SEARCH · PubMed Health

Results for “Model performance”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 613 records · Page 34Linked to original sources

PubMed2004

Population pharmacokinetics of clofarabine, a second-generation nucleoside analog, in pediatric patients with acute leukemia.

The population pharmacokinetics of plasma clofarabine and intracellular clofarabine triphosphate were characterized in pediatric patients with acute leukemias. Traditional model-building techniques with NONMEM were used. Covariates were entered into the base model using a forward selection significance level of .05 and a backwards deletion criterion of .005. Model performance, stability, and influence analysis were assessed using the nonparametric bootstrap and n-1 jackknife. Simulations were used to understand the relationship between important covariates and exposure. A 2-compartment model with weight (scaled to a 40-kg reference patient) modeled as a power function on all pharmacokinetic parameters (0.75 on clearance-related terms and 1.0 on volume-related terms) was fit to plasma clofarabine concentrations (n = 32). White blood cell (WBC) count, modeled as a power function (scaled to a WBC count of 10 x 10(3)/microL), was a significant predictor of central volume with power term 0.128 +/- 0.0314. A reference patient had a systemic clearance of 32.8 L/h (27% between-subject variability [BSV]), a central volume of 115 L (56% BSV), an intercompartmental clearance of 20.5 L/h (27% BSV), and a peripheral volume of 94.5 L (39% BSV). Intracellular clofarabine triphosphate concentrations were modeled using a random intercept model without any covariates. The average predicted concentration was 11.6 +/- 2.62 microM (80% BSV), and although clofarabine triphosphate half-life could not be definitively estimated, its value was taken to be longer than 24 hours. The results confirm that clofarabine should continue being dosed on a per-squaremeter or per-body-weight basis.

Adenine Nucleotides↗

PubMed2004

Modeling patients' acceptance of provider-delivered e-health.

OBJECTIVE: Health care providers are beginning to deliver a range of Internet-based services to patients; however, it is not clear which of these e-health services patients need or desire. The authors propose that patients' acceptance of provider-delivered e-health can be modeled in advance of application development by measuring the effects of several key antecedents to e-health use and applying models of acceptance developed in the information technology (IT) field. DESIGN: This study tested three theoretical models of IT acceptance among patients who had recently registered for access to provider-delivered e-health. MEASUREMENTS: An online questionnaire administered items measuring perceptual constructs from the IT acceptance models (intrinsic motivation, perceived ease of use, perceived usefulness/extrinsic motivation, and behavioral intention to use e-health) and five hypothesized antecedents (satisfaction with medical care, health care knowledge, Internet dependence, information-seeking preference, and health care need). Responses were collected and stored in a central database. RESULTS: All tested IT acceptance models performed well in predicting patients' behavioral intention to use e-health. Antecedent factors of satisfaction with provider, information-seeking preference, and Internet dependence uniquely predicted constructs in the models. CONCLUSION: Information technology acceptance models provide a means to understand which aspects of e-health are valued by patients and how this may affect future use. In addition, antecedents to the models can be used to predict e-health acceptance in advance of system development.

Attitude to Computers↗

PubMed2003

Evaluation of polychlorinated biphenyl bioaccumulation patterns in white sea urchins (Lytechinus pictus) using multiple approaches.

The bioaccumulation of polychlorinated biphenyls (PCBs) from three amended field-contaminated sediments (with total PCB concentrations of approximately 4, 10, and 100 microg/g dry wt) by white sea urchins (Lytechinus pictus) was evaluated using multiple statistical and theoretical approaches. Similarity analysis of the PCB bioaccumulation patterns, based on the concept of ecological communities, showed that the PCB patterns in the sea urchins and source sediments were essentially identical for all three sediment concentrations. However, affinity analysis did show some preference for bioaccumulation of higher-molecular-weight and more hydrophobic congeners by the urchins. The affinity analysis also showed that within a homologous series, bioaccumulation increased with increasing hydrophobicity. The biota-sediment accumulation factor (BSAF) profiles for the two lower concentration sediments (A and B) were found to be statistically different from the high concentration sample (sediment C) by a multivariate analysis of variance (MANOVA). The relationship between the measured apparent organic carbon-normalized partition coefficients (K(OC)) and octanol-water partition coefficient (K(OW)) (log based) suggested a significant departure from thermodynamic equilibrium. A nonequilibrium, steady-state bioaccumulation model was found to correctly predict the observed experimental bioaccumulation patterns. To improve the model performance, a hydrophobic term was introduced to account for the drop-off in BSAF profiles with log K(OW) > or = 6.5. This study showed that nonequilibrium, steady-state models are far superior to equilibrium partitioning-based models for understanding the bioaccumulation of organic chemicals by sea urchins.

Animals↗

PubMed2001

Traditional versus hazard analysis and critical control point-based inspection: results from a poultry slaughter project.

Federal meat and poultry inspection has changed little since the Federal Meat Inspection Act was passed in 1906, followed by the Poultry Products Inspection Act of 1957 and related amendments. These acts mandate sensory or organoleptic (sight, smell, and touch) inspection of all carcasses. For several decades, the U.S. Department of Agriculture's Food Safety and Inspection Service (FSIS) has been urged by various organizations to move to a scientific, risk-based inspection system. In partial response to these calls, the FSIS has developed new slaughter inspection models that are currently being tested with volunteer plants in the hazard analysis and critical control point (HACCP)-based inspection models project. To evaluate whether plants operating under the new inspection models perform at least as well as they did under the current or traditional system, microbial and organoleptic data are being collected before and after the implementation of the new inspection models. In this article, we describe the baseline and models data collection procedures and present the results of the baseline and models data collection for eight plants that slaughter young chickens. The results from the first eight volunteer plants suggest that inspection under the new models is equivalent and in some ways superior to that of traditional inspection. This pilot project suggests that new slaughter inspection systems, which rely on HACCP principles with FSIS oversight and verification services, can maintain or even improve food safety and other consumer protection conditions relative to traditional hands-on inspection methods.

Animals↗

PubMed1994

Exponential smoothing method for forecasting drug expenditures.

A model for forecasting a hospital pharmacy drug budget is described, and its results in 10 hospital pharmacy departments are evaluated. A model for forecasting inpatient drug expenditures was developed based on the method of exponential smoothing. Exponential smoothing predicts a value based on the forecast for the prior period, with adjustment for the error of that forecast. Recent data are weighted more heavily than older data; as data become older, weights decline exponentially. The model incorporates changes in workload in addition to drug expenditure data. The variable used for workload can vary from one hospital to another, depending on the statistics that are available. The model was designed to be more accurate than current methods, easy and quick to use, and usable with a minimal knowledge of statistics and forecasting theory. The model was tested on fiscal 1988-1992 drug budget data from 10 British Columbia hospital pharmacy departments. Four departments had insufficient data; of the remaining six, the forecasting model performed better than the hospitals' current methods in four departments. The mean absolute deviation between budgeted (by current methods) and actual drug expenditures was 8.70% (range, 6.19-15.16%). The forecasting model yielded a mean absolute deviation of 5.93% (range, 3.13-7.66%). Better forecasts resulted when pharmacy medication-order volume was used as a workload variable, as compared with hospital inpatient days. An exponential smoothing model improved the accuracy of drug-budget forecasts in four of six pharmacy departments.

British Columbia↗

PubMed1984

A consciousness-sampling analysis of test anxiety and performance.

In order to evaluate cognitive-interference, reassertion, and reaction-to-performance models of test anxiety, 82 students completed the Test Anxiety Scale, provided state measures of anxiety just before and after a course examination, described their preparation for the test, and reported thought content and state anxiety up to six times during the test. Test Anxiety Scale scores were predictive of pre- and posttest state anxiety but not performance or problem-solving thought frequency during the test. Thought content was significantly but weakly correlated with performance, which was well correlated with posttest state anxiety but not with pretest anxiety. Pretest state anxiety was virtually uncorrelated with posttest state anxiety, with the correlations gradually declining during the test. Question-answering thought content correlated inversely with anxiety during the test. There was no group for whom anxiety appeared to facilitate performance. Preparation correlated only with performance. The pattern of results appears inconsistent with a cognitive-interference interpretation of test anxiety and suggests that in the naturalistic setting used, anxiety is more clearly an effect than a cause of poor performance.

Achievement↗

PubMed2000

A pharmacokinetic model for predicting absorption, elimination, and tissue burden of toxaphene in rats.

A two-compartment pharmacokinetic model was formulated to predict absorption, elimination, and tissue burden of toxaphene in rats. The model was constructed based on the database of Crowder and Dindal (Bull. Environ. Contam. Toxicol. 12, 320-327, 1974) and included six tissue compartments: blood, brain, liver, muscle, fat, and carcass. The pharmacokinetically based dosimetry indicated that absorption of toxaphene was fast in fat, whole body, carcass, and blood, relatively slow in liver and muscle, and slow in brain. In contrast, the elimination rate was rapid in whole body, muscle, and blood, moderate in carcass and brain, and slow in liver and fat. Tissue burden was highest in fat, whole body, and blood, intermediate in liver, and lowest in brain. The model performance was evaluated by the data set of Pollock and Hillstrand (J. Environ. Sci. Health B 17, 635-648, 1982) on toxaphene absorption and elimination in pregnant rats. Validity of the model was confirmed by the close agreement between the predicted and observed tissue burdens of toxaphene in target tissues. Disposition of toxaphene via feces was a dominant excretory pathway while urinary excretion was a minor elimination route in male rats. However, for pregnant rats, excretion of toxaphene both in urine and feces were of similar magnitude. These characteristics of elimination are valuable for understanding the metabolism of toxaphene in pregnant rats. The model serves as a starting point for a quantitative, mechanism-based understanding of the processes that influence the pharmacokinetics of toxaphene in mammalian systems.

Algorithms↗

PubMed1998

Field evaluation of a mathematical model of PCB transfer through the freshwater aquatic food chain.

A mathematical model of the transfer of PCBs through the freshwater aquatic food chain is described. The model predicts concentrations of 11 selected individual PCB congeners in forage fish and pike, from source terms of atmospheric deposition and watershed soil concentrations. Model performance has been evaluated using data from a field study conducted in a section of the River Severn near Birmingham, UK. Results demonstrate that with the exception of congener 52, overall model predictions of individual PCB concentrations in both forage fish and pike underestimate measured concentrations by factors of between approximately 3 and 25 for individual congeners. Closer examination suggests that whilst model equations contribute to these underestimations, a significant factor is the lack of knowledge of additional PCB inputs to the waterbody.

Animals↗

PubMed2003

Numerical modeling of diazinon transport through inter-row vegetative filter strips.

A numerical simulation model of pesticide runoff through vegetative filer strips (PRVFS) was developed as a tool for investigating the effects of pesticide transport mechanisms on VFS design in dormant-sprayed orchard. The PRVFS model was developed applying existing theories such as kinematic wave theory and mixing zone theory for pesticide transport in the bare soil area. For VFS area, the model performs flow routing by simple mass accounting in sequential segments and the pesticide mass balance by considering pesticide washoff and adsorption processes on the leaf, vegetative litter, root zone and soil. Model sensitivity analysis indicated that pesticide transfer from surface soil to overland flow and pesticide washoff from the VFS were important mechanisms affecting diazinon transport. The VFS cover ratio and rainfall intensity can be important design parameters for controlling diazinon runoff using inter-row VFS in orchard. The PRVFS model was validated using micro-ecosystem simulation of diazinon transport for 0, 50 and 100% VFS cover conditions. The PRVFS model is shown to be a beneficial tool for evaluating and analyzing possible best management practices for controlling offsite runoff of dormant-sprayed diazinon in orchards during the rainy season.

Agriculture↗

PubMed2026

The Use of Deep Learning in RNA Therapeutic Development.

Ribonucleic acid (RNA)-based therapeutics have emerged as promising methods of disease treatment due to their ability to target the human genome and influence protein production, their versatility, and their relative lack of toxicity compared to other gene therapies. However, the RNA therapeutic design space is extremely large, encompassing multiple variables, including codon identities, secondary structure, and design of specific regions. RNA therapeutic optimization is difficult due to the impracticality of exploring such a vast design space experimentally. To address this limitation, deep learning methods have been employed to optimize RNA therapeutic development. In this review, we examine the application of deep learning models across three key aspects of RNA therapeutic development (RNA structure prediction, CRISPR activity, and RNA delivery), highlighting major contributions in these fields and analyzing how deep learning model architectures could affect model performance. We then discuss challenges associated with using deep learning for RNA therapeutics, such as computational and data limitations. Finally, we offer perspectives on areas for future exploration, such as emerging model architectures and methods of integration with more advanced high-throughput screening techniques. Ultimately, this review provides an overview of how deep learning is used in RNA therapeutic development and how it can evolve in the future.

Deep Learning↗

PubMed1985

Intellectual development within transracial adoptive families: retesting the confluence model.

The confluence model of intellectual development was estimated for a within-family sample of 321 children from 101 transracial adoptive families. Mental ages of the children and their parents, as well as birth or adoption intervals, were used in a nonlinear least-squares estimation procedure to obtain children's predicted mental ages. Contrary to an earlier report using these data, the confluence model performed quite well, accounting for up to 50% of the variance in mental age. When the relationship between chronological and mental age was taken into account, the predictive power of the model was reduced but not eliminated. The confluence model was also fitted separately to various subsamples. The model generated a good fit to the data from both biological and adopted children and fit the data from early-adopted children much better than the data from later-adopted children. Both findings were taken as evidence that the confluence model provides an environmental account of intellectual development within the family.

Adolescent↗

PubMed2026

Development and Validation of Machine Learning Models for Predicting Early Cognitive Decline Using Home Sensor-Derived Behavioral Data: Sensors in-Home for Elder Wellbeing (SINEW) Cohort Study.

BACKGROUND: As the global population continues to age, the prevalence of geriatric conditions, including dementia and frailty, is also increasing. Early identification of individuals at an elevated risk of these conditions, such as those presenting with mild cognitive impairment (MCI) or prefrailty, can provide a critical window for prompt intervention aimed at preventing or reversing disease progression. To promote such early identification, there is a burgeoning interest in the use of digital sensor technology and predictive modeling. OBJECTIVE: This study aimed to use a continuous, home-based monitoring sensor system for older adults to distinguish those exhibiting normal aging from those with MCI, early dementia, prefrailty, or frailty, and to predict their transition from normal aging to one of these conditions. METHODS: This longitudinal cohort study will recruit 200 community-dwelling adults aged ≥65 years with normal cognition or MCI at baseline. A multi-sensor system will be installed in participants' homes, including passive infrared motion sensors, door contact sensors, bed sensors, medication box sensors, wearable activity bands, and Bluetooth proximity beacons. These devices will continuously capture spatiotemporal activity patterns, mobility indicators, sleep behaviors, and medication-taking routines. Annual assessments will include standardized cognitive tests (eg, Montreal Cognitive Assessment, Mini-Mental State Examination, Rey Auditory-Verbal Learning Test, digit span, Color Trails Test, semantic fluency, Stroop), frailty measures (modified Fried phenotype, gait speed, grip strength), mental health scales, sleep quality, and psychosocial indicators. Sensor-derived features-such as gait variability, activity regularity, sleep fragmentation, and medication adherence patterns-will be integrated with clinical data to develop supervised machine learning models. Planned approaches include logistic regression, random forests, gradient boosting, and deep learning. Model performance will be evaluated using cross-validation and independent test sets. Primary metrics will include area under the receiver operating characteristic curve, sensitivity, specificity, precision, recall, and F1-score. Models will be benchmarked against gold-standard clinical diagnoses and validated using temporal subsets of the dataset. RESULTS: Enrollment for this study started in November 2019 and will continue until March 2030. As of June 2025, we have enrolled 138 participants. Full data analysis has yet to begin. CONCLUSIONS: We aim to develop a reliable and effective sensor system for in-home use that will facilitate the early detection of cognitive and physical decline. In so doing, it will add to our current understanding of digital biomarkers. It is common for older adults to seek clinical intervention only when their cognitive impairment has already reached an advanced stage. The implementation of readily deployable sensor systems within community settings presents us with opportunities for prompt intervention, which holds the potential for delaying or reversing disease progression and allowing for a greater number of functional and meaningful years.

Humans↗

PubMed1988

[Heterotransplantation of a human glioma and brain metastases in the athymic nude mouse--a preclinical model for radiation oncology. 1. Basic principles and methodology].

In spite of the great efforts undertaken in the different disciplines, the prognosis of patients with highly malignant gliomas, i.e. astrocytomas of degrees III and IV remains unfavorable. Up to now, new findings about an improvement of radiooncologic therapy methods are obtained retrospectively from the results of complex and time-consuming clinical studies. The heterotransplantation of human tumor tissue in immune-deficient nu/nu mice gives the opportunity to check preclinically the efficiency of different fractionation schemes and cytostatic drugs already. The author's experiences and therapy results are presented in order to demonstrate the possibilities as well as the limits of this in-vivo model performed with a view to clinical conditions.

Animals↗

PubMed2004

Prediction of aqueous solubility based on large datasets using several QSPR models utilizing topological structure representation.

Several QSPR models were developed for predicting intrinsic aqueous solubility, S(o). A data set of 5,964 neutral compounds was sub-divided into two classes, aromatic and non-aromatic compounds. Three models were created with different methods on both data sets: two regression models (multiple linear regression and partial least squares) and an artificial neural network model. These models were based on 3343 aromatic and 1674 non-aromatic compounds for training sets; 938 compounds were used in external validation testing. The range in -log S(o) is -1.6 to 10. Topological structure descriptors were used with all models. A genetic algorithm was used for descriptor selection for regression models. For the artificial neural network (ANN) model, descriptor selection was done with a backward elimination process. All models performed well with r2 values ranging 0.72 to 0.84 in external validation testing. The mean absolute errors in validation ranged from 0.44 to 0.80 for the classes of compounds for all the models. These statistical results indicate a sound ANN model. Furthermore, in a comparison with eight other available models, based on predictions using a validation test set (442 compounds), the artificial neural network model presented in this work (CSLogWS) was clearly superior based on both the mean absolute error and the percentage of residuals less than one log unit. In the ANN model both E-State and hydrogen E-State descriptors were found to be important.

Databases, Factual↗

PubMed2003

How mental health providers spend their time: a survey of 10 Veterans Health Administration mental health services.

BACKGROUND: Allocation of provider time across clinical, administrative, educational, and research activities may influence job satisfaction, productivity, and quality of care, yet we know little about what determines time allocation. AIMS: To investigate factors associated with time allocation, we surveyed all mental health providers in one Veterans Health Administration (VHA) network. We hypothesized that both facility characteristics (academic affiliation, type of organization of services, serving as a hub for treatment of severely mentally ill, facility size) and individual provider characteristics (discipline, length of time in job, having an academic appointment) would influence time allocation. METHODS: Eligible providers were psychiatrists, psychologists, social workers, physician assistants, registered or licensed practical nurses or other providers (psychology technicians, addiction therapists, nursing assistants, rehabilitation, recreational, occupational therapists) who were providing care in mental health services. A brief self-report survey was collected from all eligible providers at ten VHA facilities in late 1998 (N = 997). Data regarding facility characteristics were obtained by site visits and interviews with managers. Multilevel modeling was used to examine factors associated with three dependent variables: (i) total time allocation by activity (clinical, administrative, educational, research); (ii) clinical time allocation by treatment setting (inpatient vs. outpatient); and (iii) clinical time allocation by type of care (mental vs. physical). Licensed Practical Nurses (LPNs) were used as the reference group for all analyses because LPNs were expected to spend the majority of their time on clinical activities. RESULTS: Overall, providers spent most of their time on clinical activities (77%), followed by administrative (11%), and educational (10%). Surprisingly, research activities accounted for only 2% of their time. Multilevel analysis indicated none of the facility-level variables were significant in explaining facility variance in time allocation, but individual characteristics were associated with time allocation. The model for predicting time allocation by inpatient or outpatient settings explained 16-18% of the variance in the dependent variable. In all models, provider discipline and length of time in job played an important role. Having an academic appointment was important only in the model examining total time allocation by activity type. DISCUSSION: These simple models explained only a small amount of variance in the three dependent variables which were intended to capture issues related to time allocation; and the low number of facilities limited our power to examine effects of facility-level factors. Our models performed better in predicting allocation of clinical time to treatment setting and type of treatment than in predicting overall time allocation. Discipline and length of time in job were significant across all models. In contrast, having an academic appointment was associated with allocating significantly less time to clinical activities and more time to administrative activities but not to any significant difference in time spent in either research or education. IMPLICATIONS: While a gold standard of optimal time allocation does not exist, it is striking that research, a stated mission of the VHA, accounted for so little of providers' time. The lack of involvement of clinicians in research has implications for recruitment and retention of high-quality mental health providers in this network and for the education of future providers. Without involvement of clinicians, research conducted in the network by nonclinicians may be less relevant to "real-world" clinical issues. Reductions of funds available to mental health, coupled with increased clinical demands, may have prompted this pattern of time allocation, and these findings attest to the challenges faced by large institutions that are charged with balancing many often seemingly competing missions.

Health Services Research↗

PubMed2004

Short-term changes in cell and matrix damage following mechanical injury of articular cartilage explants and modelling of microphysical mediators.

The short-term responses of articular cartilage to mechanical injury have important implications for prevention and treatment of degenerative disease. Cell and matrix responses were monitored for 11 days following injurious compression of cartilage in osteochondral explants. Injury was applied as a single ramp compression to 14 MPa peak stress at one of three strain rates: 7 x 10(-1), 7 x 10(-3) or 7 x 10(-5) s(-1). Responses were quantified in terms of the appearance of macroscopic matrix cracks, changes in cell viability, and changes in cartilage wet weights. Loading at the highest strain rate resulted in acute cell death near the superficial zone in association with cracks, followed over the 11 days after compression by a gradual increase in cell death and loss of demarcation between matrix zones containing viable versus nonviable cells. In contrast, loading at the lowest strain rate resulted in more severe, nearly full-depth cell death acutely, but with no apparent worsening over the 11 days following compression. Between days 4 and 11, all mechanically injured explants significantly increased in wet weight, suggesting loss of matrix mechanical integrity independent of compression strain rate. Results demonstrate that short-term responses of cartilage depend upon the biomechanical characteristics of injurious loading, and suggest multiple independent pathways of mechanically-induced cell death and matrix degradation. Modifications to an existing fiber-reinforced poroelastic finite element model were introduced and the model was used for data interpretation and identification of microphysical events involved in cell and matrix injury. The model performed reasonably well at the slower strain rates and exhibited some capacity for anticipating the formation of superficial cracks during injurious loading. However, several improvements appear to be necessary before such a model could reliably be used to draw upon in vitro experimental results for prediction of injurious loading situations in vivo.

Animals↗

PubMed2026

An Exosomal Signature for Preoperative Detection of Occult Liver Metastasis in Pancreatic Cancer.

IMPORTANCE: Early liver metastasis (early-LiM) after pancreatectomy represents an aggressive biological phenotype of pancreatic ductal adenocarcinoma (PDAC) and is associated with markedly poor survival. Reliable preoperative biomarkers to identify occult hepatic micrometastasis remain lacking. OBJECTIVE: To develop and externally validate a circulating exosomal microRNA (exo-miRNA)-based machine learning model for preoperative detection of occult early-LiM in PDAC. DESIGN, SETTING, AND PARTICIPANTS: This multicenter retrospective case-control study included 3 phases: genome-wide discovery using exo-miRNA sequencing (discovery cohort), model development (training cohort), and independent external validation (2 validation cohorts). The study took place at 4 medical centers in China, Japan, and South Korea. A total of 372 patients were enrolled between 2011 and 2024. Data were analyzed from July 2024 to November 2025. EXPOSURES: Circulating plasma-derived exosomal miRNA expression profiles. MAIN OUTCOMES AND MEASURES: The primary outcome was early-LiM, defined as liver recurrence within 6 months after curative-intent resection. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and survival outcomes were assessed using Kaplan-Meier analysis. RESULTS: Among 372 patients with PDAC (median [IQR] age, 67 [59-73] years; 229 [61.6%] male and 143 [38.4%] female; median follow-up among survivors, 969 days),early-LiM was associated with significantly worse overall survival compared with other recurrence patterns (median OS, 9.1 months vs 26.6-31.8 months; log-rank P < .001). A 7-exo-miRNA extreme gradient boosting model demonstrated discrimination in the training cohort (AUC, 0.899; 95% CI, 0.822-0.976) and maintained performance in external testing cohorts (AUC, 0.876; 95% CI, 0.846-0.951 and AUC, 0.862; 95% CI, 0.744-0.981). The exo-miRNA panel score remained an independent identifier of early-LiM in multivariable analysis (odds ratio, 26.49; 95% CI, 18.45-55.28; P < .001) and stratified overall survival (log-rank P < .001). Decision curve analysis suggested improved net clinical benefit compared with conventional clinicopathologic variables. CONCLUSION AND RELEVANCE: In this multicenter study, a circulating exo-miRNA-based machine learning model enabled preoperative detection of occult early liver metastasis risk in PDAC. These findings support the potential of exosomal biomarkers to inform biology-guided treatment sequencing and warrant prospective validation.

Journal Article↗

PubMed2026

Machine Learning and Metabolomics to Characterize Warburg-Like Metabolic Subtypes in Human Retinal Endothelial Cells Exposed to Risk Factors Associated With Proliferative Diabetic Retinopathy.

PURPOSE: High glucose (HG), hypoxia (Hyp), and their combination are major risk factors for proliferative diabetic retinopathy (PDR). Although these conditions induce features of the Warburg-like metabolic reprogramming in human retinal endothelial cells (HRECs), it remains unclear whether they produce distinct metabolic and angiogenic subtypes. This study aimed to characterize the Warburg-like-associated metabolic heterogeneity induced by these PDR-related risk factors and evaluate the ability of supervised machine-learning models to distinguish these subtypes. METHODS: HRECs were cultured under normoglycemic, HG, Hyp (2% O2), and combined HG-Hyp conditions. Untargeted LC-MS/MS metabolomics quantified metabolites spanning carbohydrates, amino acids, nucleotides, and lipids. Principal component analysis (PCA) assessed overall metabolic variation, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis identified metabolic pathways associated with angiogenesis. In vitro angiogenesis assays measured endothelial tube formation and branching. Nine supervised classifiers (decision tree, logistic regression, naïve Bayes, random forest, K-Nearest Neighbors, neural network, gradient boosting, AdaBoost, and Support Vector Machine) were trained on the highest-ranked metabolites selected by the Information Gain Ratio feature-ranking approach. Model performance was evaluated using 10-fold cross-validation, leave-one-out cross-validation (LOOCV), permutation testing, and a classifier stability analysis under biologically meaningful distributional shift using an independent chemically induced hypoxia model (CoCl2). RESULTS: PCA revealed partial separation of metabolic profiles across conditions, indicating different Warburg-like metabolic subtypes. The combined HG-Hyp condition exhibited enhanced angiogenic potential relative to either HG or Hyp alone. KEGG pathway enrichment analysis identified fatty acid biosynthesis and elongation among the most significantly enriched pathways in HRECs under combined HG-Hyp conditions, alongside amino sugar and nucleotide sugar metabolism, glycerophospholipid metabolism, the pentose phosphate pathway, and glycolysis/gluconeogenesis. Supervised machine-learning classifiers distinguished these metabolic subtypes, with AdaBoost and gradient Boosting showing the most balanced, reproducible performance across 10-fold cross-validation, LOOCV, and permutation testing, and remaining the most reliable classifiers under domain-shift testing (area under the curve = 0.88, P = 0.0061). CONCLUSIONS: In this exploratory analysis, HG, Hyp, and their combination drive metabolically and functionally distinct subtypes of Warburg-like metabolic reprogramming in HRECs, with HG-Hyp in combination producing a highly angiogenic phenotype. Boosting-based ensemble classifiers provide a promising framework for detecting these subtypes even under domain-shift conditions, warranting validation in larger independent datasets. TRANSLATIONAL RELEVANCE: Integrating metabolomics with machine-learning classification offers a strategy to identify Warburg-like metabolic subtypes in retinal endothelial cells, providing insights into angiogenic mechanisms and guiding the development of targeted diagnostics or therapeutics for PDR.

Humans↗