2026-09-30

Introduction: Anemia is a common nutritional complication after bariatric surgery, but preoperative tools for individualized risk assessment remain limited. This study aims to develop and externally validate a machine learning model for predicting anemia status one year after bariatric surgery. Methods: This multicenter study included patients aged 16-65 years undergoing sleeve gastrectomy or Roux-en-Y gastric bypass. Patients from one hospital formed the derivation cohort, and patients from two independent hospitals formed the external validation cohort. Forty-five preoperative variables were evaluated using logistic regression, random forest, support vector machine, and extreme gradient boosting. Model discrimination, calibration, and clinical utility were assessed and compared with blinded predictions from 10 clinicians with different levels of experience. Results: The derivation cohort included 511 patients (409 training, 102 internal testing), and the external cohort included 300 patients. One-year anemia occurred in 39.6%, 39.2%, and 19.0% of patients in the training, testing, and external validation cohorts, respectively. Random forest showed the most favorable overall performance profile and was simplified to seven predictors. The final model achieved an area under the curve of 0.827 (95% CI, 0.746-0.896) in the testing set and 0.823 (95% CI, 0.771-0.873) in the external validation cohort, with satisfactory calibration and decision-curve net benefit. In external validation, the model outperformed junior, middle-level, and senior clinicians and remained superior in clinician-defined challenging cases. Conclusions: A seven-variable preoperative model showed good discrimination for one-year anemia status after bariatric surgery and may support individualized postoperative monitoring and nutritional care.

Background: Early mortality risk stratification remains challenging in intensive care unit (ICU) patients with acute upper gastrointestinal bleeding (AUGIB). We developed a parsimonious, interpretable ensemble model and evaluated its performance across independent ICU cohorts. Methods: MIMIC-IV was used for model development and internal validation (n=3,728), and eICU-CRD (n=7,529) and a Hainan cohort (n=200) were used for external validation. Adults with ICU stays of at least 24 hours were included. Subsequent all-cause in-hospital mortality was predicted from routinely available data collected during the first 24 hours after ICU admission. Within MIMIC-IV, predictors were prioritized according to multi-algorithm ranking and clinical feasibility, and a 17-feature subset was selected using a performance-parsimony criterion supported by knee-point analysis. Nine machine-learning algorithms were evaluated, and four complementary learners were combined using equal-weight soft voting based on cross-validated performance and prediction complementarity. Model evaluation included AUROC, PR-AUC, Brier score, calibration, and decision-curve analysis. Ensemble-level SHAP and pathway analyses supported model interpretation. Results: AUGIB Soft-Voting Ensemble (AUGIB-SVE) achieved AUROCs of 0.847, 0.837, and 0.848 in MIMIC-IV, eICU, and Hainan validation, respectively. It also achieved higher AUROCs than AIMS65, pre-endoscopic Rockall, SOFA, and SAPS II in the cohorts in which these scores could be reconstructed. Decision-curve analysis showed positive net benefit across clinically relevant threshold ranges. SHAP and pathway-support analyses linked influential predictors to four clinically coherent biological axes: hemorrhage/perfusion, coagulation dysfunction, inflammation/hypoxia, and multiorgan dysfunction. AUGIB-SVE was deployed as an online risk-prediction tool. Conclusions: AUGIB-SVE provides a parsimonious and interpretable framework for early in-hospital mortality risk stratification in ICU patients with AUGIB. It showed consistent discrimination across independent cohorts, favorable clinical utility, and higher discrimination than reconstructable conventional scores. Trial Registration: Not applicable.

Introduction: Long-term renal function is an important follow-up after radical nephrectomy for renal carcinoma. Predictive models can help identify patients at risk for chronic kidney disease (CKD) progression and enable early intervention. Methods: We retrospectively analyzed 649 patients who underwent radical nephrectomy at eight medical centers. The primary cohort from the Affiliated Hospital of Qingdao University comprised 329 patients, randomly divided into training sets (n = 229) and internal validation sets (n = 100) at a 7:3 ratio. An additional 320 patients from seven other centers constituted a multicenter external evaluation cohort. Five machine learning models were developed, including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM). Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), confusion matrices, and calibration curves. Results: 224 (34.51%) patients experienced postoperative CKD stage progression within three years. LightGBM achieved the highest discriminative performance among the five evaluated algorithms, with an AUC of 0.7508 (95% CI, 0.6399-0.8617) and an accuracy of 0.7200 in the internal validation set, and an AUC of 0.7549 (95% CI, 0.7022-0.8076) and an accuracy of 0.6813 in the external evaluation set. SHAP analysis identified preoperative eGFR, tumor size, post-to-preoperative serum creatinine ratio, preoperative serum creatinine, and age as the five most influential predictors. Conclusions: We developed and externally evaluated machine-learning models for predicting CKD stage progression after radical nephrectomy. Long-term decline in renal function is an important complication that requires urologists' attention and early prevention during follow-up.

Purpose: Although mechanical thrombectomy (MT) achieves high recanalization rates in acute ischemic stroke (AIS), functional outcomes remain highly inconsistent. We aimed to evaluate the effectiveness of radiomics and deep learning features (DLFs) extracted from both intra-thrombus and peri-thrombus regions on baseline CT in predicting prognosis following MT. Materials And Methods: This retrospective, multicenter study included 471 AIS patients who underwent MT. Patients from Center A (n = 315) constituted the training cohort, while those from Center B and C (n = 156) served as an external test cohort. Thrombus regions were manually segmented, with a subsequent 1 mm expansion to define the peri-thrombus regions. A total of 428 radiomics features and 128 DLFs were extracted from non-contrast CT (NCCT) and CTA images. After feature selection, ten machine learning classifiers were employed to build prediction models. Combined models were developed based on two strategies including early feature-level fusion and late decision-level fusion. Model performance was evaluated using ROC analysis, DeLong test, NRI and IDI. Subgroup analyses were also performed. Results: A total of 191 patients (40.6%) achieved a favorable outcome (90-day mRS ≤ 2). In the test cohort, the late-fusion combined model achieved the highest discriminative performance (AUC = 0.836, 95% CI: 0.769-0.891). It significantly outperformed both the peri-thrombus model (AUC = 0.758, p = 0.005) and the early-fusion combined model (AUC = 0.763, p = 0.009). Compared with the intra-thrombus model, the late-fusion combined model also demonstrated improved reclassification, with an NRI of 0.935 and an IDI of 0.278. Subgroup analyses showed consistent discriminative performance of the late-fusion combined model across the prespecified subgroups. Conclusion: The late-fusion combined model, integrating CT-derived radiomic and DLFs from both intra-thrombus and peri-thrombus regions, may provide an effective approach for predicting 90-day functional outcome after mechanical thrombectomy.

Background: In-hospital cardiac arrest (IHCA) represents a devastating complication among critically ill patients with traumatic spinal cord injury (TSCI), driven by profound autonomic dysfunction and systemic physiological instability. Despite the high mortality associated with these events, early recognition remains challenging, and dedicated risk stratification tools tailored to this vulnerable population are lacking. Methods: Patients from the US-based MIMIC-IV 3.1 and eICU-CRD 2.0 databases were randomly divided 7:3 into training and internal validation sets. Predictors recorded within 24 h of ICU admission were selected exclusively in the training set using univariable screening, Boruta, and group lasso. Eleven base learners were optimized by fivefold cross-validation and integrated into an L2-penalized stacking ensemble using out-of-fold probabilities. After internal validation, the best-performing model was locked and evaluated unchanged in an independent six-center Chinese cohort to assess transportability across populations and clinical settings differing in ethnic composition, geography, and practice patterns. Performance was assessed in terms of discrimination (ROC-AUC and PR-AP), calibration (calibration-in-the-large, calibration slope, Brier score, and expected calibration error), classification metrics, and clinical utility through decision-curve and lift analyses, while model predictions were interpreted using SHAP values. Results: The development cohort included 996 patients with 57 IHCA events, and the external cohort included 558 patients with 51 events. The stacking model with ten predictors retained good discrimination across the training, internal validation, and external validation cohorts, with ROC-AUCs of 0.953, 0.938, and 0.894 and corresponding PR-APs of 0.848, 0.794, and 0.774, respectively. Calibration was acceptable across all cohorts, with Brier scores of 0.013-0.020 and expected calibration errors of 0.020-0.033, and the 95% CIs for the external calibration-in-the-large (0.541) and calibration slope (1.318) included their respective ideal values, supporting the decision not to undertake post hoc recalibration. Decision-curve and lift analyses indicated favorable clinical utility, while SHAP analysis identified integrated patterns of autonomic dysfunction, perfusion compromise, metabolic disturbance, and overall physiological burden as the main contributors to prediction. Conclusion: The stacking model maintained discrimination and calibration across US and Chinese cohorts and may support interpretable early IHCA risk stratification and anticipatory management, although prospective evaluation is required before clinical implementation.

Background: Convolutional neural networks can classify thyroid nodules on ultrasound, yet published models are seldom available for independent testing, require machine learning expertise to develop and deploy, and are validated mostly on papillary thyroid carcinoma. Objective: To test whether an autonomous ("agentic"), no-code artificial intelligence (AI) agent can develop a calibrated thyroid nodule malignancy classifier and validate it internally and on an external cohort spanning multiple cancer histologies. Methods: This was a retrospective, computational diagnostic study. A no-code agent (Hugging Face ML-Intern) autonomously audited the data, selected and trained the model, and calibrated probabilities, using the open-source TN5000 dataset. The locked ResNet-18 model was externally validated on 232 nodules from the University of Colorado and benchmarked against a previously published classifier evaluated on the same source. Results: On the internal test set, the agentic model achieved an area under the receiver-operating-characteristic curve (AUROC) of 0.94 (95% CI, 0.92-0.95), sensitivity 0.90, and specificity 0.81. On external validation, the model achieved an AUROC of 0.90 (95% CI, 0.85-0.93); sensitivity was 0.92, specificity 0.68, positive predictive value 0.52, and negative predictive value 0.96, exceeding the performance of the benchmark model (AUROC 0.83; paired DeLong P = .03). Missed cancers clustered among follicular pattern tumors. Conclusion: An agentic, no-code AI workflow produced a calibrated, externally validated thyroid nodule classifier, supporting accessible, reproducible, and independently testable medical AI development. Prospective validation and local recalibration are required before clinical use.

BACKGROUND Hyperglycemic crisis is associated with substantial morbidity and mortality. Existing prognostic tools largely rely on static baseline measurements and do not exploit the longitudinal data generated during acute care. We aimed to develop a real-time dynamic prediction model for in-hospital mortality within the next 24 h in adults with hyperglycemic crisis. METHODS We performed a multicenter retrospective study using eICU-CRD and MIMIC-IV databases. We developed HCNet, a temporal deep learning architecture that integrates longitudinal trajectories with static features to generate a continuously updated 24-hour mortality risk. The eICU-CRD was used for development and internal testing; MIMIC-IV served as an external test cohort. Discrimination, calibration, and decision curve analysis were compared with those of eight machine learning models, including logistic regression (LR), random forest (RF), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), light gradient boosting machine (LightGBM), multilayer perceptron (MLP), tabular data network (TabNet), and long short-term memory (LSTM). RESULTS The study included 3,436 eICU-CRD training admissions, 1,473 eICU-CRD internal test admissions, and 1,182 MIMIC-IV external test admissions. In-hospital mortality was 1.8%, 1.9%, and 3.1%, respectively. HCNet achieved an AUC of 0.941 (95% CI: 0.939 to 0.943) in internal testing and 0.913 (95% CI: 0.911 to 0.916) in external testing, outperforming all comparators (P<0.05). The strongest baseline models according to the AUC were CatBoost (internal 0.916; external 0.871) and LSTM (internal 0.911; external 0.890). Temporal analyses showed improved discrimination with increasing length of stay and as death approached, as well as a reduced false alarm burden. CONCLUSION HCNet provides real-time 24-hour mortality risk estimates for adults with hyperglycemic crisis using routinely collected data. Its generalizable performance and interpretable temporal patterns support its potential as a clinical decision support tool for early warning and escalation of care during hyperglycemic crisis.

Background: Cardiotocography is important for continuous intrapartum fetal surveillance, yet its clinical effectiveness is limited by high inter-observer variability and a false-positive burden of approximately 60%. These limitations contribute to unnecessary intervention while delaying recognition of fetal acidemia. For clinical use, decision-support systems must not only achieve high accuracy but also define their role in guiding escalation of care. Methods: We developed and evaluated CTG-FRAME (CardioTocoGraphy-Fetal Risk Assessment and Monitoring Engine), an end-to-end deep learning decision-support framework for identifying risk of severe fetal acidemia (umbilical arterial pH < 7.05) using fetal heart rate and uterine contraction signals. The framework integrates self-supervised momentum-contrast pre-training, a hybrid ResNet-temporal convolutional encoder, bidirectional cross-attention, and a Transformer to capture global temporal context. The model was trained and internally validated on the CTU-UHB dataset (n = 552; ~8% acidemia) using three-fold cross-validation. External transferability and calibration were assessed in a separate multicentre cohort (SPaM; n = 300; ~20% acidemia) not used during model development. CTG-FRAME outputs a probability of acidemia, estimates continuous pH severity, and identifies signal segments contributing to each prediction. Training employed a difficulty-aware multi-stage curriculum to address severe class imbalance. Model performance was evaluated using AUROC, sensitivity, specificity, and agreement measures. Results: On CTU-UHB, the model achieved an AUROC of 0.888, with sensitivity 0.850 and specificity 0.964 at the predefined operating threshold (≈0.52). External evaluation of SPaM showed predicted abnormality rates consistent with cohort prevalence without re-tuning, indicating calibration preservation across sites. Ablation analysis demonstrated that the staged curriculum was necessary to prevent collapse to majority-class prediction, and that the self-supervised pre-training substantially improved pH estimation accuracy. Conclusions: CTG-FRAME is designed to facilitate the escalation of clinical review rather than autonomous clinical action by defining a clear decision-analytic operating region and generating interpretable, context-linked evidence. This work advances obstetric artificial intelligence by making clinical intent, operating trade-offs, and workflow roles explicit, moving beyond accuracy toward verifiable bedside utility.

Automated electrocardiogram (ECG) interpretation is increasingly delivered as cloud deep-learning services, trading data sovereignty, auditability and robustness for accuracy. This study tests whether an offline, explainable framework can remove that trade-off.&#xD;&#xD;Approach. Three independent interpreters feed a decision-fusion layer: a deterministic engine of 26 executable clinical criteria, each finding linked to a highlighted signal segment and a literature source; a compact one-dimensional residual network trained on PTB-XL and run on-device via the Open Neural Network Exchange runtime; and an optional agentic vision-language model reading a rendered ECG image behind a security screen with human-in-the-loop escalation. Format-agnostic ingestion, lead-aware capability tiering and a three-tier signal-quality gate precede interpretation. Evaluation covers diagnostic accuracy, on-device latency, noise resilience, modality ablation and zero-shot external validation.&#xD;&#xD;Main results. On the PTB-XL test fold (n = 2198) the network reaches a macro-averaged area under the receiver operating characteristic curve (AUROC) of 0.932 (95% confidence interval (CI) 0.918-0.943), matching published GPU baselines, while the complete pipeline runs in 108 ± 25 ms on a CPU with no network access. Performance degrades gracefully to a signal-to-noise ratio of 5 dB (AUROC 0.951) as the quality gate rejects uninterpretable signals, and rules and network agree only weakly (Cohen's kappa 0.26), indicating complementary errors. Applied without retraining to an independent cohort (Chapman-Shaoxing, n = 1599) the model retains a macro-AUROC of 0.898. In a small pilot (25 records; 30 rule/network disagreements) the general-purpose agentic model was the weakest adjudicator (accuracy 0.267, 95% CI 0.142-0.444; exact McNemar test versus rules, p = 0.09).&#xD;&#xD;Significance. Edge inference, intrinsic explainability and governed multi-modal fusion give competitive accuracy with auditable findings. The agentic layer is positioned for governance rather than diagnosis, motivating ECG-specialized models. The open-source system is a research prototype, not a medical device.

Background: Prolonged length of stay (PLOS) is common among older patients with lung cancer and is associated with increased risks of adverse outcomes, greater healthcare resource utilization, and reduced treatment efficiency. However, reliable risk stratification tools specifically designed for this population, particularly in the Chinese healthcare context, remain lacking. Methods: This multicenter retrospective study included patients aged 65 years or older with primary lung cancer from two Chinese hospitals. PLOS was defined as a length of stay exceeding the 75th percentile of the overall cohort. After least absolute shrinkage and selection operator‑based feature selection from 34 candidate variables, six machine learning models were developed in the training set and validated using both internal and external validation sets. Model performance was assessed using discrimination, calibration, and clinical utility metrics. The optimal model was interpreted using SHapley Additive exPlanations (SHAP). Results: A total of 5,836 patients were included. Among the six models, XGBoost achieved the highest discriminative performance, with area under the receiver operating characteristic curve values of 0.821 (95% CI: 0.806-0.837) in the training set, 0.773 in internal validation, and 0.727 in external validation. Calibration curves and decision curve analysis demonstrated satisfactory agreement and net clinical benefit. SHAP analysis identified pneumonia, hypoproteinemia, C‑reactive protein (CRP), uric acid (UA), and admission pathway as the top five predictors. Pneumonia, hypoproteinemia, CRP, admission pathway, and systolic blood pressure were positively associated with PLOS, whereas UA and total protein showed negative associations. Conclusion: The XGBoost model provided robust and generalizable predictions of PLOS in older Chinese patients with lung cancer. Its interpretability may facilitate early identification of high‑risk individuals and support targeted perioperative interventions in geriatric oncology care.