remain where they are. What returns to the coordinating system are model weights or parameters, which can be combined to create a model informed by all participating data sets.

For organizations reluctant to exchange sensitive or proprietary information, including competitors, this changes what must be shared. The collaboration can focus less on transferring raw records and more on what can be learned from them.

The European MELLODDY project showed that such a model can support collaboration even among competing pharmaceutical companies. Participating companies used federated approaches for drug discovery without exposing their individual data sets to one another.

Federation, in other words, changes where the analysis happens. But it does not make the underlying data easier to interpret.

Federation Does Not Fix the Data

A federated network can connect data sets without making them consistent, complete, or fit-for-purpose. Even organizations that adopt the same data standard may implement it differently. Source-to-target mappings can vary. Terminology and local conventions can differ. Problems may not become apparent until a project is well underway.

Mohammad T. Manzari, Chief Architect for Federated Systems at Deloitte, described projects in which participating institutions had all mapped their data to the same schema, only to discover later that they had done so in different ways. His advice was pragmatic: expect those problems and design projects to adapt to them rather than assuming every risk can be eliminated in advance.

A larger network does not correct that mismatch. For federated evidence to be useful, the underlying data must still be credible, interpretable, and appropriate for the question being asked.

The Real Test: Does It Improve Clinical Decision-Making?

The question of fit-for-purpose matters because federated AI is a means, not an endpoint. For Joseph Kannarkat, a physician at Johns Hopkins University School of Medicine, clinical relevance ultimately comes down to whether information can help clinicians make decisions—particularly in the gray areas where strong evidence to choose one therapy or test over another may be limited or unavailable.

Oncologists regularly face clinical questions for which randomized trial evidence is incomplete or unavailable. Rare cancers and uncommon molecular subtypes make the problem harder because individual institutions may see too few patients to generate robust evidence.

Federation could help researchers examine those distributed experiences together. Data from multiple centers might support hypothesis generation, identify patterns that are invisible in smaller data sets, or provide enough cases to develop models for rare patient populations. A hospital, for example, may see relatively few patients with glioblastoma. A network connecting imaging and clinical data across institutions could create a much larger basis for model development.

But finding a pattern is not the same as establishing evidence. AI-generated hypotheses still require validation. Models need to be shown to perform reliably in the population and setting for which they are intended, and their outputs need to be relevant to a real research or clinical decision.

The key question is therefore not simply, Can these data sets be federated? It’s: What problem can we solve by federating them?

That distinction helps separate useful applications from technology looking for a use case.

Trust, Reciprocity, and the Patient Perspective

Keeping patient-level records within the institutions that generate them can reduce some concerns about privacy and data ownership. But it does not remove the need for trust.

Organizations still need clear rules about who can participate, what analyses are permitted, how models and outputs may be used, and who is responsible for oversight.

Technology can help make those rules enforceable. Federated environments can record how data are accessed and analyzed, and machine-readable policies can define permitted uses. But the technology sits within a broader governance framework; it does not replace one. The same is true for patient trust.

Stacy Hurt, Chief Patient Officer at Parexel and a cancer survivor and caregiver, described federated AI in terms that to many are familiar from efforts to decentralize clinical research: if the goal is to meet patients where they are, federated approaches can similarly “meet data where it is.”

Patients may be willing to contribute their data to research. That does not mean they are not concerned about how the data are used. They want to know where the information goes, what it’s used for, and what comes out of that use. Those are not peripheral communications questions; they are part of building a system people are willing to participate in.

Participation also needs to offer some return of value, which does not have to be financial. It may mean providing support that makes research participation possible. It may mean returning information or explaining what researchers learned. “If I give you my data, what did you do with it?” Hurt asked.

For patients who have exhausted standard treatment options or entered a clinical trial as a last resort, that question can have particular weight. Even if a treatment does not help them, they may wish what is learned from their experience to help someone who comes after them.

The discussion also raised a practical example: data generated when a patient is screened for a clinical trial but doesn’t enroll. Could that information be returned to the patient and, with permission, used in a sponsor-agnostic federated system, potentially to identify another trial or to support future research?

Questions like this shift the conversation from extracting data to creating a more reciprocal research ecosystem.

Federation Can Scale Inequity, Too

Federating more data does not necessarily make a data set more representative. “Our data sets are only as good as what we feed into them,” Hurt said.

If a federated network primarily connects large academic medical centers, it may reproduce the same gaps already present in biomedical research. Patients treated in community settings, rural areas, federally qualified health centers, and other less research-intensive settings may remain underrepresented.

The goal is not simply to connect more institutions but to connect the data needed to answer the question, and to ensure that those data reflect the people the resulting evidence is meant to serve.

Start With the Problem

Federated AI offers a practical way to approach one of oncology’s persistent challenges: learning across information that cannot easily be brought together. But federation should not become the goal. The goal is better evidence, better decisions, and ultimately better outcomes for patients.

Getting there requires more than a technical network. Manzari described federated AI as a “sociotechnical” system in which technology, people, and governance are interdependent. That framing captures much of what emerged from the discussion:

The data must be suitable for the question, institutions need reasons to participate, governance and legal agreements must establish workable rules, models need rigorous validation, patients need transparency and a meaningful role, and the network needs to include the populations it is intended to serve.

The incentives will also differ between participants. A cancer center may contribute high-quality data but may need computational expertise. Another partner may contribute to infrastructure or analytical capabilities. Industry competitors may be willing to collaborate on questions they cannot answer alone. Patients may contribute information because they want their experience to benefit themselves or others, but they expect to know how that information is being used.

As Manzari noted, sustainable collaborations therefore need incentive models built around the specific participants and use case. So the starting question should not be “Where can we use federated AI?” It should be:

What important problem in oncology can we solve together that none of us can solve alone?

Once that problem is clear, federation may be part of the solution. Moving oncology from fragmentation to federation is not primarily about moving data, but about making it possible to learn from them together.

linical trials have grown increasingly sophisticated and increasingly difficult for participants and investigative sites to support. Protocol complexity, frequent visits, nonessential procedures, and growing demands for data can create substantial burdens that limit who is able and willing to participate. At the same time, advances in basic science, precision medicine, and artificial intelligence are accelerating therapeutic discovery faster than traditional evidence-generation systems can efficiently evaluate them.

A 2024 report summarizing Tufts CSDD findings provides supportive context, noting that the burden from questionnaires, lab and blood work, procedure frequency, and site visit requirements has increased and is linked to participant dropout.

The challenge is to redesign evidence generation so that research is easier to access, more representative, and better integrated into patients’ care, without compromising the quality of evidence needed for regulatory decision-making.

Improving enrollment, retention, representativeness, and the overall patient experience requires redesigning how evidence is generated, embedding research into clinical care where appropriate, reducing unnecessary burden, and aligning trial complexity with the specific regulatory and scientific question being addressed. Productive, collaborative patient engagement must be a commitment that runs from protocol and development planning through to sharing findings back with the participants who made the research possible. The evidence suggests that meaningful patient engagement results in tangible benefits, and pragmatic approaches may improve both efficiency and generalizability without compromising evidentiary rigor where the science supports it.

The Full Scope of the Access Challenge

Each layer of the challenge—from awareness of available studies and provider support to eligibility, ease of entry, and the participant experience—involves different stakeholders, and success at one depends on what happens at the others. So does the commitment to return useful results to the communities that made the research possible. Structural barriers across these layers contribute to lower enrollment and retention among underrepresented populations, making early patient engagement important for equitable trial design. As described in “Modernizing Site Quality in Clinical Trials,” early patient and site engagement improves feasibility and reduces operational burden.

What these layers share is their sensitivity to protocol design decisions made long before a single patient is approached. Trial complexity, unnecessary procedures, overly restrictive eligibility criteria, and excessive data collection frequently become barriers to participation, particularly in community settings and historically underrepresented populations. Bringing patient voices into the protocol planning stage, rather than consulting them after a design is substantially developed or, worse, finalized, has measurable consequences for study performance.

Research by the Tufts CSDD and the Clinical Trials Transformation Initiative found that a modest investment of $100,000 in a patient engagement activity during protocol design, such as a patient advisory board review, could yield a financial return more than 500 times that amount. This return is driven by avoided amendments, improved enrollment and retention, improved operational efficiency, better adherence, and earlier product launch. Protocols developed with patient input were also more extensively simplified and had timelines closer to or exceeding plan. These findings align with ICH E6(R3), which emphasizes participant-focused design and proportionate oversight as core principles for modernized clinical trials. That is a compelling argument for prioritizing and resourcing patient engagement from the outset.

Quality by Design and Regulatory Engagement as Enablers for Innovation

Broader participation in clinical trials can offer clear benefits: more generalizable results, stronger enrollment and retention, and more robust product labeling. EMA’s real-world evidence framework similarly stresses the role of real-world data in supporting regulatory decision-making and improving the relevance of evidence across diverse patient populations. Recent FDA guidance, particularly the emphasis on Quality by Design (QbD) embedded in ICH E6(R3), combined with the demonstration projects of FDA CDER’s Center for Clinical Trial Innovation (C3TI), gives the industry a practical mechanism to minimize perceived and actual risks and a path to be able to do this as a repeatable practice.

Two initiatives stand out for advancing study design and increasing patient involvement:

• FDA’s Streamlined Trials Embedded in Clinical Practice (STEP) project targets the integration of research into routine care settings—the most direct path to reducing participant burden and expanding trial participation beyond academic medical centers. Early C3TI-facilitated discussions suggest that cross-functional sponsor-regulator engagement can help clarify expectations around oversight, endpoint collection, and inspection readiness.
• Selective Safety Data Collection (SSDC) addresses the tendency toward overcollection of safety data for mature development assets and marketed medicines that contributes disproportionately to site and patient burden without a corresponding benefit to evidence quality. This approach aligns directly with fit-for-purpose data collection principles reflected in ICH E19 and ICH E6(R3).

What these two opportunities have in common is that they are better suited for medicines that have a reasonable amount of evidence behind them, whether from earlier indications or from large clinical trial databases where significant risk has already been discharged. Pragmatic elements such as expanded eligibility and simplified procedures and endpoints expand the potential for participation and representativeness. The potential to profile the medicine in a real-world setting is valuable to inform clinical care. SSDC, in particular, is a focused approach: it collects only the data needed to assess core safety parameters in medicines that are already well characterized, which streamlines the burden on both patients and investigators.

The ongoing VICTORION-2-PREVENT cardiovascular outcomes trial incorporated a selective safety data collection strategy. This enabled reduced visit frequency more closely aligned with routine clinical practice while maintaining appropriate safety oversight and regulatory engagement.

Sponsors’ concerns about greater heterogeneity compromising a study’s ability to meet its primary endpoint are increasingly addressable through good study design and early regulatory engagement, which turns that heterogeneity into a manageable, expectable risk. FDA, EMA, and PMDA have all signaled the need to modernize evidence generation, with growing emphasis on real-world data, AI, and the infrastructure and data standards required to achieve this at scale, including continued regulatory harmonization across regions. The importance of consistent oversight structures is further illustrated in a recent article on governance variability in clinical trials, which examines how governance differences across regions can slow adoption of innovative, patient-centered evidence-generation models. ICH E19 further reinforces harmonization by defining a selective approach to safety data collection, enabling sponsors to reduce unnecessary burden while maintaining participant safety and evidentiary integrity.

Incentives Drive Real Change

In any serious discussion of trial access, the incentive structure deserves scrutiny. A 2024 analysis found that many trial procedures do not support core endpoints and instead add participant and site burden, highlighting the need for more feasible and patient-centered protocol design. For sponsors, performance metrics centered on speed—protocol approval, first site initiated, first-patient-in—compete directly with the investment of time and resources required to finalize study designs that take full advantage of programs like STEP and SSDC. Feasibility, representativeness, and operational sustainability are still too often secondary considerations rather than core measures of trial quality and development success. For healthcare providers, overly complex studies with high investigator and patient burden are simply not feasible for many community-based practices—settings that often have access to the right patients and capable investigators, but lack the infrastructure, operational support, or prior sponsor relationships needed to participate repeatedly in research.

Better tools alone will not expand access if the incentives governing development do not reward their use.

For access to improve at scale, three shifts are necessary:

• Sponsors should build outcome metrics for enrollment feasibility, representativeness, and study performance into development team accountability frameworks. These belong alongside cycle time as core success criteria, not as supplemental key performance indicators.
• Health systems and providers need investment and support in training and data infrastructure to participate as genuine partners in research, particularly in communities that are capable but currently underserved. The government also has a role in building both the infrastructure and the incentives to modernize evidence generation, and public-private partnerships that establish models while retaining quality and addressing public health priorities can accelerate that work.
• Technology and platform companies should move toward standards-based interoperability and reusable infrastructure models that reduce study-by-study customization, improve data portability, and enable research workflows to operate more seamlessly within healthcare delivery systems.

The PACT consortium, facilitated by Tufts CSDD, drew on data from more than 100 clinical trials and found that trials using virtual and remote solutions exceeded planned enrollment timelines while also improving patient recruitment and diversity. Yet deploying these capabilities at scale requires a platform approach grounded in data interoperability and digital workflow adoption. FDA’s Digital Health Technologies guidance supports remote data acquisition in clinical investigations, while FDA’s interoperability framework reinforces safe, secure, and effective data exchange across systems.

A Call to Action Across the Ecosystem

Access to clinical trials is the front door to a learning health system, but it has to be engineered, not improvised. FDA’s learning health system initiatives reinforce that sustainable evidence generation requires intentional infrastructure investment, continuous feedback loops, and shared accountability across sponsors, regulators, and health systems. Modernizing evidence generation can advance access and rigor together. Many of the tools, technologies, and regulatory frameworks needed to improve patient access already exist. What is needed now is the resolve to apply them consistently and at scale.

Clinical trial access should be an essential design principle from the very beginning of protocol development. If we engineer research around the people and communities it is intended to serve, we can generate better evidence, develop medicines more efficiently, and build a clinical research system in which participation is a realistic opportunity for every eligible patient.

linical trials have never been more scientifically sophisticated than they are today. But many studies continue to face familiar challenges: slow recruitment, high dropout rates, protocol deviations, the informed consent decision-making process, operational burden on sites, etc. These challenges are particularly evident in complex studies such as gene or cell therapy studies that also entail long-term follow-up periods.

Why Clinical Trials Still Fail Patients

Protocols are designed to answer scientific questions; they are often not designed around the realities of patients’ lives and their specific needs.

A protocol may appear feasible on paper but becomes difficult to execute once real patients, caregivers, investigators, clinicians, nurses, and study coordinators interact with it. The result can be a gap between the way a trial is imagined during the design phase and the way it is actually executed in clinical practice. This gap has implications not only for patient experience, but also for recruitment, retention, protocol compliance, data quality, and study timelines.

SJD Barcelona Children’s Hospital designed and implemented an integrated End-to-End Patient-Centric Trial Design Method in several commercial and academic studies to address these challenges before the studies were initiated.

End-to-End Patient-Centric Trial Design Method

This methodology combines patient engagement, clinical expertise, operational feasibility assessment, human-centered design, and simulation-based testing throughout the development lifecycle of a clinical trial.

Rather than evaluating a protocol only once, this method is based on a 360-degree analysis of the study protocol and the entire participant journey from initial contact through informed consent, study participation, long-term follow-up, and study closure. The model integrates three complementary phases/approaches:

• Clinical Trials Expert Panel input
• SIMTest High-Realistic Clinical Simulations
• Patient and Caregiver Advisory Panel input

This approach goes beyond the simple combination of three individual phases by actively integrating them into a coherent framework. For example, risks identified or recommendations proposed by experts can be further explored and validated through simulation. Similarly, findings arising from these simulations can be translated into topics for discussion within the patient panel, where they can be examined from a broader and more diverse collective perspective.

Study Protocol Assessment by Clinical Experts

In this phase, the Clinical Trials Experts Panel evaluates clinical trial protocols across eight key domains (see Figure 2) that collectively influence trial feasibility, acceptability, and participant experience. The review begins with an assessment of the study rationale and objectives, followed by the study design, target population and eligibility criteria, study procedures and visits, safety monitoring requirements, investigational product and intervention characteristics, site feasibility and operational considerations, ethical aspects, and patient experience. Recommendations generated by the panel aim to optimize protocol feasibility, participant recruitment and retention, and the overall patient experience while maintaining the scientific objectives of the study.

Testing Reality Before the First Participant Is Enrolled

One of the model’s most distinctive components is the use of SIMTest, a methodology based on highly realistic clinical simulation. Professionals and patients simulate complete study visits and/or specific trial procedures in a realistic setting, generating both operational and experiential insights that are particularly meaningful because they are derived from an immersive experience.

Rather than relying solely on theoretical assessments, simulation allows sponsors and stakeholders to observe how a protocol performs in conditions that closely resemble real clinical practice. The methodology is based on the principle that there is often a difference between “work-as-imagined” and “work-as-done.” By recreating real-world circumstances, teams can identify operational bottlenecks, communication failures, cognitive overload, and workflow inefficiencies before study launch.

Beyond operational and logistic aspects, a highly realistic simulation conducted by our team also identified opportunities to improve communication strategies, use educational materials more consistently, simplify explanations of complex concepts, and distribute information across multiple interactions rather than concentrating everything into a single visit.

Listening to Patients Beyond Recruitment

Patient input is often included late in development, frequently after major protocol decisions have already been made.

This end-to-end patient-centric design method places patients at the center of protocol evaluation. Patient advisory panels review the study design feasibility and clinical trial documentation from the perspective of real users, assessing comprehension, terminology, risk communication, participant burden, decision-making requirements, accessibility, and overall experience (see below). They also work on the same topics covered by the simulation, with the goal to gather wider feedback about previously identified challenges and potential solutions.

Patient Involvement as Strategy

Patient-centricity is often discussed as a value. It must also be viewed as a development strategy: Studies that align scientific goals with participant needs, concerns, and constraints are more likely to achieve efficient recruitment, reduce dropout rates, improve adherence, support informed decision-making, and generate higher-quality data.

The SJD experience consistently showed that many protocol challenges are identifiable long before first-patient-in. Communication difficulties, burdensome schedules, unrealistic assumptions about participant behavior, workflow inefficiencies, and comprehension barriers can all be detected through structured patient involvement and simulation methodologies.

Most importantly, these challenges can often be addressed without compromising scientific rigor (e.g., by adjusting the time duration of the visits, including decentralized elements, ensuring the suitability of PROMs, etc.).

The future of patient-centered research requires moving from consultation to codesign. Sponsors, academic institutions, investigators, regulators, patient organizations, and healthcare systems should work together to test, refine, and optimize clinical trial experiences before implementation.

An End-to-End Patient-Centric Trial Design Method approach provides a practical framework for achieving this goal. By integrating patient insight, clinical expertise, operational assessment, and simulation-based validation, organizations can build trials that are not only scientifically robust, but also understandable, feasible, and meaningful for the people who ultimately make research possible: patients.

nspection readiness often fails before anyone calls it an inspection-readiness problem.

Then an audit, inspection, submission, or regulatory review gets closer, and a harder question appears: Can the organization clearly show what happened, who owned it, why decisions were made, and whether the records can stand on their own?

This is where many readiness gaps become visible. And it may be too late.

Inspection readiness is sometimes treated as final preparation: assemble the records, review open items, rehearse likely questions, and close gaps before the inspection. These activities matter. But they cannot make up for weak control earlier in the development lifecycle. In regulated development, readiness is not only something teams prepare for at the end. It is something they build while the work is happening.

Regulatory Context

The specific requirements differ by region, product type, and development stage, but major regulatory frameworks point in the same general direction: sponsors and regulated organizations must be able to show reliable records, clear oversight, traceable decisions, and controlled execution. This article focuses on clinical-development inspection readiness, particularly sponsor oversight, GCP conduct, delegated activities, and the records that support clinical, quality, and regulatory review. GMP and GLP environments have additional requirements, but the examples below are intentionally centered on clinical development to keep the discussion focused.

In the United States, clinical investigations conducted under an investigational new drug (IND) application are governed by FDA clinical investigation and good clinical practice requirements, including 21 CFR Part 312 for IND applications, 21 CFR Part 50 for informed consent, and 21 CFR Part 56 for institutional review boards. These requirements address areas that are central to inspection readiness, including sponsor oversight, investigator selection, monitoring, required records, and FDA inspection of sponsor and investigator records. When a sponsor transfers obligations to a contract research organization (CRO), the transferred responsibilities must be documented, but the need for clear sponsor oversight and reliable records remains. FDA guidance on electronic systems, electronic records, and electronic signatures also reinforces that records used in clinical investigations should be trustworthy and reliable.

In the European Union, the Clinical Trials Regulation supports a more harmonized approach to clinical trial assessment and supervision across EU and European Economic Area countries. It also increases transparency and uses the Clinical Trials Information System for trial applications, oversight, and public access to trial information.

International Council for Harmonisation (ICH) guidance, including ICH E6(R3) on good clinical practice and ICH E8(R1) on clinical-study design and quality, further supports risk-based quality management, documented processes, sponsor responsibility, and oversight of delegated activities. The same inspection-readiness principle appears in other regulatory settings as well: the United Kingdom’s Medicines and Healthcare products Regulatory Agency (MHRA) provides guidance for clinical trials and GCP inspection, while Japan’s Pharmaceuticals and Medical Devices Agency (PMDA) operates within a regulatory framework that includes GCP compliance for drug clinical trials and inspection programs for regulated product quality systems. In each setting, sponsors need to be able to explain how clinical, quality, and regulatory evidence was generated, reviewed, and controlled.

The practical message is consistent: inspection readiness depends less on last-minute cleanup and more on whether the operating system has been producing clear, reviewable evidence all along.

Where Readiness Breaks Down

Late readiness problems usually start as ordinary operating gaps. They often look manageable until the organization must explain them to someone outside the day-to-day workflow.

Records exist but are not defensible.

A document can be present and still be hard to defend. It may be incomplete, outdated, inconsistently versioned, missing context, or difficult to reconcile with related records. A record that requires a long verbal explanation may not be ready for inspection.

In clinical development, this can show up in familiar records: trial master file documents, monitoring follow-up letters, delegation logs, protocol-deviation documentation, informed-consent records, safety reconciliation outputs, vendor data-transfer files, database-lock readiness records, and regulatory-response inputs. These records may exist in the expected location, but still be difficult to defend if the decision rationale, source evidence, version history, open issues, or responsible owner is not clear.

The practical test is simple: Could an external reviewer understand the record without relying on tribal knowledge?

Ownership is assumed, not explicit.

Many development activities cross functions and partners. A clinical operations team may depend on a vendor. A regulatory team may depend on clinical, quality, safety, or chemistry, manufacturing, and controls inputs. A sponsor may delegate tasks to a CRO but still retain oversight responsibility.

When ownership is assumed rather than stated, activity can continue but accountability weakens. Teams may know who is involved, but not who owns creation, review, approval, escalation, reconciliation, or closure.

Inspection readiness requires more than participation. It requires clear ownership of specific stated activities.

Vendor and cross-functional dependencies often stay hidden too long.

Modern drug development depends on CROs, contract development and manufacturing organizations, specialty vendors, technology providers, laboratories, and internal functions. Each group may deliver its part, but readiness can still fail at the handoff.

A vendor may send the required output on time, but it may not be usable by the receiving team. For example, a database-lock milestone may appear to be on track while several supporting dependencies remain unresolved: laboratory-data reconciliation, protocol-deviation review, safety reconciliation, vendor query closure, and documentation of final study decisions. Each item may be known to the working team, but not visible to the governance forum reviewing the milestone. A timeline may show progress, while the evidence package remains difficult to defend because supporting records, owners, and open decisions are still scattered across teams or systems.

Vendor oversight should therefore focus not only on whether work is moving, but whether outputs are complete, usable, traceable, and ready for downstream review.

Practical oversight controls include an artifact-level ownership map, a handoff checklist for critical deliverables, a visible open-issue log, milestone-based readiness reviews, and predefined escalation thresholds. These practices work best when they are built into routine governance rather than added only when inspection preparation begins.

Escalations occur after options have narrowed.

Many risks are known before they are escalated and addressed. Teams may wait because the issue is still being worked on, the impact is not yet certain, or escalation feels premature. By the time the issue reaches the governance level, the remaining options may be limited. For example, repeated delays in reconciling vendor data, unresolved disagreement over a protocol deviation, or missing rationale for a key study decision may not immediately stop execution. But each can become harder to explain later if the issue is allowed to age without an owner, decision path, or documented resolution.

This is a common readiness failure. Escalation should not be reserved only for confirmed problems. It should also be used when delay, ambiguity, or repeated rework threatens the organization’s ability to defend the path forward. For example, if a monitoring issue requires repeated clarification across the CRO, clinical operations, and quality teams, escalation may be needed before the issue becomes late, because the real risk is no longer only the open action; it is whether the organization can clearly explain what happened, who reviewed it, and how it was resolved.

A practical escalation trigger is not only, “Is this late?” It is also, “If this remains unresolved, will it affect a regulatory, quality, inspection, or patient-facing commitment?”

Governance forums report activity but do not produce decisions.

Many governance meetings are full of updates: on track, in review, pending vendor response, awaiting approval, being finalized. These phrases sound useful, but they may not show whether a decision is needed, whether a risk has changed, or whether an owner has authority to close the issue.

A readiness-focused review should make decisions easier. It should clarify the issue, the options, the risk of delay, the owner, and the next action. Otherwise, the meeting may create the appearance of oversight without improving control.

A simple readiness-review template can use five prompts:

1. What is the issue or record being reviewed?
2. What decision or action is needed?
3. What evidence supports the current position?
4. Who owns closure?
5. What happens if the item remains unresolved by the next milestone?

These questions help a governance forum move from status reporting to decision-ready review.

What Can Teams Do Earlier?

Inspection readiness improves when teams test defensibility before formal inspection preparation begins. That does not require a separate bureaucracy. It requires a few practical habits.

Teams can start by asking:

• Which records would be hardest for an outside reviewer to understand?
• Which deliverables depend on vendor or cross-functional handoffs?
• Which decisions have been discussed but not clearly documented?
• Which open risks are aging without a defined escalation threshold?
• Which governance forums are reporting activity but not producing closure?

These questions help move readiness from a late-stage scramble into routine operating practice.

For example, before a database-lock or inspection-readiness milestone, a team might review whether monitoring follow-up letters, protocol-deviation documentation, safety reconciliation outputs, vendor data-transfer records, and key decision rationales can be understood without additional explanation. If that review shows that the records exist but are spread across trackers, emails, and working files, the team can assign owners, reconcile the evidence, and document the remaining decisions before the milestone is under pressure. The benefit is practical: the organization spends less time reconstructing the story later and more time confirming that the evidence is complete, usable, and ready for review.

A useful readiness review does not need to cover everything. It should focus on the records, decisions, dependencies, and risks most likely to affect regulatory review, inspection response, patient safety, data reliability, or product development continuity.

Why This Matters

Inspection readiness is often discussed as a compliance topic, but its consequences are broader. FDA’s Bioresearch Monitoring inspection metrics track inspection observations and compliance trends across clinical investigators, institutional review boards, sponsors, CROs, bioequivalence programs, and good laboratory practice settings. Individual FDA warning letters also show how weaknesses in sponsor oversight, electronic records, case histories, corrective actions, and inspection response can lead to documented inspection observations, remediation commitments, additional regulatory questions, and reduced confidence in the reliability of the evidence package.

The reverse is also true. These practices work best when they are built into routine execution rather than added only during inspection preparation. When teams define record ownership early, make dependencies visible, escalate aging risks before milestones are compressed, and use governance forums to make decisions rather than only collect updates, they can see whether the evidence package is becoming reviewable while there is still time to correct it. Leaders do not have to rely only on reassurance. They can see whether the record is coming together before the deadline.

This matters for sponsors, CROs, vendors, quality teams, regulatory professionals, and clinical operations leaders. In development work spread out across functions, systems, and partners, no single group owns readiness, which is created through the way work moves across the organization.

Practical Takeaway

The most useful readiness question may be the one teams ask before anyone is preparing for an inspection:

If an inspector, auditor, or regulatory reviewer asked tomorrow, which record, decision, vendor dependency, ownership gap, or unresolved risk would be hardest to defend?

That question is not meant to create alarm. It is meant to reveal where attention is needed while there is still time to act.

Inspection readiness fails when teams discover too late that activity was not the same as control. The earlier discipline is to make evidence, ownership, dependencies, escalation, and decisions visible while the work is still in progress.

he headline for 2026 is not that AI is getting smarter. It is that expectations are becoming defined. On January 14, 2026, the US FDA and the European Medicines Agency (EMA) published joint Guiding Principles for Good AI Practice in Drug Development. While legally nonbinding, these principles provide a window into the agencies’ current thinking and a benchmark during regulatory review. It is a signal of regulatory alignment on what “good” looks like when AI supports evidence across the drug development lifecycle, including clinical trials and safety monitoring.

The focus on whether AI can be used has now shifted to prioritizing continued oversight that guarantees that AI is implemented in a responsible and controlled way.

Shifting from Experimentation to Consistency and Accountability

Over the past 18–24 months, many life sciences organizations have explored AI in “text-heavy” and assistive use cases: summarizing long documents, classifying content, extracting data points, and drafting internal or external text. This is a logical starting point because it is fast to pilot and easy to demonstrate value.

The harder step? What comes next.

As soon as AI output starts to influence regulated evidence, decisions, or actions, “it appears to work” is no longer the standard. The standard is now evolving to become repeatable with oversight: consistent use in context, clear accountability with a human in the loop, and a record that holds up to audit scrutiny over time. That is what the EMA–FDA principles bring into sharper focus.

Context of Use Becomes Essential for “Good” AI Practice

The most consequential phrase in the regulators’ framing is “context of use.” AI is not evaluated in the abstract. It is evaluated in the context of what it is intended to do, what it influences, and what risk it introduces.

A drafting assistant used to summarize internal notes is one thing. AI that uses the same summarization method to prioritize safety case review, flag potential signals, or initiate steps in trial operations is another. The same underlying technique can move from low impact to high impact simply by where it sits in the process and how its output is used.

For example, using AI to identify trends in past health authority questions is a lower impact context of use for regulatory affairs. Building on this use case, using AI to research and then assist in building a response heightens impact and risk. And creating a final response with no subsequent human review in the process becomes an extremely high impact context of use.

This context matters because it changes what “good practice” requires in the real world. While FDA and EMA now require a structured, risk‑based credibility assessment for AI in drug development, neither agency has issued a formal risk‑classification system for specific AI use cases. But when the context of use is higher risk, the surrounding expectations should shift with it: clearer controls, stronger documentation, explicit human oversight, and more disciplined change management.

The practical implication is that AI programs will increasingly need an inventory of use cases mapped to risk, with defined boundaries and a shared understanding of accountability.

AI Delivered Natively Within Workflows Strengthens Governance

The guiding principles are technology-neutral, but they are not neutral about execution. In practice, this broadens attention beyond the model itself to the surrounding controls and documentation: data provenance, versioning, how outputs are reviewed and used, and what changed over time.

This is where a strategic design choice becomes critical: whether AI sits outside the regulated workflow as a separate tool, or whether it is natively part of workflows inside the systems where regulated work actually happens.

When AI output is delivered as part of a regulated workflow, it may trigger additional validation obligations under the FDA’s electronic records and systems expectations, but these workflows can inherit what regulated environments already require: role-based permissions, standardized steps, controlled handoffs, and audit trails by design.

When AI sits outside, teams often end up reconstructing context and controls after the fact, copying information across tools, reconciling versions, and rebuilding the “story of the record” during reviews and inspections. This is manageable in a pilot. It is fragile at scale.

This “native by design” direction is also where industry product strategy is heading. For critical processes such as eTMF intake, health authority correspondence processing, and medical, legal, and regulatory review of promotional materials, delivering AI that natively works with operational systems of record better aligns structurally with governed data, auditability, and repeatable oversight, rather than treating AI as a separate overlay.

Lifecycle Management as an Operational Discipline

Lifecycle management can sound abstract until you map it to how AI behaves in production.

Newer foundational models are released. Prompts derived from agent objectives and instructions evolve. Knowledge sources and transactional data change. Any of those changes can subtly alter AI outputs in ways that matter in a regulated context. The FDA/EMA Joint Principles’ emphasis on risk-based performance assessment and lifecycle management is, in practice, a reminder that “set it and forget it” is not an option after AI influences regulated work.

Both the FDA and EMA have now codified requirements for AI lifecycle management. FDA’s Predetermined Change Control Plan establishes expectations for managing model updates and AI drift, and the EU AI Act together with Annex 22 requires post‑market monitoring, drift detection, and change-control procedures for high‑risk AI systems.

So, what does “lifecycle management” mean in operational terms?

• Version control that is audit-relevant not only for code and models, but also for prompts, agent instructions, templates, and reference material that influences outputs.
• Defined monitoring and drift triggers that define what you monitor, how often, and what constitutes meaningful change.
• Change control proportional to risk when a change requires documentation, review, requalification, or rollback.
• Clear accountability when AI output is accepted, overridden, or escalated so that responsibility does not evaporate into “the algorithm said so.”

These are operating model disciplines. They are not optional add-ons delegated to a data science team. They sit at the intersection of quality, business process ownership, and technology, which is exactly where regulated execution already lives.

Global Regulations Converging on Responsible Use

The FDA-EMA principles arrive as part of a wider regulatory pattern that reinforces the same underlying message: Trustworthy technology depends on trustworthy execution.

The European Commission’s timeline for the EU AI Act makes governance expectations increasingly concrete. While applicability of the Act varies by article and chapter, the main application date of the Act arrived on August 2, 2026, alongside earlier milestones for prohibited practices and AI literacy (from February 2, 2026) and general-purpose AI obligations (from August 2, 2025).

In the US, FDA’s October 2024 Q&A guidance on electronic systems, electronic records, and electronic signatures in clinical investigations clarifies what it means for electronic records and systems to be trustworthy and reliable. It is not AI guidance, but it becomes directly relevant when AI output is part of evidence generation. If AI influences evidence and decisions, the integrity of the electronic record and the system that produced it becomes more central, not less.

None of the above suggests that regulators want to slow innovation. The joint principles are explicitly framed to enable responsible use. The operational bar is rising in a way that rewards organizations which make oversight practical, repeatable, and scalable.

How Leaders Can Design a Framework for AI at Scale

The practical response is not to slow down AI adoption. It is to operationalize it, so each new use case does not trigger a fresh governance debate.

For most organizations, the near-term opportunity is to use the FDA-EMA principles as a design requirement for the AI operating environment, even while product capabilities across clinical, regulatory, and safety continue to mature across the market. This means doing the foundational work to move from isolated AI applications to a governed capability in a regulated execution environment. That involves defining contexts of use, mapping them to risk, standardizing what “essential information” looks like for oversight, and putting lifecycle controls around what changes and when.

This is also the moment to push vendors for specificity. The question is not Does it do AI? but Where is AI in the workflow, what record does it produce, what is versioned, what is auditable, what is monitored, and what changes under change control? This will inform how tools meet existing regulations for validated systems and new expectations for good AI practice. The more AI works natively with the systems that already manage roles, permissions, workflows, and audit trails, the easier it becomes to scale responsibly over time without rebuilding evidence and oversight from scratch.

Raising the Bar for Operational AI

The FDA-EMA principles mark a significant shift from viewing AI as a separate tool to one that natively works within controlled workflows to support evidence generation and monitoring. This means that decisions can be reviewed and traced with clear oversight and transparent change management.

Success will rely on enabling AI within the regulated foundation of operations, turning governance and oversight from compliance requirements into an operational advantage.

ood clinical trials depend on good data, and Clinical Data Management (CDM) has long played the quiet but critical role of keeping that data reliable, usable, and ready for review. But while Electronic Data Capture (EDC) systems have largely replaced paper forms and improved primary data entry, much of the broader process is still sequential and reactive. It still depends heavily on structured data, manual checks, and downstream issue resolution.

Clinical Data Management at Inflection Point

The challenge is compounded by the explosion of data from diverse sources like EHRs, wearables, and genomics, which creates immense integration complexities. Furthermore, the rise of adaptive and decentralized trials leads to frequent midstudy changes, making traditional, reactive quality-control methods inefficient and resource intensive. The resulting data silos and lack of real-time analytics hinder the ability to gain swift, actionable insights, highlighting the need for CDM to evolve.

Together, these challenges are pushing CDM beyond traditional data cleaning toward a more insight-led CDS (Clinical Data Science) model.

Moving Beyond Data Management

Clinical organizations now need to go beyond what traditional CDM was designed to deliver, which was primarily to ensure that data is “managed the right way.” CDS expands this focus to include data risk, data meaning, and data value, ensuring that data is not only clean but also credible, reliable, and actionable for prompt decision-making. This broadens the focus from just “Are the data clean?” to “Are the data reliable and useful for decisions now?”, a shift that turns clinical data from an operational deliverable into a living asset that guides trial execution.

Industry Reality Check

The shift from CDM to CDS will be incremental, not sudden. Many clinical teams still work with fragmented systems, multiple vendors, uneven metadata practices, and workflows that were not designed for AI-enabled operations. As interest in AI grows, the real adoption question is practical: Can these outputs be trusted, validated, explained, and safely used in regulated clinical decisions?

Emergence of Agentic AI

Generative AI has already shown how quickly knowledge work can change: summarizing documents, drafting content, finding patterns, and helping teams reason across large bodies of information. The next phase is likely to be more operational. It will involve agentic AI: systems that can take a defined goal, interpret context, plan steps, and work with humans or other agents to complete a task.

In clinical data management, an agent does not need to be mysterious or fully autonomous. It can be a focused digital worker assigned to a well-bounded activity: checking incoming data for anomalies, proposing electronic case report form (eCRF) specifications from a protocol, comparing vendor data with EDC records, suggesting medical coding terms, or escalating site-level risk signals for review.

Agents will be most useful when they sit inside the workflows teams already depend on, with the right controls around them. They should recommend, document their reasoning, and keep expert review where judgment matters. Used this way, agentic AI can help teams move from periodic checks to more continuous, informed oversight. The bigger opportunity is not simply to automate tasks, but to help teams spot issues earlier, explain what they are seeing, and respond with more confidence.

Intelligent Interventions Across the Clinical Data Lifecycle

The opportunities for intelligent intervention span much of the CDM process. Protocol interpretation, eCRF design, edit-check creation, data review, reconciliation, coding, quality management, and database lock all contain repeatable work where AI can assist.

Reimagining Core CDM Functions

In study start up, AI agents can help translate protocol requirements into draft eCRFs, validation logic, metadata, and mapping suggestions. These outputs still need expert review, but they can give teams a stronger starting point and improve consistency across studies.

During study execution, agents can monitor incoming data, identify missing or unusual values, and suggest contextual queries. Instead of relying only on scheduled review cycles, teams can address issues closer to the point at which they occur.

Consider query management, one of the most resource-intensive activities within many clinical studies. Agentic AI can improve this workflow by leveraging protocol requirements, historical patterns, and clinical context to identify issues earlier, prioritize critical discrepancies, and draft contextualized queries. The value lies not just in faster query generation, but in shrinking the cycle time between detecting an issue and resolving it.

Reconciliation is another high-value area. Agents can compare data across EDC, labs, safety systems, devices, and other sources; identify discrepancies; propose likely matches; and route exceptions for human review. Medical coding and serious adverse event reconciliation can benefit in a similar way, especially where narrative text, terminology, and cross-system comparison add complexity.

The near-term value of agentic AI is not in fully autonomous trials, but in reducing repetitive work so clinical data professionals can focus on exceptions, study risk, cross-functional decisions, and patient-relevant outcomes. Used within governed workflows, agents can help teams scale expertise across process-heavy, data-rich, time-sensitive activities without compromising quality.

Building the Intelligent Clinical Trial

A few isolated AI pilots will not be enough to define the future of CDS. It will require networks of specialized agents connected to the systems, metadata, standards, and governance that already shape clinical development.

For example, a protocol intelligence agent may inform eCRF design. A data-quality agent may share signals with a risk-monitoring agent. A reconciliation agent may provide evidence to a medical reviewer or coding specialist. In each case, the value comes from joining operational steps that are often treated separately today.

Human expertise remains central. Clinical data managers, programmers, statisticians, and medical reviewers provide the judgment and accountability that AI cannot own. Their roles will increasingly shift from manual task execution to strategic oversight of AI-driven insights. Agentic AI should expand their reach, not blur responsibility. The best operating model is therefore human-led and AI-augmented.

Roadmap Toward Clinical Data Science

Moving from CDM to CDS will require more than deploying new tools. Organizations will need modern data foundations, metadata discipline, clear ownership, validation approaches, responsible AI controls, and workforce development.

Trust will become a design requirement, not an afterthought. In regulated clinical development, AI must be useful, auditable, and governed according to how much the decision matters for patient safety, data quality, and trial outcomes. Practically, this demands robust validation frameworks, immutable audit trails for every agentic action, and continuous performance dashboards. If organizations approach this well, the benefit will go beyond efficiency. They can catch problems earlier, reduce end-of-study surprises, strengthen oversight, and make better decisions while the study is still in motion.

Conclusion

Clinical Data Management has always evolved to meet new challenges, and the shift toward Clinical Data Science is the next logical step. Agentic AI offers a path forward, not by replacing human experts, but by augmenting their ability to identify issues earlier, connect siloed information, and scale their expertise. The organizations that succeed will be the ones that keep people accountable, make AI trustworthy in day-to-day use, and apply it where it genuinely helps teams make better study decisions.

can accelerate clinical work faster than organizations can redesign the decisions around it. A human in the loop is not sufficient unless that person has clear authority to act, escalate, or stop the workflow.

This same pattern shows up wherever an automated system produces a deliverable faster than the receiving team was organized to handle it. Authority questions appear when AI flags:

• a monitoring anomaly
• a protocol deviation
• a safety signal
• a data discrepancy.

These events can be flagged almost instantaneously, but the person looking at the flag often does not know what they are authorized to do next. Speed without authority creates risk.

Regulators are converging on the same expectation: the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA) jointly published Guiding Principles of Good AI Practice in Drug Development in January 2026, emphasizing human responsibility and oversight of AI. Industry’s shorthand for this expectation is to “keep a human in the loop.” But naming a human is not the same as designing what that person is authorized to decide and is ready to act upon. Unclear decision authority is not new, nor is it unique to AI. Basic automation forced the same questions in regulated industries years ago; AI simply compresses the time organizations once had to work around it.

“Authority” in this context means being explicitly given the right to decide, escalate, or stop, and knowing the rationale for each. Authority is one component of the operating model (alongside roles, timelines, and capability), but it is the component that organizations most consistently fail to design, and the component this article examines.

The Authority Gap Was Already the Hard Part

The evidence comes not from a clinical study but from regulated financial services. The banking case study: automating the manual re-entry of data from one system to another. Clinical teams are familiar with this type of work: Moving the same study information between data capture, trial management, and electronic trial master file (eTMF) systems, all with inconsistent versions of current state.

In the banking example:

• Before: 2,500 transactions per month moved manually from a legacy system into a service management platform, requiring roughly five minutes of manual effort per transaction.
• Intervention: Automated the repetitive, rules-based transfer, and deliberately redesigned the human role to actively support the automation as needed, not merely leave it unchanged.
• Redesigned role: The operations specialist monitored exceptions and was authorized to resolve defined cases, escalate others, or stop the process when the routine path no longer applied.
• Outcome: About 208 hours of routine manual transfer work removed each month, roughly 1.2 full-time equivalents of capacity returned annually.

In this example, the flagged exception that the bank operations specialist was authorized to handle can be compared to a follow-up queue in a trial that lists “CRA” as the final owner of every flag, without specifying whether the remote or onsite clinical research associate (“CRA”) owns the resolution. The system simply found the problem, but not the decision owner who can solve it.

The objective of human oversight as a governance mechanism is not to preserve manual work or require a person to click approve or reject. It is to place the appropriate human experts where judgment, authority, accountability, and contextual understanding are genuinely required. There was no AI in this banking system, yet deciding what the operations specialist was authorized to do when the automation flagged an exception was still the hard part.

While this banking case does not fully illustrate a clinical operating model, it highlights a decision pattern that clinical organizations must solve under more complex conditions. The price of weak authority design is categorically higher in a trial. A mishandled transaction exception at a bank is money, remediation, maybe a reporting obligation. A mishandled safety signal is potentially a patient death.

What Transfers, and What Does Not

The governance approach in this banking example cleanly transfers to clinical research, but the operating environment does not. In the banking case, the automation flagged a transaction it could not complete, and the operations specialist monitoring it had to be authorized to resolve, escalate, or stop it. If an eTMF dashboard flags hundreds of “missing” documents in a study, someone must be authorized to determine which alarms are real. In many organizations, that determination is re-explained in monthly communications and meetings, but no one owns the authority to ensure that the outcome and its accompanying evidence are captured for posterity.

A flagged exception in a study may also move across sponsor, contract research organization (CRO), site, investigator, and vendor boundaries, making authority design far more complex than in a single-organization setting. A clinical trial distributes responsibilities across many parties while remaining subject to external regulatory inspection, so, even though both industries are highly regulated, the clinical trial industry cannot simply import the banking playbook. It must build its authority construct natively.

Authority moments are prevalent across clinical operations: A central monitoring anomaly whose significance must be determined; a detected protocol deviation that a clinical trial manager (CTM) must assess and close; a safety signal to act on, escalate, or stand down; or a data discrepancy to resolve or query.

Audit trails may show what changed, when, and by whom. Evidence chains connect these changes to the CTM’s authority, the information considered, the rationale for the decision, and any required approval or escalation. When evidence chain elements do not travel with the decision, the organization must reconstruct them later. Reconstruction is typically more time-consuming and expensive than capturing evidence in real time. Parallel trackers are often needed when systems record an event but do not preserve the decision context, ownership, and evidence required to manage or explain it later. Introducing new tools before designing decision authority creates risk, leaving teams to maintain manual trackers that reconstruct state, ownership, and evidence after the fact.

Every authority decision in a trial is potentially inspectable. The evidence of who was authorized, who decided, and on what basis is not just internal hygiene but material that may have to be produced years later to an external party (a regulator) with enforcement power.

Pilot Paralysis and the Human Angle

The prevailing concern in public forums where clinical research professionals engage in open discourse is not fear of technology. Practitioners describe being asked to give free domain expertise to tool builders who do not even know what GCP (Good Clinical Practice) stands for. Site staff describe having to learn and juggle more than a dozen “easy” systems, each one a sponsor’s customized enterprise solution. A common refrain is that the people who understand the work must be in the room when the tools that will reshape that work are designed.

It stands to reason that AI pilots can stall when the people expected to operationalize them cannot see a clear, supported role for themselves in the new workflow. Practitioners are asking whether they are being “set up” to be accountable for something they are not authorized to control. “Pilot paralysis” is often an operating model and change management failure, not a technology failure. One example is AI-supported signal detection. It can generate follow-on investigative work, particularly when teams have not defined who reviews the signal, what they are authorized to decide, and when escalation is required. Teams see value in automation only when response roles and follow-up ownership are clearly designed upfront.

Two Halves of One Equation

Authority is one half of the equation. But to truly empower teams, there must also be capability. Practitioners (whether in industry or at clinical sites) need access to approved tools and safe environments in which to experiment, explore role-specific learning scenarios connected to real workflows, and challenge AI outputs with confidence. Structured experiments, including hackathons using approved models, can bring domain, technical, quality, and operational expertise together. As AI produces more output, human expertise becomes even more important because practitioners must judge context, consequences, and appropriate action.

Designed authority is not bureaucracy, but it is what empowers teams to move fast, because they are equipped with answers before the government inspector asks any questions.

Build the Transition Gate

To break free of “pilot paralysis” and effectively scale AI pilots into controlled use, organizations must establish a formal transition gate that requires the team to ask and answer:

1. Whose work and timelines change, and are they equipped and ready?
2. What may AI produce or recommend?
3. Who may accept, challenge, override, resolve, or escalate, to whom, and within what timeframe?
4. What evidence must accompany each decision?
5. Who can stop the workflow, and who is authorized to restart it?

Conduct this design work WITH the practitioners who will hold the authority, not TO them. For example: If the study startup acceleration had been collaboratively designed using these transition gate questions with the statistician in the room, their timeline concerns would have surfaced early and the team could have adjusted the operating model around them before the pilot stalled.

In clinical research, AI governance is not achieved by simply adding a human approval step after the system produces an answer. It is achieved when the appropriate qualified experts have the authority, capacity, capability, and evidence to act when judgment is required. That is how organizations turn AI acceleration into controlled progress.

he rapid evolution of healthcare innovation necessitates more efficient and collaborative regulatory processes. As the complexity of new therapies increases, regulatory authorities worldwide are exploring novel approaches to streamline their assessment procedures. One such approach gaining traction is the Regulatory Assessment Aid (RAA), a tool designed to facilitate multistakeholder collaboration and streamline regulatory decision-making.

Regulatory Assessment Aid: An Overview

Traditional approaches to regulatory submissions and reviews are often time-consuming and resource-intensive for both sponsors and regulatory authorities. The need for efficient, transparent, and collaborative tools has never been more apparent.

A RAA is a voluntary submission by the sponsor to the regulatory authority, designed to focus the review on critical thinking and evaluation. The primary objectives of RAAs include:

1. Increasing review efficiency and consistency
2. Decreasing review time spent on administrative tasks
3. Providing a structured format for presenting key information.

RAAs are typically standalone documents with precise formatting requirements and page limits. They are intended to present information consistently alongside the main submission, avoiding the introduction of new data or promotional language.

The first regulatory authority to introduce a RAA was the US FDA. Initiated by the FDA’s Oncology Center of Excellence (OCE), it was first used in 2018 in conjunction with the agency’s real-time oncology review (RTOR) pilot.

In the context of the FDA Assessment Aid (AAid), it is important to note that:

• It’s a stand-alone document where all tables and figures must be added in the body of the document.
• No cross-references or hyperlinks to submission (e.g., US New Drug Application [NDA]) modules are permitted.
• It must not contain any information that is not included in the NDA/Biologic License Application (BLA) submission; all content should remain consistent with the submission modules.
• Cannot include any promotional language about either the sponsor or the product, or other companies or products.
• The submitted document cannot be revised after it is submitted.
• There are strict page limits: 100 pages for New Molecular Entity (NME) initial applications and 75 pages for supplemental applications.

Other regulatory agencies to introduce a RAA include the UK Medicines and Healthcare products Regulatory Agency (MHRA), which encourages companies to complete the clinical assessment report template for innovative medicines applications. Early exploratory discussions have also started in Australia, with the Therapeutic Goods Administration (TGA) reaching out to the pharmaceutical industry association for their thoughts on RAAs.

In 2023, the European Medicines Agency (EMA) started a pilot (the Revamp Project) where participating companies pre-filled the day 80 (D80) quality, nonclinical and clinical assessment reports (AR) with factual data. A report on experience gained with the pilot has recently been published.

Benefits of RAA: Facilitating Collaboration and Regulatory Efficiency

One of the key benefits of RAAs is their ability to facilitate collaboration between sponsors, regulatory authorities, and other stakeholders. By providing a common framework for presenting and assessing information, RAAs can enhance communication and promote transparency in the decision-making process and swift availability of final assessment reports.

For instance, the FDA’s AAid has been successfully implemented in oncology submissions, allowing for more focused discussions between sponsors and reviewers. FDA has reported a reduction in time to approval for both the FDA and participating regulatory authorities in initiatives like Project Orbis, where use of the AAid is mandatory. Orbis is a collaborative framework for concurrent submission and review of oncology products and has demonstrated the potential of RAAs to accelerate patient access to novel therapies.

However, it is important to note that the impact of RAAs on overall timelines including pricing and reimbursement decisions may vary. The above-mentioned EMA Revamp Project pilot revealed variable feedback on the usefulness of this exercise. Most assessors did not find the approach beneficial and still spent time verifying the accuracy of the pre-populated sections. Applicants also found the exercise very time consuming and saw little value in terms of enhanced communication or smoother assessment processes.

One Sponsor’s Experience with FDA AAid

The FDA’s AAid in oncology provides a valuable case study for RAA implementation. Learnings from members of the Clinical Regulatory Writing (CReW) Oncology team at one multinational public biopharmaceutical company highlight several key points.

1. Document Structure: The AAid is structured so that the Sponsor provides position/information on different sections, and the FDA responds in placeholders under every section.
2. Timing of Authoring: After data readout, waiting until the main modules are almost finalized before authoring the AAid can avoid constant tweaking and reduce the chance of errors in key messages.
3. Authorship: Having a medical writer familiar with the data as AAid author is beneficial because sourcing strategic content from different modules into the AAid requires more than a simple copy and paste.
4. Review Process: Some submission project teams have opted for a streamlined review process, with brief reviews by the Global Regulatory Lead (GRL) and US Regulatory Affairs Lead, followed by GRL endorsement.
5. Cross-referencing: While hyperlinking to different modules is not allowed, cross-referencing to specific modules for more details is permitted to make locating information easier.
6. Template Updates: The FDA Assessment Aid template changes periodically, so it’s important to consult with the US Regulatory Affairs Head to confirm the current template.
7. Concise Summarization: Summarizing complex data (e.g., population pharmacokinetic (PK) and exposure response – exposure safety results) in a concise way can be challenging. Early engagement with Clinical Pharmacology colleagues is crucial for this task.

Best Implementation Practices for Sponsors

Based on experiences with the FDA AAid and feedback on the initially explored TGA RAA, several best practices have emerged:

1. Early Engagement: Prototyping the RAA before data readout can be immensely helpful, particularly for non-data-driven sections. This allows teams to agree on study details and lock them before data arrival.
2. Resource Allocation: Careful consideration should be given to the allocation of subject-matter expert (SME) resources, balancing RAA preparation with other critical activities. It’s highly advisable to have a medical writer who authored one of the modules, Clinical Study Report (CSR), or pre-submission Briefing Document to author the AAid, because familiarity with the data is a huge plus.
3. Harmonization: Where possible, RAA requirements should be harmonized across regulatory authorities to streamline resource allocation on the sponsor side.
4. Flexibility: RAA templates should allow for some flexibility in data presentation to accommodate different types of submissions and therapeutic areas.
5. Clear Guidelines: Regulatory authorities should provide clear guidelines on RAA requirements, including page limits, formatting, and content expectations.
6. Collaborative Approach: Close collaboration is required to procure content from respective SMEs for certain sections (e.g., nonclinical sections, appendices). Early engagement with nonclinical and Clinical Pharmacology team members is crucial, especially for initial NDA/BLA submissions where nonclinical sections are extensive.
7. Focused Review Process: Rather than following a full-fledged team review, a focused review by key stakeholders (e.g., Global Regulatory Lead and US Regulatory Affairs Lead) can enhance efficiency in finalizing the document.
8. Adherence to Page Limits: Strictly adhering to page limits saves time and effort; exceeding these limits can result in the document being returned by the regulatory authority.

Regulatory Challenges and Feedback

As the adoption of RAAs continues to expand globally, it is essential to examine their real-world impact and the challenges encountered during implementation, alongside the evolving regulatory feedback shaping their future use.

1. Outcomes: Since its introduction in 2018, the FDA AAid has been credited with improving review efficiency for oncology submissions. Submission of FDA AAid has facilitated earlier identification of data issues, allowed for more focused regulatory queries, and contributed to faster review cycles for approved oncology treatments, according to FDA OCE Annual Reports.
2. Challenges and Limitations: Sponsors may face resource constraints, especially under compressed timelines. In addition, updates to templates and expectations from regulatory authorities may require ongoing training.
3. Regulatory Feedback and Evolution: The FDA OCE publishes templates periodically in response to sponsor and reviewer feedback, streamlining sections and clarifying formatting expectations.

Future Considerations and Conclusions

As regulatory authorities continue to explore and refine RAA concepts, several considerations for future development emerge:

1. Global Harmonization: There is a growing need for harmonization of RAA requirements across regulatory authorities. This would allow sponsors to prepare a single, globally applicable RAA with minimal localization, potentially accelerating submissions to multiple authorities.
2. Expansion Beyond Oncology: While RAAs have been successfully implemented in oncology, potential for expansion into other therapeutic areas remains. However, this may require careful consideration of the unique challenges and data requirements in different document fields.
3. Integration with Other Regulatory Initiatives: RAAs could be integrated with other regulatory initiatives, such as parallel scientific advice from regulatory and health technology assessment (HTA) bodies, to further streamline the path from regulatory approval to patient access.
4. Technological Integration: Future iterations of RAAs could potentially leverage emerging technologies such as artificial intelligence (e.g., Microsoft Copilot) for data analysis and presentation, further enhancing efficiency.
5. Product Awareness: RAAs offer a concise orientation of the drug development and its benefit-risk considerations to assessors who have not been exposed to the data previously (e.g., during clinical trials or via scientific advice procedures).

RAAs represent a promising tool for enhancing collaboration and efficiency in the regulatory review process. While their implementation has shown benefits, particularly in oncology submissions to the FDA, there is still work to be done in optimizing and harmonizing RAA requirements across different regulatory environments.

The success of RAAs will depend on continued collaboration between sponsors, regulatory authorities, and other stakeholders. By working together to refine these tools and sharing learnings from real-world implementations, we can potentially accelerate patient access to innovative therapies while maintaining the rigorous standards necessary for ensuring safety and efficacy.

As we move forward, it will be crucial to continue sharing experiences and best practices in RAA implementation, fostering a global dialogue on regulatory innovation. The ultimate goal remains clear: to develop more efficient, transparent, and collaborative regulatory processes that can keep pace with the rapid advancements in medical science and technology, while effectively managing the additional workload these tools may initially create for both sponsors and regulatory authorities.

he clinical management of obesity has entered an era of unprecedented therapeutic efficacy. As global obesity burden continues to rise, incretin-based and multireceptor agonists capable of producing both significant and sustained weight loss have fundamentally disrupted the treatment landscape, creating new opportunities to improve long-term outcomes for patients but also raising new questions for clinical drug development.

Total body weight reduction remains a foundational endpoint, reinforced by both current European Medicines Agency (EMA) and US FDA regulatory guidance. However, gross body mass is a crude metric for metabolic health, since it may conflate desirable visceral fat loss with deleterious lean muscle loss. That is why the quality of weight loss is just as important as its magnitude.

The DIA Obesity Consortium, a neutral precompetitive public-private research partnership, focused its initial work on exploring body composition and physical function as promising efficacy endpoints, alongside methodology for evaluating multiple comorbidities within a single trial. Additional information about the insights from the collaboration have been described in detail in our September 2026 publication in the DIA Global Forum, “Obesity Drug Development: Time to Revisit Endpoints, Trial Design, and Evidence Generation.” (See History for additional information.)

Building on this work, the 2026 DIA European Executive Roundtable on Obesity Drug Development (held in Rotterdam on March 23, 2026) convened EMA regulators, industry representatives, clinicians, and academic and reimbursement experts to pressure-test emerging concepts and identify practical barriers to efficient drug development. The purpose of the meeting was not to generate formal guidance or drive consensus statements but rather to discuss and highlight key themes: the context-dependence of endpoint utility, the need to balance scientific ambition with operational feasibility, and the importance of developing collaborative evidence strategies to capture obesity’s broader metabolic, hepatic, musculoskeletal, and reproductive consequences. Below, we summarize these insights and their implications for the Consortium’s next phase.

Aligning Data Generation to Validate Label-Enabling Endpoints for Body Composition and Physical Function in Obesity Clinical Trials

Participants reviewed preliminary proposals and discussed how a more harmonized approach to endpoint selection, modality choice, and data collection could strengthen the scientific and regulatory framework for evaluating the quality of weight loss beyond total body weight reduction alone. The field is at an important inflection point: There is broad interest in moving beyond traditional measures, but progress will depend on generating consistent, high-quality evidence that is scientifically credible, operationally feasible, and relevant to patients, clinicians, and regulators.

Standardizing Data Generation for Body Composition Endpoints
The consortium discussed a draft proposal for assessing body composition via dual-energy X-ray absorptiometry (DXA), magnetic resonance imaging (MRI), and bioelectrical impedance analysis (BIA) as exploratory modality. DXA and BIA were proposed to assess whole-body fat mass and lean mass, while MRI was proposed to measure visceral fat mass, with data collected at baseline, an optional intermediate timepoint, and end of treatment.

Participants agreed that these proposals provide a useful starting point for future alignment. Total body weight loss does not distinguish between reductions in adipose tissue and lean mass. At the same time, questions remain regarding which measures are most robust, how they should be standardized across studies, and what degree of change would be clinically meaningful. Consistency in data generation will be essential if future pooled analyses are to support endpoint validation or broader regulatory relevance.

A central theme was the need to balance analytical rigor with operational feasibility. DXA was viewed as a practical option for whole-body assessment. MRI, while informative for visceral fat mass, may present operational challenges in large development programs. BIA was recognized as attractive from an implementation perspective and was included as exploratory, although further work is needed to understand its reproducibility and potential role in regulatory-grade evidence generation. Overall, modality selection should be informed not only by measurement characteristics, but also by study burden, cross-site consistency, and broader clinical applicability.

Advancing Physical Function as a Clinically Meaningful Dimension of Treatment Benefit
The roundtable also considered physical function as a complementary and clinically meaningful dimension of treatment benefit. The consortium’s draft proposal included the 6-minute walk test (6MWT), 30-second sit to stand (STS) test, and hand grip strength (HGS) test, together with IWQoL-Lite-CT, SF-36v2 Acute Form, and PGI-S Physical Function as patient-reported outcome instruments, with assessments at baseline, an optional intermediate timepoint, and end of treatment. Participants agreed that this framework reflects growing recognition that treatment benefit in obesity should not be evaluated solely through body weight reduction or body composition changes, but also through effects on how patients function and experience daily life.

There was broad agreement that physical function endpoints may help bridge physiological change and patient benefit. While body composition measures provide insight into tissue change during treatment, functional measures may help determine whether those changes translate into improvements in mobility, endurance, and quality of life. The 6MWT, STS, and HGS were generally (though not unanimously) viewed as pragmatic performance-based assessments, while patient-reported measures were recognized as important for capturing the patient perspective.

Composite Endpoints
Composite endpoints, encompassing body composition, physical function, and cardiometabolic effects consistent with the June 2026 Cardiovascular-Kidney-Metabolic clinical guidelines, were briefly discussed at the roundtable. However, the group considered these endpoints challenging to construct, as the individual components may differ in clinical relevance, measurement characteristics, and responsiveness to treatment. Their interpretation may also be difficult, particularly if the treatment effect is not consistent across all components of the composite. Further discussion is therefore needed to assess the feasibility, clinical rationale, and appropriate methodology for developing such endpoints.

Hot Topics in Obesity Drug Development

Alternatives to Large Placebo Arms
Obesity now has multiple highly effective pharmacotherapies with modern GLP-1 receptor agonists and dual/triple agonist therapies achieving up to 15%-25% body weight reductions, creating an unprecedented ethical dilemma in research design.

The Declaration of Helsinki states that placebo controls are justifiable only when no proven effective intervention exists or when patients face no serious harm. Yet patients in multiyear placebo arms face progressive weight gain, worsening comorbidities, declining quality of life, and lost opportunities for disease modification during critical treatment windows. Crucially, this ethical dilemma deepens when considering that modern anti-obesity medications have proven benefits extending far beyond weight loss alone, specifically robust cardio-protection. Having high-risk patients remain on a prolonged placebo regimen when active therapies are proven to significantly reduce major adverse cardiovascular events (MACE) and mortality is increasingly difficult to justify ethically. This prolonged placebo exposure in the context of available effective therapies raises serious ethical concerns about balancing scientific rigor with patient welfare.

The most recent 2025 FDA draft guidance for Overweight and Obesity, referenced above, requires a minimum of 1,500 subjects randomized to placebo for at least one year, powered to detect a 50% increase in background adverse event rates (3% to 4.5%). These parameters were never designed to detect rare, serious events ( <1%), which generally require robust post-market surveillance. Notably, as described in the 2022 publication, the EMA has previously acknowledged that active-control designs may be more relevant, and that large placebo arms may increase dropout rates, thereby compromising trial integrity. Participants at the roundtable therefore discussed a proposal for a two-pillar framework to reduce unnecessary exposure to placebo in clinical studies of agents in well-understood classes, such as GLP-1 receptor analogs. That framework included:

– A reduced concurrent placebo strategy component, targeting 500–750 patients (versus 1,500), which represents a 50%–75% reduction. It was agreed that this type of cohort still provides directional confirmation, detects the most prevalent safety signals, and preserves assay sensitivity while minimizing ethical concerns.
– An active comparator bridging component, incorporating indirect treatment comparisons via shared active comparators and historical studies to generate placebo-anchored estimates, while maintaining statistical rigor with reduced placebo exposure.

Other proposed solutions included capping placebo duration at ≤52 weeks, supported by open-label extensions, adaptive titration, or rescue medication. Continued consortium work will develop further and more detailed recommendations for alternatives to full placebo arms in the next year.

Obesity requires lifelong management, yet clinical trials focus on acute phases of the disease. Modern frameworks and guidance must address safety when transitioning from high-dose induction to lower-dose maintenance. Participants agreed that the key remaining question is whether an indirect comparison via active comparators could reduce placebo requirements. Unpublished proof-of-concept studies show strong concordance between direct and indirect estimates, supporting rigor while enabling FDA–EMA harmonization. The validation pathway for this issue would require standardizing cross-trial comparability, developing statistical frameworks, and defining bridging acceptance criteria. Managing placebo duration offers multiple options: rescue therapy, open-label extensions, adaptive designs, and robust ethical oversight.

Combination Drug Products and Treatment Regimes for Obesity and Related Conditions
Traditional fixed-dose combination (FDC) evaluation has previously focused on additive weight loss. But “meaningful contribution” should expand beyond kilograms lost to include muscle preservation, improved fat-to-lean ratio, and cardiometabolic benefits like insulin sensitivity and lipid profiles. Moreover, enhanced tolerability with fewer side effects, better titration, and lower discontinuation is known to directly impact real-world effectiveness. Frameworks should recognize FDCs achieving similar weight loss with improved tolerability as meaningful advances. Therefore, robust phase 2 trials could establish component contributions and dose optimization, allowing phase 3 to focus on FDC versus standard care without replicating all monotherapy arms.

Participants concluded that the field stands at a critical juncture where outdated requirements may potentially harm patients while providing diminishing value. This evolution represents a fundamental shift in balancing scientific rigor with ethical responsibility, demanding courage to challenge paradigms and compassion for the patients who benefit from research.

Next Steps

The next stage of the DIA Obesity Consortium will translate the multistakeholder scientific alignment developed during the initial portion of the collaboration, and further advanced during the 2026 DIA–EMA Executive Roundtable, into a concrete work plan.

The overall obesity field has largely accepted that total body weight alone is an incomplete measure of treatment benefit. The more difficult task now is to determine what data need to be collected and utilized to support the broader interpretation of new clinically meaningful endpoints in obesity drug development.

A central priority will be the continued refinement of data elements needed to support future pooled analyses that can inform existing global regulatory guidance and start potential biomarker qualification efforts. This work will focus on two closely related areas: (1) quality of weight loss and body composition, and (2) physical function. For body composition, the consortium will continue to evaluate which measures are most appropriate for consistent use across studies, including assessment of fat and lean mass, as well as visceral adiposity. For physical function, the focus will be on identifying measures that are feasible to implement, interpretable across populations, and clinically meaningfully to patients’ quality of life.

The goal is to develop a disciplined evidence-generation framework that can clarify whether changes in body composition and function reflect true clinical benefit. We are aligning our work to be complementary to other ongoing public-private partnerships in the space, including the FNIH REAL BODY project.

A second major deliverable will be the development of a draft protocol design for evaluating multiple comorbidities related to obesity. The roundtable discussions highlighted the potential value of master protocol or basket trial approaches, particularly for conditions such as metabolic, hepatic, musculoskeletal, and reproductive comorbidities, but also outcomes related to vascular complications or heart or kidney failure, as a relevant area to explore. This builds on discussions first introduced at the 2025 Obesity Roundtable, summarized in our September 2026 article referenced above.

The next stage will begin to define how such a protocol may be structured, including population selection, endpoint differentiation across comorbidity cohorts, shared or condition-specific comparators, and statistical considerations that are needed to preserve interpretability. Ongoing regulatory engagement will remain essential. Future discussions with FDA, EMA, and other global regulators will be staged around specific scientific questions, proposed data frameworks, and draft protocol concepts rather than broad theoretical issues. The consortium will also expand regional dialogue through anticipated executive roundtables in Asia and the Middle East, creating opportunities to understand how global regulators view endpoint development, trial design innovation, placebo use, and evidence requirements for obesity-related comorbidities. Communication and dissemination will also be built into the future work plan. A planned public DIA webinar in 2027 will communicate progress, share emerging themes, and broaden awareness of the consortium’s work. Strategic in-person sessions aligned with major DIA global meetings will provide additional opportunities for member engagement, regulator interaction, and refinement of deliverables.

The DIA Obesity Consortium will produce updated consensus frameworks, white papers, and guidance concept submissions to global regulatory agencies, as well as publish them through DIA channels and other appropriate scientific outlets that would use this collaborative partnership to improve the efficiency of obesity drug development. The outputs of the next stage will establish foundational elements for future phases of the work beyond 2027. Ultimately, this work will move the field from general agreement to usable tools, clearer data standards, and focused global regulatory dialogue leading to unified approaches around the world.

his past August, Tufts CSDD (Center for the Study of Drug Development) and Medable jointly announced the results of a study that asked questions about the economic value of implementing an AI monitoring agent in a clinical trial and answered these questions with empirical evidence. Based on a benchmarked oncology program, Medable contract and experience data, and Tufts CSDD analysis of key activities in oncology trials that directly impact study timelines, resource utilization, and efficiency, this study demonstrates that AI agents genuinely accelerate study timelines and improve productivity, which will ultimately deliver more cures to more patients.

Q: How did this study originate? Who first proposed the research question, and how did Medable and Tufts CSDD divide their respective roles in designing the analysis, supplying the data, choosing the assumptions, performing the calculations, and interpreting the results?

A: Tufts CSDD has a long history of conducting independent empirical studies quantifying the net financial value of various innovative solutions (e.g., patient engagement, Decentralized Clinical Trials [DCTs], digital endpoints) to inform industry-wide adoption. Medable approached Tufts CSDD to conduct a similar, independent study. The Tufts CSDD research team, led by Dr. Joseph DiMasi, directed and managed all aspects of the research effort. Medable provided data on actual use cases associated with its agentic AI study monitoring solution including clinical trial outcomes data from actual use experience and the investment required to deploy the solution.

Q: The announcement describes the analysis as using both “actual use” data and benchmark data, while the methodology says Medable supplied operational data from its Clinical Monitoring Agent “modeled in an oncology development program.” Can you walk through the underlying data set? Which improvements were actually observed in trials using the agent, which came from Tufts benchmarks, and which were assumptions or modeled extrapolations?

A: The underlying data populating the model are oncology clinical trial- and program-level benchmarks compiled and maintained by Tufts CSDD. We also gathered data on global travel costs and global salary levels of clinical research associates (CRAs). As mentioned earlier, Medable provided clinical trial outcome data (study timelines) from actual agentic AI use cases as well as the financial investment required to deploy the solution per individual CRA.

In modeling impact, Tufts CSDD used data from recent sponsor-reported operational changes to on-site and off-site monitoring in transitioning from traditional to risk-based/centralized monitoring. This data informed the change in CRA workload and number of visits under an agentic AI-supported approach. The primary assumption modeled (a 20% reduction in study enrollment) came from published reports (e.g., McKinsey).

Q: What clinical-monitoring activities did you actually evaluate? What tasks were performed or augmented by the AI agent, what would a CRA or other study personnel ordinarily have done instead, and what parts of clinical monitoring were outside the scope of this analysis?

A: The agentic AI solution largely addresses the very high time- and labor-intensive tasks handled by a study monitor, including 100% source data review and verification, query management, eTMF (electronic Trial Master File) and ISF (Investigator Site File) document review and filing, site communications and notifications, and action item and follow-up tracking.

Q: One of the most consequential findings is roughly 18 weeks of acceleration, including a 109–119-day reduction in enrollment time and about two weeks to database lock. How did you establish that the monitoring agent was responsible for those improvements? Were those differences directly observed against comparable trials or sites, or are they estimates based on changes in workflow? What other factors could account for some of that difference?

A: Enrollment time improvements were informed by published reports (e.g., McKinsey, MIT) and corroborated elsewhere. It largely reflects an agentic AI solution accelerating signal detection and alerting the site sooner so that it can implement remedial steps. Although we modeled time savings, we did not include cost reductions due to lower drop-out rates. This is an area that we will be pointing out in the forthcoming peer-reviewed manuscript.

Q: You estimate reductions in on-site monitoring costs of about $4.4 million for phase 2 and $5.6 million for phase 3, plus another $600,000 and $1.7 million respectively in off-site administrative efficiencies. What portion represents cash that a sponsor would actually stop spending, versus CRA time or organizational capacity that is freed up and could potentially be redeployed elsewhere?

A: The $4.4 million and $5.6 million ROI figures represent both cost savings (e.g., reduced CRA travel costs associated with on-site visits) plus time savings that can be reallocated. On average, CRAs typically oversee sites participating in three studies. For the CRA, the reduced work effort on the agentic AI-enabled study frees up their time to focus on stronger relationships with site personnel and to work on other studies.

Q: The numbers likely to attract the most attention are the estimated 64x ROI in phase 2 and 82x in phase 3. What exactly is the numerator and denominator in that calculation? Which costs of implementing and operating the agent are included—software, integration, validation, training, change management, and ongoing human oversight—and which costs are excluded? How sensitive are those ROI figures to different assumptions?

A: The eNPV values and multiples represent the difference between the eNPV of a traditional oncology development program compared to the eNPV of the agentic AI-enabled oncology program. The eNPV model is risk-adjusted and includes all cash-in and outflows associated with program development expenses, the full program investment required to deploy the innovation for the agentic AI-enabled program, development durations, and the commercial performance of approved oncology therapies.

The program-level eNPV value and ROI multiples represent expected outcome if an agentic AI solution were implemented effectively. We describe this as average expected net financial impact because we have not taken into account order-of-entry effects and individual program commercial potential. We have averaged the commercial performance of a large number of marketed cancer therapies.

Q: This analysis is based on an oncology development model. What characteristics of the modeled phase 2 and 3 trials—number of sites, patients, countries, monitoring intensity, enrollment duration, and so forth—drive these results? How confident are you that the findings would translate to a smaller oncology study or to therapeutic areas with different enrollment and monitoring requirements?

A: All the oncology development data is drawn from Tufts CSDD benchmarks published in the literature (most often in Therapeutic Innovation & Regulatory Science [TIRS]). The oncology area was deliberately selected for several reasons: Not only is it the most active area in drug development, but also because investigational oncology therapies are sponsored by a highly diverse community of emerging, small, and large sponsor companies. Also, these investigational therapies are highly personalized, targeting very narrowly defined patient communities.

We’re confident that the results translate well across oncology programs. We have conducted sensitivity analysis to directionally show how the eNPV delta changes with a much larger targeted patient community (e.g., chronic diseases). However, we encourage sponsor companies to adapt and customize the model to best reflect their unique portfolio experience.

Q: Is the $21 million/82x result best understood as a representative expectation, a plausible upper-end scenario, or something closer to a best-case situation? That last distinction matters because the press release says that AI agents “can deliver up to” $21 million and 82x ROI rather than presenting those figures as average expected performance.

A: The $21 million eNPV is for phase 3 trials, so when you look at phase 2 trials, the eNPV is $7.5M for the program, as the risk to the development success is greater.

Q: The detailed paper has not yet gone through peer review. If you were reviewing this work yourselves, what would you identify as its two or three most important limitations? And what additional evidence—prospective studies, controlled comparisons, independent data sets, or replication with other technologies—would you want to see before concluding that these economic benefits are broadly achievable in clinical development?

A: The detailed paper has been submitted for peer review, and it touches on two really important limitations. First is that our financial modeling is based on averages and benchmarks. Companies need to customize the model with their own data to get a more representative measure of the value proposition. The second limitation is that our model assumes that an organization will effectively deploy the innovation. But for many reasons, organizations often struggle to fully leverage an innovation’s impact.

Global Forum thanks Data/Digital Editor Lisa Barbadora for coordinating this Q&A.

art 1 of this Commentary argued that the risk sponsors should worry about is not concentration but variability, and that the benchmarking data shows Multisite Clinical Research Corporations, or MCRCs, compress that variability without sacrificing quality. If the diagnosis holds, the practical question follows: What should sponsors do to successfully manage this variability, and what must MCRCs offer in return? The answer starts where regulators have been pointing for a decade.

Regulatory Tailwind

Globally, regulators have pushed sponsors toward prospective, risk-based quality management. ICH E6(R3), the FDA’s work on quality by design, and the Clinical Trials Transformation Initiative quality framework all ask the same thing of sponsors: Build quality in at the design stage and manage risk across the trial rather than inspecting for it at the end. This is where MCRCs earn their keep.

When a sponsor works with 50 independent sites, it manages 50 separate quality systems and 50 separate risk profiles. When it allocates the same volume to an MCRC, it manages one of each. Centralized processes give the sponsor a single point of accountability, which is exactly what a risk-based quality framework is designed to reward. Rather than fight the direction of regulation, network-level allocation leans into it.

Power of Scale, Read Honestly

The concentration-risk objection has an empirical answer, and it is worth stating objectively rather than overclaiming. Velocity’s internal confidential data from 2024 onwards shows that adding more in-network sites to a study increases contracted enrollment and decreases screen-fail rates. As protocol complexity climbs year on year, sponsors appear to be realizing more value from organizations that can deliver scale, standardized processes, and operational consistency.

This honest version of the power-of-scale argument is more persuasive than a sweeping claim precisely because it survives scrutiny. Scale within one accountable entity improves realized enrollment. It is the same narrowing of variabilities described in part 1, seen from the recruitment side.

That evidence does not simply rest on one sponsor’s data. Across seven blinded Velocity case studies spanning six therapeutic areas—closed between 2024 and 2026—the same pattern holds regardless of sponsor or indication:

Exhibit: Velocity Case Study Portfolio, 2024–2026 (Blinded)

*Site-share and patient-share figures were not available in the source data for this case and are left blank rather than estimated.

F1 and F2 are two related studies within the same hemagglutinin vaccine program.

*Site-share and patient-share figures were not available in the source data for this case and are left blank rather than estimated.

F1 and F2 are two related studies within the same hemagglutinin vaccine program.

The range tells the story as much as any single number: Velocity’s share of total trial sites in these studies ran from roughly one-tenth to one-half, yet realized enrollment against contracted goal never fell below 100% and reached as high as 400%. This is the reduction of variability from part 1 made visible in delivery: Allocate meaningful volume to one accountable network—in modest or large measure—and the outcome is dependably at or above commitment, not a coin flip that depends on which fraction of sites a sponsor happened to pick.

Two honesty checks, in the same spirit as the single-sponsor data we chose not to use above:

1. This is a curated set of studies chosen to illustrate the pattern, not a random or exhaustive sample of Velocity’s full portfolio; readers should view it as illustrative rather than statistically comprehensive.
2. Three of these seven studies are vaccine trials, a study type that can enroll patients faster in more accessible populations; the pattern holds across the nonvaccine studies too, but the vaccine-heavy skew is worth naming rather than obscuring.

Trust Must Run Both Ways

If allocation is going to replace selection, it needs the scaffolding that made the CRO partnerships work. The strategic relationships that succeeded in the 2000s were the ones that built in metrics, governance, and shared accountability from the first day. Many of the relationships that failed were built on little more than goodwill and a handshake. The same is true here.

Sponsors should ask for reliability scoring, standardized key performance indicators, transparent visibility into process and infrastructure, and bilateral commitments that put obligations on both sides. But transparency cannot run in only one direction. Much of the industry conversation treats the site as the party that must open its books while the sponsor watches. Real partnership does not work that way. Sponsors need to trust sites as much as sites trust sponsors.

Bringing a network in early, sharing the shape of the incoming work, and discussing protocol design with sites before they are locked can dramatically alter the success of a subsequent trial. MCRCs can hire and train ahead of need, confirm they have the sites, therapeutic expertise, and patient access the program will require, and flag capacity constraints while there is still time to address them. A network that knows what is coming can prepare for it; a network that is kept in the dark can only react. Trusting a partner with sensitive information is a real shift in behavior, but this is the type of trust that makes network-level allocation deliver excellent results.

Watching Quality in Real Time

One underrated benefit of the entity model (MCRC) is visibility. Because an MCRC runs common systems across its sites, it can take the pulse of a trial’s quality while the trial is still running, rather than discovering problems at database lock. Deviations surface faster and get resolved faster when the same processes and the same dashboards span the whole network. For a sponsor, this means fewer surprises and a shorter distance between a problem appearing and its solution being found. This has shown up in screening success with a recent diabetes program where highly varied screen failure rates among sites revealed entirely different local practices to pre-screening that could be fine-tuned for better patient selection; eDiary completion rates in the 80 percents at one site compared to more than 90% at 10 other sites revealed an intervention opportunity to improve patient engagement, and, in turn, data compliance. Real-time quality oversight is difficult to achieve across a scatter of independent sites but is a natural property of a network.

Velocity’s own data illustrate the point. Its centralized Electronic Data Capture model is now entering 97% of clinical trial data within one business day of the patient visit—more than four times faster than the five-to-10-day industry average reported in independent benchmarking research from Veeva’s industry survey on clinical data management. That speed comes from replacing the traditional model—where entry speed and quality vary from coordinator to coordinator, site to site—with a single, purpose-built data-entry team trained to one standard and measured against one benchmark, network-wide. The payoff for sponsors is well-prepared monitoring visits, faster query resolution, and a shorter path to database lock: real-time visibility turned into a measured number instead of an abstract promise. Less variability.

Risks of Getting This Wrong

None of this is automatic, and it would be dishonest to pretend that this model has no failure modes. They are, tellingly, the same traps into which the CRO market fell:

• Overstandardization can strip out the local flexibility that complex trials sometimes need.
• Network-level allocation, applied bluntly, can shut out smaller sites that bring genuine access to specific patient populations.
• The language of quality by design can decay into a buzzword that decorates conference slides without changing how anyone works.

An MCRC earns the right to allocated volume by understanding where the model breaks and building to avoid it. This starts by naming these risks.

Practical First Step

Sponsors do not need to overhaul their outsourcing strategy to test any of this. But, in the next 12 months, they can allocate a defined slice of volume to a single MCRC under a bilateral agreement with shared KPIs and measure the results against their own site-by-site baseline: activation speed, deviation rates, realized enrollment, and, above all, the predictability of each. The comparison will be its own argument. What MCRCs must do is make that conversation straightforward enough: publish the metrics, commit to them, and stand behind network-level outcomes.

The drug development industry spent the 1990s learning that buying reliability rather than capacity was the smarter way to engage CROs. The site layer is now at the same threshold. Sponsors that move first, and MCRCs that give them the metrics and commitments to move with confidence, will spend the rest of this decade running more predictable trials than their competitors. This is not a marketing claim. It is what happens when you manage the risk that actually matters.

I’d welcome the chance to compare notes with any sponsor already thinking this through.

n the 2026 Sir Alasdair Breckenridge Lecture, EMA Executive Director Emer Cooke set out her vision for fundamentally reimagining medicines regulation in response to accelerating scientific, technological, and societal change. Her central argument is focusing on changes for those who are at the interface between regulations and developer: the Regulators. In her view, Regulators must evolve from primarily being gatekeepers of medicines to becoming enablers of responsible innovation, while preserving the principles on which credible regulation depend: independence, scientific rigor, transparency, and trust.

EMA Executive Director Emer Cooke (courtesy of European Medicines Agency)

Cooke frames this challenge around a paradox: Biomedical science has never been more capable, yet significant patient needs remain unmet. mRNA vaccines, cell and gene therapies, personalized medicines, artificial intelligence, advanced manufacturing, and treatments for extremely small patient populations are impressive examples of the extraordinary pace of innovation. The question is therefore no longer whether science will innovatively advance, but whether regulatory frameworks can evolve quickly enough to translate that innovation into benefits for patients.

Medicines regulation needs adaptable individuals and communicators who can work on tasks across different domains and in multidisciplinary teams, whether regulatory, healthcare, or technology-based, and can collaborate with business. AI ethics, knowledge, and judgment become incredibly important skills in a future that is dominated by this technology.

Why the regulatory model must change

Much of today’s medicines regulation was designed for an era of small-molecule drugs, large-scale manufacturing, relatively predictable development pathways, and conventional clinical trials. Today’s environment includes gene editing, individualized cell therapies, platform technologies, decentralized manufacturing, device-drug combinations, digital twins, and AI-supported discovery and development.

For Cooke, incremental improvements to existing processes are insufficient. The “regulator of the future” must be as innovative and adaptable as the science it assesses. It must be able to deal with complexity, new technologies, and new approaches. And the outputs must remain anchored in the values that give regulation its legitimacy: independence, rigor, transparency, and trust.

COVID-19 demonstrated what is possible

Looking back through her tenure at EMA to date, Cooke finds that the COVID-19 pandemic provides the clearest evidence that regulatory systems can operate differently. EMA and Europe’s regulatory network mobilized more than 4,000 scientific experts, adapted working methods, intensified international cooperation, and evaluated vaccines and treatments at unprecedented speed.

Crucially, speed did not require abandoning regulatory standards. Instead, regulators collaborated more closely, worked in parallel rather than sequentially where appropriate, and concentrated resources on the issues that mattered most.

Cooke argues that this agility must become part of normal regulatory practice. Future infectious-disease outbreaks, antimicrobial resistance, medicine shortages, climate-related health threats, and geopolitical instability will demand regulatory systems that are adaptive, resilient, and capable of moving quickly without compromising standards.

Europe’s pharmaceutical reform

The proposed overhaul of EU pharmaceutical legislation represents, in Cooke’s view, the most important reform of European medicines regulation in a generation. It is intended to support the vision of a system that is faster, simpler, more flexible, and more supportive of innovation, while continuing to protect patients.

The reforms include simplifying EMA’s committee structure, reducing assessment times, embracing digitalization, and strengthening engagement with academia, biotechnology hubs, technology-transfer offices, start ups, and small and medium-sized companies.

Europe has been creating a more attractive environment for clinical research. Initiatives such as ACT EU and FAST-EU are underway to make clinical-trial assessment faster and more coordinated and facilitate innovative trial methodologies.

New tools will enable a new way of working: an EMA pilot to connect to tech transfer offices and biotech hubs across Europe. The sandbox concept will allow Regulators to learn and adapt safely together with Developers.

The objective is proportionality: AI as regulatory challenge

Regulators must simultaneously facilitate responsible AI innovation and safeguard the quality, safety, and efficacy of medicines developed or monitored using AI. Regulation that is too rigid may suppress useful innovation; regulation that is too weak may expose patients to unacceptable risks.

Cooke identifies three requirements to achieve this:

• Clear, predictable regulatory expectations across the medicines lifecycle;
• Robust governance from the beginning, including early agreement between developers and regulators on evidence requirements; and
• Deep international collaboration, with common principles emphasizing human-centered design, transparency, and multidisciplinary oversight.

Cooke’s vision of the ideal Regulator is one that understands which evidence genuinely matters and asks for neither too much nor too little.

One globally accepted single development plan for medicines

Future regulation should involve more joint scientific advice, collaborative assessments, reliance and recognition between regulators, and harmonized approaches to emerging technologies. Her long-term aspiration is particularly ambitious: a globally accepted single development plan for medicines.

International networks such as the International Coalition of Medicines Regulatory Authorities (ICMRA) remain an excellent model for a more integrated global regulatory ecosystem.

New technology requires new regulatory skills

Reimagining regulation also means reimagining the regulatory workforce. EMA is investing in digital skills, AI literacy, and stronger connections between data scientists and professionals with clinical and traditional regulatory expertise.

Future regulators will need multidisciplinary capabilities combining scientific judgment, data expertise, technological literacy, communication, and ethical awareness. Importantly, Cooke argues that greater reliance on AI will make human judgment more—not less—important.

Trust is ultimately the foundation

Cooke’s final—and perhaps most fundamental—theme is public trust. Regulatory reform cannot succeed through faster assessments, better data, or AI alone. In an environment of misinformation and declining confidence in institutions, regulators must communicate evidence clearly and with empathy, acknowledge uncertainty honestly, and engage people through channels they actually use.

EMA is therefore strengthening social listening, collaboration with trusted experts and stakeholders, multimedia communication, and efforts to counter misinformation and improve vaccine confidence.

Cooke’s ultimate vision is not simply a faster regulator. It is a learning regulatory system: agile Regulators, scientifically sophisticated, digitally capable, internationally connected, and responsive to patients, firmly anchored in independence, rigor, transparency, and public trust.

The full text of the lecture has been published on the EMA website.

n the last 10 years, Clinical Data Management (CDM) has grown exponentially in Japan and throughout the world, in both scope and activities. Although we use the term CDM in job and project titles, today’s work is essentially the expansion of CDM: from recording, collecting, reporting, and analyzing clinical research data to broader Clinical Data Science (CDS).

DCTs require a unique approach to clinical trials, where the personnel and participants are not at a clinical trial site; rather they are part of the community. We looked at how a DCT can meet all regulatory requirements and monitor the trial appropriately. Just as in the United States, DCTs in Japan are part of academic/nonindustry collaborations. Thus, we provide lessons learned from a DCT collaboration example (international agreement) involving National Cancer Center Japan (NCC) partner sites in Japan and Thailand. This example illustrates the increasing complexity of data flow, oversight, and source diversity in modern clinical trials.

Related information about this example was shared in a session at the DIA Japan Annual Meeting 2025. This session was organized as a collaborative effort between the DIA Japan and Global DIA Clinical Data Management (CDM) communities, with members participating from both.

DCTs move away from the traditional study model, where a patient comes into a clinic and is followed in a prescribed manner with the type of data to be collected spelled out beforehand, to a model that focuses on the logistics of data collection and the integrity of the data flow.

The roles of data managers and data management have expanded substantially. While traditional CDM focused on the logistics of data collection and the integrity of the data flow, CDS expands this focus to include the meaning and value of the data, and the associated risks. The data manager is not only in charge of the data management function but is integral in making sense of the whole clinical data universe, which includes a growing variety of eSource types of data; e.g., electronic health records (EHRs), electronic medical devices, e-diaries, e-PROs, eCOAs, and Digital Health Technologies (DHTs).

DCT Benefits and Challenges for CDM/CDS

Decentralized clinical trial study elements such as telemedicine, wearables, and other remote data capture methods offer many benefits. These include the ability to query the data in real time, leading to less missing data and faster access to the trial data compared with traditional data capture. Through the inclusion of well-planned DCT elements, trials can now answer important scientific questions with high-quality data not possible even a few years ago, while simultaneously reducing trial burden for study participants and trialists.

Progress in designing and conducting DCTs can be attributed to a complex combination of factors, including the adoption of new technologies and progress in drug development science.

However, these changes pose challenges for clinical data scientists, whose primary mission is to ensure the quality and integrity of data. Clinical trial data goes through a complex process, including the collection, management, analysis, and reporting of findings. As a result, ensuring data quality and integrity cannot be the sole responsibility of a single group; rather, consideration of the overall data flow and the objectives of the trial, and cross-role collaborations, is necessary and extremely important.

New Landscape Needs New Knowledge and Capabilities

The complexities inherent in the adoption of DCTs have fundamentally transformed the landscape for CDM/CDS professionals, demanding a re-evaluation of established practices and the cultivation of new knowledge and capabilities.

Collective insight focuses on several critical areas that necessitate immediate attention and strategic thinking:

Diversity of Data Sources and the Assurance of Data Quality

The expansion of DCT tools has led directly to a greater diversity of data sources. It has become necessary to develop and establish robust methods for ensuring data quality and integrity that are specifically tailored to each scientific question and its associated data source and adhere to applicable regulatory requirements and scientific standards. Increasingly, this involves assuring reliability, traceability, completeness, and timeliness of data from nontraditional sources such as home visits, telemedicine applications, wearables, and wearable devices, and communicating the best way to collect this information.

Such complexity may become even more visible in DCT models involving multiple partner sites or international collaboration, but the underlying CDM/CDS challenges are broadly relevant to DCTs in general.

Regulatory Modernization: The Impact of ICH E6(R3) and ICH E8(R1)

The most significant recent regulatory change that affected the CDM/CDS role is the official adoption of the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) E6(R3) Good Clinical Practices and ICH E8(R1) General Considerations for Clinical Studies guidelines. ICH E6(R2) Good Clinical Practice: Integrated Addendum to ICH E6(R1) represents modernization of GCP to address innovations such as DCTs, Digital Health Technologies (DHTs), and data-driven oversight models. E6(R3) emphasizes trial quality, participant protection, and data integrity within a risk-based framework.

Quality by Design (QbD) and Risk-Based Quality Management (RBQM)

QbD and RBQM illustrate long-term industry/regulatory initiatives to modernize the clinical trial ecosystem and go hand-in-hand with ICH E6 and E8. These modifications are important to CDM/CDS professionals across both pharmaceutical industry and academic research settings. Because QbD offers a systematic approach to development that begins with predefined objectives and emphasizes process understanding and control, its benefit is not limited to a single setting or role. For clinical data scientists specifically, QbD means extracting and using knowledge from diverse sources to assess quality attributes and improve data strategies from the initial discussions of the design.

Understanding Operational Challenges in Assessing Data Risks

To truly safeguard data integrity, it is essential to understand the operational challenges of DCT trials. This comprehensive understanding of challenges is fundamental to appreciating the inherent risks associated with the data generated in these novel, nontraditional trial settings. It may be expected, for example, that partner medical institutions will have a lower level of expertise regarding the collection of data for the trial in question. This may require a comprehensive medical/educational support process together with monitoring and data checks.

Learning from Practical Clinical Research Examples

Finally, we wished to examine how a complex DCT model involving partner-site collaboration was planned. We felt that there was significant value in sharing practical insights from DCT planning and preparation that could pose quality and data issues, and how such difficulties can be addressed. Specifically, we looked at a study agreement between sites in Japan and Thailand’s Department of Medical Services that highlighted invaluable lessons detailed below.

Through this planned partnership, we identified study-related tasks that can appropriately be delegated to a partner site, including examinations, drug dispensing, and electronic data capture, while the investigator responsible for the study retains oversight, including remote participant interviews, through DCT tools. Unlike a conventional international study, this model allows the investigator at NCC to remotely provide study-related medical care and oversight to participants located in another country through DCT tools, while collaborating with and being supervised by local physicians and partner institutions. This approach is expected to help address barriers such as limited access to investigators with specialized expertise, funding constraints, and restricted opportunities for eligible patients to participate in academic trials.

This framework illustrates how DCTs can create greater diversity in data sources and operational responsibilities, requiring careful coordination of IRB review, data sharing, and local medical licensure. It increases access to study participation for potential participants in countries where availability of qualified investigators in the relevant therapeutic area is limited.

Now What? Strategic Shift from CDM to CDS

The transformation from CDM to CDS is a foundational change underlying modern clinical data development activity. This transition requires that clinical data professionals move beyond thinking about and following checklist-driven processes to responsibilities and activities based upon more critical thinking and proactive risk management.

The core responsibility of CDM/CDS in ensuring data quality and integrity demands extending traditional data management/system knowledge and capabilities into greater proficiency in digital health technologies, including knowledge of inherent biases in each system, such as incomplete audit trails, pre-sorted adverse event terms based on a MedDRA pick list in an EDC system, incorrect summation of too many detailed data generated by wearables, and a keen awareness of evolving regulatory expectations.

Consequently, CDM/CDS, which has traditionally focused on streamlining and standardizing data management operations, must face even greater challenges in robust cross-functional collaboration with key stakeholders, such as experts in RBQM, Clinical Operations, IT, Biostatistics, Project Management, and the clinical sites themselves while moving towards the strategic oversight of complex data ecosystems that require new communication and education.

This expanded oversight includes the inclusion and description of CDM/CDS leadership roles on drug study teams. For example, sponsors must specify each activity that is permitted remotely in trial protocols and ensure that all personnel involved, including local healthcare providers, are qualified for and understand their contributions to the collection of quality clinical trial data. CDM/CDS professionals who have the primary responsibility for data integrity must contribute to protocol development from the start and remain engaged through all activities associated with trial conduct, analysis/reporting, and regulatory communication. The CDS, like every other key role in clinical trials, must have the education and training even about adjacent areas of the role, to ensure efficient operations and communication along the entire project.