The scale of the problem is stark. According to the review, led by Ryan Kilpatrick and Jonathan Davis of Tufts University School of Medicine together with Rachel Greenberg of Duke University, Danielle Boyce of Johns Hopkins University, and Kanwaljit Singh of the Critical Path Institute, most medications administered in the neonatal intensive care unit lack adequate safety, efficacy, and dosing data. Newborns have been described as one of the last therapeutic orphans: drugs are routinely used off-label, at doses extrapolated from older children or adults, in patients whose developing kidneys, livers, and brains handle medications in profoundly different ways. No major therapeutic advance has improved preterm survival since the introduction of surfactant therapy in the early 1990s.

Why has this population been left behind? The review points to a convergence of ethical and practical barriers that make traditional randomized clinical trials extraordinarily difficult in ELGANs. Parents are asked to consent to research for a child they have just met, often in the midst of a medical crisis, and studies of enrollment decision-making show how heavy that burden is. The population is small relative to adult disease areas, fragmented across hundreds of intensive care units, and heterogeneous in ways that demand enormous sample sizes to detect meaningful effects. Regulatory incentives have historically done little to encourage investment in neonatal drug development, and pediatric trials more broadly suffer from high rates of early discontinuation and incomplete reporting.

And yet, paradoxically, ELGANs generate more clinical data than almost any other patient group. A single stay in the neonatal intensive care unit can stretch for months, during which continuous physiologic monitors stream heart rate, oxygen saturation, and respiratory measurements second by second. Laboratory values are drawn daily or more often. Every medication is recorded with dose, route, and timing, adjusted to a body weight measured on a daily or twice-daily schedule. Imaging studies, ventilation settings, nutrition records, and neurodevelopmental assessments accumulate in the electronic health record. The review’s authors argue that this extraordinary density and granularity makes ELGANs ideal candidates for real-world data research—if the infrastructure exists to use it.

Real-world data, in the language of regulators, refers to information relating to patient health collected outside the confines of conventional randomized trials: electronic health records, claims databases, registries, and device data. When such data are analyzed with rigorous methods, they become real-world evidence, and agencies including the United States Food and Drug Administration and the European Medicines Agency have published frameworks describing how such evidence can support regulatory decision-making. The FDA’s 2024 guidance on assessing electronic health records and claims data, and the EMA’s real-world evidence framework for regulator-led studies, signal a regulatory environment far more receptive to these approaches than a decade ago.

The technical challenge, however, is formidable. The review examines the current landscape of common data models—standardized structures such as the OMOP model developed by the Observational Health Data Sciences and Informatics collaboration, and interoperability standards like HL7’s FHIR—that allow different institutions to pool and compare their data. These models were not designed with neonatal complexity in mind. They lack native support for linking mothers and infants, a critical requirement when maternal exposures, medications, and pregnancy complications shape neonatal outcomes. They struggle with continuous physiologic waveforms, with weight-based medication dosing that changes daily, and with the long-term neurodevelopmental follow-up that defines success in this population. Extensions are emerging, including perinatal adaptations of OMOP and standardized representations of ventilation data, but the review concludes that neonates remain poorly served by existing infrastructure.

To address this, the authors propose a minimum neonatal-specific variable set: a core collection of data elements that every neonatal network should capture in a standardized way. This includes gestational age and birth weight, maternal and delivery characteristics, medication exposures with precise dosing, respiratory support parameters, laboratory values, diagnoses such as bronchopulmonary dysplasia defined using evidence-based criteria, adverse events graded with neonatal-specific severity scales, and neurodevelopmental outcomes extending well beyond discharge. Defining bronchopulmonary dysplasia, the chronic lung disease of prematurity, illustrates the difficulty: multiple competing definitions exist, and without harmonization, one center’s outcome cannot be meaningfully compared with another’s. Data quality assessment frameworks, the authors note, must be applied at every step, because a model is only as trustworthy as the messy clinical reality it encodes.

The regulatory applications of well-structured real-world data are where the review’s vision becomes most consequential. External controls—comparison groups drawn from existing patient data rather than randomized within the trial—could allow new therapies to be tested in ELGANs without withholding promising treatment from a control group, an approach the FDA has formally addressed in guidance for externally controlled trials. Target trial emulation, a framework developed by epidemiologist Miguel Hernán and colleagues, allows researchers to design an observational analysis that mimics a hypothetical randomized trial, specifying eligibility, treatment strategies, and follow-up in advance to avoid the biases that plague naive analyses of registry data; emulation studies across 32 clinical trials have shown that such designs can reproduce randomized results when executed carefully. Pharmacovigilance—the ongoing surveillance of drug safety—could operate continuously across networks rather than waiting for rare events to surface in small trials. Natural history studies, mapping how diseases progress without intervention, remain foundational for a population in which the course of illness is still incompletely described.

The review is equally clear that technology alone will not deliver this transformation. Building real-world data infrastructure for ELGANs requires coordinated investment from a broad coalition of stakeholders. Neonatologists and bedside caregivers must ensure that the data entered into electronic health records are accurate and complete, often under time pressure at the cot side. Informaticians must design models and vocabularies that respect neonatal physiology, extending frameworks such as the GA4GH Phenopacket schema for computable clinical representation. Regulators must continue to clarify when real-world evidence is acceptable, and industry must see a viable path to neonatal drug development. Families, the authors emphasize, belong at the table as well: international priority-setting partnerships led by parents of the most premature infants have already identified research questions that matter most to those who live with the consequences.

The stakes extend beyond individual units. Wide practice variation across neonatal intensive care units is well documented, and outcomes for very preterm infants differ substantially across health systems even after adjustment for case mix. That variation is often treated as a problem of quality; the review reframes it as an opportunity, because differences in practice, captured in standardized data, become natural experiments that can reveal which strategies genuinely improve survival without severe morbidity. Cohorts such as the European EPICE study have shown that greater use of evidence-based practices is associated with better outcomes, but identifying those practices requires the kind of harmonized, granular data infrastructure the authors describe. If the vision is realized, the data exhaust of intensive care—currently a byproduct of survival—could become the engine of the first genuine therapeutic progress for extremely preterm infants in more than three decades.

Subject of Research: Use of neonatal real-world data to generate clinical evidence for extremely preterm infants

Article Title: Enabling evidence generation for extremely preterm neonates: the use of neonatal real-world data

Article References: Kilpatrick, R., Greenberg, R. G., Boyce, D., Singh, K., & Davis, J. (2026). Enabling evidence generation for extremely preterm neonates: the use of neonatal real-world data. Pediatric Research. https://doi.org/10.1038/s41390-026-05561-9

Image Credits: AI Generated

DOI: 10.1038/s41390-026-05561-9

Keywords: real-world data, extremely preterm neonates, neonatal intensive care, ELGAN, clinical trials, electronic health records, OMOP common data model, external controls, target trial emulation, pharmacovigilance, drug dosing, neurodevelopmental outcomes

The research, published in BMC Health Services Research, is the first attempt to measure anaphylaxis-related knowledge and self-reported practice among physicians in the West Bank. A team led by Azzam Zrineh of Al-Quds University surveyed 317 doctors across 12 hospitals in five governorates between October 2025 and January 2026. The participants, drawn from seven clinical specialties through convenience sampling, had a median age of 29 years and were predominantly male. The questionnaire, adapted from an instrument previously validated in an Egyptian physician cohort, probed four critical domains: recognition of valid anaphylactic scenarios, identification of the first-line drug, knowledge of the correct route of administration, and command of the weight-based dosing regimen that determines whether the treatment works or falls short.

The headline numbers look reassuring at first glance. A striking 85.5 percent of respondents correctly identified epinephrine as the first-line treatment for anaphylaxis, and 88.0 percent knew it should be given intramuscularly rather than intravenously or subcutaneously. Those figures suggest that decades of guideline dissemination have penetrated at least the most basic layer of clinical awareness. But the study’s finer-grained results tell a more unsettling story. Only 36.3 percent of physicians could identify the correct weight-based dose of epinephrine, meaning nearly two out of three doctors surveyed would be at risk of underdosing or overdosing a patient in the very scenario where precision matters most. And only 24.0 percent accurately recognized all four valid anaphylactic presentations presented in the questionnaire without also endorsing a distractor case that did not meet diagnostic criteria.

That diagnostic figure deserves particular scrutiny. Anaphylaxis is notoriously easy to miss when it does not announce itself with the classic combination of hives, airway swelling, and collapse. Guidelines define it through patterns involving skin involvement with respiratory or cardiovascular compromise, or isolated severe respiratory or cardiovascular symptoms after a likely allergen exposure. A physician who cannot distinguish these patterns from less dangerous reactions may withhold epinephrine when it is needed, or conversely, label every urticaria outbreak as anaphylaxis and dilute the urgency of the real thing. The study’s finding that three-quarters of physicians failed to correctly classify all presented scenarios suggests that diagnostic discrimination, not drug recognition, is the weakest link in the chain of survival in these hospitals.

Knowledge on paper does not always translate into action, and the self-reported practice data made that plain. Among the 89.3 percent of physicians who said they had personally encountered anaphylaxis cases, 13.8 percent reported never administering epinephrine to those patients. In other words, roughly one in seven doctors who had faced a genuine anaphylactic emergency admitted that the lifesaving drug never reached the patient under their care. The reasons are not captured in detail by the survey, but the institutional context offers clues. Fewer than half of the workplaces represented, 46.7 percent, had both a written anaphylaxis protocol and readily accessible epinephrine. A physician standing in an emergency bay without a protocol to consult and without an ampoule within reach faces structural barriers that no amount of individual knowledge can overcome.

The statistical analysis uncovered patterns in who knew what. Younger physicians significantly outperformed their older colleagues on identifying the first-line agent, a difference the authors reported at a p-value of 0.017, and the generational gap on route knowledge was even wider, with p below 0.001. This likely reflects the recency of medical training among junior doctors, who have more recently absorbed current guideline recommendations into their education. Yet seniority told the opposite story for dosing: specialists demonstrated significantly higher dose knowledge than residents and general practitioners, with p below 0.001. The picture that emerges is of a knowledge landscape where different physician groups hold different pieces of the puzzle, with juniors versed in modern recommendations but lacking the accumulated clinical judgment of specialists, and seniors retaining dosing expertise while lagging on updated route and agent guidance.

Perhaps the most sobering finding concerned training itself. Self-reported prior exposure to anaphylaxis-related training was not significantly associated with better knowledge performance in the unadjusted analyses. If attending a training session does not measurably improve a physician’s grasp of diagnosis, dosing, or administration, then the region’s educational interventions may be failing in their design or delivery rather than in their intent. Passive lectures and one-off workshops, the authors imply, are not the answer. What is needed instead, they argue, are targeted educational interventions that emphasize the two areas of greatest deficit: weight-based epinephrine dosing and the recognition of atypical anaphylactic presentations that deviate from the textbook picture.

The study’s methods carry the usual caveats of cross-sectional survey research. Convenience sampling may have skewed recruitment toward physicians more interested in the topic, and self-reported practice is an imperfect proxy for observed behavior; doctors may overstate or understate what they actually do under pressure. The questionnaire was adapted from an Egyptian validation cohort, and while this lends it prior credibility, cultural and institutional differences between health systems could affect how questions land. Still, the scale of the survey, spanning 12 hospitals, five governorates, and seven specialties, gives the findings a breadth that single-center studies cannot match, and the consistency of the gaps with patterns documented globally suggests the West Bank is not an outlier but a case study in a worldwide problem.

Indeed, knowledge deficits around anaphylaxis have been reported among physicians in many countries, from emergency departments in Europe to primary care networks in Asia, making this a structural weakness in medical education rather than a local failing. What the Palestinian data add is a clear roadmap for remediation. The authors call for institutional measures alongside individual education: written anaphylaxis protocols in every workplace and guaranteed access to epinephrine where patients are treated. The logic is straightforward. Even a perfectly informed physician cannot inject a drug that is not in the building, and a well-stocked emergency cart is of little use to a clinician who cannot calculate the dose. Closing the gap between knowing that epinephrine saves lives and actually delivering it, at the right dose, through the right route, to the right patient, will require fixing both the minds and the medicine cabinets of the region’s hospitals.

Subject of Research: Physician knowledge and self-reported practice in the diagnosis and management of anaphylaxis in West Bank hospitals

Article Title: Physicians’ knowledge and self-reported practice concerning diagnosis and management of anaphylaxis: a multi-hospital cross-sectional study in the West Bank, Palestine

Article References: Physicians’ knowledge and self-reported practice concerning diagnosis and management of anaphylaxis: a multi-hospital cross-sectional study in the West Bank, Palestine. (n.d.). https://doi.org/10.1186/s12913-026-15874-4

Image Credits: AI Generated

DOI: 10.1186/s12913-026-15874-4

Keywords: anaphylaxis, epinephrine, physician knowledge, emergency medicine, cross-sectional study, West Bank, Palestine, clinical guidelines, medical education, drug dosing, allergy, health services research

Scrofuloderma occupies a peculiar place in the spectrum of tuberculosis. Most people imagine the disease as a lung infection marked by cough and fever, but Mycobacterium tuberculosis is a versatile pathogen capable of seeding nearly any organ. When it settles in structures lying just beneath the skin, such as lymph nodes, bones, or joints, the infection can spread by direct extension into the overlying tissue. The result is a chronic, progressive lesion: the skin above the infected node becomes firm, then inflamed, then ulcerated, eventually forming draining sinuses that ooze a mixture of pus and necrotic material. Among the cutaneous forms of tuberculosis, scrofuloderma is the most common in developing countries, precisely because it tracks so closely with untreated or inadequately treated lymph node tuberculosis, which itself remains widespread in high-burden settings. Its rarity in dermatology clinics elsewhere is not a sign that the organism has changed, but that public health systems have, in many places, succeeded in interrupting the chain of infection before it reaches the skin.

The Indonesian adolescent’s story began with bilateral cervical nodules, swollen lymph nodes on both sides of the neck, together with a mass on the right side of his chest wall. These are classic territories for tuberculous lymphadenitis, the most frequent form of extrapulmonary tuberculosis in children and adolescents. He received antituberculosis therapy, but the dose was inadequate, a deceptively simple error with profound consequences. Underdosed treatment fails to achieve the drug concentrations needed to kill the bacteria reliably, and it creates conditions in which partially resistant bacterial populations can flourish. The infection was suppressed but not eliminated. When he presented with a fresh cervical nodule after six months of this insufficient regimen, his clinicians faced a patient whose disease had effectively been given a chance to adapt and to burrow deeper into the tissues of his neck and chest.

Diagnosis in such cases demands a layered approach, and the team proceeded methodically. Fine-needle aspiration biopsy of the cervical nodules, a minimally invasive technique in which a thin needle withdraws cells and fluid from a suspicious mass, confirmed tuberculous lymphadenitis. This mattered enormously, because scrofuloderma can mimic a long list of other conditions, from bacterial abscesses and fungal infections to malignancies, and treating the wrong disease wastes precious time. A thoracic CT scan then revealed an abscess within the chest wall, mapping the hidden anatomy of the infection and showing that the disease had extended well beyond what the skin lesions suggested. Imaging of this kind is essential in suspected chest-wall tuberculosis, because the abscess cavity often communicates with infected ribs, costal cartilage, or pleural space in ways that determine the surgical plan.

With the anatomy defined, the team turned to the operating room. A thoracotomy with aspiration was performed, allowing surgeons to access the chest wall abscess directly, drain its contents, and obtain tissue for examination. Histopathological analysis of that tissue confirmed tuberculous inflammation of the chest wall, closing the diagnostic loop. This dual role of surgery, both therapeutic and diagnostic, is a recurring theme in the management of complicated extrapulmonary tuberculosis. Draining a large abscess removes a reservoir of bacteria and necrotic material that antibiotics penetrate poorly, while the excised tissue provides the histology and culture material that laboratory tests alone cannot always supply. In an era of increasing drug resistance, obtaining organisms directly from the lesion also opens the door to drug susceptibility testing, which identifies which medications will actually work against the specific strain infecting the patient.

Once the diagnosis was secured, the treatment pivoted decisively. The adolescent was restarted on adult fixed-dose combination antituberculosis therapy, this time at an adequate dose, and the course was extended to a full twelve months. Fixed-dose combination pills, which bundle multiple first-line drugs into a single tablet, are a cornerstone of modern tuberculosis control because they simplify adherence and reduce the risk that patients will take some drugs but not others, the exact pattern that breeds resistance. The extended duration reflected the severity and location of the disease: chest-wall and disseminated lymph node involvement generally require longer treatment than uncomplicated pulmonary tuberculosis, and a patient who has already failed one course of therapy warrants a more aggressive and more closely supervised approach.

The outcome vindicated the strategy. The nodules resolved, the skin lesions healed, and the patient experienced significant weight gain, a marker of recovering health in a disease notorious for wasting its hosts. For a teenager who had endured months of ineffective treatment, recurring lesions, and major surgery, the return to normal growth and healing represents more than a clinical success; it is the restoration of an adolescence that tuberculosis had been steadily eroding. The authors emphasize that this favorable result depended on the combination of early diagnosis using histopathology and molecular testing, surgical intervention, and appropriate antituberculosis therapy, with none of these elements sufficient on its own.

Embedded in this single case are lessons that reach far beyond one patient. First, clinicians in high tuberculosis-burden settings must recognize that the disease wears many faces, and that skin lesions, neck masses, and chest-wall swellings in adolescents should always raise the question of tuberculosis, particularly when there is a history of prior treatment. Second, inadequate dosing is not a benign compromise; it is a setup for treatment failure, relapse, and potentially drug resistance, and therapeutic drug monitoring, in which blood levels of antituberculosis drugs are measured and adjusted, is an increasingly available tool to prevent it. Third, diagnosis should never rest on appearance alone. Fine-needle aspiration biopsy, histopathology, and molecular testing each contribute independent confirmation, and together they distinguish tuberculosis from the infections and tumors that imitate it.

The case also underscores the value of multidisciplinary care for adolescents with complicated tuberculosis. A pediatric respirologist, surgeons, pathologists, radiologists, and microbiologists all contributed essential pieces to this patient’s recovery, and the authors argue that close monitoring by such a team is key both to achieving good outcomes and to preventing recurrence. Adolescents occupy an awkward position in tuberculosis programs: too old for straightforward pediatric dosing schemes, too young for the assumptions built into adult care, and often poorly adherent to long medication courses. This patient’s initial underdosing likely reflected exactly that gap, and his eventual cure required deliberately bridging it with adult fixed-dose combination therapy given at full strength under supervision.

For the wider public, the story is a reminder that tuberculosis has not retreated into history. It remains one of the world’s deadliest infectious diseases, and in the regions where it thrives, its reach extends into lymph nodes, bones, joints, and skin, producing presentations that can baffle clinicians who expect only a cough. The sixteen-year-old at the center of this report was fortunate in one respect: his disease, even after inadequate treatment, remained treatable with first-line drugs and responded to a carefully orchestrated combination of surgery and a full year of properly dosed medication. Not every patient with underdosed therapy is so lucky, and the spread of multidrug-resistant tuberculosis means that each treatment failure carries risks for the community as well as the individual. The authors’ closing message is straightforward but urgent: in settings where tuberculosis is common, early recognition of its varied manifestations, rigorous diagnostic workup, adequate dosing from the very first prescription, and sustained multidisciplinary follow-up are the difference between a healed adolescent and a chronic, spreading, drug-resistant infection.

Subject of Research: Cutaneous tuberculosis (scrofuloderma) extending to the chest wall in an adolescent following inadequate antituberculosis therapy

Article Title: Lessons learned from the management of unusual scrofuloderma extending to the chest wall in an adolescent male: case report

Article References: Nataprawira, H. M., & Hadi, R. S. (2026). Lessons learned from the management of unusual scrofuloderma extending to the chest wall in an adolescent male: case report. BMC Pediatrics. https://doi.org/10.1186/s12887-026-07823-1

Image Credits: AI Generated

DOI: 10.1186/s12887-026-07823-1

Keywords: tuberculosis, scrofuloderma, cutaneous tuberculosis, chest-wall tuberculosis, adolescent health, tuberculous lymphadenitis, fine-needle aspiration biopsy, antituberculosis therapy, drug dosing, extrapulmonary tuberculosis, pediatrics, case report

The distinction at the heart of the study is what the authors call mechanism-aware benchmarking, and it matters because the two error types carry different dangers. A true hallucination occurs when a model invents a contraindication that does not exist, for example claiming a drug cannot be used in a situation where it is actually the recommended choice. A false contraindication, by contrast, is a real-world clinical caution that the model applies incorrectly, refusing to recommend an appropriate intervention because it misreads the context. Both can lead a trainee astray, but they stem from different failure mechanisms inside the model, and separating them allows educators to understand precisely where and why these systems go wrong rather than simply tallying a single error score.

To conduct the evaluation, the researchers presented fifty cases of pediatric difficult airway management to five large language models using a standardized prompt anchored in the 2022 American Society of Anesthesiologists Difficult Airway Guidelines. Two anesthesiologists then rated the responses on a modified 0 to 5 scale in a blinded, comparative assessment. The reliability of that human scoring was extraordinary: the inter-rater agreement, measured by a quadratic weighted kappa, reached 0.982, a value that essentially approaches perfect concordance and lends considerable statistical weight to the findings. The differences between the models themselves were also highly significant, with a Friedman test returning a chi-squared value of 100.12 across four degrees of freedom and a p-value below 0.001, confirming that the performance gaps were not statistical noise.

In this single-pass evaluation, GPT-5.2 Thinking emerged as the strongest performer, achieving both the highest mean score and the largest proportion of responses judged acceptable. At the other end of the spectrum, Claude Opus 4.5 recorded the highest combined critical-error rate at 23.0 percent, compared with 6.5 percent for both GPT models tested. Those numbers alone would be striking, but the mechanism-level analysis revealed something even more consequential: the models fail in characteristically different ways, and those failure signatures matter enormously when deciding whether such tools belong anywhere near a training environment.

True hallucinations, the fabrication of nonexistent contraindications, were overwhelmingly concentrated in one model. Claude Opus 4.5 produced them in 11.8 percent of cases, while the other four models ranged from just 0.5 to 2.2 percent. False contraindications, however, proved to be a universal weakness, appearing across every model at rates between 5.0 and 13.0 percent. Perhaps most tellingly, these false contraindications clustered around a single drug: rocuronium, the neuromuscular blocking agent that is central to rapid sequence intubation and a cornerstone of emergency airway management. A model that hesitates to recommend rocuronium, or wrongly flags it as contraindicated, is not making a harmless stylistic error; it is steering a learner away from a potentially lifesaving intervention in the very scenarios where seconds count.

The study also uncovered a gradient of failure that tracks directly with clinical complexity. The proportion of responses judged inadequate, defined as a score of 2 or below, rose steadily as scenarios became harder. For difficult intubation cases, inadequacy ranged from 32 to 98 percent across the models. For difficult ventilation, it climbed to between 46 and 94 percent. And for the most dire scenario of all, cannot intubate cannot oxygenate, known in the field as CICO, inadequacy spanned 52 to 92 percent. In other words, the situations where a trainee most needs accurate, guideline-concordant guidance are precisely the situations where these models are most likely to fall short, a pattern that inverts the usual assumption that AI assistance is most valuable in the hardest cases.

The authors’ conclusion is carefully calibrated but unambiguous. Contemporary LLMs exhibit model-specific performance characteristics and error patterns in pediatric difficult airway management, and dosing-related failures manifest through distinct mechanisms of true hallucination versus false contraindication. The findings reveal what the researchers describe as a marked inadequacy of the scenarios employed, particularly as complexity and critical error rates increase. Their verdict on deployment is equally measured: current LLMs may hold value as supervised supplementary learning tools, but they should not function as autonomous sources of clinical or educational guidance. That framing positions these systems as something closer to a study partner whose answers must always be checked, rather than a reference whose word can be trusted.

Why does pediatric difficult airway management stress these models so severely? The clinical domain itself offers clues. Children are not small adults; airway anatomy, drug dosing, and equipment sizing all scale with age and weight in ways that demand precise, patient-specific calculation. The ASA’s 2022 difficult airway guidelines embed a structured decision tree that models must navigate step by step, and any drift at an early node cascades into a wrong endpoint. Dosing errors are especially hazardous in this population because the therapeutic window for neuromuscular blockers and induction agents is narrow, and the study’s finding that errors concentrate around rocuronium suggests the models struggle most where weight-based calculation intersects with urgency and contraindication logic.

The study’s methodology also deserves attention as a template for future AI evaluation in medicine. Rather than asking whether a model’s answer is simply right or wrong, the mechanism-aware approach asks what kind of wrong it is, and that granularity has practical consequences. An educator deploying an LLM as a teaching aid can now anticipate that one model may invent contraindications out of thin air while another may reflexively withhold appropriate drugs, and can design supervision and debriefing around those known tendencies. The near-perfect inter-rater agreement achieved by the two blinded anesthesiologist raters further demonstrates that expert human judgment can reliably and reproducibly grade the quality of AI clinical reasoning, providing a credible benchmarking standard.

For the broader conversation about artificial intelligence in medicine, the study lands at a moment of intense enthusiasm and equally intense anxiety. It neither condemns LLMs outright nor licenses their casual use; instead it draws a precise boundary. As informal just-in-time learning tools for trainees, under the eye of an experienced supervisor, these models may stimulate reasoning and provide a starting point for discussion. As autonomous advisors in pediatric airway emergencies, they are not ready, and the data show why: critical error rates as high as 23 percent, universal false contraindications around essential drugs, and inadequacy rates that spike exactly when stakes peak. The message for medical educators, and for the developers of these systems, is that safety in high-stakes clinical education must be demonstrated, not assumed, and that the next generation of benchmarks should measure not just whether models answer, but how they fail.

Subject of Research: Benchmarking large language models for safety and error mechanisms in pediatric difficult airway medical training

Article Title: Mechanism-aware benchmarking of large language models as learning aids in pediatric difficult airway training: true hallucinations vs. false contraindications

Article References: Yediyıldız, M. B., & Durmuş, İ. (2026). Mechanism-aware benchmarking of large language models as learning aids in pediatric difficult airway training: true hallucinations vs. false contraindications. BMC Medical Education. https://doi.org/10.1186/s12909-026-10509-y

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10509-y

Keywords: large language models, pediatric anesthesia, difficult airway, hallucination, false contraindication, rocuronium, ASA difficult airway guidelines, medical education, simulation-based training, drug dosing, CICO, artificial intelligence

The team, led by co-first authors Judith Verdonk and Leila-Sophie Otten under the supervision of pharmacist Rob ter Heine, focused on non-small cell lung cancer, the setting in which pembrolizumab is most commonly used. Three approved dosing regimens were compared in silico: 2 milligrams per kilogram every three weeks, a flat 200 milligrams every three weeks, and a flat 400 milligrams every six weeks. These regimens were established through modelling and simulation during drug development, but the authors argue that the pharmacodynamic surrogate underpinning them, peripheral receptor occupancy, has never been rigorously validated as a predictor of clinical benefit.

To build their model, the researchers started from a physiological pharmacokinetic framework describing how pembrolizumab moves through plasma and penetrates tumour tissue. Antibody tumour penetration is a notoriously inefficient process, governed by convective transport, diffusion and antigen-mediated binding that can slow or even reverse the movement of drug into tumour interstitium. The model accounted for these dynamics, extending earlier work on antibody tumour transport, and then added two crucial layers: receptor occupancy kinetics and the induction of interleukin-2, a cytokine whose production reflects genuine T cell activation rather than mere receptor blockade.

The interleukin-2 parameters were not invented from thin air. They were derived from ex vivo experiments in which patient T cells were exposed to pembrolizumab and their functional responses measured. This experimental grounding matters, because an earlier study by the same group had already revealed a critical gap between receptor occupancy and T cell functionality: cells whose PD-1 receptors were fully occupied by the drug did not necessarily show the downstream signalling and cytokine production that clinicians would want to see. The new study translates that laboratory observation into a quantitative, patient-scale simulation framework.

The simulations produced a striking and, for the field, uncomfortable result. All three approved regimens achieved near-complete PD-1 receptor occupancy, both in the circulation and within the tumour microenvironment. On the traditional view, this means all three doses should be pharmacodynamically equivalent, and indeed this saturation logic is precisely why regulators and developers accepted the regimens as interchangeable. Yet when the researchers looked at predicted interleukin-2 concentrations, a very different picture emerged. The cytokine levels were predicted to rise and fall throughout the dosing interval, tracking the concentration of pembrolizumab itself rather than the occupancy of its receptor.

Even more provocatively, the 400 milligram every-six-weeks regimen, the highest total dose per interval, produced the highest predicted interleukin-2 levels. If immune activation rather than receptor saturation is what drives anti-tumour efficacy, then the longest-interval regimen may not be a mere convenience option but potentially the most immunologically active of the three. The authors are careful to note that interleukin-2 is a marker, not a proven mediator of clinical response, and that more activation is not automatically better; excessive immune stimulation can contribute to toxicity. But the finding decisively breaks the assumed link between occupancy and pharmacodynamic effect.

A second important nuance concerned the tumour itself. Changes in intratumoural interleukin-2 concentrations were predicted to be considerably less pronounced than those in plasma. In other words, even if systemic immune activation fluctuates substantially between doses, the immune microenvironment inside the tumour may experience a more buffered response. This has implications for how biomarkers are interpreted: a blood sample showing robust cytokine swings may not faithfully represent what is happening at the site where the immune system meets the cancer. It also raises the possibility that intratumoural pharmacodynamics, not plasma pharmacokinetics, should be the target of future dose optimisation efforts.

The broader context makes these findings timely. Pembrolizumab is expensive, and health systems worldwide have scrutinised whether lower doses or longer intervals could deliver the same benefit at reduced cost. Pharmacokinetic simulation studies and even randomised trials of low-dose immunotherapy in head and neck cancer have suggested that less drug may suffice. Meanwhile, other lines of research have explored whether the timing of infusion within the day, circadian chronotherapy, influences survival. What has been missing from all these debates is a validated pharmacodynamic marker that actually reflects the biological effect the drugs are meant to produce. This study argues that receptor occupancy, the marker the field has relied upon, fails that test.

The authors conclude that receptor occupancy is an inadequate pharmacodynamic surrogate for pembrolizumab and, by extension, for the broader class of PD-1 inhibitors. They call for the development of robust pharmacodynamic markers and, crucially, a clearer definition of the level and pattern of immune activation required for optimal efficacy. Until such markers exist, dose selection for checkpoint inhibitors will remain guided by surrogates that may bear little relationship to the biology that matters. The study, funded by Stichting Treatmeds as part of the NVALT30/DEDICATION-1 trial programme, was approved by the ethics committee METC Oost-Nederland and conducted according to the Declaration of Helsinki. Its message is likely to resonate far beyond lung cancer: for a drug class that has transformed oncology, the question of how much drug patients really need may have been asking the wrong question all along.

Subject of Research: Translational pharmacokinetic-pharmacodynamic modelling of pembrolizumab dosing, receptor occupancy and immune activation in non-small cell lung cancer

Article Title: Translational modelling challenges receptor‑occupancy‑based dosing of PD‑1 inhibitors like pembrolizumab

Article References: Verdonk, J. D. J., Otten, L.-S., Montaseri, G., van den Heuvel, M. M., Smeets, R. L., Koenen, H. J. P. M., & ter Heine, R. (2026). Translational modelling challenges receptor‑occupancy‑based dosing of PD‑1 inhibitors like pembrolizumab. British Journal of Cancer. https://doi.org/10.1038/s41416-026-03612-w

Image Credits: AI Generated

DOI: 10.1038/s41416-026-03612-w

Keywords: pembrolizumab, PD-1 inhibitors, receptor occupancy, cancer immunotherapy, pharmacokinetic modelling, interleukin-2, non-small cell lung cancer, dose optimisation, pharmacodynamics, immune checkpoint inhibitors, tumour penetration, drug dosing

Hydroxyurea has long been the cornerstone disease-modifying therapy for sickle cell anemia, the homozygous HbSS or HbSβ0-thalassemia genotypes that produce the most severe disease course. The drug works partly by reactivating fetal hemoglobin production, reducing the polymerization of sickle hemoglobin that drives red cell sickling, painful crises, and organ damage. Yet despite decades of clinical use, the precise relationship between dose and both clinical benefit and hematologic harm has remained uncertain, particularly across the diverse populations in which the disease is now treated.

To resolve this, the researchers conducted a systematic review and one-stage multilevel dose-response meta-analysis encompassing randomized trials, nonrandomized comparative studies, and longitudinal cohorts of children and adults with sickle cell disease. The quantitative synthesis focused principally on sickle cell anemia, with hydroxyurea dose expressed in mg/kg/day and achieved or maintenance doses preferred over starting doses. The primary benefit outcome was the rate of recurrent vaso-occlusive crises, while the primary safety outcomes were annual risks of severe or treatment-limiting neutropenia and thrombocytopenia, the two cytopenias that most often force dose reductions or discontinuation.

The modeling strategy was unusually rigorous. The team fitted Emax maximum-effect dose-response models as their primary analysis, testing linear and dose-independent alternatives in sensitivity analyses, and prespecified a decision rule requiring predicted annual neutropenia risk of no more than 5 percent and thrombocytopenia risk of no more than 3 percent, with at least an 80 percent simulation-derived probability that both criteria would be met simultaneously. Any eligible dose also had to retain near-optimal crisis control.

The results revealed a striking asymmetry between benefit and risk across the dose range. Empirical support for the primary analysis clustered between 20 and 28 mg/kg/day, and within that window recurrent vaso-occlusive crisis rates continued to fall, from a predicted 39.8 events per 100 person-years at 20 mg/kg/day to 32.5 at 28 mg/kg/day. Critically, the data showed no efficacy plateau: the Emax ED50 hit the upper limit of the prespecified search grid, and a simpler linear model actually fit marginally better, suggesting that whatever additional crisis reduction higher doses confer had not yet been exhausted within the observed range.

Safety told a different story. Predicted annual risk of severe or treatment-limiting neutropenia climbed steeply and early, from 1.98 percent at 20 mg/kg/day to 3.74 percent at just 21.5 mg/kg/day and 5.63 percent at 22.5 mg/kg/day. Thrombocytopenia, by contrast, remained uncommon and showed little dose dependence. Under the prespecified joint safety rule, the probability that both criteria were met was 87.1 percent at 20 mg/kg/day, 85.6 percent at 21 mg/kg/day, but only 77.8 percent at 21.5 mg/kg/day, failing the 80 percent threshold. The authors cautioned that because safety evidence was sparse and the joint classification depended partly on the relation between the two safety models, the low-20-mg/kg/day range should be read as an approximate safety transition rather than a precise toxicity threshold.

Within the empirically supported range, the lowest modeled recurrent crisis rate occurred at 28 mg/kg/day, and the unconstrained composite-benefit maximum at 25.5 mg/kg/day, but neither could clear the safety bar. Twenty-one mg/kg/day emerged as the highest dose satisfying the analytical rule, and the rounder conventional increment of 20 mg/kg/day was selected as the population reference dose, chosen from standard 2.5-mg/kg/day increments for practical prescribing.

The certainty of evidence behind these estimates was graded as very low for all three outcomes—recurrent crises, neutropenia, and thrombocytopenia—underscoring how much uncertainty remains. The review was prospectively registered with PROSPERO, and the authors emphasize that their decision framework, rather than any single trial, is what allows conflicting and sparse evidence to be converted into an actionable population-level recommendation.

Perhaps the most important clinical message is what the reference dose is not. The authors explicitly state that 20 mg/kg/day is neither a maximum tolerated dose nor a ceiling for treatment. Because crisis rates kept declining above 20 mg/kg/day, higher doses may well be appropriate for individual patients who need further clinical benefit, provided hematologic tolerance, adherence, and laboratory monitoring permit careful escalation. In resource-limited settings where laboratory surveillance is constrained, however, the population-level framework offers a defensible default that maximizes expected benefit while keeping predicted severe cytopenia risks within prespecified limits.

For the millions of people living with sickle cell anemia worldwide, most of them in sub-Saharan Africa, the findings translate decades of fragmented trial evidence into a single, transparent, quantitatively justified starting point for therapy—and a clear signal that the field still needs better safety data to push the benefit curve further.

Subject of Research: Dose-response relationship of hydroxyurea therapy in sickle cell anemia

Article Title: Hydroxyurea dosing in sickle cell anemia: a systematic review, dose-response meta-analysis, and population-level benefit-risk assessment

Article References: Hydroxyurea dosing in sickle cell anemia: a systematic review, dose-response meta-analysis, and population-level benefit-risk assessment. (n.d.). https://doi.org/10.1186/s12887-026-07679-5

Image Credits: AI Generated

DOI: 10.1186/s12887-026-07679-5

Keywords: hydroxyurea, sickle cell anemia, dose-response meta-analysis, vaso-occlusive crisis, neutropenia, thrombocytopenia, benefit-risk assessment, systematic review, fetal hemoglobin, pediatric hematology, BMC Pediatrics, drug dosing