Clinical Trials Brief | May 18, 2026
This week's five signals cover a new global standard that defines how clinical trial protocols should be structured and exchanged electronically for the first time, FDA's proposal to withdraw a vasculitis drug over alleged pivotal data manipulation and rising liver injury concerns, a French AI company signing a multi-year agent-building deal with AstraZeneca while others flee EU regulation, a cancer drug pulled from market after causing new blood cancers in trial patients, and Novo Nordisk deploying an AI reasoning agent for clinical data analysis on Microsoft Azure.
ICH publishes the first global standard for how clinical trial protocols should be structured and exchanged electronically
What changed
On May 5, 2026, ICH published its official explainer for ICH M11: Clinical Electronic Structured Harmonised Protocol (CeSHarP), adopted for implementation on November 19, 2025. Prior to this guideline, there was no globally harmonized standard for protocol format and content, leading to significant variability across sponsors and regions that created inefficiencies in searching, reviewing, and assessing protocols. ICH M11 establishes a standardized protocol structure with detailed guidance on essential protocol elements, presentation of scientific rationale, and risk-based approaches to protocol design. It also introduces an interoperable standard for common terminologies to enable electronic protocol exchange between sponsors, regulators, and ethics committees.
Why it matters
Protocol variability has been one of the invisible costs of global drug development. Different sponsors structure protocols differently, different regulators expect different formats, and no machine can reliably search across protocols from different companies or regions. ICH M11 eliminates this problem at the structural level. For sponsors running multi-regional trials, it means one protocol format accepted everywhere. For regulators, it means protocols that can be computationally searched, compared, and assessed. For the broader ecosystem, it is the foundational layer that makes AI-assisted protocol review and automated compliance checking technically feasible.
Takeaway
ICH M11 is not just a formatting standard. It is the infrastructure layer that makes electronic protocol exchange, automated regulatory assessment, and AI-assisted protocol review possible at global scale. Sponsors should begin aligning internal protocol templates with M11 structure now.
Sources
This item is grounded in the ICH M11 explainer document published May 5, 2026 on ich.org.
FDA proposes to withdraw Tavneos from the market over alleged pivotal data manipulation and rising liver injury signal
What changed
In mid-May 2026, FDA's Center for Drug Evaluation and Research (CDER) formally proposed withdrawing Tavneos (avacopan) from the U.S. market. The proposal rests on two grounds. First, new information shows that unblinded personnel manipulated results of the pivotal ADVOCATE trial so the drug appeared effective when the original analysis did not support that conclusion, and ChemoCentryx (now an Amgen subsidiary) did not disclose the original analysis to FDA. Second, CDER has identified 76 post-marketing cases of drug-induced liver injury (DILI) with reasonable evidence of causal association, including 7 cases of vanishing bile duct syndrome (VBDS) and 8 deaths. Acting CDER Director Tracy Beth Høeg issued a Notice of Opportunity for Hearing to Amgen; the drug remains on market pending a hearing or voluntary withdrawal.
Why it matters
This is the first time in years that FDA has proposed withdrawing an approved drug primarily on data integrity grounds combined with a post-marketing safety signal. The case matters beyond Tavneos because it tests several precedents simultaneously: what happens when post-approval data integrity reviews reveal manipulation, how the risk-benefit calculation changes when both efficacy and safety are in question, and whether the NOOH process can move quickly enough to protect patients while preserving due process for the manufacturer. Amgen has signaled it disagrees with FDA's position and is weighing its options.
Takeaway
The Tavneos case is now a live test of FDA's ability to enforce data integrity standards post-approval. For sponsors, it is a reminder that the pivotal trial record is never truly closed: if post-approval evidence surfaces that pivotal data was manipulated, the entire approval can unravel regardless of how many patients are currently on the drug.
Sources
This item is grounded in the FDA CDER withdrawal proposal page, the March 31, 2026 Drug Safety Communication on liver injury, and Healio reporting from May 15, 2026.
A French AI company signs a multi-year deal to build pharma decision agents for AstraZeneca, while others flee EU regulation
What changed
On May 13, 2026, Owkin announced a three-year licensing agreement with AstraZeneca to build biopharma AI agents on its K Pro platform. Under the deal, Owkin will develop end-to-end AI agents integrated within AstraZeneca's IT infrastructure and decision workflows, designed to help decision-making teams access data-rich insights for complex competitive intelligence questions. The agreement builds on previous collaboration where Owkin developed an AI-based BRCA pre-screening solution (BRCAura) for breast cancer, published at ESMO with 93% sensitivity. Owkin CEO Thomas Clozel stated the company believes 'the future of the pharmaceutical industry is agentic' and that K Pro enables building complex agents for pharmaceutical partners.
Why it matters
This matters in the context of the EU AI Act because Owkin is a Paris-headquartered company that has chosen to build its entire AI infrastructure within the European regulatory framework rather than flee from it. While OpenEvidence withdrew from Europe citing the EU AI Act's high-risk classification, Owkin is actively deepening its European-origin AI platform into global pharma operations. The contrast is not coincidental: companies building specialized B2B pharmaceutical AI agents face a different regulatory calculus than clinical decision support tools that interface directly with physicians. The EU AI Act's high-risk rules may be driving a structural split between clinical-facing AI (leaving) and R&D-facing AI (staying and scaling).
Takeaway
The EU AI Act is not uniformly hostile to clinical AI. It is creating a structural separation between physician-facing decision tools (high-risk, heavy compliance burden) and pharmaceutical R&D platforms (potentially lower classification, enterprise-governed). European AI companies operating in the R&D layer can use regulatory clarity as a competitive advantage rather than a reason to leave.
Sources
This item is grounded in Owkin's May 13, 2026 press release announcing the AstraZeneca K Pro licensing agreement.
A cancer drug is pulled from market after causing new blood cancers in 5.7% of trial patients, with zero cases in the control arm
What changed
In March 2026, Ipsen voluntarily withdrew tazemetostat (Tazverik) from all approved indications in follicular lymphoma and epithelioid sarcoma after an Independent Data Monitoring Committee recommended immediate discontinuation based on adverse events in the Phase 3 SYMPHONY-1 trial. FDA subsequently issued a formal public alert underscoring the risk. In the trial, 18 of 318 patients (5.7%) receiving tazemetostat plus lenalidomide and rituximab (R2) developed hematologic secondary primary malignancies (SPMs) at a median treatment duration of 15.8 months. Zero SPMs were reported in the control arm receiving R2 plus placebo. Most SPMs were myelodysplastic syndrome and acute myeloid leukemia. SPMs appeared as early as 7.5 months after treatment start and occurred in some patients even after stopping treatment. Three patients with SPMs died.
Why it matters
This is a textbook example of the accelerated approval system working as intended but with real consequences. Tazemetostat received accelerated approval in 2020 when the SPM incidence appeared to be 1.7%. The confirmatory Phase 3 trial then revealed a rate more than three times higher at 5.7%, with zero events in the control arm, making the signal impossible to dismiss as background noise. The fact that SPMs occurred after treatment cessation raises a particularly difficult question for patients who received the drug outside the trial: the risk does not necessarily end when treatment stops. For the broader field of epigenetic cancer therapies (EZH2 inhibitors), this withdrawal raises questions about mechanism-related malignancy risk.
Takeaway
The tazemetostat case demonstrates that accelerated approvals carry a genuine obligation: the confirmatory trial must be run, and if it reveals risks that outweigh benefits, withdrawal must follow regardless of commercial commitments. It also underscores why post-cessation safety monitoring matters for therapies that modify epigenetic machinery.
Sources
This item is grounded in the FDA's formal alert on Tazverik withdrawal, OncLive and Medscape reporting, and Ipsen's March 9, 2026 announcement.
Novo Nordisk deploys a governed AI reasoning agent for clinical data analysis, cutting exploratory analysis from weeks to minutes
What changed
Novo Nordisk, working with Microsoft's AI Acceleration Studio, developed and deployed a governed reasoning agent on Azure designed specifically for pharmaceutical R&D. The system allows researchers and medical experts to independently explore hypotheses against Novo Nordisk's proprietary clinical datasets using AI-generated code execution and statistical analysis with human validation built into the process from the beginning. The company reports that the system reduces time to insight from weeks to minutes for many exploratory analyses while increasing the number of scientific questions teams can evaluate. Novo Nordisk is now expanding the solution beyond its initial deployment into trial design optimization, portfolio intelligence, and additional decision-support workflows.
Why it matters
This is not a generic AI chatbot deployed for efficiency. It is a purpose-built reasoning agent operating inside a regulated pharmaceutical R&D environment with governance, compliance, and human oversight designed in from the start. The significance is in the specificity: the agent executes statistical code against real clinical data, which means it is doing the analytical work that previously required dedicated biostatisticians and weeks of queue time. For the broader industry, Novo Nordisk's deployment establishes a reference architecture for how AI agents can operate inside pharmaceutical R&D without compromising regulatory rigor. The roadmap includes expanding from descriptive analytics into predictive trial design and portfolio intelligence.
Takeaway
The pharma industry's AI adoption is moving from conversational tools to reasoning agents that execute quantitative analysis against proprietary clinical data. Novo Nordisk's deployment shows how this can work inside regulatory constraints. The next frontier is predictive: agents that do not just answer questions about past data but help design the next trial.
Sources
This item is grounded in the Microsoft customer story on Novo Nordisk's Azure deployment.