The study included 1,255 adults: 251 people with lung cancer and 1,004 controls. Researchers fitted a multivariable logistic regression model and assessed internal performance with 10-fold cross-validation and 2,000 bootstrap resamples. The abstract reports that performance was consistent across the bootstrap resampling and cross-validation analyses.

The final model included eight predictors: age, sex, household solid fuel use, asbestos exposure, years of smoking, years of exposure to secondhand smoke, prior tuberculosis, and family history of cancer. Investigators also derived a simplified, point-based risk score intended to improve practical applicability. The abstract states that the score had discriminatory performance and classification accuracy comparable with the full model. The supplied abstract does not provide confidence intervals for the reported performance measures.

These findings describe a prediction model and its performance within the studied population, rather than evidence that using the tools improves clinical outcomes or produces benefits from screening or prevention interventions. The case-control design and internal validation limit the conclusions that can be drawn about performance in other populations. External validation is required to establish generalizability. The authors further state that recalibration to population-level incidence would be needed if the model is used to estimate absolute risk or define population-based clinical or public health thresholds. The article was published in BMC Cancer as open-access research; the version supplied was identified as subject to further edits before replacement by the final Version of Record.

Subject of Research: Internal validation of a locally developed lung cancer risk prediction model in Ethiopia

Article Title: Development and validation of a lung cancer risk prediction model in Ethiopia

Article References: Estifanos, N., Egata, G., Addissie, A., Yusuf, H., Bitew, A., Getachew, A., & Deyessa, N. (2026). Development and validation of a lung cancer risk prediction model in Ethiopia. BMC Cancer. https://doi.org/10.1186/s12885-026-16997-x

Image Credits: AI Generated

DOI: 10.1186/s12885-026-16997-x

Keywords: lung cancer, risk prediction, case-control study, internal validation, logistic regression, AUC, Ethiopia, risk score, Development, validation, lung, cancer

The research team set out to answer a deceptively simple question: once a clinician knows a child’s age, Pediatric Appendicitis Score, and C-reactive protein level, how much additional diagnostic value does an appendicitis-compatible initial ultrasound report actually provide? This question matters because ultrasound is widely used as the first-line imaging modality in children with suspected appendicitis, particularly given concerns about radiation exposure from computed tomography in young patients. Yet the incremental value of the initial ultrasound report, considered alongside established clinical variables, has remained surprisingly unclear in the literature. Prior studies have integrated ultrasound findings with clinical scores, refined sonographic criteria using age-based appendiceal measurements and secondary signs, and explored the meaning of partial appendix visualization, but few prospective cohorts have directly measured the statistical gain that ultrasound adds to a baseline clinical model.

To address this gap, the investigators conducted a prospective single-centre cohort study enrolling children aged 3 to 17 years who were evaluated for suspected acute appendicitis between May 2025 and March 2026. The primary outcome was histopathologically confirmed appendicitis, a rigorous standard that anchors the diagnostic analysis in tissue-level truth rather than clinical impression alone. The study design compared two statistical models. The first, a clinical model, included age, the total Pediatric Appendicitis Score, and log2-transformed C-reactive protein values. The second, an ultrasound-extended model, added an appendicitis-compatible initial ultrasound report to those same variables. Model performance was assessed using the area under the receiver operating characteristic curve, known as the AUC, along with the Brier score, likelihood-ratio testing, and the DeLong method for comparing correlated receiver operating characteristic curves.

The cohort comprised 135 children evaluated through an ultrasound-first pathway. Of these, 81 had histopathologically confirmed appendicitis, while 54 formed the comparison group of children in whom appendicitis was ultimately excluded. The clinical model alone demonstrated moderate apparent discrimination, achieving an AUC of 0.771 with a 95 percent confidence interval spanning 0.689 to 0.852, and a Brier score of 0.188. These figures describe a model that performs meaningfully above chance but still leaves substantial diagnostic uncertainty in individual cases, a familiar reality for clinicians who know that scores and inflammatory markers can overlap considerably between children who need surgery and those who do not.

The headline result emerged when ultrasound entered the equation. Among the 132 children whose initial ultrasound reports could be classified, the addition of an appendicitis-compatible ultrasound report increased the AUC from 0.772 to 0.912 and reduced the Brier score from 0.187 to 0.113. In practical terms, the AUC jump represents a substantial improvement in the model’s ability to separate children with confirmed appendicitis from those without it, moving discrimination from the moderate range toward levels typically associated with highly reliable diagnostic tools. The Brier score, which measures the mean squared difference between predicted probabilities and actual outcomes, fell by nearly 40 percent, indicating that the ultrasound-extended model produced not only better ranking of patients but also better-calibrated probability estimates. The likelihood-ratio framework confirmed that the improvement was statistically meaningful rather than a product of chance variation within the sample.

These numbers carry weight for anyone who has followed the evolution of pediatric appendicitis diagnostics. The Pediatric Appendicitis Score, originally described by Samuel in 2002, combines clinical findings such as migration of pain, anorexia, fever, and rebound tenderness into a structured assessment. C-reactive protein, an acute-phase reactant measurable within hours, has been validated through systematic reviews and meta-analyses as a useful but imperfect marker of inflammation in pediatric appendicitis. Previous work by Bachur and colleagues in 2015 demonstrated the value of integrating ultrasound findings with clinical scores, and subsequent studies have refined sonographic criteria by combining age-based appendiceal size thresholds with secondary findings such as periappendiceal fluid and echogenic fat. The new study extends this lineage by prospectively quantifying, within a single cohort, the exact statistical increment that a clear ultrasound report delivers on top of those established variables.

The study also examined the role of post-ultrasound computed tomography within the same ultrasound-first pathway. CT was performed in 52 of the 135 children, representing 38.5 percent of the cohort. Among those children, 29 had appendicitis, while 23 did not undergo appendectomy. This figure illustrates a persistent tension in pediatric imaging strategy: ultrasound is favored as the initial modality because it avoids ionizing radiation, but equivocal or negative ultrasound findings in children with persistent clinical suspicion often prompt a second-line CT examination. The ACR Appropriateness Criteria for suspected appendicitis in children endorse this staged approach, and prior implementation studies have shown that ultrasound and fast magnetic resonance imaging pathways can reduce CT utilization. The new data suggest that when the initial ultrasound report is clearly appendicitis-compatible, its diagnostic contribution is strong enough to anchor clinical decisions, while the residual group with inconclusive imaging continues to require careful escalation.

Methodologically, the choice of log2-transformed C-reactive protein reflects the skewed distribution typical of inflammatory markers, where a small number of children with very high values would otherwise dominate a linear model. The use of the DeLong comparison, a standard method for testing whether two correlated AUCs differ significantly, strengthens the inference that the ultrasound-extended model genuinely outperforms the clinical baseline rather than merely appearing to do so in this particular sample. The Brier score decomposition further distinguishes discrimination from calibration, two properties of a predictive model that can diverge: a model may rank patients correctly yet assign probabilities that are systematically too high or too low. The observed reduction from 0.187 to 0.113 indicates improvement on both fronts, which is precisely what clinicians need when translating a statistical model into bedside decisions about observation, imaging escalation, or surgical consultation.

The authors frame their conclusion carefully: a clear appendicitis-compatible initial ultrasound report provided substantial incremental diagnostic information when added to age, Pediatric Appendicitis Score, and C-reactive protein, supporting the role of ultrasound as an integrated component of the diagnostic assessment rather than a standalone test. This framing resists the common temptation to treat any single test as definitive. Appendicitis in children remains a condition in which history, physical examination, laboratory data, and imaging each contribute partial information, and the art of diagnosis lies in combining them coherently. The study’s prospective design, histopathological gold standard, and formal statistical comparison give its central claim unusual solidity for a single-centre cohort.

The implications reach beyond the radiology suite. For emergency physicians and pediatric surgeons, the results support protocols in which ultrasound findings are interpreted in the context of validated scores and inflammatory markers, rather than in isolation. For health systems weighing imaging investments, the data quantify the value of maintaining high-quality pediatric ultrasound services as the backbone of an ultrasound-first pathway. And for researchers, the study provides a template for measuring incremental diagnostic value with the rigor that the field increasingly demands. As pediatric appendicitis diagnostics continue to evolve, with standardized reporting templates, refined sonographic criteria, and alternative second-line modalities all under active investigation, this prospective cohort from Kütahya offers a clear quantitative answer to a question that clinicians have long asked in qualitative terms: ultrasound matters, and now the size of its contribution can be stated with confidence.

Subject of Research: Diagnostic value of ultrasound in pediatric suspected acute appendicitis

Article Title: Incremental diagnostic value of ultrasound in children with suspected acute appendicitis

Article References: Ibrahimoglu, H., & Sahin, S. (2026). Incremental diagnostic value of ultrasound in children with suspected acute appendicitis. Pediatric Radiology. https://doi.org/10.1007/s00247-026-06812-x

Image Credits: AI Generated

DOI: 10.1007/s00247-026-06812-x

Keywords: acute appendicitis, ultrasound, pediatrics, Pediatric Appendicitis Score, C-reactive protein, diagnostic accuracy, computed tomography, pediatric radiology, AUC, emergency medicine, histopathology, ultrasound-first pathway

The research team enrolled 460 patients with pathologically confirmed granulomatous lobular mastitis or invasive ductal carcinoma, splitting them randomly into a training cohort of 322 patients and an internal-validation cohort of 138 patients at a 7:3 ratio. From grayscale ultrasound images of each lesion, the investigators extracted hundreds of quantitative radiomic features, mathematical descriptors of texture, intensity distribution, and spatial patterns that are invisible to the naked eye. Using least absolute shrinkage and selection operator regression, commonly known as LASSO, they compressed this high-dimensional feature space into a single radiomics score, or Radscore, designed to capture the subtle image signatures that distinguish inflamed glandular tissue from malignant tumor.

But the study did not stop at image texture. The researchers built what they call a multimodal ultrasound framework, combining conventional grayscale features with clinical variables and elastography, a technique that measures how stiff a lesion is by applying gentle compression and observing tissue deformation. Cancers tend to be stiffer than inflammatory masses, and elasticity scoring has long been used as an adjunct in breast ultrasound. By integrating age, lesion shape, margin characteristics, the presence of calcifications, and elasticity score alongside the Radscore, the team constructed four competing models: an age-only baseline, a clinical multimodal ultrasound model without radiomics, a radiomics-only model, and a combined model incorporating everything.

The statistical engine behind the final tool was multivariate logistic regression, a method that estimates the independent contribution of each predictor while adjusting for the others. Six predictors survived the selection process and were woven into a nomogram, a graphical scoring chart that allows clinicians to assign points to each feature and read off a probability that a given lesion is granulomatous lobular mastitis rather than cancer. Nomograms have become increasingly popular in oncology because they translate complex statistical models into something a clinician can actually use at the bedside or in the imaging suite.

The performance figures in the internal-validation cohort were striking. The clinical multimodal ultrasound model, which excluded the radiomics signature entirely, achieved an area under the receiver operating characteristic curve, or AUC, of 0.978, with a 95 percent confidence interval stretching from 0.956 to 0.999. Its sensitivity for detecting the condition reached 0.972, meaning it correctly identified nearly all true cases, while its specificity stood at 0.867. In practical terms, a clinician using this model would miss very few cases of granulomatous lobular mastitis while correctly ruling out the condition in the vast majority of patients who do not have it.

When the radiomics score was added, the combined model pushed the AUC marginally higher to 0.985, but the improvement came with an unexpected trade-off. Sensitivity dropped to 0.907 while specificity rose to 0.967, a redistribution the authors describe as a sensitivity-specificity rebalancing rather than a clear gain. The difference in AUC between the two models, at minus 0.0070, was not statistically significant, with a p-value of 0.153. In other words, the sophisticated texture analysis extracted from ultrasound images added no measurable diagnostic value beyond what the clinical and conventional ultrasound features already provided.

This finding carries real weight for the field of radiomics, which has generated enormous enthusiasm in recent years. Radiomics promises to unlock subvisual information from medical images, and hundreds of studies have reported impressive classification performance when radiomic features are combined with clinical variables. Yet the Hangzhou results suggest that in this particular diagnostic problem, the conventional multimodal ultrasound assessment was already so informative that the machine-extracted texture features had little left to contribute. Decision curve analysis, which evaluates the net clinical benefit of a model across a range of probability thresholds, showed largely overlapping curves for the two models between 0.1 and 0.5, confirming that the radiomics signature offered no consistent incremental benefit in real-world decision-making terms.

Perhaps the most surprising result came from the age-only model. Using nothing but patient age, this minimalist model achieved a validation AUC of 0.884, a figure that would be considered respectable in many diagnostic contexts. The clinical multimodal ultrasound model significantly outperformed this baseline, with a difference in AUC of 0.094 and a p-value of 0.005, confirming that the imaging and clinical features genuinely add diagnostic information beyond demographics alone. Still, the fact that age alone carries such discriminative power underscores how important it is to include simple clinical variables in any predictive model before reaching for more complex tools.

Calibration analysis, which assesses how well predicted probabilities match observed outcomes, showed acceptable agreement in the internal-validation cohort for both the multimodal ultrasound and combined models. This matters because a model can discriminate well between groups yet still produce poorly calibrated probabilities, leading clinicians astray when they try to interpret the numerical output. The Brier score and calibration slope metrics used in the study suggested that the nomogram’s probability estimates were reasonably trustworthy within the cohort in which it was developed.

The authors are careful to emphasize the limitations of their work. The cohort was pathology-enriched and binary, meaning every patient had a definitive tissue diagnosis of either granulomatous lobular mastitis or invasive ductal carcinoma, a population that does not reflect the full spectrum of breast lesions encountered in clinical practice. The model cannot be generalized to unselected breast lesions, and its utility in cases with equivocal conventional ultrasound findings remains a hypothesis for future research. External validation in independent cohorts is required before any clinical adoption. Nevertheless, the study demonstrates that a carefully constructed multimodal ultrasound nomogram, built from features already available in a routine breast ultrasound examination, can achieve near-perfect discrimination between a benign inflammatory condition and one of the most common breast malignancies, potentially sparing patients with granulomatous lobular mastitis from unnecessary invasive procedures while maintaining a high level of diagnostic confidence.

Subject of Research: A multimodal ultrasound and radiomics nomogram for differentiating granulomatous lobular mastitis from invasive ductal carcinoma

Article Title: Development and internal validation of a multimodal ultrasound and Radiomics‑Based nomogram for differentiating granulomatous lobular mastitis from invasive ductal carcinoma

Article References: Yao, C., Yao, J., Shan, Y., Yao, Y., Zeng, F., Hu, Z., & Huang, P. (2026). Development and internal validation of a multimodal ultrasound and Radiomics‑Based nomogram for differentiating granulomatous lobular mastitis from invasive ductal carcinoma. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02913-z

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02913-z

Keywords: granulomatous lobular mastitis, invasive ductal carcinoma, breast ultrasound, radiomics, nomogram, elastography, LASSO regression, diagnostic accuracy, AUC, decision curve analysis, BMC Medical Imaging, machine learning

A team led by William J. Waldock, Ahmad Guni, Ara Darzi, and Hutan Ashrafian of Imperial College London, together with Cesar de la Fuente-Nunez of the University of Pennsylvania, has published a systematic review and meta-analysis in Frontiers in Drug Discovery that dissects the predictive performance of machine learning models for peptide–target interaction (PTI) prediction. The verdict is a study in contrasts: the best models achieve genuinely impressive scores, yet the field as a whole suffers from inconsistent metrics, opaque reporting, and a reproducibility deficit that makes honest comparison nearly impossible.

The researchers scoured PubMed, Cochrane, and arXiv for studies published through 1 August 2025, casting a deliberately wide net that included preprints, where cutting-edge transformer and graph neural network architectures often debut before formal peer review. From 4,175 unique records, the team winnowed the field down to 23 studies that met strict inclusion criteria: original machine learning models predicting peptide–target binding, bioactivity, or docking, each reporting at least one of three standardized performance metrics—area under the ROC curve (AUC), the concordance index (CI), or precision.

The pooled results tell a nuanced story. Across 14 studies reporting AUC for binary classification of binders versus non-binders, the meta-analytic estimate was 0.87, with a 95% confidence interval of 0.83 to 0.90. Ten of those models cleared the 0.90 threshold, signaling strong discriminatory power, though two studies fell below 0.75, a reminder that model simplicity, class imbalance, and small datasets can drag performance down sharply. For regression-style affinity ranking, five studies reported a concordance index, pooling at 0.90 (95% CI 0.89–0.91)—remarkably tight and consistently high, suggesting that modern deep learning models can rank peptide affinities with real reliability when fed structural descriptors or rich sequence embeddings. Precision, reported by six studies, pooled at a more modest 0.75 (95% CI 0.69–0.81), reflecting the difficulty of correctly identifying true binders among predicted positives.

Yet the headline numbers come with a heavy asterisk. The between-study heterogeneity was extreme—I-squared values hovered near 99% for all three metrics—meaning the pooled estimates describe a wildly diverse landscape rather than a single coherent signal. Reported AUC values ranged from 0.60 to 0.98, and the authors note that most studies provided no confidence intervals at all, making formal statistical synthesis of uncertainty impossible. The very fact that only 23 of thousands of screened studies could be pooled, the researchers argue, is itself a symptom of the reporting practices they set out to document.

Architecture mattered. The top-performing models predominantly employed transformer networks or graph neural networks (GNNs) fed with structural input features. Transformers, which use attention mechanisms to weigh relationships across sequence positions, and GNNs, which represent molecules as graphs of atoms and bonds, allow end-to-end learning from raw molecular inputs—reducing reliance on costly high-throughput screening. Studies that incorporated three-dimensional structural data, balanced or stratified datasets, and complex architectures tended to report higher scores than simpler models like logistic regression or random forests, which underperformed particularly in low-data or imbalanced settings. The authors also highlight the concordance index as an underused gem: unlike Pearson correlation, which can stay high even when a model misorders its top candidates, the CI directly estimates the probability that a randomly chosen pair of compounds is ranked correctly—precisely the quantity that matters when prioritizing candidates for experimental follow-up. Only one-fifth of the reviewed studies reported it.

The reproducibility picture is bleaker still. Using a modified version of the PROBAST risk-of-bias tool adapted for computational modeling, the reviewers found that the majority of studies were rated unclear or high risk in the domains of reproducibility and validation methodology. Very few performed external test-set validation; most relied on internal cross-validation without stratified splitting. Only about one-third described their hyperparameter tuning in detail, and several were vague about how they generated negative samples—the synthetic non-binding pairs that classification models need to learn what non-binding looks like. Because biological databases are extremely imbalanced, the choice of negative sampling strategy can artificially inflate metrics, a known pitfall in protein–protein interaction prediction that the authors say persists in the PTI literature.

To fix this, the team proposes a framework called STRIDE—Standardization, Transparency, Representativeness, Integration, Discovery, and Evidence—modeled loosely on the community benchmarks that propelled AlphaFold to fame. Standardization calls for unified evaluation protocols: temporal train/validation/test splits, defined clustering thresholds, biologically plausible negative sampling, and mandatory confidence intervals, following reporting standards like TRIPOD+AI and REFORMS. Transparency demands calibrated uncertainty estimates that separate aleatoric uncertainty (experimental noise) from epistemic uncertainty (incomplete training data), along with mechanistic interpretability tools such as Integrated Gradients and SHAP attribution. Representativeness confronts dataset bias head-on: current repositories overrepresent model organisms and crystallizable protein folds while underrepresenting modified peptides and intrinsically disordered regions, which comprise more than 30% of human proteomes. Integration urges multimodal fusion of protein language model embeddings, structural features, and physicochemical properties. Discovery pushes for symmetry-respecting architectures—SE(3)- and E(n)-equivariant networks that respect rotational and translational physics—while Evidence insists that high AUC alone is insufficient for clinical deployment, where calibration, not raw discrimination, is often the Achilles heel.

The authors are candid about their own limitations. The search was restricted to PubMed, Cochrane, and arXiv, potentially missing models published in engineering venues indexed by IEEE Xplore or Scopus, though a validation search against a broader database surfaced no additional eligible studies. And the retrospective, metric-based scope means the analysis cannot speak to prospective, hit-rate-benchmarked validation of the kind demonstrated by de novo protein design pipelines. Still, the review is positioned as a prerequisite for that next step: standardized retrospective reporting must come before any meaningful assessment of clinical applicability.

The stakes are considerable. Peptide drugs are notoriously fragile—plagued by poor oral bioavailability, rapid enzymatic degradation, and structural instability—so any computational method that can reliably triage candidates before synthesis promises to save enormous time and money. The meta-analysis confirms that machine learning can deliver strong classification and ranking performance when trained on well-curated datasets. But it also delivers a warning that the field cannot benchmark its way to clinical translation on inflated, incomparable numbers. Whether STRIDE becomes the CASP of peptide–target prediction will depend on whether the community embraces shared benchmarks, open code, and honest uncertainty reporting—the same ingredients that transformed protein structure prediction from an intractable grand challenge into a solved problem.

Subject of Research: Systematic review and meta-analysis of machine learning models for peptide–target interaction prediction in drug discovery

Article Title: Predictive accuracy of peptide–target interaction models in drug discovery: a systematic review and meta-analysis

Article References: Waldock, W. J., Guni, A., Darzi, A., de la Fuente-Nunez, C., & Ashrafian, H. (2026). Predictive accuracy of peptide–target interaction models in drug discovery: a systematic review and meta-analysis. Frontiers in Drug Discovery, 6, Article 1894745. https://doi.org/10.3389/fddsv.2026.1894745

Image Credits: AI Generated

DOI: 10.3389/fddsv.2026.1894745

Keywords: machine learning, peptide-target interaction, drug discovery, systematic review, meta-analysis, deep learning, transformers, graph neural networks, reproducibility, STRIDE framework, AUC, benchmarking

The study arose from a genuinely puzzling clinical setup. Indirect inguinal hernia is the most common condition treated in elective pediatric surgery, and children undergoing hernia repair routinely receive a preoperative chest X-ray even though the inguinal canal lies entirely outside the image’s field of view. That routine created an unusual natural experiment: a large set of chest radiographs from children with a confirmed diagnosis that the images themselves cannot show. If deep models trained on such images appeared to detect hernias, the finding would either suggest something remarkable about what these networks perceive or expose a subtle methodological trap. The researchers, led by Jin Luo and Chaosheng He of the First Affiliated Hospital of Xiamen University, together with Hao Liang of Rice University and Fang Lin of Xiamen, chose to interrogate the question with unusual rigor.

Their dataset comprised 1,960 preoperative chest radiographs from one center, one image per child, split almost evenly between 981 children with inguinal hernias and 979 surgical controls, plus 100 images from a second center to test whether any signal traveled beyond the hospital where it was learned. Rather than training models from scratch, the team used four frozen image encoders — pretrained networks whose internal representations of an image can be reused for new tasks — and probed those representations with simple logistic regression under repeated cross-validation. This approach, common in modern medical imaging research, asks whether the features a foundation model has already learned are sufficient to distinguish the groups without any task-specific retraining of the encoder itself.

On the surface, the results looked like a discovery. All four encoders separated the hernia group from the controls, with areas under the receiver operating characteristic curve, or AUCs, ranging from 0.645 to 0.709 — modest by clinical standards but clearly above chance, and two of the encoders held up at the second center with AUCs of 0.654 and 0.714. In many published studies, numbers like these would be reported as evidence that a deep model can detect a condition from an image that does not depict it, feeding a narrative about AI perceiving what humans cannot. The Xiamen team instead treated the result as a suspect to be interrogated, and the interrogation is where the study becomes genuinely important for anyone interpreting AI claims in medicine.

The first and largest confounder was age. Although age was absent from the structured records, it was printed on the images themselves, and the researchers used optical character recognition to read it for 1,178 children. The hernia group turned out to be younger than the control group by a median of three years — a property of which children get which surgeries, not of hernia biology. Strikingly, age alone, read off the image as text, produced an AUC of 0.754 with a 95 percent confidence interval of 0.725 to 0.782, higher than any of the deep encoders achieved from the actual radiographic content. In other words, the single most predictive feature in the entire pipeline was a number printed in the corner of the image, not anything in the lungs, heart, or bones.

To quantify how much of the encoders’ apparent skill was explained by demographics and acquisition, the team applied three complementary controls. First, within each cross-validation fold, they regressed the model outputs on age, sex, and 29 image-derived variables. Holding age fixed statistically cost the models between 0.093 and 0.137 AUC, leaving a residual of only 0.546 to 0.587 — barely above the 0.5 that denotes chance. Sex accounted for at most 0.012, and five variables describing how the images were acquired cost at most 0.044. That acquisition component mattered more than it might seem: an image-free model built from those five technical variables alone achieved an AUC of 0.673, statistically indistinguishable from the two best acquisition-adjusted encoders. The machines, it turned out, were partly reading the scanner, not the patient.

The acquisition signal itself tracked the child’s body. Image height correlated with age at a Spearman coefficient of 0.78, because older children present larger bodies that fill more of the detector. A chest radiograph, in this sense, encodes the size of the child being imaged, and body size encodes age, and age in this dataset encoded which operation the child was scheduled for. Each link in that chain is mundane; together they manufacture a spurious diagnostic signal of exactly the magnitude the deep models displayed. This is confounding in its most modern costume — not a variable in a spreadsheet but a physical and technical fingerprint baked into the pixels themselves, invisible to anyone who evaluates the model only on aggregate accuracy.

The decisive test came from matching. Instead of adjusting statistically after the fact, the researchers paired each hernia case with a control child matched on age, image acquisition characteristics, and sex, yielding 257 well-matched pairs. When the fitted model was evaluated on this balanced set, its performance collapsed to an AUC of 0.492 to 0.512 — indistinguishable from a coin flip. A probe retrained from scratch inside the matched set fared little better, reaching 0.502 to 0.580. A weak residual difference may persist, the authors acknowledge, but it is far too small to attribute to anything specific, and certainly too small to support any clinical use. The contrast between the regression adjustment, which left a residue that could masquerade as signal, and the matching, which erased it, is the methodological heart of the paper.

The team’s conclusion is blunt: nothing in this work supports using a chest radiograph to screen for inguinal hernia, and findings of this kind should be tested by matching rather than by regression alone, because regression can leave a residue that looks like discovery. The study was approved by the Clinical Research Ethics Committee of the First Affiliated Hospital of Xiamen University, conducted under the Declaration of Helsinki, and reported with the CLAIM and TRIPOD+AI checklists that promote transparency in artificial intelligence research. The authors also disclosed that they used a large language model assistant to help write analysis code and draft text, with all data processed on their own institutional server and no patient information sent to external services — a level of disclosure still rare in the field.

For a general audience, the lesson extends well beyond hernias. As foundation models are deployed across medicine, they will inevitably latch onto whatever variables co-vary with disease in the training data: which hospital a patient visited, which machine scanned them, how old they are, which conditions prompt which workups. A model that appears to diagnose one condition may simply be recognizing the demographic or institutional profile of the patients who receive that diagnosis. The Xiamen-Rice collaboration offers a template for catching such illusions — read the metadata, measure the acquisition variables, match the cohorts, and demand that the signal survive when the confounders are physically removed. The most impressive thing these models detected from chest X-rays was not a hernia. It was the shape of the study that produced them, and knowing that difference is what separates genuine medical AI from an expensive coincidence.

Subject of Research: Confounding in deep learning detection of pediatric inguinal hernia from chest radiographs

Article Title: Can chest radiographs detect inguinal hernia in children? A study of confounding

Article References: Luo, J., Liang, H., He, C., & Lin, F. (2026). Can chest radiographs detect inguinal hernia in children? A study of confounding. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02923-x

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02923-x

Keywords: chest radiography, deep learning, confounding, inguinal hernia, pediatric surgery, medical imaging, foundation models, model validation, case-control matching, AUC, optical character recognition, BMC Medical Imaging

The research team, led by co-first authors Fei Zheng and Wenjia Zhang under the senior authorship of Ping Yin and Nan Hong, set out to address a specific and practical problem. Pre-operative differentiation between benign and malignant sacral tumors is essential for guiding treatment strategy, yet the similarity of clinical symptoms and imaging characteristics reduces the specificity and accuracy of conventional assessments of bone malignancy. Their solution does not require new scanners, injected contrast agents, or additional radiation exposure. Instead, it extracts far more information than radiologists can perceive by eye from the non-contrast computed tomography images that these patients often already receive.

The study enrolled 112 patients whose sacral tumors were ultimately confirmed by pathology between July 2021 and September 2023, providing a ground-truth backbone against which every algorithmic prediction could be measured. The investigators drew regions of interest around each tumor on the CT scans and then applied a technique known as habitat radiomics. The underlying premise is that a tumor is not a uniform blob of tissue; it is a patchwork of biologically distinct sub-regions, each with its own density, texture, and microarchitecture. On a CT image, these sub-regions express themselves as subtle differences in Hounsfield units, the standard measure of radiodensity, that are invisible to casual inspection but statistically robust when quantified systematically.

To capture that internal heterogeneity, the team used K-means clustering, a classic unsupervised machine learning algorithm, to partition each tumor into three spatial habitats. K-means works by iteratively grouping voxels, the three-dimensional pixels of the CT volume, into clusters based on how similar their intensity patterns are, so that each habitat represents a compartment of relatively homogeneous imaging behavior within the tumor. The number of clusters was selected with statistical validation, referenced in the paper through the Calinski-Harabasz criterion, which rewards partitions that are tight within clusters and well separated from one another. The result is a digital map of the tumor’s internal architecture rather than a single averaged measurement of the whole lesion.

The innovation did not stop at the tumor’s edge. The researchers also deliberately extended their analysis outward into the tissue surrounding the lesion, dilating the tumor regions by thicknesses of 3 millimeters and 6 millimeters to create peri-tumoral shells. This choice reflects a growing recognition in imaging science that malignant processes remodel their surroundings, altering the density and texture of adjacent bone and soft tissue in ways that benign lesions typically do not. By comparing models built on the narrower 3-millimeter band with those built on the wider 6-millimeter band, the team could test empirically how much diagnostic signal resides just beyond the visible tumor margin. Their findings favored the tighter band: a Multi-Layer Perceptron, a type of neural network, based on the 3-millimeter peri-tumoral region exhibited enhanced performance relative to the one using the 6-millimeter region.

The feature extraction pipeline drew on the established vocabulary of radiomics, quantifying first-order statistics of voxel intensities, textural patterns that describe how densities are spatially arranged, and finer structural descriptors. To keep the models honest and generalizable, the researchers used Least Absolute Shrinkage and Selection Operator, or LASSO, regularization, a technique that penalizes model complexity and drives the coefficients of uninformative features toward zero, leaving a compact set of genuinely predictive measurements. Multiple machine learning architectures were compared, including logistic regression, random forests, support vector machines, and the Multi-Layer Perceptron, with reproducibility of feature measurements checked through correlation coefficients before modeling began.

The headline result concerned the habitats model. A Multi-Layer Perceptron trained on the three-habitat partition of the tumor achieved an area under the receiver operating characteristic curve, or AUC, of 0.877 in the test cohort, outperforming a model based on the conventional whole-tumor region of interest. An AUC of that magnitude indicates that the model ranks a randomly chosen malignant tumor as more suspicious than a randomly chosen benign one nearly nine times out of ten. That the habitat-based approach beat the standard whole-lesion analysis is the study’s central conceptual message: the spatial arrangement of a tumor’s internal sub-regions carries diagnostic information that averaging across the entire lesion destroys.

But the researchers went one step further, assembling what they call an ensemble nomogram. This fusion model integrated three streams of information: the patient’s age, a clinical indicator with recognized relevance to sacral tumor biology; the peri-tumoral radiomic features from the 3-millimeter shell; and the habitat-derived signatures from within the tumor. The ensemble achieved the study’s best performance, an AUC of 0.892 in the test sets, and it did so by a statistically meaningful margin, significantly outperforming a model built on clinical data alone, which managed only an AUC of 0.619 with a P value of 0.02. For context, a clinical-only model at 0.619 is only modestly better than a coin flip, while the fused model approaches the performance threshold at which a test becomes genuinely actionable in the clinic.

Accuracy alone, however, is not enough for a predictive model to earn clinical trust, and the authors addressed this with two complementary assessments. Calibration curves, evaluated with Hosmer-Lemeshow statistics and mean standard error, demonstrated good concordance between what the nomogram predicted and what pathologists actually observed, meaning that when the model says a tumor has a seventy percent chance of being malignant, roughly seventy percent of such tumors truly are malignant. Decision curve analysis, meanwhile, revealed a high overall net benefit across a clinically relevant range of threshold probabilities, indicating that patients and physicians acting on the nomogram’s predictions would, on balance, make better decisions than they would under default strategies of treating everyone or treating no one.

The practical implications reach well beyond the radiology reading room. A reliable pre-operative classification can influence whether a surgeon pursues wide en bloc resection, a morbid but potentially curative procedure for malignancies, or a more conservative intralesional approach appropriate for benign lesions, as well as informing the decision to obtain a biopsy beforehand. Because the method relies on non-contrast CT, it avoids the cost, contrast exposure, and contraindications associated with enhanced imaging, making it deployable in settings where advanced modalities are scarce. The authors conclude that NCCT-based habitat and peri-tumoral radiomics markedly improve preoperative prediction of benign and malignant sacral tumors, with habitat signatures offering precise predictions capable of guiding personalized treatment. As with any retrospective single-center study of 112 patients, external validation across larger and more diverse cohorts will be the necessary next step, but the work demonstrates that a routine CT scan, read through the eyes of a machine, can surface biological distinctions that clinicians have spent decades trying to see.

Subject of Research: Pre-operative classification of benign and malignant sacral tumors using habitat and peri-tumoral radiomics from non-contrast CT

Article Title: NCCT-based habitats and peri-tumoral radiomics enhance pre-operative prediction of benign and malignant sacral tumors

Article References: Zheng, F., Zhang, W., Liang, K., Liu, T., Yin, P., & Hong, N. (2026). NCCT-based habitats and peri-tumoral radiomics enhance pre-operative prediction of benign and malignant sacral tumors. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02806-1

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02806-1

Keywords: sacral tumor, radiomics, habitat imaging, non-contrast CT, machine learning, Multi-Layer Perceptron, K-means clustering, peri-tumoral micro-environment, nomogram, tumor heterogeneity, AUC, BMC Medical Imaging

The research, led by Brian D. Nicholson and colleagues at the University of Oxford, drew on the Clinical Practice Research Datalink, one of the world’s richest repositories of primary care records. The team identified 275,205 adults aged 18 and over who presented to English general practices with unexpected weight loss between January 2000 and December 2018. Within this cohort, 13,798 patients, or 5.0 percent, were subsequently diagnosed with cancer, with the diagnosis confirmed through linkage to the National Cancer Registrations Data. That relatively modest overall cancer rate is precisely the problem clinicians face: for every twenty patients with worrying weight loss, nineteen do not have cancer, yet each one currently triggers the same difficult judgment about whether to investigate further.

The methodological heart of the study lies in its comparison of two ways of reading the same laboratory data. The conventional approach treats the most recent result as a binary signal: a platelet count above the reference range, or an albumin level below it, counts as abnormal and raises suspicion. The alternative approach, tested here, examines the trajectory of each of 26 quantitative blood tests over the one, three, five, and ten years preceding the weight loss, asking not whether a value is abnormal but whether it is drifting in a concerning direction. To capture these trajectories, the researchers used Cox proportional hazards models for single abnormalities and joint models for trends, and they evaluated performance using the area under the receiver operating characteristic curve, or AUC, a standard measure of how well a test separates those with disease from those without.

A crucial refinement ran through the entire analysis: every model was estimated both unadjusted and adjusted for age and sex. This matters because many blood values shift systematically with age and differ between men and women, so a haemoglobin level that is perfectly normal for an 80-year-old man might be a red flag in a 40-year-old woman. The adjustment consistently improved discrimination for both abnormalities and trends. Indeed, adjusted blood test abnormalities and trends were more discriminative than their unadjusted equivalents across the board, a finding with immediate practical relevance for how reference ranges are applied at the point of care.

The headline result is that trends beat single thresholds most of the time once age and sex were taken into account. Adjustment resulted in higher AUCs for the trend compared with the equivalent blood test abnormality on 34 occasions, with the highest AUC reaching 0.82 (95 percent confidence interval 0.81 to 0.83). In clinical prediction terms, an AUC of 0.82 represents useful, though not definitive, discrimination, meaning the model ranked a randomly chosen cancer patient above a randomly chosen non-cancer patient roughly 82 percent of the time. For a cheap, routinely collected set of laboratory values, that level of performance is noteworthy, particularly in a symptom as non-specific as weight loss.

The site-specific findings add texture to the overall picture. Trends in mean cell volume, a measure of the average size of red blood cells, improved discrimination for bowel cancer and lymphoma. White blood cell count and neutrophil trends carried signal for lung cancer, consistent with the systemic inflammation that often accompanies solid tumours. For prostate cancer, trends in aspartate aminotransferase, red blood cell count, haematocrit, and the platelet-to-lymphocyte ratio all outperformed the corresponding single abnormality. These patterns make biological sense: cancers can remodel blood chemistry slowly over months or years, and a value that has slid steadily toward the edge of the normal range may be more telling than one that sits just inside it on a single occasion.

The sheer scale of routine testing in primary care made the analysis feasible. Among the cohort, the median number of blood tests per person over the ten-year window ranged from two for some test types to four for others, with interquartile ranges stretching from two to twelve tests depending on the analyte. The median interval between a patient’s first and last blood test was 5.2 years for those later diagnosed with cancer and 4.3 years for those who remained cancer-free. Critically, the last test before weight loss typically occurred close to the presentation itself, a median of 0.1 years beforehand in cases and 0.3 years in non-cases, meaning the trend information was genuinely available to the GP at the moment the decision to refer had to be made.

The authors are candid about the limitations of their modelling strategy. Because the study was retrospective and observational, it could not fully account for bias in who was tested; patients who receive frequent blood tests differ systematically from those who do not, often because they already have chronic conditions or concerning symptoms, and this selection can distort apparent associations. The trend analysis was also restricted to patients with at least two results within each trend window, which excludes people with sparse testing histories and may limit how the findings generalise to the least well-monitored patients. Discrimination, moreover, is only one dimension of a diagnostic tool; calibration, clinical utility, and the downstream consequences of more referrals all require prospective evaluation before any change to guidelines.

Even with those caveats, the implications for primary care are concrete. The first and most actionable message is that blood test abnormalities accompanying unexpected weight loss should be interpreted after adjustment for age and sex, since this alone improved discrimination without requiring any new data collection. The second message is that electronic health records already contain the longitudinal data needed to compute trends, and for a meaningful set of test-cancer combinations those trends add discrimination beyond a single threshold. In health systems where decades of laboratory results sit in searchable databases, algorithms that read the trajectory rather than the snapshot could be built into existing clinical software.

For patients, the study offers a reason to feel that the humble full blood count and its chemical cousins hold more diagnostic power than traditionally assumed. Unexpected weight loss will always demand careful clinical judgment, and no blood test pattern can replace that. But as machine-readable histories grow longer and statistical models grow more refined, the difference between a single abnormal value and a decade of drifting values may become the difference between catching a cancer early and catching it late. This study, grounded in nearly three decades of English primary care data, provides the strongest evidence yet that when it comes to blood tests and cancer, the trend is the thing.

Subject of Research: Blood test trends versus single threshold abnormalities for detecting undiagnosed cancer in primary care patients with unexpected weight loss

Article Title: Blood test trend versus single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss: A retrospective cohort study

Article References: Nicholson, B. D., Bankhead, C. R., Zhu, S., Oke, J. L., Wright Drakesmith, C., Perera, R., Hobbs, F. D. R., & Virdee, P. S. (2026). Blood test trend versus single threshold abnormality to discriminate cancer from non-cancer in patients with unexpected weight loss: A retrospective cohort study. PLOS Medicine, 23(9), e1004956. https://doi.org/10.1371/journal.pmed.1004956

Image Credits: AI Generated

DOI: 10.1371/journal.pmed.1004956

Keywords: unexpected weight loss, cancer diagnosis, primary care, blood tests, retrospective cohort study, Clinical Practice Research Datalink, diagnostic discrimination, AUC, PLOS Medicine, early detection, laboratory medicine, joint models

The research team, led by Tong Xiao and Yiru Wang, who contributed equally to the work, with Yanhong Zhang as corresponding author, searched ten databases spanning the English-language and Chinese biomedical literature, including PubMed, Embase, Web of Science, CINAHL, the Cochrane Library, PsycINFO, and four major Chinese databases. Their search covered everything published from database inception through August 15, 2025. From that sweep they identified 27 studies containing 58 distinct prediction models for adolescent non-suicidal self-injury. Seventeen of the source studies were conducted in hospital settings and ten in community settings, a split that matters because the predictors available and the base rates of self-injury can differ substantially between a psychiatric clinic and a school population.

The centerpiece of the analysis is the pooled area under the receiver operating characteristic curve, or AUC, the most widely used summary of how well a model separates those who will experience an outcome from those who will not. An AUC of 0.5 reflects performance no better than a coin flip, while 1.0 represents perfect discrimination. Across the 17 models for which training-set data could be pooled, the meta-analysis found a combined AUC of 0.850, with a 95 percent confidence interval of 0.810 to 0.890. For the far smaller set of five models evaluated on independent validation data, the pooled AUC was 0.840, with a wider interval of 0.760 to 0.930. In conventional terms, that places these tools in the moderate to good range of predictive performance, comparable to risk models used in other areas of medicine.

But the authors are quick to point out that a strong AUC on a training set is not the same as a trustworthy clinical instrument. Only a handful of the 58 models had ever been externally validated, meaning tested on data not used to build them, and the confidence interval for the validation estimate is wide enough to encompass everything from mediocre to excellent performance. Discrimination is only half the story in prediction research. A model must also be well calibrated, meaning that when it says a patient has a 30 percent risk, roughly 30 percent of similar patients should actually experience the outcome. Calibration was rarely assessed or reported in the studies the team reviewed, a gap that has plagued prediction modeling across medicine and that the authors single out as a priority for future work.

Methodological quality emerged as the study’s most sobering finding. Using the PROBAST + AI tool, a risk-of-bias instrument designed specifically for prediction models, including those built with artificial intelligence, the reviewers judged 24 of the 27 studies to be of low overall quality and 23 to be at high risk of bias. Two studies were rated at high risk of concerns regarding applicability, meaning their data or predictors might not transfer well to the populations clinicians actually want to serve. Common sources of bias in prediction research include small sample sizes relative to the number of predictors, inappropriate handling of missing data, optimistic performance estimates from evaluating models on the same data used to develop them, and selective reporting of the best-performing model among many that were tried.

What drives the risk, according to the models themselves? Across the included studies, a consistent cluster of predictors surfaced again and again. Childhood trauma was among the most powerful signals, echoing decades of research linking early adversity to self-directed harm. Depressive mood and anxiety, the emotional workhorses of adolescent risk screening, appeared frequently, as did a history of prior non-suicidal self-injury, which is unsurprising given that past behavior is one of the strongest predictors of future behavior in nearly every domain of mental health. Sleep disorders, stressful life events, and female gender rounded out the list of common predictors. The recurrence of these variables across diverse settings suggests a reasonably stable risk architecture, even if the models that combine them vary widely in construction.

The implications for clinical practice are double-edged. On one hand, the finding that pooled AUCs hover around 0.85 suggests that a data-informed approach to identifying at-risk adolescents is feasible. Schools, pediatric clinics, and emergency departments could, in principle, deploy brief screening tools that combine a small number of well-validated predictors to flag young people who warrant deeper assessment. On the other hand, the authors conclude that the existing models perform poorly in methodological quality and clinical applicability, and they explicitly call for future research to strengthen validation and calibration of existing models and to improve methodological rigor in line with PROBAST criteria. Until that happens, they caution, these tools should not be treated as ready-made clinical decision instruments.

The review also highlights a structural weakness of the field: fragmentation. With 58 models across 27 studies, many built on small, single-center samples and using different predictor sets, statistical techniques, and outcome definitions, there is little opportunity to compare approaches head to head or to pool their strengths. The authors suggest that the path forward lies less in building yet more novel models and more in rigorously validating, calibrating, and refining the ones that already exist, ideally in large, diverse, prospectively collected cohorts. Reporting standards such as TRIPOD, the transparency checklist for prediction studies, and the PROBAST framework for assessing bias offer a shared language for judging whether a model deserves clinical trust.

For a behavior as hidden and as consequential as adolescent self-injury, the promise of early, data-driven identification is genuinely compelling. Non-suicidal self-injury is associated with serious adverse effects on individuals, families, and society, and it frequently goes undetected until it has become entrenched. A validated screening model could shift the clinical posture from reaction to prevention, directing scarce mental health resources toward the adolescents who need them most. This meta-analysis provides the clearest picture yet of what that future might look like: a field with real predictive signal, a recognizable set of risk factors, and performance statistics that look respectable on paper, but one that still must clear the harder hurdles of external validation, honest calibration, and methodological discipline before its algorithms earn a place in the clinic.

The study, funded by the Nanjing Health Science and Technology Development Special Fund Project and published open access, arrives at a moment when machine-learning tools are being proposed for nearly every corner of mental health care. Its message generalizes well beyond self-injury. Impressive AUCs are easy to produce and hard to trust; the difference lies in validation, calibration, and transparent reporting. For the researchers, clinicians, and families hoping that prediction models can help protect vulnerable teenagers, the new review offers both a map of what has been achieved and a candid account of how much work remains before those achievements can be safely put to use.

Subject of Research: Risk prediction models for non-suicidal self-injury among adolescents

Article Title: Risk prediction models of non-suicidal self-injury among adolescents: a systematic review and meta-analysis

Article References: Xiao, T., Wang, Y., Xiong, X., Zou, Y., Jia, Y., Xu, B., Wang, R., Bian, C., & Zhang, Y. (2026). Risk prediction models of non-suicidal self-injury among adolescents: a systematic review and meta-analysis. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08719-1

Image Credits: AI Generated

DOI: 10.1186/s12888-026-08719-1

Keywords: non-suicidal self-injury, adolescents, risk prediction models, systematic review, meta-analysis, machine learning, PROBAST, AUC, childhood trauma, depression, anxiety, mental health screening

Doxorubicin is an anthracycline antitumor antibiotic and one of the most widely used injectable chemotherapeutics in veterinary medicine. It acts on cancer cells through a battery of mechanisms, including DNA intercalation, inhibition of DNA and RNA polymerases, inhibition of topoisomerase II, DNA alkylation, and the generation of reactive oxygen species. That versatility makes it a staple against many canine malignancies, both as a single agent and within multi-drug protocols. But its toxicity profile is equally well known. Gastrointestinal adverse events and myelosuppression, the suppression of bone marrow activity, are the dose-limiting toxicities, and dogs receiving doxorubicin as a single agent experience gastrointestinal side effects more frequently than with most other intravenous chemotherapeutics.

The problem with the current standard lies in its assumptions. Doses of doxorubicin, like many chemotherapies, are calculated using body surface area, a practice originally adopted to normalize doses across species on the grounds that surface area better reflects basal metabolic rate than body weight alone. Yet body surface area cannot account for the factors that actually determine how much drug circulates in a patient: distribution, metabolism, and excretion. The evidence that something is amiss has been accumulating for years. Smaller dogs suffer more severe adverse events when dosed by body surface area, which is why dogs under 15 kilograms are typically treated with a weight-based dose of 1 milligram per kilogram instead. The formula, in short, is imperfect.

Pharmacokinetics offers a way out. Rather than asking how much drug was administered, pharmacokinetic-guided dosing asks how much drug the patient actually absorbed and retained over time, quantified as the area under the drug concentration-time curve, or AUC. Previous work in canine patients demonstrated that predicted doxorubicin AUC correlates with neutropenia, the dangerous drop in neutrophils that marks bone marrow suppression, as well as with total white blood cell counts and the surviving fraction of neutrophils at the nadir, the lowest point of the count. Human medicine has already capitalized on this logic. Pharmacokinetic-guided dosing of docetaxel reduced variability in bone marrow suppression by up to 50 percent, and individualized fluorouracil dosing in metastatic colorectal cancer significantly improved objective response rates compared with conventional dosing.

The Davis team, led by Sridhar Madan Veluvolu together with Robert B. Rebhun, Jaeyoung Kim, and Luke Anthony Wittenburg, built on a tool developed earlier in their laboratory: a limited sampling model that predicts total doxorubicin exposure from just three blood samples. Under the protocol, dogs received a standard 30 milligrams per square meter dose by 20-minute intravenous infusion, and blood was drawn at five, 45, and 60 minutes after the infusion ended. Serum doxorubicin concentrations were measured with a validated liquid chromatography tandem-mass spectrometry assay, linear from 10 to 750 nanograms per milliliter, and plugged into a weighted equation that estimates the AUC over the first six hours. The elegance of the approach is its practicality: three time points instead of an exhaustive sampling series makes prospective clinical use feasible.

The prospective study ran from June 2021 to August 2022 at the UC Davis Veterinary Medical Teaching Hospital, enrolling tumor-bearing dogs at least 15 kilograms in weight with favorable performance scores and no cardiac dysfunction or multidrug-resistance gene mutations. Ten dogs met the criteria, but three were withdrawn after a single dose because of owner-perceived adverse events, leaving seven patients who completed three full cycles of doxorubicin with complete blood counts at baseline and at the seven-day nadir after each dose. Owners filled out standardized questionnaires about gastrointestinal signs after every treatment and were asked to keep their pets on a consistent diet to make those observations as comparable as possible.

The headline result is stark. Measured and dose-normalized doxorubicin concentrations varied nearly tenfold between patients at the sampled time points, and there was no significant correlation between the total milligram dose a dog received and its predicted AUC. When the coefficient of variation was calculated after normalizing for dose, the average within-patient variability across three cycles was just 4.7 percent, while the average between-patient variability was 25.4 percent, a difference that was highly statistically significant. Before dose normalization the within-patient figure was 10.7 percent against 22.5 percent between patients. Notably, the within-patient variability observed in these dogs, ranging from 2.8 to 9 percent, is lower than the 6 to 59 percent reported in humans, while the between-patient variability in dogs was tighter than the 37 to 93 percent seen in human studies.

The model also proved its worth as a predictive instrument. Although no significant relationship emerged between predicted AUC and absolute neutrophil counts across all data points, the picture sharpened when the analysis focused on dogs whose seven-day neutrophil counts were genuinely lower than baseline, suggesting the true nadir had been captured. For those patients, higher exposure correlated significantly with a lower surviving fraction of neutrophils. The limited sampling model, using baseline neutrophil count and predicted AUC together, successfully predicted the absolute nadir neutrophil count with a mean prediction error of just 3.3 percent. Three neutropenic events were captured during the study, two of which occurred at the highest predicted exposures, and one patient who received a 25 percent dose reduction after severe neutropenia subsequently showed a 25 to 32 percent drop in exposure, an encouraging hint that the model tracks dose changes faithfully.

The gastrointestinal findings add a patient-centered dimension. Using owner-reported scores, the researchers found that higher predicted AUC correlated significantly with decreased appetite and increased nausea, though not with vomiting or diarrhea, and no significant relationship emerged for composite gastrointestinal scores. The authors caution that owner scoring of subjective signs carries inherent bias, but the correlations suggest that the patients with the highest drug exposure are also the ones feeling the worst, a pattern that personalized dosing could potentially flatten. The study had clear limitations, including its small size of seven completers, strict eligibility criteria that may not represent the broader population of treated dogs, and unexplored genetic factors, since doxorubicin disposition involves influx and efflux transporters and metabolizing enzymes whose variants in dogs remain poorly characterized.

What comes next is the crucial question of targets. The researchers emphasize that no one yet knows what AUC a dog should hit to maximize tumor response while minimizing toxicity, and that target may differ from patient to patient. In human docetaxel therapy, AUC-guided dose adjustment dramatically increased the proportion of patients within the therapeutic window and improved disease control rates. Neutropenia itself may serve as a double-edged biomarker, since in canine lymphoma it has been associated with longer remission and survival times, hinting that the right amount of bone marrow suppression may signal effective dosing. For now, the Davis team’s message is simple and provocative: the same dose produces wildly different exposures in different dogs, but each dog is remarkably consistent with itself. That consistency is precisely what makes individualized dosing achievable, and the first treatment cycle may be all the information a clinician needs to get there.

Subject of Research: Pharmacokinetic variability of doxorubicin exposure in dogs with cancer using a limited sampling model

Article Title: Comparison of interpatient and intrapatient variability in doxorubicin exposure using a validated limited sampling model in dogs with cancer

Article References: Veluvolu, S. M., Rebhun, R. B., Kim, J., & Wittenburg, L. A. (2025). Comparison of interpatient and intrapatient variability in doxorubicin exposure using a validated limited sampling model in dogs with cancer. Veterinary Oncology, 2(1), Article 24. https://doi.org/10.1186/s44356-025-00038-z

Image Credits: AI Generated

DOI: 10.1186/s44356-025-00038-z

Keywords: doxorubicin, veterinary oncology, pharmacokinetics, limited sampling model, chemotherapy dosing, body surface area, neutropenia, AUC, dogs, adverse events, personalized medicine, canine cancer