iFeed Weekly: the evidence architecture shift.
Nine signals for the week of 11–17 May 2026. FDA pilots real-time clinical trials with AstraZeneca TRAVERSE and Amgen STREAM-SCLC, declares one pivotal trial the new default, launches one-day inspectional assessments. EU opens AI Act Article 50 transparency consultation, updates its responsible-GenAI-in-research guidelines, reaches political agreement on the Critical Medicines Act. WHO sizes the climate footprint of clinical trials. FDA and EMA align on 10 Good AI Practice principles. Vanderbilt shows LLMs can detect immune-related AEs from clinical notes — with a notable over-prediction caveat.
FDA announced two major steps on 28 April 2026: successful initiation of two proof-of-concept real-time clinical trials (RTCTs), and a Request for Information on a broader pilot program — AI-Enabled Optimization of Early-Phase Clinical Trials — launching this summer.
FDA met with each sponsor on the establishment of criteria for reporting signals in real time. The agency has received and validated signals for AstraZeneca's trial through Paradigm Health, establishing the feasibility of the technical framework required for real-time signal sharing.
"Real-time trials have been talked about for years. We demonstrated that it is not only possible, but also potentially transformative." — Jeremy Walsh · Chief AI Officer, FDA
- Remain ultimately responsible for scientific output. Researchers accountable for content; aware of AI limitations (bias, hallucinations, invented citations).
- Use generative AI transparently. Detail substantially-used tools, account for stochastic nature, aim for reproducibility, disclose limitations.
- Pay particular attention to privacy, confidentiality and IP rights when sharing sensitive information with AI tools.
- Trial planning + design with sustainability principles
- Reduce long-distance travel for feasibility visits
- Improve local trial capacity when selecting sites
- Optimise logistics; reduce packaging
- Conduct virtual trial meetings when appropriate
- Sample + data collection aligned to principle of sufficiency
- Warming up ultra-low-temperature freezers when feasible
- Reducing on-site monitoring + in-person audits where possible
- Reduce plastic waste
- Repurpose / repair equipment
- Adapt archiving length to risk level
- Require environmental-consideration disclosure in publications
Elizabeth Miller, Pharm.D., FDA Associate Commissioner for Inspections and Investigations: the agency is "closely analyzing the operational and compliance data from these assessments — including trends in outcomes, risk signals, and investigator feedback." FDA is developing evaluation metrics on inspection duration, escalation rates, and utility of findings.
FDA + EMA define AI as "system-level technologies used to generate or analyze evidence across the drug product life cycle" — covering nonclinical, clinical, post-marketing, and manufacturing.
The models showed a systematic bias toward overpredicting immune-related adverse events. Operationalising this methodology for pharmacovigilance requires triage workflows that anticipate false-positive load, not just precision/recall claims.
Why this matters in the bioanalytical / PV intersection: the work tests LLMs against both real-world EHR notes (Vanderbilt + UCSF) and industry-sponsored clinical-trial notes (Roche ICI trials, 272 patients). The cross-dataset variation (F1 56–66%) and the over-prediction bias are the operative findings for anyone considering LLM-on-notes as a PV signal-detection layer.
One pivotal trial as default — not exception
Single-trial pathway has always existed for breakthrough · accelerated · priority review. FDA leadership (Makary + Prasad) now positions it as the default for novel-product approval — published in NEJM, Feb 2026. Hummel (Cytel): "modern drug development provides much more clarity on a precise mechanism of action… supporting statistical with biologic inference."
FDA Bayesian Methodology Draft Guidance
Use of Bayesian Methodology in Clinical Trials of Drug and Biological Products · CDER + CBER · 12 Jan 2026 · Docket FDA-2025-D-3217. "The primary focus is on the use of Bayesian methods to support primary inference in clinical trials intended to support the effectiveness and safety of drugs."
Cytel's Savina Jaeger on the Bayesian guidance: "The strategic question is not whether Bayesian methods are acceptable in principle — it is whether the efficiency gains justify the additional complexity and review burden for your specific program."
- NEJM · Makary + Prasad · Feb 2026 · nejm.org → (paywalled)
- FDA Bayesian draft guidance · CDER + CBER · 12 Jan 2026 · fda.gov →
- Cytel · Jürgen Hummel · 26 Feb 2026 · cytel.com →
- Cytel · Savina Jaeger · 5 Feb 2026 · cytel.com →
This week's nine signals point to one system-level movement: regulators on both sides of the Atlantic are rewriting the evidence architecture of drug development.
FDA Real-Time Clinical Trials (TRAVERSE + STREAM-SCLC), FDA's single-pivotal-trial default (NEJM Feb 2026) and FDA's draft Bayesian guidance (Jan 2026) collapse the evidence stack from "two adequate and well-controlled trials" toward "one pivotal + Bayesian inference + real-time signal sharing." FDA + EMA's 10 Good AI Practice principles and EU AI Act Article 50 set the cross-Atlantic AI baseline. WHO sizes the environmental footprint with hard numbers. Vanderbilt's LLM-PV shows the methodology layer that makes real-time review scalable — and honestly flags its over-prediction bias. EMA's Critical Medicines Act opens a new EU supply-chain regime. FDA's One-Day Inspectional Assessments add a targeted surveillance tool for lower-risk facilities.
The integration question for W20: which organisations can adapt their evidence-generation, AI-governance, and operational design before the regulator and the procurement function start asking?
Snapshot as of 18 May 2026. Every claim on this page is anchored to a primary source actually read (PDF) or content-verified (WebFetch / direct curl). iFeed publishes synthesis and methodology — not legal advice, not a substitute for full guideline reading or your own SOPs and validation plans. Always confirm against the latest official texts for your submission. Cross-domain regulatory intelligence — bioanalytical · bioequivalence · clinical trials · governance · AI in healthcare — in one place. This is the constant. Every Monday.