Area Under The ROC Curve Research Articles (Page 1)

In intensive care units (ICU), malnutrition is very common and closely related to severe inflam­matory states. The aim of this study was to develop and validate a nutritional inflammatory prognostic score (NIPS) predictive of short-term mortality, and to verify the validity of existing scores. A total of 606 ICU patients were included in a longitudinal study. The population was randomly divided into two groups: development (383) and validation (223). The NIPS score was developed from nutritional and inflammatory parameters using multi-adjusted Cox proportional regression. Validation of NIPS as a prognostic score was tested using the area under the ROC curve (AUC), Cox proportional regression, and the Kaplan-Meier curve. The validity of five selected scores was also assessed. The NIPS score was developed from C-reactive protein to prealbumin ratio, total cholesterol, procalcitonin and neutrophil to lymphocyte ratio. With an AUC of 0.89 and a cutoff of 5.0, NIPS predicted short-term mortality with a sensitivity of 80.5% and a specificity of 93.8%. The risk of mortality was fivefold higher in the high nutritional risk group (RR= 5.0, [3.1–7.9], p < 0.0001). The crude cumulative incidence of mortality was significantly higher in the high-risk group (pLog-Rank < 0.0001). The five selected scores had a significant, but, lower prognostic value, compared with the developed score (AUC between 0.59 and 0.70). This study provides a new nutritional risk score, based on widely available biological parameters, with proven efficiency in predicting ICU mortality risk. Nutritional risk screening should be routinely performed at the earliest admission stages.

Cuproptosis is a type of cell death caused by copper imbalance associated with the growth and proliferation of cancer cells. Long noncoding RNAs (lncRNAs) play a crucial role in hepatocellular carcinoma (HCC) development. Here, we aimed to investigate the role of cuproptosis-related lncRNAs in the clinical prognostic prediction and immunotherapy for HCC. A correlation network between lncRNAs and cuproptosis-related genes in HCC was constructed by conducting co-expression and Cox regression analyses. LASSO-Cox analysis was used to obtain lncRNAs that constitute the cuproptosis-associated lncRNA signature, which was then used to predict patient prognosis and immunotherapy response. To verify the clinical applicability of the risk model, a nomogram was constructed and an anti-neoplastic drug sensitivity analysis was performed. Four cuproptosis-related lncRNAs (AL590705.3, SPRY4-AS1, AC135050.5, and AL031985.3) were identified and used to develop a prognostic signature for HCC. Based on the four lncRNAs, a risk prediction model was established. For overall survival (OS) in HCC patients, the area under the ROC curve (AUC) for the risk score was 0.715, outperforming traditional clinical factors such as age (AUC=0.531), gender (AUC=0.509), and stage (AUC=0.671). High-risk patients had worse overall survival, progression-free survival, and higher mortality. Independent ROC analysis, nomogram-based modeling, and concordance index analysis indicated that the risk model has high predictive accuracy for HCC. This model demonstrates potential in predicting prognosis, and offers novel insights to the treatment and management of HCC. Our study suggests that lncRNAs may serve as a novel biomarker and potential therapeutic target for HCC.

Lysine 2-hydroxyisobutyrylation (Khib) has emerged as a crucial Post-Translational Modification(PTM) with significant roles in diverse biological processes ranging from gene expression to metabolic regulation. Despite its importance, computational approaches for accurately predicting Khib sites remain limited. This study introduces BLOS-Khib, a deep-learning framework that utilizes evolutionary information encoded in the BLOSUM62 matrix within a Convolutional Neural Network (CNN) architecture for cross-species Khib site prediction. Through systematic optimization, we found that a 43-amino acid peptide length captures the optimal sequence context for prediction across six taxonomically diverse organisms. Comprehensive comparative analyses demonstrated BLOS-Khib competitive performance compared to existing methods, achieving notable Area Under the ROC Curve (AUC) values on independent test sets: human (0.913), wheat (0.892), T. gondii (0.893), rice (0.887), Candida albicans (0.885), and Botrytis cinerea (0.903). Our framework showed improved performance compared to state-of-the-art approaches, including traditional machine learning classifiers and alternative deep learning architectures. Sequence signature analysis revealed both conserved lysine-rich regions preceding modification sites and species-specific amino acid preferences at positions immediately flanking the target residue. Notably, our cross-species applicability experiments identified high transferability between evolutionarily distant organisms, ensuring the potential convergent evolution of Khib determinants. BLOS-Khib demonstrates competitive performance for PTM prediction, while providing evolutionary insights into the sequence determinants governing this emerging regulatory mechanism across diverse species.

Background Vitiligo is a chronic autoimmune disorder with profound psychosocial implications. Methods The paper propose a multimodal artificial intelligence (AI) framework that combines and integrates YOLOv11 for the detection of dermatological lesion and a BERT-based sentiment classifier for the monitoring of mental health, supported by questionnaire data sets (DLQI, RSE). Results YOLOv11 achieved mAP = 98.8%, precision = 95.6%, recall = 97.0%; the mental health module uses a BERT-based sentiment classifier, fine-tuned in the GoEmotions corpus, reaching F1 = 0.83. A simulated fusion score that integrates the Dermatology Life Quality Index (DLQI) and Rosenberg Self-Esteem (RSE) scores, resulting in an area under the ROC curve (AUC) of 0.82 for the identification of high-risk patients. Conclusion The implemented prototype establishes the feasibility of AI-assisted psychodermatology, allowing early diagnosis, emotional monitoring, and real-time alerting by physicians.

Colorectal neoplasia contributes substantially to morbidity and mortality worldwide. While fecal immunochemical test (FIT) has been adopted as a screening method, its limitations (suboptimal adherence to follow-up colonoscopy among FIT-positive population and missed diagnosis in FIT-negative population) hamper its effectiveness in reducing colorectal cancer (CRC). To address these issues, a combination of FIT with circulating tumor cell (CTC) enumeration for colorectal neoplasia prediction and CRC risk evaluation was proposed. Participants (n=113) underwent FIT, colonoscopy examination, and CTC enumeration. For the latter, CD45neg EpCAMpos CTC and CD45neg EpCAMneg cell counts were assayed. For statistical analysis, multivariable logistic regression and area under the ROC curve (AUROC) analysis were used. For colorectal neoplasia prediction, a better performance (AUROC 0.8371) was achieved when the variables (FIT results and CTC enumeration) were combined, which was significantly improved compared to that of the FIT result-only model (AUROC 0.7555). For its further use for CRC risk assessment, among FIT-positive individuals, those with either ≥5 CD45negEpCAMpos CTCs or ≥300 CD45negEpCAMneg cells had significantly increased CRC risk. For the FIT-negative population, 80% of the FIT-negative CRC cases in this study could be detected if the thresholds were set at ≥2 CD45negEpCAMpos CTCs and ≥500 CD45negEpCAMneg cells. Its application could thus rescue missed diagnoses due to false-negative FIT results. The method was able to enhance the prediction of colorectal neoplasia. The assessment of CRC risk was also successfully demonstrated. The method is promising as a supportive method tackling the problems encountered in current FIT-based screening.

Background: Cancer therapy–related cardiac dysfunction (CTRCD) is a well-recognized adverse effect of HER2-targeted therapies, particularly trastuzumab. While current guidelines recommend serial left ventricular ejection fraction (LVEF) monitoring every 3 months during treatment, such uniform surveillance may lead to overutilization in low-risk patients. Objective: This study aimed to externally validate a nomogram from Memorial Sloan Kettering to predict CTRCD in a contemporary cohort of patients with HER2-positive breast cancer. Methods: This retrospective study included women with HER2-positive breast cancer treated with trastuzumab between 2013&amp;2022 at a large cancer center. The primary endpoint was 1-year CTRCD-free survival, with CTRCD defined as an LVEF decline ≥10% to &lt;53% or a ≥16% reduction from baseline. Patients were stratified into risk groups based on total nomogram points. Model performance was evaluated in the validation cohort using Kaplan-Meier estimates, calibration plots, concordance index (C-index), and area under the ROC curve (AUC). Results: The derivation and validation cohorts demonstrated notable differences in baseline characteristics. Patients in the validation cohort (n=356) had a higher mean BMI (39.6% vs. 22.9%) and greater prevalence of cardiovascular comorbidities, including hypertension (34.8% vs. 23.5%), diabetes (11.5% vs. 7.2%), and hyperlipidemia (24.7% vs. 17.1%) but with substantially lower anthracycline exposure (31.5% vs. 77.8%) compared to the derivation cohort. 9.6% of patients developed CTRCD within 395 days of trastuzumab initiation, which was defined as the 1-year CTRCD event. The 1-year CTRCD event rates across increasing quartiles of the nomogram score were 6.0%, 4.4%, 9.6%, and 19.5%, respectively, indicating effective risk stratification. Kaplan-Meier–estimated 1-year CTRCD-free survival was 90.1%, closely aligned with the nomogram-predicted probability of 87.8%. The model demonstrated excellent discriminative performance, with a C-index of 0.68 (95% CI, 0.49–0.86) and an AUC of 0.69 (95% CI, 0.58–0.78) (Figure A, B). Conclusion: The novel CTRCD nomogram demonstrated excellent discriminatory power and generalizable to an independent, real-world patient population. These findings support its utility for personalized CTRCD risk stratification and suggest its potential to guide less frequent LVEF monitoring in low-risk patients receiving HER2-targeted therapy.

Background: Approximately 50% of patients with angina who undergo invasive coronary angiography (ICA) do not exhibit obstructive epicardial coronary artery disease. A significant subset of these patients has coronary microvascular dysfunction (CMD). The current gold standard for diagnosing CMD is the invasive Index of Microcirculatory Resistance (IMR), which requires administration of adenosine and pressure wire instrumentation, carrying procedural risks. Angiography-derived IMR (Angio-IMR) is a novel, non-invasive, machine learning-based method that applies computational fluid dynamics to standard ICA images to emulate hyperemic microvascular assessment. Hypothesis: Angio-IMR is an accurate predictor of CMD when compared with invasively measured IMR. Methods: This prospective, multicentre study was conducted across three tertiary Australian hospitals. Patients undergoing ICA were assessed using both invasive IMR and Angio-IMR. Angio-IMR analyses were performed centrally and blinded, using manual frame counting with Medis QFR software (version 3.1). Patients were stratified into derivation and validation cohorts. Diagnostic performance of Angio-IMR was assessed using ROC analysis, with invasive IMR &gt; 25 was used as the threshold for significant CMD. Results: A total of 335 patients (63 ± 10.6 years; 56% male), encompassing 394 vessels, were included. In the derivation (n = 124) and validation (n = 124) cohorts, significant CMD (IMR &gt; 25) was present in 27% and 23% of vessels respectively. Angio-IMR showed moderate correlation with invasive IMR (r = 0.47 in derivation; r = 0.33 in validation). Angio-IMR predicted significant CMD with an area under the ROC curve (AUC) of 0.82 and 0.76 (both p &lt; 0.001) in the derivation and validation cohorts, respectively. An Angio-IMR threshold of &gt; 21 optimized sensitivity and specificity. At this cutoff, sensitivity and negative predictive value (NPV) were 94.9% and 95.9% in the derivation cohort, and 89.0% and 92.0% in the validation cohort, respectively. Conclusion: Angio-IMR is a promising non-invasive tool for the detection of CMD, demonstrating strong diagnostic performance, particularly in ruling out disease. Its high sensitivity and NPV support its use as a gatekeeper to invasive IMR, potentially reducing need for pressure wire-based assessments in low-risk patients. This study establishes Angio-IMR as a clinically valuable adjunct in the assessment of microvascular coronary disease.

Background: C1q/TNF-related protein 7 (CTRP7), a significant adipokine primarily secreted by adipocytes, serves as a key effector molecule in vascular remodeling signaling pathways and inflammatory responses. This study aimed to investigate the association between serum CTRP7 levels and short-term outcomes following return of spontaneous circulation (ROSC). Methods: A retrospective cross-sectional study was conducted using clinical data from in-hospital cardiac arrest (IHCA) patients treated at West China hospital between 00:00 on January 31, 2021, and 23:59 on January 31, 2022, in combination with an established biobank. Serum CTRP7 concentrations were compared between neurological outcome groups 28 days post-ROSC to evaluate their correlation with short-term neurological prognosis. Results: A total of 129 IHCA patients were included. Serum CTRP7 concentrations were significantly higher in the 28-day mortality group (P&lt;0.05) and in patients with poor 28-day neurological outcomes (P&lt;0.05). Among survivors at 28 days, CTRP7 levels were positively correlated with Cerebral Performance Category (CPC) scores (r=0.415, P&lt;0.001). The area under the ROC curve (AUC) for CTRP7 in predicting 28-day mortality was 0.713 (95% CI: 0.625-0.791), with 58.49% sensitivity and 83.10% specificity. The AUC for predicting poor 28-day neurological outcomes was 0.682 (95% CI: 0.561-0.787), with 66.67% sensitivity and 72.00% specificity, indicating that serum CTRP7 has moderate predictive value for short-term neurological prognosis after CPR. Conclusions: In IHCA patients, elevated serum CTRP7 levels after CPR are associated with increased mortality and worse neurological outcomes, suggesting its potential as a prognostic biomarker.

Background and Aims: Carotid plaque is an early manifestation of atherosclerosis and is closely associated with the risk of myocardial ischemia, ischemic stroke, and other atherosclerotic cardiovascular diseases (ASCVDs). This study aims to identify new protein biomarkers associated with carotid plaque, which will enhance early warning of ASCVD. Methods: We launched a nested case-control study based on the blood samples at baseline and repeated carotid ultrasound measurements during follow-ups in the ChinaHEART cohort. Among participants without carotid plaque at baseline, 145 with incident carotid plaque within two-year follow-up were selected as cases, and 147 without incident carotid plaque during follow-up were matched for demographic characteristics and traditional risk factors as controls. After the Meso Scale Discovery test for 28 biomarkers, Least absolute shrinkage and selection operator (LASSO) regression was used to select potential predictors and constructed a logistic regression model for predicting carotid plaque. Furthermore, the incremental predictive value was validated in the UK Biobank of 30,800 subjects. Results: A total of 11 biomarkers, including thrombomodulin, ICAM-3, P-Selectin, GDF-15, adiponectin, MCP-1, IL-10, PlGF, Tie-2, VEGF-D, and VCAM-1 were selected by LASSO regression and used to construct a prediction model for the carotid plaque. The area under the ROC curve (AUC) of the eventual model is 0.778 and it showed good calibration capability graphically with a Brier score of 0.192. In the UK Biobank cohort, when these biomarkers were added to a traditional predictive model, a better predictive power was generated, with an AUC improvement of 0.021 (P &lt;0.001, Delong test), Brier score of 0.093, a continuous NRI of 0.259 (0.223-0.294, P &lt;0.001), IDI of 0.017 (0.015-0.019, P &lt;0.001) reference to the traditional model. Conclusions: We found and validated the biomarkers, including thrombomodulin, ICAM-3, P-Selectin, GDF-15, adiponectin, MCP-1, IL-10, PlGF, Tie-2, VEGF-D, and VCAM-1, can predict the incidence of carotid plaque in ChinaHEART, and except for Tie-2, these biomarkers have additional value for the prediction of incident ASCVD in UK Biobank.

Background: TAVR is increasingly offered for treatment of aortic stenosis with over 100,000 procedures performed annually in the US. Advanced heart block is a known complication that requires permanent pacemaker implantation. While disparities in cardiovascular outcomes by race and sex are well-documented, an analysis of both variables simultaneously offers a more nuanced and practical description of these disparities. In this study, we introduce a composite race-sex variable—using White males as the reference category—to assess differential outcomes in TAVR patients developing advanced atrioventricular block (AVB) requiring a permanent pacemaker (PPM). Methodology: We conducted a retrospective cohort study using the 2021 national inpatient sample database, a survey weighted, nationwide sample with over 6 million observations across the US. Post TAVR patients (identified using ICD-10 codes) comprised our patient population. The primary outcome was AVB (Type II mobitz and third degree) necessitating pacemaker insertion and all-cause mortality. The primary exposure variable was a 12-level composite of gender and race values (1= white male to 12=other female). Logistic regression was employed to calculate crude and adjusted OR calculated for AVB and all cause mortality (primary outcome) adjusting for demographics and comorbidities using Charlson comorbidity index. Results: Among the 86,840 patients who underwent TAVR, 4.32% (n=3751) developed heart block requiring pacemaker insertion. Mean age of the cohort was 77.8 years with 42.1% identifying as female. Overall in-hospital mortality was 1.05%. Adjusted for age and comorbidity the all-cause mortality was 1.5 times higher for white females (OR:1.48, CI:1.05-2.07, p=0.02) and 3.5 times higher for hispanic females (OR:3.5, CI:1.7-6.8, p=0.00) when compared to white males. However, no statistically significant difference was found between the groups in the incidence of AVB requiring PPM placement. The logistic regression model demonstrated moderate discrimination with an area under the ROC curve (AUC) of 0.69. Conclusion: White and Hispanic females experience significantly increased all-cause mortality following TAVR compared to white males, despite similar rates of heart block and PPM implantation. These findings underscore the importance of considering intersectional race-sex variables when evaluating outcomes and may inform more equitable preprocedural risk stratification and post-procedure care strategies

BackgroundPolygenic risk scores (PRSs) are increasingly being used to predict disease risk from genetic data. While promising in research, their clinical utility—especially when combined with non-genetic (NG) data such as lab results, physical measurements, and diagnostic history—remains uncertain. Myocardial infarction (MI), a leading cause of morbidity and mortality, is a key use case for assessing the incremental value of PRSs in risk models.MethodsUsing UK Biobank data, we evaluated the added value of PRSs for 10-year MI risk prediction. We trained models with NG data alone and in combination with PRSs, varying model complexity and the NG feature space. Two modeling frameworks were used: logistic regression and a neural network. NG data was defined using two feature sets: NG1, which included established MI risk factors from structured fields; and NG2, a high-dimensional dataset derived from millions of diagnostic codes across five linked UK Biobank electronic health records (EHR) datasets combined with NG1 features. NG2 was generated using a deep representation learning approach that produced low-dimensional embeddings capturing latent medical concepts and disease co-occurrence patterns. Each model was trained with and without PRSs and evaluated using metrics such as the area under the ROC curve (AUC).ResultsPRSs add minimal predictive value when used alone. In contrast, diagnostic data from EHRs significantly improve performance. The best results are achieved using a multimodal neural network combining NG1, NG2, and PRSs.ConclusionsPRSs provide limited standalone utility for MI prediction compared to detailed diagnostic data. Their clinical value likely lies in integration with EHR-based models. Future work should focus on multi-modal approaches that contextualize PRS information within broader clinical data.

Development and validation of a machine learning model to predict cognitive behavioral therapy outcome in obsessive-compulsive disorder using clinical and neuroimaging data.

Electrocardiographic LVH criteria: Poor diagnostic accuracy even with optimized cutoffs. Insights from MESA study.

BackgroundAlthough many drugs have been associated with drug-induced movement disorders (DIMDs), the associated risks are unclear. This study aimed to identify high-risk drugs for DIMDs through disproportionality analysis of the Food and Drug Administration Adverse Event Reporting System (FAERS) database and to explore risk factors for DIMDs through sensitivity analysis.MethodsFour disproportionality analysis methods were used to assess the risk signals of drugs that may induce DIMDs from the first quarter of 2004 to the fourth quarter of 2024. One-way analyses, LASSO analyses, and logistic regression analyses were performed to explore the risk factors associated with DIMDs.ResultsThere are 138,081 reports related to DIMDs. This study identified 148 suspected drugs. Age under 33 years, male gender, and 62 medications, including METOCLOPRAMIDE, ARIPIPRAZOLE, CARBIDOPA, LEVODOPA, RISPERIDONE, and QUETIAPINE, are all independent risk factors for drug-induced movement disorders. The area under the ROC curve (AUC) reflecting model predictive accuracy was 0.724.ConclusionOur disproportionality analysis and sensitivity analysis of the FAERS database identified drugs potentially associated with DIMDs. These findings can provide valuable information for clinicians to be more cautious when prescribing these drugs and to monitor patients for the development of movement disorders closely. Additionally, the results can help regulatory agencies make informed decisions regarding the safety of drugs.

This study aims to evaluate the ability of an artificial intelligence (AI) model developed for use in the field of orthodontics to accurately and reliably classify skeletal maturation stages of individuals using hand–wrist radiographs. A total of 809 grayscale hand–wrist radiographs (250 × 250 px; pre-peak n = 400, peak n = 100, post-peak n = 309) were analyzed using four complementary image-based feature extraction methods: Local Binary Pattern (LBP), Histogram of Oriented Gradients (HOG), Zernike Moments (ZM), and Intensity Histogram (IH). These methods generated 2355 features per image, of which 2099 were retained after variance thresholding. The most informative 1250 features were selected using the ANOVA F-test and classified with a stacking-based machine learning (ML) architecture composed of Light Gradient Boosting Machine (LightGBM) and Logistic Regression (LR) as base learners, and Random Forest (RF) as the meta-learner. Across all evaluation folds, the average performance of the model was Accuracy = 83.42%, Precision = 84.48%, Recall = 83.42%, and F1 = 83.50%. The proposed model achieved 87.5% accuracy, 87.8% precision, 87.5% recall, and an F1-score of 87.6% in 10-fold cross-validation, with a macro-average area under the ROC curve (AUC) of 0.96. The pre-peak stage, corresponding to the period of maximum growth velocity, was identified with 92.5% accuracy. These findings indicate that integrating handcrafted radiographic features with ensemble learning can enhance diagnostic precision, reduce observer variability, and accelerate evaluation. The model provides an interpretable and clinically applicable AI-based decision-support tool for skeletal maturity assessment in orthodontic practice.

This study explores the use of radiomic features extracted from preoperative T2-weighted MRI and CT images, combined with machine learning models, to predict the risk of vertebral refracture after percutaneous kyphoplasty (PKP) in postmenopausal women. We retrospectively collect data from 156 postmenopausal women with osteoporotic vertebral compression fractures (OVCFs) who underwent PKP (35 refracture cases, 121 non-refracture controls). All patients had preoperative lumbar T2-weighted MRI and CT scans. We extract MRI and CT radiomic features and constructed radiomic signatures through feature selection. Key clinical factors (age, body mass index [BMI], vertebral CT Hounsfield unit [HU] values, smoking history, diabetes history, alcohol use, etc.) are used to build clinical prediction models. Various machine learning classifiers (Support Vector Machine [SVM], K-Nearest Neighbors [KNN], Random Forest [RF], ExtraTrees, XGBoost, LightGBM, Multi-layer Perceptron [MLP]) are trained on the radiomic signatures and clinical factors. Model performance was evaluated on an independent test set using area under the ROC curve (AUC) as the primary metric. Accuracy, sensitivity, specificity, and other measures on the test set were compared between radiomic models, clinical models, and a combined model. The refracture group (n = 35, 22.4%) is significantly older (72.09 ± 4.25 vs 70.11 ± 3.31 years, P = 0.002) with lower vertebral bone density (97.00 ± 6.31 vs 102.49 ± 4.68 HU, P < 0.001). Among individual algorithms, the KNN clinical model achieves optimal performance (AUC = 0.74), while the SVM radiomics model demonstrates the best accuracy (AUC = 0.798, accuracy = 0.839, sensitivity = 0.857, specificity = 0.833). The combined model achieves superior performance (AUC = 0.886), significantly outperforming both standalone models. Multi-modal radiomics combined with key clinical factors provides superior prediction of refracture risk after PKP. This approach offers clinicians an objective tool for individualized risk stratification, representing a meaningful step toward precision medicine in managing osteoporotic fractures.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-21894-7.

This study aims to achieve accurate differentiation of malignant pleural mesothelioma (MPM) from metastatic pleural disease (MPD) and to predict the overall survival of MPM. This IRB-approved retrospective study included 385 subjects in total (85 patients with malignant mesothelioma and 290 with MPD secondary to lung adenocarcinoma). A ResNet-3D-18 model was trained on annotated pretreatment CT scans to distinguish MPM from MPD. Using chronological segregation, the training cohort included 70 histologically confirmed mesothelioma and 258 MPD cases, with an independent test cohort of 15 MPM and 32 MPD cases for validation. A multivariate logistic regression model served as the clinical benchmark for comparison. Deep learning features extracted from the trained ResNet model were then assessed for their prognostic utility in MPM patients using a random forest classifier. Model performance was evaluated at both lesion- and patient-levels, with metrics including the area under the ROC curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value. The ResNet-3D-18 model demonstrated excellent discriminative performance in differentiating MPM from MPD, with mean AUCs of 0.972 (95% CI 0.947-0.990) and 0.840 (95% CI 0.757-0.929) in the training and independent test cohorts. Compared to the clinical model, the deep learning approach showed higher sensitivity (0.867 vs. 0.533) in the independent test dataset. For overall survival prediction in MPM patients, the random forest classifier achieved an AUC of 0.829 (95% CI 0.663-0.943) in 5-fold cross-validation. ResNet-3D-18 classification model has excellent abilities in differentiating MPM from MPD, and morphological distinctions between MPM and MPD also contain prognostic information. Question The rising global incidence of malignant pleural mesothelioma contrasts with persistent diagnostic challenges. Findings Deep learning-derived discriminative features simultaneously contain prognostic information. Clinical relevance This study bridges the gap between radiological findings and clinical decision-making in MPM, offering a reproducible tool for early diagnosis and personalized prognosis prediction based on CT imaging alone.

Guillain-Barré syndrome (GBS) exhibits clinical heterogeneity and variable progression. In low-resource settings, malnutrition and limited treatment worsen prognosis, underscoring the need for a simple prognostic tool. This study evaluated the Prognostic Nutritional Index (PNI) and Nutritional Risk Index (NRI) in relation to GBS severity and long-term outcomes, comparing their predictive value with standard prognostic indicators. An observational cohort study of 252 GBS patients enrolled between 2019 and 2024 was conducted. PNI and NRI were calculated using serum albumin, lymphocyte count, and body weight. The GBS-disability score (GBS-DS) assessed baseline severity and 26-week outcomes. Statistical analysis included Chi-square tests, Mann-Whitney U tests, Spearman's ρ, and logistic regression to identify predictors. ROC analysis determined optimal PNI cut-offs, confirmed by Kaplan-Meier survival curves. PNI, unlike NRI, was significantly reduced in severe GBS (GBS-DS > 3) compared to mild/moderate GBS (GBS-DS ≤ 3). PNI correlated with GBS-DS (ρ = -0.62), MRC sum score (ρ = 0.5), hemoglobin (ρ = 0.53), and neutrophil count (ρ = -0.35) (all p < 0.0001). PNI independently predicted disease severity (odds ratio [OR] = 0.91; p = 0.036) and 26-week outcomes (OR = 0.93; p = 0.033). Area under the ROC curve (AUC) was 0.769 for severity and 0.719 for 26-week outcomes. PNI cut-offs of 49.395 and 45.72 predicted severe GBS and long-term poor outcome, respectively. Kaplan Meier analysis confirmed patients with PNI < 45.72 required a longer time to gain independent locomotion (p < 0.0001). Lower PNI, but not NRI, is associated with greater GBS severity and poor long-term outcomes. PNI independently predicted disease severity and 26-week outcomes, with specific cut-offs identifying patients requiring longer recovery, supporting its prognostic utility.

This study evaluates the predictive performance of various machine learning (ML) algorithms for postpartum hemorrhage (PPH), peripartum hysterectomy, and severe coagulopathy using routinely collected pre-delivery clinical and biochemical data. In this retrospective cohort study, data from 566 deliveries at a tertiary obstetric center between 2019 and 2025 were analyzed. A total of 283 patients with PPH and 283 matched controls were included. Twenty maternal variables, including hemoglobin, body mass index, uterine scar, and comorbidities, were used to develop ML models: support vector machine (SVM), logistic regression, random forest, gradient boosting, and naive Bayes. Model performance was evaluated using accuracy, F1 score, and area under the ROC curve (AUC). Reduced-feature models with ten predictors were also assessed. The SVM model demonstrated the highest performance for PPH prediction (accuracy: 83.3%, AUC: 0.903), followed closely by logistic regression (AUC: 0.902). Reduced-feature models maintained high performance (AUCs >0.88), indicating feasibility for practical deployment. Random forest achieved the best performance for predicting hysterectomy (AUC: 0.88) and coagulopathy (AUC: 0.90). Key predictors included low pre-delivery hemoglobin, prolonged active labor phase, uterine scar, and preterm delivery. Machine learning models can reliably identify patients at risk for postpartum hemorrhage and its complications using accessible pre-delivery data. The robustness of reduced-variable models enhances their clinical utility, especially in resource-limited settings. Integration of such algorithms into electronic health record systems might support early intervention and improved maternal outcomes.

Preoperative diagnosis of the nature of PCNSL is crucial for distinguishing it from commonly confused intracranial gliomas, such as glioblastoma multiforme (GBM). The present research aims to sequence peripheral blood exosomal miRNAs from PCNSL and GBM patients, and to identify the discriminative miRNAs with high feasibility. Plasma exosomal RNAs were extracted, and miRNA sequencing was performed. The miRNAs with significant differences were validated by RT-qPCR. A total of 67 miRNAs were significantly different between the PCNSL and GBM groups. Twenty-seven PCNSL and 27 GBM patients were enrolled for the RT-qPCR validation of 10 selected differentially expressed miRNAs. The expression levels of plasma exosomal hsa-miR-148a-3p, hsa-let-7f-5p, hsa-miR-345-5p and hsa-miR-4433b-5p were upregulated in the PCNSL group compared with those in the GBM group (P = 0.019, P = 0.036, P = 0.009, and P = 0.001, respectively). The combined panel comprising hsa-miR-148a-3p, hsa-miR-345-5p, and hsa-miR-4433b-5p demonstrated significantly enhanced diagnostic performance, with an area under the ROC curve (AUC) of 0.791. Immunohistochemistry analysis revealed that the expression level of EGFR in PCNSL was significantly lower than that in GBM. Western blot and RT-qPCR analysis of EGFR expression levels in LN229 cells revealed that miR-148a-3p and miR-4433b-5p downregulated EGFR expression. The results of luciferase reporter assay showed that the relative luciferase activity of HEK293T cells transfected with EGFR-WT was notably suppressed by miR-4433b-5p (P < 0.001). In conclusion, the plasma exosomal hsa-miR-148a-3p, hsa-miR-345-5p, and hsa-miR-4433b-5p might be identified as novel biomarkers for differentiating PCNSL and GBM. The increased expression level of EGFR in GBM may be achieved by the negative regulatory effects of miR-4433b-5p. The diagnostic performance of the miRNA biomarkers still needs to be further verified in larger sample size cohorts.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12885-025-14933-z.