The answer, published by Volkan Duran, Abdulhakim Tuğluk and Murat Ali Karavelioğlu, is sobering and instructive in equal measure. Of the five content-analytic dimensions the team tracked across the novel’s 44 chapters, only two met a pre-specified reliability criterion. Emotional valence—the dominant positive or negative tone of each chapter—reached an intraclass correlation coefficient of 0.79, with a 95% confidence interval from 0.64 to 0.88. Empirical reasoning, the concrete sensory and bodily mode of thinking, came in at 0.70, with a wider but still acceptable interval of 0.48 to 0.83. The remaining three dimensions—metaphorical reasoning at 0.50, hypothetical reasoning at 0.42 and analogical reasoning at 0.31—fell below the 0.60 threshold and were formally withdrawn from interpretation.

That act of withdrawal is the study’s most distinctive feature. In much AI-assisted content analysis, dimensions that score poorly on reliability are quietly reported anyway, their numbers presented as if they described the text rather than the quirks of a particular scoring procedure. Here, the authors did the opposite. The values for the three failed dimensions are reported only as properties of the adjudicated dataset—the specific set of scores produced by the AI model and then reviewed and finalized by two human experts. They are explicitly not treated as findings about the novel itself, because no independent reader could reproduce them.

The theoretical motivation for the analysis comes from psychological accounts of how affective state couples to mode of reasoning. Pain, according to this framework, biases processing toward the immediate sensory present: the body’s signals crowd out abstraction, and construal becomes concrete, empirical, anchored in what can be felt. Distance from threat, by contrast, supports abstract, comparative and possibility-oriented thinking—the mind is free to imagine alternatives, construct analogies and entertain hypotheticals. An extended first-person illness narrative, the researchers reasoned, offers a complementary source of description for this coupling: if the theory holds, chapters dominated by negative affect should also be chapters dominated by concrete, sensory and bodily representation.

The methodological design was built to test that idea without contaminating it. Each of the 44 chapters was characterized for dominant emotional valence and for four reasoning modes—metaphorical, analogical, hypothetical and empirical—by an AI-assisted first-pass scoring procedure using a GPT-4-class model. All AI-generated candidate scores were then reviewed, adjudicated and finalized by two human experts, so the final dataset reflected human consensus judgments. But because that adjudicating procedure exposes the human readers to the model’s proposals, it risks a subtle form of anchoring: the experts may unconsciously gravitate toward the machine’s candidates, a phenomenon psychologists call automation bias. To guard against it, the team recruited two further readers who rescored all 44 chapters working independently and blind to both the model’s candidates and the adjudicated values.

The results of that blind replication are the study’s core contribution. The association between the two surviving dimensions replicated: emotional valence and empirical, sensory-bodily content co-varied inversely. In the blind scoring, the Spearman correlation was −0.54, with a Pearson coefficient of −0.49 and a value of −0.57 once corrected for the unreliability of the measures. In the adjudicated dataset, the same association was stronger, at −0.78. The direction and rough magnitude held across both scoring procedures; only the strength differed, which is exactly the pattern one expects when the blind scores are noisier but unbiased. The negative affective baseline of the novel was likewise reproduced by every score set: 32, 39 and 37 of the 44 chapters were scored negative across the three procedures, a striking consensus about a text whose narrator lives under the shadow of amputation.

The fate of the three withdrawn dimensions tells the complementary story. Associations involving metaphorical, hypothetical and analogical reasoning differed substantially across scoring procedures: one changed sign, one increased in magnitude, and one approached zero. In other words, whatever relationship the adjudicated dataset suggested for these dimensions was an artifact of the scoring process, not a stable property of the text. Had the team reported only the adjudicated results, readers would have been presented with findings that independent, blind raters could not reproduce—an outcome that mirrors, in miniature, the replication crisis that has reshaped experimental psychology over the past decade.

The authors are careful about what their one secure finding does and does not mean. The inverse coupling of negative affect and concrete sensory-bodily representation in this text is a replicated descriptive pattern, not a measured psychological association, and no causal or generalizing claim follows from it. A single novel, however canonical, cannot establish how pain shapes cognition in general; it can only show that in this particular narrative, written by an author who himself experienced chronic illness, the two features co-occur chapter by chapter in a way that independent readers agree on. The finding is idiographic—about one text, one narrator, one illness trajectory—and the study’s honesty about that scope is part of its methodological message.

The transparency surrounding the use of AI is equally noteworthy. The authors provide a detailed generative AI disclosure: a GPT-4-class model served strictly as a structured first-pass scoring instrument, Claude was used for language editing of author-drafted English text, for organizational assistance in reordering arguments and improving transitions, and for brainstorming and stress-testing the framing of the discussion. No generative AI tool contributed an original scientific idea, hypothesis, theoretical construct or interpretive conclusion, the authors affirm, and all decisions about content and interpretation rest with them. The blind readers scored the chapters without AI assistance of any kind, and their score files are released unmodified as supplementary material, alongside the analysis scripts and the adjudicated dataset. Pre-editing author drafts are retained and available to the editorial office on request, so the scope of AI editing can be inspected directly against the submitted text.

For a field increasingly tempted to let language models do the reading, the study offers a template rather than a warning shot. The wider contribution, the authors write, is methodological: AI-assisted, human-adjudicated annotation produced a coherent coefficient structure of which only one association survived blind rescoring. Reliability, they argue, should be established before substantive interpretation, not after it. Applied to the growing literature in which machines score texts for sentiment, style, metaphor or argument, the lesson is blunt: a coefficient structure that looks impressive in an adjudicated dataset can dissolve under blind replication, and only the findings that survive both stages deserve the name of result. In this case, what survived is a quiet, humanly verifiable pattern in a century-old Turkish novel—that where the pain is sharpest, the language turns toward the body.

Subject of Research: AI-assisted content analysis and blind replication of emotional valence and reasoning styles in a Turkish illness narrative

Article Title: AI-assisted content analysis of a literary illness narrative: a case study in reliability and replication in Peyami Safa’s Dokuzuncu Hariciye Koğuşu (The Ninth External Ward)

Article References: Duran, V., Tuğluk, A., & Karavelioğlu, M. A. (2026). AI-assisted content analysis of a literary illness narrative: a case study in reliability and replication in Peyami Safa’s Dokuzuncu Hariciye Koğuşu (The Ninth External Ward). BMC Psychology. https://doi.org/10.1186/s40359-026-05773-y

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05773-y

Keywords: AI-assisted content analysis, inter-rater reliability, blind replication, emotional valence, reasoning styles, illness narrative, Peyami Safa, Turkish literature, medical humanities, automation bias, intraclass correlation, BMC Psychology

The research team, led by Chloe R. Dupleix with colleagues Jacob P. Rastas, Qianqian Zhao, Robert Young, and Rebecca A. Johnson, set out to answer a deceptively simple question: can a sensor stuck to the skin over a major blood vessel track the core temperature of an anesthetized dog as accurately as a rectal probe? The answer, according to their preliminary data, is a qualified yes, with performance depending on both the sedative drug used before anesthesia and whether the fur at the sensor site has been clipped away.

Core body temperature monitoring is a vital part of physiological evaluation during anesthesia. Depending on the type of procedure and the position in which the patient is placed, obtaining consistent core temperature measurements can be surprisingly difficult. Long surgical drapes, limited access to the patient, and the practical challenges of keeping a rectal probe in place all complicate continuous monitoring. Anesthetized animals cannot regulate their own body temperature effectively, and sedative drugs themselves interfere with the body’s thermoregulatory mechanisms, making accurate and frequent measurement essential for patient safety.

The study enrolled three healthy beagle dogs in a pilot design. Before anesthesia was induced, each dog received a premedication of either dexmedetomidine, a potent alpha-2 adrenergic agonist known to cause significant drops in body temperature, or acepromazine, a commonly used phenothiazine tranquilizer. Anesthesia was then induced with intravenous propofol and maintained with isoflurane, a typical inhalant anesthetic regimen in veterinary practice. Throughout the procedures, the dogs were monitored every five minutes with pulse oximetry, non-invasive blood pressure measurement, electrocardiography, and capnometry, providing a full picture of their anesthetic status.

Temperature data were collected every three minutes for up to 120 minutes following induction. The researchers compared three measurement approaches: conventional rectal thermometry, the CORE sensor placed on clipped skin over the jugular veins, and the same sensor placed on unclipped skin at the same location. Testing the sensor both with and without fur removal was a key design choice, because clipping is time-consuming and may not always be desirable in clinical cases, particularly in patients presenting for non-surgical procedures where an owner may object to shaving a patch of fur.

The CORE sensor works on a heat flux principle. Instead of assuming that skin temperature reflects core temperature, which it often does not, the device measures the actual flow of heat through the skin and uses that measurement to calculate a deeper body temperature estimate. This approach has gained attention in human sports and health monitoring, where athletes and researchers use the sensor to track core temperature during exercise without invasive probes. Applying the same physics to a furry, four-legged patient is far from trivial, because fur adds a variable insulating layer that alters heat flux in unpredictable ways.

To evaluate agreement between the wearable and the rectal reference, the team used two complementary statistical tools. The correlation coefficient for repeated measures, denoted r_rm, captures the strength of the association between two measurement methods when multiple readings are taken from the same subjects over time. The Shrout-Fleiss intraclass correlation coefficient, or ICC, goes further by assessing not just whether two methods move together, but whether they agree closely enough to be used interchangeably in practice. Values approaching 1.0 indicate excellent reliability, while values below roughly 0.5 suggest that agreement is too weak for confident clinical use.

In dogs premedicated with acepromazine, the results were encouraging across the board. There was good, statistically significant correlation and reliability between rectal temperatures and CORE readings on clipped skin, with r_rm equal to 0.81 and a P value below 0.0001, paired with an ICC of 0.78. Performance on unclipped skin was equally strong, showing r_rm of 0.79 with P below 0.0001 and an ICC of 0.85. These findings indicate that when acepromazine is the premedication, the sensor can track rectal temperature closely whether or not the fur is clipped, which is a meaningful practical advantage in a busy clinic.

The picture changed under dexmedetomidine. Although the clipped-skin CORE readings were still statistically correlated with rectal measurements, with r_rm of 0.30 and P below 0.0001, the reliability was poor, at an ICC of only 0.28. Dexmedetomidine produces pronounced peripheral vasoconstriction and alters heat distribution in the body, which likely changes the heat flux signal the sensor depends on. Notably, reliability improved substantially when the sensor was placed on unclipped skin in these same dogs, with r_rm rising to 0.48 and the ICC climbing to 0.88, suggesting that the insulating effect of intact fur may in some circumstances stabilize the heat flux measurement rather than disrupt it.

Beyond the correlation statistics, the researchers also assessed clinical agreement using Bland-Altman analysis, which plots the difference between two methods against their average to reveal systematic bias and limits of agreement. Clinically acceptable Bland-Altman agreement was found between the rectal method and the unclipped CORE method, meaning the wearable’s readings fell within a range veterinarians would consider acceptable for guiding patient care. Taken together, these preliminary results suggest that the CORE sensor may provide a viable alternative method for assessing core body temperature in unclipped, anesthetized dogs. As a pilot study involving three animals, the work cannot yet support universal clinical recommendations, and the authors’ findings will need validation in larger and more diverse patient populations. But the prospect of continuous, wireless, non-invasive core temperature monitoring that does not require shaving or a rectal probe represents an appealing step forward for veterinary anesthesia, where every practical improvement in monitoring translates directly into safer procedures for patients.

Subject of Research: Validation of a wearable heat flux body temperature sensor against rectal thermometry in anesthetized dogs

Article Title: Evaluation of a novel body temperature sensor (CORE) in dogs under general anesthesia following premedication with dexmedetomidine or acepromazine: A pilot study

Article References: Dupleix, C. R., Rastas, J. P., Zhao, Q., Young, R., & Johnson, R. A. (2026). Evaluation of a novel body temperature sensor (CORE) in dogs under general anesthesia following premedication with dexmedetomidine or acepromazine: A pilot study. PLOS One, 21(10), e0360808. https://doi.org/10.1371/journal.pone.0360808

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0360808

Keywords: veterinary anesthesia, dogs, core body temperature, wearable sensor, heat flux thermometry, dexmedetomidine, acepromazine, hypothermia, rectal thermometry, Bland-Altman analysis, intraclass correlation, PLOS One

The concept entered formal law through Directive 2010/63/EU, the European Union’s cornerstone legislation on the protection of animals used for scientific purposes, which calls on institutions to foster a climate of care. Guidance from the RSPCA, the Laboratory Animal Science Association and the EU’s own working documents on animal welfare bodies go further, defining a culture in which all staff, from principal investigators to animal technicians, are encouraged and enabled to take responsibility for animal welfare and ethical standards that go beyond the bare minimum the law requires. Since 2016, an International Culture of Care Network has connected institutions pursuing that ambition. What has been missing, the Utrecht team argues, is a clearly described, literature-based way to actually assess whether such a culture exists on the ground.

The study unfolded in two phases. The first was a scoping review conducted according to the PRISMA-ScR reporting guidelines, the standard framework for transparent scoping reviews. The researchers searched PubMed, Embase and Web of Science with a Boolean strategy combining terms for the Culture of Care and related concepts, such as caring culture, climate of care and organizational culture, with terms for the laboratory animal context, including animal research, in vivo and animal facility. The search, run on April 10, 2025, yielded 155 unique records after deduplication, which were screened independently and blinded by two reviewers. Twenty-one publications survived full-text screening, published between 2007 and 2024 with a median publication year of 2022, and drawn mostly from the United Kingdom, followed by the United States and Germany.

What the review revealed was striking. Most of the included articles described the Culture of Care at a purely conceptual level, and not a single one used a validated assessment instrument. Eleven articles included some form of assessment, six using surveys and five using qualitative methods such as interviews, but the approaches varied widely in scope and rigor. A quality assessment of the four survey-based studies, scored against simplified criteria drawn from the SURGE and CHERRIES guidelines, found that formal reliability reporting remained limited and bias control was generally modest. One study developed a 12-item scale evaluated with principal component analysis; others reported internal wellbeing surveys or questionnaires distributed through emails and mailing lists with little methodological detail. In short, the field had plenty of vision and very little instrumentation.

From the 21 publications, supplemented by policy documents from the EU and the International Culture of Care Network, the researchers distilled nine recurring thematic domains. Compliance with organizational and legal standards was the most universal, explicitly mentioned in 19 of the 21 articles, reflecting the view among many authors that meeting legal requirements is the starting point rather than the endpoint. Next came a proactive attitude that goes beyond compliance, mentioned in nine articles, followed by strong and caring leadership and sufficient resources and conditions to do a proper job, each named in 13 articles. Empathy and respect towards colleagues and animals, shared responsibility and accountability, and transparency and open communication each appeared in 15 to 17 articles, while a supportive work environment and team collaboration appeared in 16, and sharing knowledge and continuous reviewing in 15.

Several of these themes carry a distinctly human dimension that has gained prominence over the past decade. The Culture of Care builds on animal welfare science by emphasizing organizational culture and staff wellbeing, and the review noted that attention has shifted from animal care alone to the people who provide it. Compassion fatigue, the emotional exhaustion that arises from ongoing exposure to animal suffering, ethical tension and high workload, can erode vigilance, empathy and professional judgment, and thereby feed back into animal welfare. A related scoping review cited in the study found that the most common strategy suggested to mitigate work-related stress among laboratory animal personnel was acknowledgement and validation of their feelings. The new questionnaire reflects this dual focus, with items on feeling safe to share emotional reactions, knowing where to find support, and still having energy after a working day.

The second phase translated the nine themes into a measurable instrument. The researchers drafted 50 positively worded statements in the first person and present tense, each rated on a five-point Likert scale from strongly disagree to strongly agree, a design chosen to reduce cognitive load and encourage reflection on personal experience. Four questions were adopted directly from the NHS Culture of Care Barometer, a validated tool from human healthcare, using both the original English version and a published Dutch translation. Eleven staff members of the Animal Welfare Body Utrecht were invited to rate each item’s relevance and provide open-text feedback on clarity, redundancy and translation accuracy; eight responded, with mean relevance scores ranging from 3.8 to 4.8. After discussion with three experts, items with low relevance, overlap or vague wording were removed or merged, leaving a final 30-item questionnaire covering all nine themes.

The team then put the instrument through a test-retest study, asking participants to complete the same questionnaire twice, two weeks apart. The first round targeted 50 laboratory animal coordinators at Utrecht University and UMC Utrecht, of whom 15 completed the first round and 5 the second. An open call in October 2025 on MyDALAS, the platform of the Dutch Association for Laboratory Animal Science with roughly 700 active members, and on the International Culture of Care Network added 12 more test respondents, 11 of whom completed the retest. In total, 16 participants completed both timepoints. Internal consistency was very high at both timepoints, with Cronbach’s alpha of 0.917 at test and 0.898 at retest, well above the commonly cited acceptable range of 0.70 to 0.95, though values above 0.90 can hint at some redundancy among items.

Test-retest reliability, assessed with two-way absolute-agreement single-measure intraclass correlation coefficients, told a more nuanced story. The total-scale ICC was 0.73, with a 95 percent confidence interval of 0.38 to 0.90, which the authors characterize as moderately good, and the wide interval is expected given the small sample. At the level of individual items, ICCs ranged from poor, below 0.50, to good, between 0.75 and 0.90, with broad confidence intervals. The researchers explain this variability through classical test theory: single items carry more random error, while aggregated scores gain reliability as that error cancels out. They also identified ceiling effects, noting that one item was answered almost exclusively with agree or strongly agree at both timepoints, leaving too little variance between respondents to detect meaningful stability. Item-level results, they caution, should be interpreted with care.

The study’s limitations are candidly acknowledged. The two recruitment rounds differed in timing and incentives, with the second offering a randomly awarded 30 euro gift card, and participants recruited through a network already engaged with Culture of Care initiatives may have been more positively inclined toward the concept, introducing selection bias. Response rates could not be calculated because the number of people who saw the open calls is unknown, and no demographic data were collected to preserve anonymity in the small sample. Still, the authors argue, the work delivers what the field has lacked: a structured overview of how the Culture of Care is described, a nine-theme analytical framework, and a questionnaire with initial evidence of internal consistency and preliminary test-retest reliability. The next challenges are construct validity, factor structure, subgroup differences and performance in larger, more diverse samples, including whether the tool can detect change over time. If those steps succeed, institutions may finally be able to measure the invisible climate they have long been asked to cultivate.

Subject of Research: Development and testing of a Culture of Care questionnaire for laboratory animal institutions

Article Title: From concept to measurement: a scoping review and development of a Culture of Care questionnaire for laboratory animal institutions

Article References: Weber, O. M., van de Kolk, S. A. C., & Van Loo, P. L. P. (2026). From concept to measurement: a scoping review and development of a Culture of Care questionnaire for laboratory animal institutions. Frontiers in Ethology, 5, Article 1796584. https://doi.org/10.3389/fetho.2026.1796584

Image Credits: AI Generated

DOI: 10.3389/fetho.2026.1796584

Keywords: Culture of Care, laboratory animals, animal welfare, staff wellbeing, scoping review, questionnaire development, test-retest reliability, Cronbach's alpha, intraclass correlation, Directive 2010/63/EU, 3Rs, organizational culture

The central problem the researchers set out to solve is one that has long troubled evaluation and assessment research in health communication. When scientists or clinicians create visual materials—illustrations, artworks, or diagrams intended to convey health information—there is often no standardized, construct-based way to judge whether those images actually communicate biological meaning. Judgments tend to be subjective, varying from one reviewer to the next, and the criteria behind them are rarely made explicit. The team grounded their solution in two established theoretical frameworks: the Biopsychosocial Model, which frames health as an interplay of biological, psychological, and social factors, and Visual Literacy Theory, which describes how viewers construct meaning from images. From these foundations, they built a 20-item scale designed to measure how biologically interpretable a visual stimulus is.

Developing the scale was not a solitary act of authorship but a structured consensus exercise. The researchers employed a two-stage modified Delphi process, a method widely used in instrument development to distill expert opinion into stable, agreed-upon items. In the first stage, a joint panel considered both the images and the candidate items together, allowing reviewers to see how proposed criteria behaved when applied to actual stimuli. In the second stage, an independent panel evaluated the items in isolation, without the images, to test whether each item stood on its own as a clear and meaningful criterion. This dual-panel design was intended to strip out items that only made sense in context and to retain those that could function as reliable, reusable measurement tools.

The final instrument produces scores across four dimensions. Three of these are biomedical subscales, capturing different facets of biological interpretability in health-related imagery, while the fourth, Non-Medical Interpretability, assesses meaning that lies outside the biomedical domain. With the scale in hand, the team turned to the question at the heart of the study: could multimodal AI systems—models that process both images and text—apply this framework in a reproducible way, functioning as standardized raters that do not fatigue, disagree idiosyncratically, or drift in their criteria?

To answer it, the researchers assembled a small but carefully controlled stimulus set: six artworks, four of them health-themed and two serving as controls. Three multimodal AI models and four researchers, who served as an internal human reference, evaluated the images using a standardized protocol. The protocol’s design reveals how seriously the team took the methodological pitfalls of AI assessment. Each model was accessed through independent researcher-operated accounts, memory-isolation procedures were used to prevent the systems from carrying information from one evaluation into the next, and assessments were repeated across two separate days. This allowed the team to measure not only whether different accounts produced similar scores, but whether the same system produced stable scores over time.

The statistical approach was deliberately matched to the limitations of the data. Because the image type variable spanned only six stimuli—four health-themed and two controls—the researchers summarized health-versus-control differences at the image level using Cohen’s d, a standardized measure of effect size, and treated human–AI differences descriptively at the rater level rather than claiming statistical significance. Reproducibility was assessed with intraclass correlation coefficients, or ICCs, which quantify how strongly scores agree across raters or occasions, supplemented by Spearman correlations. This restraint matters: with so few images, the study cannot deliver definitive population-level conclusions, and the authors are explicit that their findings are preliminary methodological evidence rather than established generalizations.

Within that constraint, the pattern of results is instructive. The AI-generated scores were descriptively higher for health-themed images than for control images on the three biomedical subscales, with image-level Cohen’s d values ranging from 0.91 to 1.38—effect sizes that, in a larger sample, would typically be considered large. On the Non-Medical Interpretability subscale, the difference was smaller, at d = 0.49. With only four health-themed and two control images, none of these differences reached statistical significance, with p values ranging from 0.19 to 0.60. In other words, the machines appeared to distinguish health-themed from control material in the expected direction, particularly on biomedical criteria, but the sample was too small to confirm that the distinction was real rather than chance.

Reproducibility told a more nuanced story. Across researcher-operated accounts, agreement was excellent for the three biomedical subscales, with ICCs between 0.90 and 0.97—figures that suggest the AI systems, when run under controlled conditions, applied the biomedical criteria with remarkable consistency. Agreement on the Non-Medical Interpretability subscale was moderate, at an ICC of 0.57, indicating that the machines were considerably less consistent in judging meaning outside the biomedical domain. Most strikingly, agreement across assessment days was poor, with ICCs ranging from just 0.04 to 0.46. Whatever produces this day-to-day instability—whether model updates, sampling variability in the models’ generative processes, or subtle differences in prompt handling—it represents a genuine obstacle to using AI raters as stable measurement instruments without further safeguards.

The comparison with human raters added another layer of complexity. The four researchers’ internal reference scores ran higher than the AI-generated scores on the biomedical subscales for health-themed images, with mean differences of 1.41 to 1.96 points, suggesting the humans saw more biological interpretability in the health-themed artworks than the machines did. On Non-Medical Interpretability, the pattern reversed: the human scores were lower than the AI scores by 2.02 to 2.19 points, meaning the machines detected more non-medical meaning than the human panel. This divergence is intriguing, because it hints that AI systems and trained human raters may weight different aspects of an image differently, and that calibration between human and machine judgment is not automatic. It also flags the Non-Medical Interpretability subscale as the weakest component of the instrument; the authors report that it showed substantial internal-consistency problems, meaning its items did not hang together as a coherent measure, and it requires further refinement.

The authors are careful about what their study does and does not establish. The findings, they conclude, provide preliminary methodological evidence that multimodal AI can reproducibly apply a structured interpretability framework to controlled visual stimuli, supporting the feasibility of using AI as standardized non-human raters within controlled methodological settings. But the caveats are substantial: day-to-day stability must be improved, the Non-Medical Interpretability subscale needs rework, the stimulus set was tiny and artificial, and the human reference was an internal panel rather than a diverse group of independent external raters. Broader validation using authentic health communication materials—real patient education graphics, public health campaigns, clinical illustrations—and independent external human raters remains necessary before AI raters could be trusted in applied settings. Even so, the study opens a provocative line of inquiry for a field that increasingly relies on visual media: if the criteria for biological interpretability can be made explicit enough for a machine to apply them consistently, then the machines may eventually become not just tools that make images, but disciplined judges of what those images actually communicate about health.

Subject of Research: Using multimodal artificial intelligence as standardized non-human raters to assess biological interpretability of health-related visual stimuli

Article Title: Operationalizing biological interpretability in health-related visual stimuli using multimodal artificial intelligence as standardized non-human raters

Article References: Ayman, D., Ozkan, H., Ayman, D., & Tanriverdi, O. (2026). Operationalizing biological interpretability in health-related visual stimuli using multimodal artificial intelligence as standardized non-human raters. Scientific Reports. https://doi.org/10.1038/s41598-026-75309-2

Image Credits: AI Generated

DOI: 10.1038/s41598-026-75309-2

Keywords: artificial intelligence, multimodal AI, health communication, visual literacy, biopsychosocial model, psychometrics, Delphi method, intraclass correlation, medical humanities, image evaluation, Scientific Reports, non-human raters

Non-persistent chemicals are those that do not linger in the body for months or years, unlike legacy pollutants such as persistent organic pollutants or heavy metals. Instead, they are metabolized and excreted within minutes, hours, or days. This category includes an enormous range of compounds that pregnant women encounter daily: phthalates from food packaging and personal care products, parabens from cosmetics, benzophenones from sunscreens, bisphenols from plastics, antimicrobials from soaps, pesticide residues, flame retardant metabolites, polycyclic aromatic hydrocarbons from combustion sources, melamine derivatives, and aromatic amines from tobacco smoke and dyes. Because these chemicals clear the body so rapidly, urinary biomarkers essentially reflect very recent exposure. Whether a single measurement can stand in for exposure over an entire pregnancy depends entirely on how stable an individual’s concentrations remain over months of gestation.

The research team, led by Giehae Choi of Johns Hopkins University together with colleagues from institutions across the United States, tackled this question by analyzing three urine samples collected at different time points during pregnancy from 90 participants at three ECHO Cohort sites: New York, Illinois, and Puerto Rico. The scale of the chemical analysis was remarkable. The laboratory measured 154 analytes representing 163 chemicals across 12 chemical classes, using liquid chromatography-tandem mass spectrometry methods at the Wadsworth Center Human Health Exposure Analysis Resource laboratory. Of these, 84 analytes were detected in at least 20 percent of samples at a given site, providing sufficient data for variability analysis.

The central statistical tool was the intraclass correlation coefficient, or ICC, which quantifies how reliably a single measurement represents a person’s typical level across repeated measurements. ICCs range from 0 to 1, with values approaching 1 indicating that concentrations are highly stable within an individual over time, and values near 0 indicating that measurements are essentially unpredictable from one time point to the next. The team calculated ICCs for 58 analytes with detection frequencies above 60 percent, using log2-transformed concentrations standardized for urinary dilution by specific gravity. They also computed kappa statistics for 56 analytes with more moderate detection rates, assessing whether the mere presence or absence of a chemical above the detection limit was consistent across the three pregnancy samples.

The results were strikingly heterogeneous. ICCs ranged from −0.12 to 0.91, and the pattern of high versus low reliability differed substantially by cohort site for most chemicals. Only two analytes bucked this trend: benzophenone-1 and propyl paraben showed consistently high ICCs above 0.5 at all three sites, indicating stable exposure throughout pregnancy, while bisphenol F showed consistently low ICCs below 0.1 everywhere. Beyond these exceptions, the site-specific patterns were revealing. All parabens showed high reliability in Illinois but lower reliability in Puerto Rico and New York. Benzophenones were more stable in New York. Antimicrobials were more stable in Puerto Rico. Most phthalate and polycyclic aromatic hydrocarbon metabolites were moderately to highly reliable in Illinois but less so in Puerto Rico, while most aromatic amines showed the opposite pattern, with moderate to high reliability in New York but lower reliability in Illinois.

One of the study’s most technically insightful contributions was its analysis of how concentration levels relative to detection limits shape reliability estimates. The researchers hypothesized that measurement error exerts a proportionally greater influence when concentrations sit close to the limit of detection, obscuring the true rank-ordering of individuals. They therefore calculated the ratio of each analyte’s site-specific median concentration to its detection limit. Chemicals with median concentrations more than 50 times the detection limit tended to show greater stability across pregnancy, with notable exceptions among certain phthalate metabolites. This exception is itself informative: high concentrations paired with low ICCs suggest that exposure to the source products occurs sporadically but intensely, rather than continuously. Conversely, chemicals with median concentrations within tenfold of the detection limit generally showed lower ICCs, indicating that low-level measurements may reflect recent, sporadic exposures rather than a chronic exposure state. Benzophenone-8 provided a fascinating exception, maintaining moderate ICCs despite low concentrations, which the authors interpret as evidence of near-continuous low-level exposure.

The consistently high ICCs for benzophenone-1 deserve particular attention given the widespread use of benzophenones as ultraviolet filters in sunscreens, cosmetics, and personal care products, as well as their use as UV stabilizers in food packaging. Because these compounds do not bioaccumulate and have short biological half-lives, stable urinary concentrations can only mean that exposure itself is nearly continuous throughout pregnancy. Benzophenone-1 may appear more stable than benzophenone-3 partly because it is a major metabolite of benzophenone-3, integrating exposure from multiple sources, and because only a small fraction of benzophenone-3 is excreted unchanged in urine. The finding that benzophenone-3 was stable in Puerto Rico and New York but not Illinois may reflect seasonal differences in sunscreen use, a reminder that exposure context shapes biomarker reliability.

The site-specific differences also point to the influence of demographic, geographic, and behavioral factors on exposure chronicity. The New York site, where a majority of participants were classified in the high cotinine category suggesting substantial tobacco smoke exposure, showed moderate to high ICCs for aromatic amines such as aniline, which is found in tobacco smoke. Continuous exposure to tobacco through active or passive smoking could plausibly explain this stability. Similarly, moderate ICCs for melamine analogs like ammelide in New York may reflect ongoing exposure through contaminated foodstuffs and water. These observations suggest that individual, cultural, and structural factors, rather than pure physiology alone, determine whether a person’s chemical exposures remain constant or fluctuate across the months of gestation.

The practical consequences for environmental epidemiology are substantial. For the majority of non-persistent chemicals examined, a single pregnancy urine sample appears to represent exposure near the time of collection rather than exposure across the entire pregnancy. This matters enormously because different gestational windows carry different vulnerabilities: organogenesis in the first trimester, neurodevelopment throughout gestation, and fetal growth in later trimesters may each be sensitive to chemical exposures at distinct time points. If researchers link a single third-trimester measurement to a birth outcome and interpret it as reflecting first-trimester exposure, they risk substantial exposure misclassification, typically biasing estimated health associations toward the null. The authors note that simulation work suggests a chemical with an ICC below 0.2 would require more than 35 samples per person to keep this bias below 10 percent, an approach that is neither affordable nor practical for most studies.

Fortunately, the study outlines realistic alternatives. Researchers can incorporate validation sub-studies with repeated measurements and apply measurement error correction techniques such as regression calibration or simulation extrapolation. They can time urine collection to align with known critical windows of developmental susceptibility rather than defaulting to arbitrary collection schedules. They can collect multiple samples and pool them within each subject, which reduces assay costs while improving the representation of typical exposure. Or they can turn to alternative biospecimens such as hair, which integrates exposure over longer periods. The study also provides some of the first evidence on within-pregnancy stability for understudied chemical classes, including halogenated phenols such as the carcinogen pentachlorophenol, melamine and its analogs, and aromatic amines, filling critical gaps in the exposure science literature. As chemical production and consumer patterns evolve, the authors caution that reliability estimates from earlier studies may not transfer to current exposures, making ongoing characterization of biomarker variability an essential, not optional, component of rigorous environmental health research. For now, the message is clear: when it comes to non-persistent chemicals in pregnancy, one sample tells you where a woman was this week, not where she has been for nine months.

Subject of Research: Intra-individual variability of urinary non-persistent chemical biomarkers measured at three time points during pregnancy

Article Title: Intra-individual variability of urinary non-persistent chemical concentrations during pregnancy in the ECHO Cohort

Article References: Choi, G., Xun, X., Bennett, D. H., Meeker, J. D., Sathyanarayana, S., Schantz, S. L., Trasande, L., Watkins, D. J., Pellizzari, E., Li, W., Kannan, K., Woodruff, T. J., Buckley, J. P., & for the ECHO Cohort Consortium (2026). Intra-individual variability of urinary non-persistent chemical concentrations during pregnancy in the ECHO Cohort. Journal of Exposure Science & Environmental Epidemiology. https://doi.org/10.1038/s41370-026-00979-z

Image Credits: AI Generated

DOI: 10.1038/s41370-026-00979-z

Keywords: ECHO Cohort, pregnancy, non-persistent chemicals, urinary biomarkers, intraclass correlation, phthalates, parabens, benzophenones, exposure assessment, environmental epidemiology, aromatic amines, melamine

The research team, led by Hannah Lundh of the Centre for Physical Activity in Region Västra Götaland and colleagues at the University of Gothenburg, recruited 39 children aged 6 to 12 who had been referred to outpatient paediatric clinics for obesity treatment, along with one parent each. The children, 19 of them girls, had a mean age of 9.7 years and a mean body mass index standard deviation score of 3.0, indicating severe obesity relative to Swedish reference populations. Their parents, 20 of them mothers, averaged 42.2 years of age. Fifty-seven percent of the parents were born in Sweden, 16 percent in other European countries, and 27 percent outside Europe, giving the sample a broadly diverse socioeconomic and cultural profile.

Each child-parent pair wore a small Axivity AX3 accelerometer on an elastic belt around the waist, positioned over the right hip close to the body’s centre of mass, for seven consecutive days and nights. The devices sampled movement at 50 Hz with an acceleration range of plus or minus 8 g, and participants kept diaries of bedtime, waking time, cycling, water activities and strength training to help researchers identify sleep and non-wear periods. A valid day required at least 10 hours of wear time, and a valid week required at least two such days — a deliberately lenient threshold that sensitivity analyses showed made no difference compared with the traditional four-day requirement, but which maximised the number of pairs that could be included.

The data processing was where the study distinguished itself technically. Rather than applying conventional low-pass filters that can misclassify vigorous movement as noise, the team used the 10 Hz Frequency Extended Method, which has been shown in calibration work to capture higher-intensity activity more faithfully. Data were reduced into 3-second epochs, a window short enough to register the brief, sporadic bursts that characterise children’s movement. Perhaps most notably, the researchers set the cut-point for moderate-to-vigorous physical activity at 4 metabolic equivalents rather than the traditional 3 METs, arguing that 3 METs corresponds to a slow walking pace of just 3 to 4 kilometres per hour and shows minimal association with cardiometabolic health, whereas clinically meaningful benefits begin around 4 METs, roughly a brisk walk of 4 to 5 kilometres per hour. Cut-points of 52 mg and 388 mg for children and 39 mg and 331 mg for adults marked the boundaries between light activity and moderate-to-vigorous activity.

The headline numbers were sobering. Children spent an average of 9.8 hours per day in sedentary behaviour and just 0.7 hours — about 42 minutes — in moderate-to-vigorous activity, with another 3.9 hours in light activity. Their parents sat even longer, averaging 12.3 sedentary hours daily, with only 0.3 hours of moderate-to-vigorous activity and 3.4 hours of light activity. Only six of the 39 children, or 15 percent, met the World Health Organization recommendation of at least 60 minutes of moderate-to-vigorous activity per day for ages 5 to 17. Among parents, the picture was bleaker still: just two of 38, or 5 percent, achieved the adult guideline of 150 minutes of moderate activity per week.

The central analytical question was whether children’s and parents’ behaviours tracked together within each pair. Using intraclass correlation coefficients in two-way mixed-effect models, the team found a statistically significant but weak correlation for sedentary time, with an ICC of 0.14 and a p-value of 0.017. By the conventional interpretation scheme of Koo and Li, any value below 0.5 counts as weak, so the shared sitting habit was real but modest. The correlation for light activity was stronger, at 0.24, but fell just short of significance with a p-value of 0.050. For moderate-to-vigorous activity, the correlation was weaker still, at 0.11, and clearly non-significant. Bland-Altman plots visualising the within-pair differences showed enormous variability: some pairs moved in near lockstep, while others diverged dramatically.

Interestingly, the generations differed systematically in how they spent their days. Children accumulated significantly more moderate-to-vigorous activity than their parents — a mean paired difference of roughly 27 minutes per day — while parents sat significantly more than their children. This inversion of the intuitive picture, in which adults might be expected to be the more active party, underscores how deeply sedentary adult lifestyles can become even in families actively engaged with paediatric obesity services. It also suggests that children in this clinical population, despite falling far short of guidelines, remain more physically capable of spontaneous movement than their parents.

The sedentary finding carries particular weight because it contradicts much of the existing parent-child literature. Previous systematic reviews, covering predominantly healthy, non-obese children, have generally reported weak to moderate positive associations between parent and child physical activity, and a large meta-analysis found a small effect of parental modelling on children’s activity. Only one of the 65 studies across two major reviews had examined accelerometer-measured associations in an obese paediatric population. The Swedish authors suggest several reasons their results may diverge: earlier studies often relied on self-report, which is prone to bias; many used the lower 3 METs threshold, inflating apparent activity levels; and crucially, previous cohorts tended to include younger children, aged 2 to 10, who are more dependent on their parents and more likely to share their activities. Children aged 6 to 12, by contrast, increasingly move in school and peer contexts their parents never see.

The study’s limitations are candidly acknowledged. The cross-sectional design rules out any causal inference — sitting together may reflect shared environment, modelling, genetic predisposition, or all three. The sample of 39 pairs fell short of the original target of 60, partly because of pandemic-related staffing problems, and no formal power calculation was performed, so a larger sample might have detected significant correlations that here hovered near the threshold of significance, particularly for light activity. The team did not collect parents’ body mass index, which likely influenced parental activity levels given the genetic component of obesity, nor did they record living arrangements, meaning some children may have spent the measurement week with the non-participating parent. Selection bias is also possible, since families with positive attitudes toward activity may have been more willing to enrol.

Still, the implications for treatment are concrete. The WHO recommends family-based, multi-professional interventions as first-line treatment for childhood obesity, and the Swedish physical activity on prescription model places parental involvement at its core. If sedentary time is the one behaviour where parent and child genuinely mirror each other, the authors argue, then interventions for this population might profitably begin by targeting sitting time together — a change that over time could open the door to increased activity — while recognising that role modelling through parental exercise may matter more for some families than others. Facilitating access to activity opportunities, initiating outings and offering encouragement may ultimately prove more powerful levers than a parent’s own workout routine. Larger samples, other settings and longitudinal designs, the researchers conclude, are needed to untangle whether shared sedentary habits are cause, consequence or coincidence in the trajectory of childhood obesity.

Subject of Research: Parent-child associations in accelerometer-measured physical activity and sedentary behaviour among children with obesity

Article Title: Physical activity and sedentary behaviour amongst children with obesity – exploring cross-sectional associations between child and parent

Article References: Lundh, H., Arvidsson, D., Greven, C., Fridolfsson, J., Börjesson, M., Boman, C., Lauruschkus, K., Lundqvist, S., Melin, K., & Bernhardsson, S. (2025). Physical activity and sedentary behaviour amongst children with obesity – exploring cross-sectional associations between child and parent. Journal of Activity, Sedentary and Sleep Behaviors, 4(1), Article 2. https://doi.org/10.1186/s44167-025-00072-0

Image Credits: AI Generated

DOI: 10.1186/s44167-025-00072-0

Keywords: childhood obesity, physical activity, sedentary behaviour, accelerometry, parent-child dyads, moderate-to-vigorous physical activity, paediatric health care, Sweden, family-based intervention, intraclass correlation, WHO guidelines, behavioural modelling