The paper, published on 18 August 2026 in Frontiers in Drug Safety and Regulation by Leihong Wu, Joshua Xu, Oanh Dang, and Robert Ball, draws its provocative framing from history. The authors invoke Francis Fukuyama’s famous “end of history” thesis from the close of the Cold War, which borrowed Hegel’s dialectical process of thesis, antithesis, and synthesis. In their analogy, the historical thesis of pharmacovigilance is the human expert who reads narrative case reports and applies introspective clinical judgment. The contemporary antithesis is the claim that drug safety surveillance should be fully computable, independent of human reviewers. The emerging synthesis is the human-AI pharmacovigilance system, in which neither pole alone is sufficient.

The stakes are enormous. Pharmacovigilance revolves around individual case safety reports, or ICSRs, submitted in vast volumes by healthcare professionals, patients, and pharmaceutical companies. Each report contains structured fields such as age, sex, suspect drug, and outcome, alongside sprawling free-text narratives of variable quality. Reviewers must verify completeness, detect duplicate reports, assign priority, and integrate information across cases and external data sources to identify potential safety signals. Back in 2022, Ball and Gregory Dal Pan asked whether artificial intelligence for pharmacovigilance was “ready for prime time” and concluded that while machine learning could support parts of the workflow, full automation of case processing was not realistic. Human review remained essential, and the field lacked a “pharmacovigilance cognitive framework,” a computable description of the rules of inference that safety reviewers apply when judging whether a drug likely caused an adverse event.

The new paper asks whether large language models, trained on massive text corpora to summarize, extract, and generate language, have changed that conclusion. The authors’ answer is a framework they call “computable pharmacovigilance,” in which tasks are evaluated for their computational tractability. A task is considered computable when it can be translated into explicit, auditable computational steps with clearly defined inputs and outputs, processing details precise enough for traceable execution, and outputs that can be judged by reliable, reproducible criteria. By that standard, computability is a spectrum, shaped by input standardization, output verifiability, tolerance for error, auditability, reliance on clinical judgment, and the regulatory consequences of mistakes.

By this lens, many routine tasks are increasingly ripe for automation. Completeness checks that identify missing minimum report elements, detection of duplicate reports, and structured information extraction from narratives all have well-defined requirements and fixed parameters. Real-world results support this. The ALIGN system, a compositional large language model architecture that generates candidates and then self-evaluates them, achieves 87 to 90 percent accuracy in MedDRA medical coding. Deduplication tools using probabilistic record linkage on structured and unstructured data can reduce manual workloads by up to 60 percent, and these tools have seen relatively high adoption because their algorithms transparently mirror existing human cognitive processes.

At the other end of the spectrum sits case-level causality assessment, the judgment of whether a specific drug actually caused a specific adverse event. The authors argue this remains profoundly difficult to formalize. It demands integration of temporality, rechallenge information, and alternative etiologies against a background of pharmacological mechanism, along with abstract reasoning whose structure depends on the particular drug-event pair in question. Human adjudicators themselves reach full agreement on binary causality classifications in only about 61 percent of cases, and current benchmarks robust enough to measure AI performance against expert judgment are scarce, complicated by the nuances of individual drug-event combinations and the continual emergence of rare or novel adverse events.

The paper also weighs the choice between “big” and “small” models, a distinction that shapes architecture and governance. Large foundation models can be repurposed across many tasks through prompting or light fine-tuning, summarizing long narratives, extracting structured fields, and even translating natural-language reviewer questions into executable database queries. But they bring hallucinations, prompt sensitivity, variable outputs, and higher validation burdens, and running them across every report in a large safety database is expensive, with external APIs raising data protection questions. Small, task-specific models, by contrast, have constrained inputs and well-defined outputs, more stable failure modes, and can be trained on internal historical safety data that commercial models cannot access, making them easier to document and audit.

Illustrating the trade-off, the authors report a pilot study benchmarking a fine-tuned BioBERT named entity recognition model against a zero-shot Llama-4 large language model on safety concept extraction from roughly 800 annotated FAERS reports and an existing VAERS corpus. The fine-tuned BioBERT outperformed the large model on both datasets, achieving an overall F1-score of 0.54 versus 0.42 on FAERS narratives and 0.53 versus 0.31 on the VAERS vaccine corpus. Error analysis showed the large language model prone to hallucinated character positions, boundary drift, and label confusion, while the BERT model behaved as a conservative, schema-consistent workhorse. The authors stress the comparison illustrates the value of task-specific fine-tuning rather than settling any definitive ranking of model capability.

Their preliminary work on causality assessment reinforces caution. When an LLM was applied to a well-documented case series, the model acted too confidently, labeling ambiguous cases as “causality assessable” where human reviewers concluded no reliable determination could be made. While AI and humans agreed on clear cases at the extremes, the machine was overly optimistic in the muddled middle, accepting cases experts would reject. The authors also caution that explainable AI techniques such as SHAP and LIME, though capable of improving trust, can increase cognitive load and misalign with established clinical reasoning, so transparency alone does not guarantee that outputs are accurate, contextually relevant, and actionable.

The authors conclude that generative AI does not signal the end of history for pharmacovigilance but the end of purely manual, labor-intensive case processing. They propose treating computability as a design principle, not an endpoint: decompose workflows into steps with established inputs and outputs, match each step to an appropriate level of automation, deploy in phased and risk-based stages, and embed AI within sociotechnical systems that preserve accountability and human control. In the hybrid architecture they envision, large models act as flexible interfaces and orchestrators calling on small models, rule engines, terminologies, and retrieval tools, while experts provide strategic and ethical oversight. Trustworthiness, they argue, is itself the core of the pharmacovigilance cognitive framework, something that for now only matters to humans and only humans can judge. The future, in their view, lies in replacing AI-averse workflows with AI-enabled ones that remain accountable to human experts.

Subject of Research: Automating pharmacovigilance workflows with generative AI and large language models under human oversight

Article Title: Does generative AI mean the “end of history” for pharmacovigilance automation? towards a framework for the future of human-AI systems

Article References: Wu, L., Xu, J., Dang, O., & Ball, R. (2026). Does generative AI mean the “end of history” for pharmacovigilance automation? towards a framework for the future of human-AI systems. Frontiers in Drug Safety and Regulation, 6, Article 1846339. https://doi.org/10.3389/fdsfr.2026.1846339

Image Credits: AI Generated

DOI: 10.3389/fdsfr.2026.1846339

Keywords: pharmacovigilance, generative AI, large language models, drug safety, individual case safety reports, causality assessment, FDA, human-AI systems, medical coding, deduplication, computability, regulatory science

The economic toll is staggering. A frequently cited estimate suggests biofilms contribute to losses approaching 5 trillion US dollars globally across healthcare, industrial, environmental and infrastructure sectors, although the review’s authors caution that this is a broad aggregate figure and should be interpreted with care. Much of the cost stems from biofouling and corrosion, but healthcare and medical applications represent a substantial share. Biofilms drive foreign-body related infections such as prosthetic joint infections and polymicrobial catheter infections, as well as non-surface associated conditions including cystic fibrosis lung disease, chronic wound infections and recurrent vulvovaginal candidiasis. The consequences include rising healthcare costs, lost working days, increased mortality and significantly diminished quality of life for patients.

What makes biofilm infections so difficult to treat is a layered arsenal of defence mechanisms. On top of conventional genetic resistance, which arises through mutation and horizontal gene transfer, a process that occurs readily in biofilms because densely packed cells are in constant close contact and can exchange DNA through conjugation, biofilms exhibit physical and physiological tolerance unique to the community lifestyle. Antimicrobials often simply fail to penetrate the protective matrix. Neutrophils, the immune system’s frontline phagocytes, can only engulf bacteria smaller than about 10 micrometres, whereas biofilms shielded beneath the matrix can exceed 100 micrometres and are effectively out of reach. Dormant, slow-growing persister cells form readily as oxygen and nutrients become depleted in mature biofilms, rendering cell-wall-targeting drugs such as beta-lactams, glycopeptides and echinocandins ineffective because their targets are simply not being synthesised. Slower cellular metabolism further dampens drug susceptibility. Overlaid on all of this is quorum sensing, a sophisticated chemical communication system in which bacteria and fungi release autoinducer molecules, such as acylated homoserine lactones and autoinducing peptides in bacteria, and farnesol and tyrosol in fungi, to coordinate gene regulation governing biofilm formation, antimicrobial resistance and virulence.

These formidable defences have fuelled an explosion of research into anti-biofilm technologies, from antimicrobial peptides and monoclonal antibodies for prosthetic joint infection to antimicrobial coatings, wound care products, matrix-degrading enzymes, bacteriophage therapies, quorum sensing inhibitors and repurposed drugs. Yet remarkably few human medicines carry a licence with an explicit anti-biofilm claim. Dornase alfa, marketed as Pulmozyme and approved by the US Food and Drug Administration, degrades extracellular DNA and is used in cystic fibrosis, a clinically biofilm-driven disease, but its formal indication is not framed in biofilm terms. Azithromycin, similarly, was never licensed with an anti-biofilm claim despite demonstrating excellent activity against Pseudomonas aeruginosa biofilms under some conditions. The gap between laboratory promise and regulatory approval is wide, and the review argues that a large part of that gap is procedural rather than scientific.

At the heart of the problem lies a reproducibility crisis. Biofilm science depends on a sprawling array of experimental models, from simple high-throughput microtitre plates and the Calgary Biofilm Device to physiologically realistic flow chambers, drip-flow reactors and fermenters. Each model embodies a compromise between mimicking real-world biofilm behaviour and maintaining reproducibility and throughput. Simple assays such as crystal violet staining of biomass are attractive for routine screening but often produce heterogeneous, poorly reproducible results that carry little weight in regulatory decision-making. More complex systems offer physiological relevance but require specialised equipment that differs between laboratories. Growth conditions, including nutrient composition, temperature, flow and shear forces, and surface properties, vary enormously depending on whether the model is meant to represent clinical infection, industrial contamination or an environmental biofilm, and even minor differences in oxygen availability or temperature can alter gene expression, metabolic activity and matrix production. Harvesting and neutralisation methods add further variability, since some antimicrobials alter biofilm adhesion or structural integrity, affecting how cells detach and are counted.

Compounding the methodological chaos is a conceptual one: there is no universally accepted definition of what a biofilm actually is. Descriptions generally involve surface-attached microbial communities embedded in self-produced matrix, but different definitions emphasise different features, from surface attachment and matrix production to community structure, spatial organisation and phenotypic characteristics. This lack of consensus thwarts the comparison of data across studies, the establishment of reference materials and the development of performance standards. The review suggests that simply reaching greater agreement on the defining characteristics of a biofilm would be an important early step towards standardisation, and proposes that biofilm maturity might be better defined through quantitative thresholds of matrix-associated biomarkers, such as carbohydrate content, extracellular DNA and matrix proteins, rather than through arbitrary incubation times like 24 or 48 hours.

Several international initiatives are attempting to impose order. The National Biofilm Innovation Centre, through its International Biofilm Standards Task Group, convenes stakeholders from academia, industry and regulation, including the USA Center for Biofilm Engineering and the Singapore Centre for Environmental Life Sciences Engineering, to collate evidence and disseminate best practice across sectors from human health to water systems. The European COST Action on Building Consensus on Biofilm Regulatory Decision Making is developing a regulatory toolbox through working groups on standardised technologies and data translation. The Biofilm Alliance networks researchers, metrologists, regulators and standardisation bodies with the same goal. Formal written standards are also emerging through a dedicated British Standards Institution committee on biofilms and products and through ASTM standards housed by the American National Standards Institute. But each effort has limitations: academic centres depend on fluctuating funding, COST Actions have fixed four-year lifespans, and neither the collaborative networks nor the standards committees can themselves define regulatory frameworks. Written standards can be inflexible, slow to publish and paywalled, and because biofilms differ fundamentally by species, environment, growth conditions and maturity, a single rigid standard risks being either too simplistic to reflect reality or too complex to reproduce. The review argues instead for performance-based standards that specify key parameters, model system type, organism, growth conditions, contact times, replication strategy and efficacy thresholds, while leaving flexibility in model selection.

Perhaps the most ambitious proposal concerns physical reference materials, the biological equivalent of the calibration weights used in metrology. A universal biofilm reference material is difficult to envisage because biofilms are living, dynamic, context-dependent systems. One approach would be biological reference biofilms, such as well-characterised strains of Pseudomonas aeruginosa grown under tightly controlled flow-cell conditions, with reference values for biomass, thickness, viability, matrix composition and cell density, though such materials would suffer from short shelf-life and species-specificity. Another would be artificial matrix-based materials composed of representative polysaccharides, proteins and extracellular DNA, offering stability and ease of distribution for calibrating imaging systems and evaluating matrix-disrupting technologies. The authors suggest the most realistic path may be analytical reference materials that standardise measurements rather than the biofilm itself. Importantly, they caution that the goal should not be a universal biofilm model, since polymicrobial communities show emergent properties unpredictable from monospecies systems, but rather fit-for-purpose methodologies underpinned by common principles for cultivation, characterisation, reporting and validation.

On the regulatory side, the picture is less bleak than developers often assume. Neither the UK’s MHRA, the European Medicines Agency nor the FDA has biofilm-specific guidance, and anti-biofilm products are instead classified implicitly within broader medicines, medical devices or combination product frameworks according to their primary function. The review argues that many perceived barriers stem from uncertainty rather than genuine obstruction, and it draws a direct parallel with therapeutic bacteriophages. That community believed regulation was incompatible with phage therapy, but engagement with the MHRA revealed that phages fit comfortably within the biological medicines framework; the agency published regulatory considerations and interpretive guidance without creating new regulation, and the EMA and European Pharmacopoeia followed with quality guidelines. The lesson is that regulators routinely evaluate innovative products and welcome early scientific advice, which benefits both parties through horizon scanning. The authors sketch a roadmap: within one to three years, consensus on biofilm definitions, better reporting standards and candidate reference materials; within three to seven years, harmonised performance-based standards and regulatory guidance on evidence for anti-biofilm claims; and within seven to ten years, validated biofilm-specific endpoints embedded in product evaluation worldwide. Achieving that will require sustained collaboration among researchers, industry, metrology institutes, standards bodies and regulators, but if it succeeds, the trillion-dollar microbial cities that currently defeat our antibiotics may finally face a coordinated defence.

Subject of Research: Standardisation and regulatory science challenges for antibiofilm research and innovation

Article Title: Unravelling biofilm complexity: a regulatory science perspective

Article References: Unravelling biofilm complexity: a regulatory science perspective. (n.d.). https://doi.org/10.3389/fbrio.2026.1886299

Image Credits: AI Generated

DOI: 10.3389/fbrio.2026.1886299

Keywords: biofilms, regulatory science, antimicrobial resistance, standardisation, reproducibility, extracellular matrix, quorum sensing, reference materials, MHRA, bacteriophage therapy, cystic fibrosis, medical devices

The problem ARTEMIS addresses is familiar to anyone who has watched a large language model answer a specialized scientific question with confident fluency and quietly wrong facts. When the researchers asked a base model to produce analytical reports on monoclonal antibodies, the 3Rs framework of Replacement, Reduction and Refinement, and animal-use reduction, the output was polished but generic, untraceable to source evidence, and occasionally incorrect. In one striking example, the model asserted that the FDA had only 12 reviewers with specific alternative-methods expertise across all centers, a precise quantitative claim that was simply false. Static training data and the risk of hallucination make raw model knowledge unreliable in specialized domains, and even retrieval-augmented approaches that fetch raw passages often fall short when a question requires integrating findings scattered across dozens of studies.

ARTEMIS takes a different route: rather than handing the model raw documents, it first reorganizes the entire corpus into a hierarchical, evidence-grounded memory. The workflow operates at two linked levels. At the granular level, each source article is segmented into overlapping chunks of 2,000 characters, and each chunk is distilled by a language model into a concise observation capturing its most salient evidence. At the higher level, the system generates reflections by synthesizing observations from each chronological batch of five articles together with the immediately preceding reflection, so that new findings are always interpreted in light of previously accumulated knowledge. This recursive process builds an evolving memory that preserves both local evidence and broad patterns, and because the corpus is processed in publication order, the memory retains a timeline that supports historical analysis.

The scale of the compression is considerable. Processing 3,096 full-text articles produced 31,757 raw chunks, which were distilled into a memory of simple observations and 620 reflections. The median word counts tell the story: observations averaged just 24 words, reflections 599, and raw chunks 805. Crucially, the compression did not flatten the literature into a few oversimplified themes. Topic modeling with BERTopic found a similar number of topics in the reflections as in equally sized random samples of raw chunks, suggesting broad thematic coverage was preserved. But the reflections were far more conceptually convergent: their average inter-topic distance was roughly 0.2, compared with about 0.5 for raw-chunk samples, and they occupied a markedly smaller semantic volume in embedding space, approximately 0.02 versus 0.045 to 0.055. In other words, ARTEMIS does not merely shorten the literature; it reorganizes distributed evidence into a tighter, more coherent analytical representation.

To test whether this synthesized memory actually helps, the team assembled 30 high-level questions drawn from real analytical problems in the monoclonal antibody literature, including questions about reducing animal use and the evidence base for emerging alternative methods. Because these questions demand synthesis across studies, they are poorly served by retrieving isolated passages. For each question, answers were generated with the same Llama 3.3-70B model under different context conditions, and quality was scored on a 1-to-10 rubric by three independent LLM judges: GPT-4o, Claude 4.0 and Claude 4.6, each operating in isolated sessions to avoid conversational carryover. The results were consistent. ARTEMIS-based contexts produced complete answers for all 30 questions, whereas raw-chunk retrieval explicitly declared insufficient information in 5 of 30 cases. Across judges, ARTEMIS configurations received higher or comparable scores than raw-chunk retrieval, achieving 100 percent answerability versus 83 percent for the baseline.

The efficiency gains were just as notable. A context of 10 observations and 10 reflections represented only about 77.4 percent of the word count of 10 raw chunks yet matched or beat their answer quality, and a leaner ARTEMIS-lite configuration of 10 observations and 5 reflections, just 40.2 percent of the raw word count, still scored similarly or higher in many cases. Simply throwing more text at the problem did not help: when the team expanded retrieval to 50 raw chunks on a subset of questions, performance generally failed to improve and sometimes degraded, and on one item both approaches returned not-enough-information responses. The lesson, the authors argue, is that context structure and information density matter more than volume.

A series of component comparisons clarified why the combined design works. Observation-only contexts were often too compressed, failing to answer 16 of 30 questions with a mean score of 3.47. Reflection-only contexts were better but sometimes too abstract, with 8 insufficient responses and a mean score of 5.23, slightly below raw-chunk retrieval at 5.53. A single-pass summarization baseline, in which the top 10 retrieved chunks were flattened into one summary, performed poorly as well, with 14 insufficient answers and a mean score of 3.73, showing that one-shot compression without recursive accumulation is not the secret. The full ARTEMIS configurations topped the table with mean scores of 6.05 for ARTEMIS-lite and 6.28 for full ARTEMIS, and were at least as good as raw-chunk retrieval on 25 and 26 of the 30 questions respectively. The two layers, granular evidence and higher-level synthesis, are complementary, and neither alone suffices.

Perhaps the most intriguing capability is steerability. Because the reflection prompt can be modified, the same corpus can be organized under different analytical lenses without building separate evidence bases. The researchers generated three distinct sets of reflections from the identical monoclonal antibody literature using three expert personas: a 3Rs Ethical Reviewer, a Clinical Development Risk Strategist, and a Novel Alternative Methods Technology Analyst. Each persona produced distinct analyses emphasizing different signals and concerns, much as different human experts would. When the team probed the memories with a NAMs-specific query about organ-on-a-chip validation challenges, the NAMs persona’s reflections showed clearly higher cosine similarity at the top of the relevance ranking, confirming that the steering process enriches the memory with signals pertinent to the target lens. The framework also supports timeline-aware historical synthesis: using publication-ordered reflections as context, ARTEMIS generated an evidence-grounded narrative of Aducanumab’s development in four chronological phases, tracing how animal model usage evolved through the principles of Replacement, Reduction and Refinement.

The authors are careful about scope and limitations. LLM-as-a-Judge evaluations, they note, should be read as supportive rather than definitive; the judges differed in scoring stringency, with GPT-4o generally assigning higher marks than the Claude models, though relative rankings were consistent and repeated Claude 4.0 scoring in isolated sessions yielded identical results. Synthesis is also inherently lossy, and observations and reflections may omit low-frequency technical details that matter for some questions, which is why ARTEMIS is positioned as a support framework for complex literature analysis rather than a replacement for consulting source documents when full detail is required. The study centers on a single monoclonal antibody use case, and the team suggests future work could explore alternative corpus-ordering strategies, such as thematic grouping or citation-based progression, that might shape the analytical perspective embedded in the accumulated memory. Funded by the FDA and released as open access, the work arrives at a moment when regulatory agencies, drug developers and researchers alike are wrestling with literature too large and fast-moving for any human team to synthesize alone. If the approach generalizes beyond this first domain, the idea of building reusable, steerable, evidence-grounded memories from scientific corpora could become a standard layer between the world’s papers and the models asked to make sense of them.

Subject of Research: A memory-based AI framework for generating reusable analytical contexts from scientific literature for regulatory science

Article Title: ARTEMIS enables reusable analytical context generation from scientific literature

Article References: Li, D., Gong, B., Tong, W., & Xu, J. (2026). ARTEMIS enables reusable analytical context generation from scientific literature. Discover Artificial Intelligence, 6(1), Article 1465. https://doi.org/10.1007/s44163-026-02485-5

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02485-5

Keywords: ARTEMIS, large language models, regulatory science, monoclonal antibodies, retrieval-augmented generation, literature mining, 3Rs, novel alternative methods, LLM-as-a-Judge, evidence synthesis, FDA, knowledge curation

Hutt joined C-Path’s Board in 2015, bringing with him decades of unparalleled leadership and expertise in food and drug law. At that point, he had already spent the better part of a lifetime shaping the very field in which C-Path operates. The institute, founded in 2005 as a public-private partnership in response to the FDA’s Critical Path Initiative, exists to lead collaborations that advance better treatments for people worldwide. Its consortia and programs currently include more than 1,600 scientists and representatives drawn from government and regulatory agencies, academia, patient organizations, disease foundations, and pharmaceutical and biotech companies. For a body like that, having a board member who had personally helped construct the statutory foundations of drug regulation was an extraordinary asset, and colleagues say Hutt used that position to push the organization toward ever more rigorous thinking about how science and policy meet.

The milestones of Hutt’s career read like a history of modern American food and drug law. From 1971 to 1975, he served as Chief Counsel for the U.S. Food and Drug Administration, a tenure during which he helped transform the agency’s approach to regulation and left a lasting imprint on the modern legal architecture of the field. That period was pivotal for the FDA, and Hutt’s innovative approach to regulation during those years cemented his reputation as a foundational figure in regulatory science. After leaving government, he did not retreat from public service; instead, he became a bridge between the agency, Congress, industry, and the academy, a role he would occupy for the rest of his professional life.

Perhaps no statistic captures his influence better than his legislative record. Over a career spanning more than six decades, Hutt served as a principal architect for most major legislation amending the Federal Food, Drug, and Cosmetic Act, the foundational statute under which the FDA regulates drugs, devices, and food. Every time Congress has modernized that law, in most cases Hutt’s hand was somewhere in the drafting, negotiation, or intellectual groundwork. He also testified before the U.S. House and Senate more than 100 times, a level of engagement with the legislative process that is almost unheard of in any field of law. Few attorneys in any discipline can claim to have shaped their area of practice so directly, across so many decades, and at so many points of leverage.

Hutt was equally committed to education and scholarship. He taught food and drug law at Harvard Law School for more than three decades, training generations of lawyers who would go on to lead regulatory agencies, pharmaceutical companies, public interest organizations, and law firms. He co-authored the leading casebook in the field, Food and Drug Law: Cases and Materials, which became the standard text through which students encountered the subject. Through that combination of classroom teaching and written scholarship, he effectively defined how the discipline is understood and practiced. He was also a longtime member of the Institute of Medicine, now known as the National Academy of Medicine, and maintained an active legal practice and teaching schedule throughout his final years, a testament to the energy he brought to his work well into the later stages of his life.

At C-Path, colleagues emphasize that Hutt’s contribution went far beyond his legal credentials. Klaus Romero, M.D., M.S., FCP, the institute’s CEO, said he had the privilege of knowing Hutt for many years and could say without exaggeration that few people have shaped the practice of food and drug law, and by extension the course of regulatory science, the way Hutt did. Romero described a board member who brought a rare combination of historical perspective and forward-looking judgment, someone who had seen regulatory science evolve across different sectors for over five decades and who used that experience to help guide C-Path through some of the most important years of its growth. According to Romero, Hutt was generous with his time, sharp in his counsel, and genuinely invested in the organization’s mission, always pushing it to think rigorously about how science and policy intersect.

Board Chairwoman Karen Bernstein, Ph.D., offered an equally striking assessment of Hutt’s legacy. She described him as a towering figure in food and drug law whose influence will be felt for generations, noting that he did more than interpret the laws governing medicines, medical devices, and food. In her view, he helped shape the modern regulatory framework itself while educating and mentoring generations of leaders who would carry that work forward. Bernstein credited Hutt with bringing extraordinary intellectual rigor, historical perspective, and a deep commitment to public health to everything he did, and said C-Path was fortunate to benefit from his wisdom and service on its board. She expressed profound gratitude for his enduring contributions to the organization, to the FDA, and to the field of regulatory science.

The significance of Hutt’s work becomes clearer when considered against the backdrop of the institution he served. The Federal Food, Drug, and Cosmetic Act is the legal backbone of American consumer and medical protection, and the FDA’s authority to require evidence of safety and efficacy before a drug reaches patients rests on amendments to that statute. Regulatory science, the discipline of developing the tools, standards, and methods that make such oversight scientifically meaningful, depends on a legal framework that is both robust and adaptable. Hutt’s career embodied that dual demand: as a government lawyer he strengthened the agency’s regulatory approach, as a legislative architect he modernized the statute itself, and as an educator he ensured that the next generation could carry the work forward. His presence on C-Path’s board connected that long legal tradition directly to the institute’s scientific collaborations, which aim to speed the development of better treatments through data sharing, biomarker qualification, and other science-driven approaches.

C-Path itself occupies a distinctive position in the drug development ecosystem. Established in 2005 as a public-private partnership in response to the FDA’s Critical Path Initiative, the institute has grown into a globally recognized pioneer in accelerating drug development, with international consortia and programs spanning regulators, academia, patient advocates, and industry. Its headquarters in Tucson, Arizona, and its European subsidiary in Amsterdam, Netherlands, anchor a network of dedicated team members located throughout the world. The organization is supported by the Food and Drug Administration of the Department of Health and Human Services, which funds 43 percent of its work, totaling $20,724,703, with the remaining 57 percent, totaling $27,346,613, coming from non-government sources. The institute notes that its contents do not necessarily represent the official views of, nor an endorsement by, FDA/HHS or the U.S. Government. Within that structure, Hutt’s decade on the board provided legal and historical ballast for an organization whose work depends on navigating exactly the regulatory terrain he helped build.

Tributes from across C-Path reflect a consistent theme: gratitude. The institute has extended its deepest condolences to Hutt’s family, friends, and colleagues during this time of loss. For the broader community of regulatory scientists, drug developers, and public health advocates, his passing represents the loss of a singular figure, someone who stood at the junction of law, science, and medicine for more than sixty years and who shaped each of them in turn. From his years as FDA Chief Counsel in the 1970s, through his decades at Harvard Law School and his role in drafting the nation’s most important health legislation, to his final years of active practice and service on the C-Path board, Peter Barton Hutt leaves behind a regulatory landscape that bears his imprint at nearly every turn. The organization he served for a decade says it is grateful for everything he gave, and the field he helped create will continue to feel his influence for generations to come.

Subject of Research: The passing of food and drug law pioneer and Critical Path Institute board member Peter Barton Hutt

Article Title: C-Path mourns the passing of Board Member Peter Barton Hutt

Article References: C-Path mourns the passing of Board Member Peter Barton Hutt. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Peter Barton Hutt, Critical Path Institute, FDA, food and drug law, regulatory science, Federal Food Drug and Cosmetic Act, Harvard Law School, drug development, public health, C-Path Board of Directors, obituary, pharmaceutical regulation

Brenda Betancourt, a Term Associate Professor of Statistics in George Mason’s College of Engineering and Computing, has received $286,130 from the U.S. Food and Drug Administration for a project titled Regulatory Grade Frameworks for Assessing Record Linkage Uncertainty in Real-World Evidence. The funding began in September 2026 and runs through September 2028. Over those two years, Betancourt will evaluate how uncertainty in record linkage affects the reliability, interpretability, and evidentiary confidence of analyses based on linked real-world data that are used to support FDA regulatory decision-making. In other words, the project asks a deceptively simple question: if the links between records are not certain, how certain can the conclusions drawn from them be?

The stakes are considerable. Record linkage is the invisible plumbing of real-world evidence. When a researcher wants to know whether patients exposed to a medication later experienced a particular adverse event, the exposure records typically live in one database, perhaps pharmacy claims, while the outcome records live in another, perhaps a hospital discharge registry. Joining those datasets requires deciding which rows refer to the same individual, often without a shared unique identifier such as a national patient number. Probabilistic matching algorithms compare names, dates of birth, addresses, and other fields, assigning scores that indicate how likely two records are to describe the same person. Every one of those assignments carries a probability of error, and those errors propagate silently into every downstream estimate.

Statisticians have long understood that linkage errors can bias results in either direction. False matches, in which records from two different people are merged, can dilute genuine signals or manufacture spurious associations. Missed matches, in which records from the same person remain split, can fragment a patient’s medical history and undercount exposures or outcomes. The direction and magnitude of the bias depend on the nature of the analysis, the matching variables, and the population being studied. What has been missing, particularly in the regulatory context, is a standardized framework that quantifies this uncertainty in a way that meets the evidentiary bar regulators demand when a decision affects public health.

That is the gap Betancourt’s project is designed to fill. The phrase regulatory grade in the project title signals the ambition: not merely to measure linkage uncertainty in an academic sense, but to develop frameworks rigorous and transparent enough to be used in submissions that inform FDA decisions. Regulatory science operates under constraints that academic research often does not. Methods must be reproducible, their assumptions explicit, their outputs interpretable by reviewers who must weigh evidence under statutory standards. A framework that cannot communicate how much of a reported effect might be attributable to linkage error is of limited use to an agency deciding whether to approve, restrict, or warn about a therapy.

The mathematical core of the problem is rich. Probabilistic linkage models, descendants of the classical Fellegi-Sunter framework, estimate the likelihood that a pair of records matches based on agreement and disagreement patterns across comparison fields. More modern approaches bring in Bayesian modeling, clustering, and machine learning classifiers, each producing not a single deterministic assignment but a posterior distribution over possible linkages. The challenge is propagation: how does uncertainty in those assignments flow through data cleaning, cohort construction, statistical modeling, and ultimately into confidence intervals and p-values for the regulatory question at hand? Ignoring that propagation can yield analyses that appear far more precise than they actually are, a phenomenon statisticians describe as understated uncertainty.

Several strategies exist in principle for handling this propagation. One is to condition on the most probable linkage and then adjust estimates using sensitivity analyses that vary the assumed error rates. Another is multiple imputation over plausible linkages, generating several alternative versions of the linked dataset and combining the resulting estimates so that the final inference reflects linkage ambiguity. A third embeds the linkage model and the analysis model in a joint Bayesian framework, allowing uncertainty to be integrated out rather than fixed. Each approach carries computational costs and modeling assumptions, and each performs differently depending on data quality, record volume, and the discriminative power of the matching fields. A central task for the project is evaluating when these methods deliver trustworthy inference and where they fall short.

The timing of the work reflects a broader shift in how evidence reaches regulators. The FDA has invested heavily in real-world evidence programs, including the RWE framework established under the twenty-first Century Cures Act, which directed the agency to clarify how real-world data can support approvals and label changes, particularly in areas such as rare diseases, oncology, and post-market safety surveillance where randomized trials are impractical or unethical. As the volume of linked real-world analyses grows, so does the need for standards governing every step of the pipeline. Linkage is among the earliest and least visible of those steps, which makes its errors especially insidious: they are baked in before most analysts ever see the data.

For patients and clinicians, the practical payoff of this research is confidence. If a safety signal emerges from linked claims and hospital data, clinicians and patients deserve to know whether that signal could plausibly be an artifact of mismatched records. Conversely, if a real-world analysis finds no elevated risk, regulators should know how robust that null finding is to plausible linkage errors. Quantified linkage uncertainty turns these questions from matters of intuition into matters of measurement, allowing reviewers to weigh evidence with a clearer sense of its fragility or strength. In an era when real-world evidence increasingly shapes drug labels, coverage decisions, and clinical guidelines, that clarity is not a technical luxury but a foundation of trustworthy medicine.

The project also highlights the growing role of statisticians in regulatory science and the position of universities near the federal policy apparatus in shaping it. George Mason, Virginia’s largest public research university, enrolls more than 40,000 students and sits near Washington, D.C., placing its researchers in close proximity to the agencies whose standards define the evidentiary landscape. Over the next two years, Betancourt’s work will contribute to a question that touches every linked dataset behind every regulatory submission: how sure can we be that the records we joined belong together, and how should that residual doubt be reflected in the evidence we act upon. The answer may help determine how much trust real-world evidence ultimately earns.

Subject of Research: Quantifying record linkage uncertainty in real-world evidence for FDA regulatory decision-making

Article Title: Betancourt studying regulatory grade frameworks for assessing record linkage uncertainty in real-world evidence

Article References: Betancourt studying regulatory grade frameworks for assessing record linkage uncertainty in real-world evidence. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: record linkage, real-world evidence, FDA, regulatory science, statistics, data uncertainty, George Mason University, probabilistic matching, pharmacovigilance, electronic health records, Bayesian methods, drug safety

The drug at the center of the study, roxithromycin, is a macrolide antibiotic related to erythromycin, used against respiratory tract and other bacterial infections. Like many macrolides, it is given orally and its absorption into the bloodstream depends on a chain of events that includes dissolution of the tablet in the gastrointestinal tract, permeation across the intestinal wall, and distribution into tissues. Any change in the formulation, the manufacturing process, or even the excipients used to bind and coat the tablet can in principle alter how quickly and completely the drug reaches the circulation. That is why regulators require bioequivalence testing whenever a generic version or a modified formulation is brought to market. The standard metrics are the area under the concentration-time curve, which reflects total drug exposure, the maximum plasma concentration, which reflects peak exposure, and the time at which that maximum occurs. If the ratios of these parameters between test and reference products fall within a ninety percent confidence interval of eighty to one hundred twenty-five percent, the products are considered bioequivalent.

Conducting such trials in humans is expensive, slow, and ethically burdensome, since volunteers are exposed to a drug without any personal therapeutic benefit. This is where physiologically based pharmacokinetic modeling, or PBPK, enters the picture. Unlike simpler empirical models that treat the body as a set of mathematical compartments with fitted parameters, a PBPK model attempts to represent the actual anatomy and physiology of the body: organs and tissues are connected by blood flows, and drug movement between them is governed by physicochemical properties such as solubility, permeability, lipophilicity, and protein binding, together with physiological parameters like organ volumes, blood flow rates, and enzyme abundances. Because the underlying structure reflects real biology, a well-constructed PBPK model can be adapted to different populations, doses, and even formulations simply by swapping in the relevant inputs, making it a powerful tool for predicting drug behavior in scenarios that have never been tested clinically.

The research team, led by Danfeng Yu, Qin Li, and colleagues including corresponding authors Jiyu Chen and Yan He, assembled roxithromycin’s physicochemical, in vitro, and in vivo parameters from established databases and the published literature, and constructed an adult PBPK model using the open-source software platform PK-Sim. The initial model was developed and calibrated against clinical pharmacokinetic data from a 150 milligram dose of the reference formulation. The critical test of any model, however, is not how well it fits the data used to build it, but how well it predicts data it has never seen. To address this, the researchers performed a comparative evaluation using clinical data from the test formulation, checking whether the model could reproduce concentration-time profiles for a product whose behavior had not informed the model’s construction.

To quantify the agreement between prediction and observation, the team applied the mean fold error approach, a standard metric in pharmacokinetic modeling that measures the ratio of predicted to observed values. For the three key parameters, the area under the curve from zero to the last measurable time point, the maximum concentration, and the time to maximum concentration, more than ninety percent of the mean fold error ratios fell within the range of 0.5 to 2. This means that for the overwhelming majority of comparisons, the model’s predictions were within a factor of two of the measured values, and in most cases considerably closer. With the model thus validated, the researchers imported formulation-specific in vitro dissolution data for both products, allowing the simulations to distinguish between the two tablets based on how rapidly each released its drug payload in the digestive tract.

The virtual bioequivalence trial itself was conducted in a simulated cohort of ninety-nine East Asian subjects, a population chosen to match the demographic context in which the formulations are intended to be used. Virtual subjects in PBPK platforms are not identical clones; the software samples from realistic distributions of physiological parameters such as body weight, gastric emptying time, and intestinal transit, so that the simulated population exhibits the same variability that a real trial population would. Each virtual subject received both formulations, and the resulting pharmacokinetic profiles were analyzed exactly as they would be in a conventional bioequivalence study. The outcome was unambiguous: the ninety percent confidence intervals for the geometric mean ratios of the primary pharmacokinetic parameters all fell within the regulatory acceptance range of eighty to one hundred twenty-five percent, indicating that the two formulations would be judged bioequivalent.

The implications of this result extend well beyond a single antibiotic. Bioequivalence studies are among the most numerous clinical trials conducted worldwide, required for every generic approval and for many post-approval manufacturing changes. If validated PBPK models can reliably substitute for some of these trials, the savings in cost, time, and human exposure could be substantial. Regulators, including the United States Food and Drug Administration and the European Medicines Agency, have already issued guidance on when modeling and simulation can support or replace certain clinical studies, and virtual bioequivalence is an active area of regulatory science. The present study adds to a growing body of evidence that, for drugs with well-characterized pharmacokinetics and formulations whose dissolution behavior can be measured reliably in vitro, a computer can indeed stand in for the clinic.

The study also illustrates the methodological discipline that makes such virtual trials credible. The researchers did not simply assume the model was correct; they built it from one dataset, tested it against an independent dataset from a different formulation, and reported a formal quantitative measure of predictive performance before trusting it for the bioequivalence question. The work was registered retrospectively at ClinicalTrials.gov under number NCT06798051, the underlying clinical study was approved by the Medical Ethics Committee of the Affiliated Hospital of Guizhou Medical University, and all participants in the source trials provided informed consent. The modeling was funded by the Science and Technology Planning Project of Guizhou Province and related institutional programs, and the authors declared no competing interests.

Certain caveats remain. A virtual trial is only as good as its underlying model, and PBPK predictions for drugs with complex metabolism, food effects, or narrow therapeutic windows demand even more extensive validation than was possible here. The ninety-nine subject cohort, while adequate for this comparison, is a simulation of a specific population rather than a demonstration across all patient groups. Nevertheless, the study offers a concrete template for how virtual bioequivalence can be conducted and documented: assemble the drug’s parameters, build and validate the model against real data, import formulation-specific dissolution profiles, and simulate a demographically realistic cohort under the same statistical criteria that govern human studies. As more models of this kind accumulate and are independently verified, the day when a generic drug can win approval on the strength of a computer trial, with human studies reserved for cases the models cannot yet handle, moves measurably closer.

Subject of Research: Virtual bioequivalence assessment of two roxithromycin formulations using a physiologically based pharmacokinetic model

Article Title: Virtual bioequivalence assessment for two roxithromycin formulations using a physiologically based pharmacokinetic model

Article References: Yu, D., Li, Q., Zeng, L., Liao, M., Lu, X., Chen, J., & He, Y. (2026). Virtual bioequivalence assessment for two roxithromycin formulations using a physiologically based pharmacokinetic model. BMC Pharmacology and Toxicology. https://doi.org/10.1186/s40360-026-01244-7

Image Credits: AI Generated

DOI: 10.1186/s40360-026-01244-7

Keywords: roxithromycin, PBPK modeling, virtual bioequivalence, PK-Sim, pharmacokinetics, generic drugs, drug dissolution, macrolide antibiotics, clinical trial simulation, regulatory science, East Asian population, BMC Pharmacology and Toxicology

The review, conducted by Abir Hamze and Emran Alotaibi of Canadian University Dubai, is structured as a PRISMA-ScR-informed evidence map rather than a conventional narrative survey. The authors searched bibliographic databases, publisher platforms, Google Scholar and citation-chasing sources, identifying 134 corpus entries before duplicate removal. After screening, 126 unique records formed a broad map of AI applications to PFAS, spanning detection, environmental fate, treatment, exposure and toxicity. The team then applied a stricter, endpoint-based reclassification to isolate the toxicological core of the literature. Of 46 records initially coded as toxicity or human and ecological health, only 39 survived as genuine AI-PFAS toxicology studies, and detailed full-text extraction was possible for 35 of them. The distinction matters: many studies that mention PFAS and health in the same breath are actually modeling environmental occurrence or exposure classification, not biological hazard.

Within that core subset, the methodological landscape is strikingly heterogeneous. Molecular simulation combined with machine learning emerged as the family producing the strongest-validated examples, while QSAR, QSPR and read-across studies, the traditional workhorses of computational toxicology, often clustered at intermediate validation levels. Tree ensembles, boosted-tree algorithms, Bayesian probabilistic models, neural networks and graph-based approaches all appear in the literature, each suited to different tasks. The authors emphasize that no algorithm is inherently decision-ready: a random forest may serve tabular biomonitoring data well if externally validated, a graph neural network may excel at structure-based prediction when scaffold-split validation is reported, and Bayesian models are particularly attractive for sparse exposure data because they make uncertainty explicit rather than hiding it behind a single deterministic number.

One of the review’s most valuable contributions is its grounding of AI prediction in mechanistic toxicology. Drawing on recent specialist reviews, the authors identify several biological anchors that any credible PFAS model must respect. Reproductive toxicity work points to reactive oxygen species as a molecular initiating event that cascades through oxidative stress, mitochondrial dysfunction, steroidogenic disruption and thyroid-hormone changes, ultimately impairing gametogenesis and fertility. Transporter-focused research highlights serum albumin, solute carrier and ATP-binding cassette transporters, and nuclear receptors as critical but underdeveloped determinants of where PFAS go in the body. Early-life studies identify the placenta as a sensitive target that may mediate latent disease decades after exposure. Omics research, mixture toxicology and underrepresented endpoints such as neurotoxicity and ocular toxicity round out the picture. An AI model whose predictions contradict these established pathways, or whose feature-importance scores ignore chain length, functional group or protein-binding chemistry, should raise immediate red flags.

The clearest practical role for AI, the authors argue, is prioritization. It is simply impossible to test thousands of PFAS comprehensively in animal studies or in vitro batteries, so QSAR models, read-across frameworks, molecular descriptors and generative-AI workflows can triage which compounds and endpoints deserve earlier experimental attention. But the review insists this only works as a triage tool, never as a substitute for evidence. Before any such model is built, chemical data curation must come first: structure standardization, salt stripping, explicit handling of acids, salts, branched isomers and tautomers, duplicate removal, harmonization of chemical identifiers and careful treatment of missing values. Because the same PFAS substance may appear in databases as an acid, a salt, a precursor or a transformation product, inconsistent representation can silently distort descriptor values and inflate apparent model performance.

Validation design receives particularly sharp criticism. The authors show that random train-test splitting, the default in much of the machine learning literature, can seriously overstate performance for PFAS because structurally similar homologues, samples from the same laboratory or geographically clustered cohorts can end up in both training and test sets. More informative alternatives include scaffold splits, leave-family-out or leave-subclass-out validation, temporal and spatial validation, and independent-cohort testing for biomonitoring models. Applicability-domain assessment should likewise move beyond vague disclaimers, using tools such as Williams plots, distance-to-model measures, conformal prediction and ensemble disagreement to specify whether a prediction actually applies to long-chain legacy compounds, short-chain acids, ether PFAS, polymers or specific biological matrices. These requirements align directly with the OECD’s long-standing QSAR validation principles, which the authors argue apply with full force to modern deep learning and graph-based models.

The audit of the broader 126-record corpus quantifies how far the field still has to go. Using the full map as denominator, the authors found that 73 records, or 57.9 percent, showed no or limited external validation; 49 records, or 38.9 percent, lacked clear experimental or field validation; 41 records, or 32.5 percent, reported no formal uncertainty or applicability-domain analysis; and 28 records, or 22.2 percent, had no clearly available code. Bias compounds these gaps in specific directions: training data concentrate heavily on PFOA, PFOS and a few legacy analogues; available assays cover hepatotoxicity and endocrine activity far better than immune, developmental, neurological or reproductive endpoints; and monitoring data cluster in known hotspots and wealthy jurisdictions, leaving children, pregnant people, occupational groups, private-well users and under-monitored communities underrepresented.

On the exposure side, the review acknowledges genuine promise alongside real peril. Machine learning models that predict which adults carry high PFAS burdens, or that integrate biomonitoring with environmental data, can help target confirmatory sampling, drinking-water treatment, dietary guidance and public-health follow-up. Yet serum concentrations reflect a tangle of drinking water, diet, dust, occupation, geography, renal function, pregnancy and protein binding, and a model may classify high exposure accurately while learning proxies for monitoring intensity rather than true exposure determinants. Interpretable association studies linking PFAS to diabetes and cardiovascular-kidney-metabolic outcomes illustrate both the analytical potential and the danger of causal overinterpretation. Mixture context adds another layer: because humans encounter shifting PFAS cocktails rather than isolated compounds, the authors argue that regulatory-grade models must support cumulative assessment and propagate uncertainty, with Bayesian frameworks offering a natural fit.

To translate these findings into practice, the review proposes a tiered regulatory-readiness framework that scales evidence requirements to decision consequence. Exploratory screening can tolerate uncertainty as long as it triggers further testing. Prioritization demands fit-for-purpose validation, descriptor transparency and domain checks. Supporting evidence requires external or biological validation, mechanistic plausibility and reproducible workflows. Decision-critical uses, such as regulatory read-across, testing waivers, restrictions, cleanup priorities or public-health communication, demand independent validation, calibrated uncertainty, explicit applicability domains, audit trails, expert review and stakeholder transparency. The authors also propose a minimum reporting checklist covering data provenance, chemical identifiers, validation, uncertainty, interpretability and code availability, and they sketch a closed-loop workflow in which AI generates hypotheses, human-relevant new approach methodologies such as organoids, receptor assays and omics platforms test them, and results feed back to update models and shrink uncertainty.

The review’s bottom line is neither technophobic nor uncritically enthusiastic. AI can genuinely accelerate PFAS hazard identification, bioactivity prediction, toxicokinetic inference, protein-binding assessment and ecological screening at a pace conventional testing cannot match. But prediction should never be mistaken for proof: a docking result suggests binding without demonstrating harm, an omics signature shows association rather than mechanism, and a low predicted risk is not evidence of safety. The authors are explicit about what AI must never be used to do, including declaring a compound safe because it falls below a predicted threshold, waiving testing outside a model’s domain, or treating absent monitoring as absent exposure. With transparent reporting, mechanistic anchoring, equity-sensitive validation and human oversight, they conclude, artificial intelligence can make PFAS toxicology faster, broader and more trustworthy, supporting rather than replacing the measurement, biological validation and expert judgment on which environmental-health protection ultimately depends.

Subject of Research: Artificial intelligence applications in PFAS toxicology and environmental-health risk assessment

Article Title: Artificial intelligence for PFAS toxicology and risk assessment

Article References: Hamze, A., & Alotaibi, E. (2026). Artificial intelligence for PFAS toxicology and risk assessment. Discover Artificial Intelligence, 6(1), Article 1291. https://doi.org/10.1007/s44163-026-02266-0

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02266-0

Keywords: PFAS, artificial intelligence, machine learning, toxicology, QSAR, read-across, risk assessment, biomonitoring, adverse outcome pathways, uncertainty quantification, regulatory science, environmental health

The origins of device regulation in Europe are unusual, and arguably the root of many current problems. In the United States, legislation controlling pharmaceutical safety followed deaths from sulfanilamide syrup in 1937, and the first European Economic Community directive on medicines was a direct response to the thalidomide disaster of 1957 to 1962. American device legislation likewise grew out of a 1970 expert report chaired by Theodore Cooper, which attacked an unquestioning acceptance of safety claims unsupported by scientific fact. Europe’s device rules, by contrast, were not precipitated by any scandal and were not designed after objective analysis. They emerged from requests to the European Commission by manufacturers’ trade associations, and when the first Medical Device Directive was drafted in the late 1980s, policymakers applied the so-called New Approach adopted by the Council of Ministers in 1985, a framework intended to harmonise standards for industrial products across the common market, not to safeguard patients.

Under that New Approach, the evaluation of evidence demonstrating that a device conformed with requirements was outsourced to independent, private inspection companies known as notified bodies. Those organisations remain central to the system today: manufacturers must submit technical documentation, quality management systems and clinical evaluation reports to a notified body for all medium-risk Class II and high-risk Class III devices, and for nearly all in vitro diagnostics, before receiving the certificates of conformity required for CE-marking and market access. With hindsight, the authors argue, applying an industrial harmonisation model to high-risk medical devices was a poor decision. Notified bodies effectively act as quasi-regulators, deciding whether devices are approved, yet as private companies operating in a competitive marketplace they are not bound by European requirements for public access to documents. Their dual role also limits the advice they can give to the very companies whose products they assess, because of inherent conflicts of interest, and oversight by national authorities has often been inadequate.

The original directive’s rules were deliberately written in rather abstract terms, leaving manufacturers to decide for themselves what clinical studies were needed. Concerns accumulated over decades: standards varied between notified bodies, some devices reached patients with insufficient evidence, and serious complications and deaths were reported from high-risk implantables approved on claims of equivalence to existing devices without any new clinical investigation. A public consultation launched by the Commission in 2008 eventually produced the Medical Device Regulation, which came into force in 2021, alongside the In Vitro Diagnostic Medical Devices Regulation. The MDR aimed to raise standards of evidence for high-risk devices, but its central requirement remained vague, asking only for clinical data providing sufficient clinical evidence. Common specifications, which would have set mandatory scientific standards for particular device types, were never realised, apparently due to limited regulatory capacity. Instead, guidance documents issued by the Medical Device Coordination Group are voluntary, interpretation varies between member states, and the type of clinical evidence required for new high-risk devices, whether observational or experimental, is still not defined.

The MDR also created unintended consequences for patients. Massive recertification requirements and unpredictable, protracted conformity assessments drove up costs, and manufacturers withdrew some essential products from the market, including well-established technologies and devices designed for children. A perception spread that Europe had become hostile to device innovation, and the Austrian National Public Health Institute is now monitoring device availability on behalf of the Commission. The scheduled five-year review of the legislation was brought forward after intense lobbying from industry and other stakeholders, culminating in the December 2025 draft regulation. At a High-Level Conference on Medical Devices: Innovation and Patient Safety held in Brussels in March 2026, European Commissioner for Health Olivér Várhelyi declared that a major overhaul was absolutely necessary, but insisted it must come without compromising on patient safety. How to reconcile those goals is the central question of the reform.

One project has supplied much of the evidence base for answering it. CORE–MD, Coordinating Research and Evidence for Medical Devices, was funded by the EU Horizon Europe programme from 2021 to 2024 and led by the European Society of Cardiology together with the European Federation of National Associations of Orthopaedics and Traumatology, with support from the Biomedical Alliance in Europe. Every stakeholder group, including national regulatory agencies and notified bodies, contributed to an extensive review of the system. The consortium confirmed deficiencies in the evidence available for high-risk devices, a dearth of randomised trials and a high risk of bias in many published studies. It proposed a progressive framework matching observational studies and controlled trials to each stage of a device’s life-cycle, and separately described alternative methodologies under the banner of large simple trials, including registry trials and pragmatic studies, which could make rigorous randomised evidence feasible at realistic cost.

The contrast with medicines regulation is instructive. The US Food and Drug Administration and the European Medicines Agency both implement scientific approaches to support regulatory decision-making, whereas for devices the EU has emphasised interpretation of legal rules applied by a patchwork of independent organisations. The authors argue that reform should set explicit standards for demonstrating clinical efficacy where needed for high-risk therapeutic implants, while retaining simpler safety and performance requirements for low-risk devices. Clinical evidence should represent women as well as men, and safe, effective technologies should be equitably accessible to all patients, including children and those with rare conditions. The in vitro diagnostics regulation needs its own tailored policies rather than a copy of the device framework: the authors call for relaxed restrictions on sharing laboratory-developed tests between health institutions, and for the abolition of IVDR Article 5.5 d, which forbids use of an in-house test if an equivalent CE-marked product exists, a rule they say stifles diagnostic innovation in favour of market monopolies.

The European Medicines Agency is set to play a much larger role. The MDR already established Expert Panels of independent clinicians and scientists who provide non-binding opinions on notified bodies’ clinical evaluations of selected high-risk devices. The Screening Panel has judged that roughly ten per cent of high-risk submissions require a full scientific opinion; sixteen have been published, most identifying deficiencies in clinical evidence, yet every reviewed device was still granted market access, prompting calls for transparency about how notified bodies act on expert recommendations. A recent analysis of thirty-four opinions found that nearly half of expert comments concerned the sufficiency of clinical evidence supporting benefit–risk determinations. Under the draft regulation, the panels’ remit will expand to advising the Commission, member states, notified bodies and manufacturers, contributing to guidance and common specifications, and designating breakthrough and orphan status for promising devices. The authors welcome this but warn that delivery is impossible without major investment: the Commission unit coordinating the regulations has only one medically qualified officer, while notified bodies employed 4,099 full-time staff in conformity assessment in 2025.

The deeper prescription is cultural as much as legislative. The authors endorse a definition of regulatory science as the use of scientific approaches to determine which policies and standards best assure safety and efficacy, so that approved technologies improve individual outcomes and public health without disproportionate over-regulation. They want regulatory science to become a recognised subspecialty for clinicians, laboratory scientists, epidemiologists, statisticians and engineers, supporting the EMA and notified bodies alike, with medical associations supplying high-quality evidence from comprehensive device registries. CORE–MD insights already shaped consensus recommendations submitted to the Commission by the Biomedical Alliance on behalf of thirty-five professional associations. Among their warnings: easing conditions for establishing equivalence would repeat a mistake that caused major problems under the old directive, and deleting the standard of sufficient clinical evidence risks more subjectivity rather than consistency. A new high-risk implantable, they argue, should not enter an existing class without a randomised trial against an active comparator or evidence from objective performance criteria. With equivalent standards promoted globally through the International Medical Device Regulators Forum, Europe has a chance to break free from recurrent legislative revisions, if it commits to funding regulatory science as a public good.

Subject of Research: Reform of the European Union regulatory framework for medical devices and in vitro diagnostics

Article Title: Empowering regulatory science for medical devices in Europe

Article References: Empowering regulatory science for medical devices in Europe. (n.d.). https://doi.org/10.1016/j.lanepe.2026.101876

Image Credits: AI Generated

DOI: 10.1016/j.lanepe.2026.101876

Keywords: medical devices, EU regulation, MDR, IVDR, notified bodies, European Medicines Agency, regulatory science, CORE-MD, clinical evidence, in vitro diagnostics, CE-marking, patient safety

The appeal of lung-on-a-chip systems is straightforward. Conventional two-dimensional cell cultures strip away the organised tissue interfaces, fluid dynamics and mechanical forces that define the respiratory microenvironment, while animal models suffer from interspecies differences that often fail to predict human responses. By combining microfabrication and microfluidic control with pulmonary cell culture, chip platforms can reproduce epithelial-endothelial interfaces, air-liquid interface culture, vascular perfusion and even breathing-related strain, features first brought together in the landmark 2010 human lung-on-a-chip by Huh and colleagues at Harvard. Since then, the design space has exploded, encompassing immune-competent models, patient-derived cells, high-containment infection systems and inhalation exposure devices.

Yet the review’s central message is provocative: more complexity does not automatically mean better science. Membrane thickness, pore size, porosity, stiffness and extracellular matrix functionalisation are not passive structural details; they actively shape transport, cell attachment, mechanical deformation and signalling between compartments. Small-pore polyester membranes, for example, favour a compartmentalised epithelial barrier, while larger pores are needed if researchers want immune cells to migrate across the interface. Barrier fidelity and leukocyte recruitment therefore demand different, sometimes conflicting, interface properties, and choosing between them depends entirely on the question being asked.

Fluidic strategy introduces a similar set of trade-offs. Perfusion delivers nutrients, removes waste, transports drug compounds and mechanically stimulates endothelial cells, but volumetric flow rate alone does not define the cellular environment, because wall shear stress also depends on channel geometry and fluid viscosity. Pump-driven systems offer precise control over flow profiles but increase dead volume, bubble risk and operational burden. Pump-free, gravity-driven rocking platforms simplify operation and reduce contamination risk in high-containment viral infection studies, at the cost of less precise control over instantaneous flow. The authors argue that studies should routinely report the flow-generation method, channel dimensions and calculated or measured shear stress so that perfusion-mediated transport can be distinguished from endothelial mechanostimulation.

Mechanical actuation, the defining capability of breathing chips, receives particularly careful treatment. The review emphasises that breathing motion is not a single condition but a family of variables spanning strain magnitude, frequency, waveform and actuation mechanism. Recent studies show these parameters matter enormously: cyclic stretch combined with airflow can accelerate mucociliary maturation of airway epithelium, and dynamic strain can modify how cells respond to inhaled toxicants. In one striking example, an orthotopic lung cancer chip demonstrated that breathing-related deformation altered tumour proliferation, invasion, dormancy and even responsiveness to tyrosine kinase inhibitors, implicating mechanical cues in epidermal growth factor receptor and MET signalling. Therapeutic response, in other words, cannot always be attributed to cellular genotype and drug exposure alone when the mechanical microenvironment itself modifies phenotype.

The same principle applies to exposure delivery and cellular composition. A next-generation breathing chip integrated with whole-cigarette-smoke exposure reported an approximately sixty percent reduction in barrier resistance and a 4.5-fold increase in interleukin-8 expression, with dynamic strain accelerating barrier disruption relative to static conditions, results that submerged chemical extracts could never reproduce. Cellular complexity, meanwhile, should be justified functionally rather than maximised blindly. An immune-competent lung-on-a-chip modelling severe influenza found that adding tissue-resident macrophages generated cytokine profiles more closely matching patient bronchoalveolar lavage data, but reproducing the broader inflammatory storm of severe infection required resident and circulating immune populations, perfusable microvasculature and a five-micrometre-pore interface that permitted immune migration. Complexity earned its place by enabling a predefined function.

Manufacturing and quality control emerge as underappreciated determinants of credibility. Small variations in channel dimensions, membrane position, bonding or actuator geometry can alter shear stress, strain transfer and barrier behaviour, and such technical variation can easily masquerade as biological variability. The review calls for predefined tolerances on critical device attributes, functional quality-control criteria including leak testing and sensor calibration, batch traceability, and documented fabrication parameters. It also cautions that switching fabrication methods, from soft lithography in polydimethylsiloxane to thermoplastic replication for scalable production, can change gas permeability, surface chemistry and compound adsorption, and should therefore trigger re-verification of critical engineering and biological endpoints rather than being treated as a neutral process change.

To organise the path from bench to regulatory relevance, the authors propose a five-stage framework anchored to a clearly defined context of use. The first stage, engineering verification, confirms that the device reproducibly generates its intended physical environment, from verified strain fields to delivered smoke exposure. The second, biological qualification, demonstrates that this environment supports the specific biological functions the application requires, such as barrier integrity, mucociliary activity or immune-cell migration. The third stage, disease or pharmacological validation, requires reproducible detection of the target response against appropriate comparators, as illustrated by a human alveolus chip that reproduced radiation-induced DNA damage, inflammation and barrier disruption and then evaluated prednisolone and lovastatin as candidate countermeasures.

The final two stages demand the most stringency. Human concordance requires direct comparison of chip outputs with patient-derived or clinical reference data, a standard met in part by the influenza chip’s cytokine benchmarking against patient samples, but the authors warn that such evidence should not be stretched into broader claims of clinical prediction. Predictive performance, needed when a context of use involves therapeutic or toxicological decision-making, requires predefined metrics such as sensitivity, specificity or response discrimination against characterised reference datasets. Deployment and regulatory readiness then asks whether qualified performance survives manufacturing scale-up, automation and transfer between independent laboratories. The framework deliberately distinguishes physiological resemblance from demonstrated concordance, and concordance from clinically anchored prediction, so that each claim carries only the weight its evidence supports.

The review’s conclusion is a call for discipline over dazzle. Prioritising validated, fit-for-purpose performance over maximal complexity, the authors contend, offers a more credible route toward reliable respiratory research tools, drug development platforms and eventual regulatory decision support. Standardisation, they argue, should target terminology, reporting of critical parameters, reference controls and inter-laboratory reproducibility rather than forcing identical device architectures, since legitimate applications genuinely require different designs. As regulators, including the United States Food and Drug Administration in recent draft guidance, increasingly frame new approach methodologies around context of use, the framework provides developers with a practical checklist: define the question and acceptance criteria before experimentation, report engineering parameters with tolerances, benchmark against prespecified comparators, incorporate human datasets when predictive claims are intended, and prove that performance holds across batches, operators and laboratories. If the field follows that path, the miniature lungs growing on chips today may earn the trust needed to influence tomorrow’s drug decisions.

Subject of Research: Engineering and translational validation of lung-on-a-chip microphysiological devices for respiratory disease modelling and drug testing

Article Title: Engineering Lung-on-a-chip microdevices for respiratory disease modelling and drug testing: A fit-for-purpose framework for design and translational validation

Article References: Cheong, S. Y., Liew, T. I. Z., Wong, C. K., Teng, X. X., Cha, X. Y., Ng, N. C.-S., Wong, R. S.-Y., & Goh, B. H. (2026). Engineering Lung-on-a-chip microdevices for respiratory disease modelling and drug testing: A fit-for-purpose framework for design and translational validation. Biomedical Microdevices, 28(3), Article 62. https://doi.org/10.1007/s10544-026-00849-3

Image Credits: AI Generated

DOI: 10.1007/s10544-026-00849-3

Keywords: lung-on-a-chip, microfluidics, organ-on-chip, respiratory disease modelling, drug testing, microphysiological systems, translational validation, mechanical strain, inhalation toxicology, quality control, regulatory science, context of use

NAMs span complex in vitro and in silico systems designed to generate efficacy and safety evidence while reducing reliance on animal testing. Although many pharmaceutical companies already use these tools during discovery, translating them into regulatory science has lagged. A key challenge remains that NAMs often lack standardized workflows and shared benchmarks, making context-specific validation and qualification essential before regulators can rely on them for specific contexts of use.

Launched in spring 2026, NAMs-DC convenes model developers and engages regulators to build a qualification framework suited to advanced, human-relevant model systems. Rather than treating NAMs as generic assays, the coalition emphasizes reproducibility, performance characterization, and evidence mapping that align scientific readiness with regulatory expectations.

Crown Bioscience brings capabilities centered on patient-derived xenograft (PDX) models, tumor organoids, ex vivo patient tissue platforms, and biomarker analysis. With a large commercially available oncology model library, the company supports translational research by integrating bioinformatics and biomarker services across discovery and clinical development stages.

Importantly, these patient-derived systems can help capture tumor heterogeneity and treatment-response dynamics that are difficult to represent with conventional models. In parallel, biomarker-centric readouts can strengthen the linkage between observed model behavior and measurable biological mechanisms relevant to clinical outcomes.

NAMs-DC Executive Director Nicholas King, M.S., said the coalition’s progress depends on collaboration among developers, end-user companies, and regulators. He added that Crown Bioscience’s experience in patient-derived models and biomarker capabilities offers developers and regulators a clearer route to implementation.

Crown Bioscience Vice President, Research and Innovation Ludovic Bourré, Ph.D., emphasized that validating and qualifying NAMs requires coordinated work with regulators. He noted the coalition’s aim to advance shared qualification approaches that support broader adoption across the drug development pipeline.

As the coalition grows, NAMs-DC will continue bringing together developers and regulators to establish standards that enable confident use of complex NAMs. The overarching goal remains to make model selection and evaluation more efficient—so validated NAM tools can realize their potential for improving treatment decisions.

Subject of Research: New approach methodologies (NAMs) for drug discovery and regulatory qualification, including patient-derived models and biomarker analysis.
Article Title: C-Path NAMs-DC Adds Crown Bioscience as Ninth Coalition Member.
News Publication Date: July 16, 2026.
Web References: https://c-path.org/program/new-approach-methodologies-developer-coalition-nams-dc/ , https://c-path.org/ , http://www.crownbio.com/
References: None provided beyond the coalition and organizational descriptions.
Image Credits: None provided.

Keywords: New approach methodologies, NAMs, C-Path, NAMs-DC, Crown Bioscience, patient-derived xenografts, PDX, tumor organoids, ex vivo tissue, biomarker analysis, regulatory qualification, drug discovery, translational models.