共找到 20 条结果
This study aimed to evaluate the association between LAA metabolic parameters-particularly lactic acid, glucose, and calcium-and spontaneous echo contrast, and to develop and externally validate a multivariable prediction model incorporating these indicators. Consecutive patients with AF undergoing radiofrequency catheter ablation and/or left atrial appendage occlusion were retrospectively enrolled. All patients underwent preprocedural transesophageal echocardiography with direct LAA blood sampling for metabolic analysis. An internal cohort was used for feature selection by LASSO regression and multivariable logistic regression. Model performance was assessed using ROC analysis, calibration, and decision curve analysis, with external validation in an independent cohort. A total of 272 patients were included in the internal cohort, among whom 96 (35.3%) had SEC. Patients with SEC showed higher LAA lactic acid levels and lower LAA glucose and calcium levels. Age, persistent AF, LAA blood flow velocity, LAA lactic acid, LAA glucose, and LAA calcium were independently associated with SEC. The resulting nomogram demonstrated excellent discrimination in the internal cohort (AUC 0.895) and maintained robust performance in the external cohort (AUC 0.947). Decision curve analysis indicated a positive net clinical benefit across a wide range of threshold probabilities. LAA metabolic characteristics, particularly elevated lactic acid levels, are independently associated with SEC in AF. A prediction model integrating metabolic, clinical, and echocardiographic parameters provides robust and externally validated risk stratification for SEC.
Deep learning (DL) chest radiograph (CXR) models are often trained on downsampled images to reduce computational overhead, despite clinical workflows operating at high resolution. Previous studies have investigated the impact of input resolution on CXR classification accuracy, yet two fundamental pillars of safe and trustworthy AI, explainability and generalizability, remain underexplored. In this retrospective study, we evaluated how training image resolution affects CXR classification performance and explanation quality in internal versus external testing. We trained Convolutional Neural Networks (CNN) for disease classification on the SIIM-ACR Pneumothorax and RSNA Pneumonia datasets at six resolutions (ranging from 64×64 to 1024×1024) using five-fold cross-validation and evaluated models on internal and external test sets. Internal performance was high across resolutions (AUROC >0.85), but external testing showed substantially worse generalizability at lower training resolutions, with internal-to-external drops >20% versus 4.2%-10.7% at higher resolutions (512×512 to 1024×1024). Higher resolutions also produced more concise explanations, with the tightest saliency-map coverage at 1024×1024 (<4%) across models and datasets, and improved explanation quality on external data (peak precision plateauing at 768×768 for pneumothorax). Overall, training at higher CXR resolutions improved both generalizability and explainability, providing practical guidance for radiology AI design beyond internal test performance.
Hypothalamic hamartoma (HH) causes drug-resistant epilepsy, with gelastic seizures (GS) progressing to non-gelastic seizures (nGS). MRI-guided laser interstitial thermal therapy (MRgLITT) is first-line treatment, but robust prognostic models with external validation are lacking. To evaluate MRgLITT outcomes for HH-related epilepsy and develop externally validated predictive nomograms for GS and nGS recurrence. This retrospective study included 169 patients (training: n = 121; validation: n = 48). Multivariate Cox regression identified prognostic factors. Nomograms were developed in the training cohort and externally validated in the independent cohort. Seizure freedom rates were 76.4% (GS) and 81.8% (nGS). GS recurrence correlated with larger HH volume (HR = 4.32), incomplete ablation (HR = 0.268/0.212), and cognitive impairment (HR = 2.51). nGS recurrence associated with bilateral PET hypometabolism (HR = 60.53), nGS duration (HR = 1.14), automatisms (HR = 7.90), and sGTCS (HR = 3.84). External validation confirmed GS and nGS nomogram generalizabilities (C-index = 0.731, 0.844). Complications included transient hyperphagia (26.7%), persistent obesity (12%), and permanent hypothyroidism (1.8%). Caregiver-reported cognitive outcomes were exploratory. MRgLITT is safe and effective for HH-related epilepsy, with lesion-centric (GS) versus network-level (nGS) prognostic pathways. Externally validated nomograms offer actionable tools for clinical decision-making.
Emergency care in the emergency department (ED) requires continuous, multifaceted decision-making based on evolving clinical information. This study aimed to develop and externally validate a comprehensive ED decision-support model based on a Temporal Fusion Transformer (TFT) with multi-modal time-series data to jointly predict testing, treatment, diagnosis, and disposition. Adult ED visit data from one hospital were used for model development, and data from another hospital for external validation. The static inputs included patient characteristics, ED visit-related information, and triage notes, while the time-varying inputs included vital signs, laboratory results, management, and clinical notes. The primary outcomes were the areas under the receiver operating characteristic curves (AUCs) for predicting computed tomography, magnetic resonance imaging, echocardiography, gastrointestinal endoscopy, mechanical ventilation (MV), antibiotic administration, oxygen therapy, vasopressor use, transfusion, primary diagnosis, and ED disposition. A single TFT model was trained to predict all targets jointly, and separate Random Forest (RF) models were developed for comparison. The development and external validation datasets included 272,058 and 138,343 patients, respectively (median age, 61-62 years; females, 51.3%-51.7%). Across internal and external validation, the TFT demonstrated strong discrimination for tests (AUC 0.877-0.961), treatments (AUC 0.912-0.990), and disposition outcomes (AUC 0.811-0.905). The top-5 accuracy for diagnosis prediction was 73.7% and 65.4% in the internal and external validation, respectively. The TFT outperformed RF models for most targets and showed comparable performance in predicting MV, oxygen therapy, and vasopressor use. The TFT model achieved high accuracy across multiple ED decisions, demonstrating its potential as a comprehensive and temporally aware decision-support tool throughout the ED trajectory.
This study aimed to evaluate the accuracy of an ultrasound imaging-based deep transfer learning model for predicting human epidermal growth factor receptor 2 (HER2)-positive breast cancer and to explore its potential advantages and clinical applications. This retrospective study included ultrasound images from 492 patients with breast cancer who were treated at Gongli Hospital (Shanghai, China) between December 2023 and January 2025. The dataset was randomly divided into a training dataset (n = 343), a validation dataset (n = 73), and an internal test dataset (n = 76) in a ratio of 7:1.5:1.5 for model development, parameter optimization, and internal evaluation. In addition, an independent external cohort consisting of 72 patients from another hospital (January 2025 to June 2025) was used as an external test dataset to evaluate the generalizability of the model. Based on a transfer learning framework, classification models were developed using six convolutional neural network backbone architectures, involving GoogLeNet, ResNet-18, ResNet-50, DenseNet-161, MobileNetV2, and EfficientNet-B0. Model performance was comprehensively evaluated using sensitivity, specificity, accuracy, positive predictive value, negative predictive value, the area under the curve (AUC), calibration curves, and decision curve analysis (DCA). The model demonstrating the best overall performance in the internal test dataset was subsequently validated using the external test cohort to further assess its generalization capability. In the internal test dataset, the AUC values of the six models ranged from 0.913 to 0.989, with DenseNet-161 (AUC = 0.989) and EfficientNet-B0 (AUC = 0.987) demonstrating the strongest discriminative capacity. Among these architectures, EfficientNet-B0 achieved the most favorable overall performance, characterized by superior accuracy and specificity. When evaluated on the external test dataset, EfficientNet-B0 maintained excellent predictive performance, yielding an AUC of 0.972. DCA further indicated that this model provided a greater net clinical benefit across a wide range of threshold probabilities, while calibration analysis demonstrated good agreement between predicted probabilities and observed outcomes. In conclusion, ultrasound-based deep transfer learning model demonstrated remarkable potential for the noninvasive prediction of HER2-positive breast cancer. Among the evaluated architectures, EfficientNet-B0 exhibited the most robust overall performance and strong generalizability, indicating noticeable promise for future clinical translation in supporting precision diagnosis and individualized management of breast cancer. As a non-invasive preoperative auxiliary screening tool, this tool promotes the preliminary identification of high-risk, HER2-positive patient cohorts prior to tissue biopsy or surgical intervention, thereby establishing a substantive radiological basis to guide clinical decision-making.
The prognostic value of postoperative chemotherapy for non-metastatic right-sided colon neuroendocrine carcinoma (RCNEC) remains unclear. This study aimed to evaluate its association with survival outcomes and to develop a prognostic nomogram for surgically treated RCNEC patients. Surgically treated non-metastatic high-grade RCNEC patients were identified from the Surveillance, Epidemiology, and End Results (SEER) database and were classified into chemotherapy and non-chemotherapy groups according to SEER-recorded chemotherapy status. Propensity score matching (PSM) was performed to balance baseline characteristics between groups. An age-exact matched sensitivity analysis was additionally reported in the supplementary materials. Kaplan-Meier survival analysis and Fine-Gray competing risk models were used to assess the association between chemotherapy and survival outcomes. A prognostic nomogram based on independent predictors was subsequently constructed to estimate individualized CSD probabilities and was evaluated using internal and external validation cohorts. A total of 216 surgically treated non-metastatic RCNEC patients were identified from SEER, including 111 in the non-chemotherapy group and 105 in the chemotherapy group. After 1:1 PSM, neither overall survival nor cancer-specific survival differed significantly between groups (P = 0.598 and P = 0.422, respectively). In competing-risk analyses, chemotherapy was not associated with a lower cumulative incidence of cancer-specific death (CSD) after matching (P = 0.219), whereas other-cause death remained significantly lower in the chemotherapy group (P = 0.002), a finding likely reflecting treatment selection rather than a protective effect of chemotherapy. The age-exact matched sensitivity analysis yielded consistent findings. In multivariable Fine-Gray analysis, chemotherapy was not independently associated with CSD (sHR = 1.21, P = 0.440). The competing risk-based nomogram showed apparent discrimination, with 1-, 3-, and 5-year AUCs of 0.785, 0.819, and 0.755 in the training cohort, 0.895, 0.856, and 0.864 in the internal validation cohort, and 0.795, 0.891, and 0.897 in the external cohort; however, these estimates should be considered preliminary because the training cohort was modest and the external cohort included only 22 patients. In this SEER-based cohort of surgically treated non-metastatic RCNEC, chemotherapy as recorded in SEER was not independently associated with lower CSD in competing-risk analyses; confirmation in clinically annotated datasets is warranted. Given the limitations of registry-coded chemotherapy data, the absence of association should not be interpreted as lack of efficacy of chemotherapy.
The purpose of this study is to develop and validate a multimodal, multitask prediction framework for clear cell renal cell carcinoma (ccRCC) by integrating preoperative CT radiomics, pathology-derived biomarker data from preoperative biopsy specimens, and clinical variables. The model was built for pathologically confirmed ccRCC and excluded other RCC histologic subtypes (e.g., papillary and chromophobe). In this multicenter retrospective study, ccRCC patients were enrolled and data were collected from preoperative CT scans, AMACR/P504S immunohistochemistry results, pathology reports, and follow-up records. Patients were assigned by hospital site into a training cohort and an independent external test cohort. Radiomic features were extracted from CT tumor regions of interest (ROIs), while deep learning features were derived from pathology images. Clinical variables were incorporated as additional inputs. A post-feature fusion strategy enabled simultaneous prediction of tumor classification and postoperative survival. Model performance was assessed using AUC for diagnosis and C-index for survival, together with calibration curves, decision curve analysis, and bootstrap confidence intervals. SHAP analysis was applied to quantify feature contributions. In external validation, the integrated multimodal model achieved strong diagnostic discrimination for P504S/AMACR (AUC = 0.983) and demonstrated improved prognostic performance for postoperative outcomes (C-index = 0.804). Subgroup analyses by grade and stage further supported model robustness, while SHAP-based interpretation indicated complementary contributions from imaging, pathology, and clinical variables. Overall, the proposed multimodal, multitask fusion framework enables reliable preoperative P504S/AMACR-based classification and postoperative prognostic prediction in ccRCC, supporting more refined risk stratification and individualized postoperative management using noninvasive imaging and clinical information. Calibration and decision curve analyses further support its potential clinical utility. Larger prospective and external multicenter validations are still needed to confirm generalizability.
Patients hospitalized for acute exacerbation of chronic obstructive pulmonary disease (AECOPD) remain at risk of readmission and death after discharge. Opportunistic chest computed tomography (CT) body-composition metrics may provide additional prognostic information beyond conventional clinical scores. To develop and validate a 90-day adverse-outcome prediction model for hospitalized AECOPD using admission clinical variables and opportunistic chest CT body-composition metrics. A retrospective modelling cohort of 203 AECOPD admissions from the index centre was analysed. Admissions from 2021 to 2024 formed the development cohort (n = 152), and 2025 admissions formed the temporal-validation cohort (n = 51). An external-validation cohort from another centre included 103 admissions from records screened between 1 January 1 and 1 January 2025 after the model was locked. The primary outcome was 90-day readmission or death. LASSO was used for variable screening in the development cohort. A feature-count AUC plateau analysis and prespecified multialgorithm screening were used to lock the final model before validation. DECAF and BAP-65 were retained as comparator scores. The 90-day adverse outcome occurred in 66 of 152 development patients (43.4%), 18 of 51 temporal-validation patients (35.3%) and 35 of 103 external-validation patients (34.0%). LASSO retained prior AECOPD admissions, home oxygen before admission, diabetes mellitus, intermuscular adipose tissue area, long-term NIV before admission, heart rate and coronary artery disease. Feature-count analysis supported this seven-predictor set, and multialgorithm screening selected HistGradientBoosting for validation. In temporal validation, the locked model achieved AUC 0.80 (0.64-0.95), sensitivity 0.78 (0.59-0.94), specificity 0.88 (0.76-0.97) and Brier score 0.15 (0.09-0.21), with imperfect calibration (Hosmer-Lemeshow p < 0.001). In external validation, the locked model achieved AUC 0.77 (0.66-0.87), sensitivity 0.66 (0.49-0.80), specificity 0.79 (0.71-0.88) and Brier score 0.18 (0.15-0.21); the Hosmer-Lemeshow p value was 0.209. A 90-day AECOPD prediction model combining clinical and opportunistic CT body-composition variables showed consistent discrimination across validation cohorts, but calibration remained a key implementation boundary. Formal multicentre validation and calibration updating are needed before routine clinical use.
Cardiac arrest (CA) is a critical clinical event associated with extremely high mortality and long-term disability. This study aimed to identify risk factors associated with ICU mortality and develop a practical nomogram for risk prediction to support prognostic assessment and risk stratification among patients with CA. A retrospective analysis was conducted using two large public critical care databases. The MIMIC-IV database was used as the training cohort for model development, while the eICU-CRD database served as the external validation cohort. Variables were initially screened using least absolute shrinkage and selection operator (LASSO) regression, and a predictive nomogram was subsequently established using binary logistic regression analysis. The discriminative performance of the model was evaluated using the receiver operating characteristic (ROC) curve. Calibration was assessed using calibration curves and the Hosmer-Lemeshow goodness-of-fit test. Clinical applicability and net benefit were further analyzed using decision curve analysis (DCA). A total of 3,029 patients from the MIMIC-IV database and 2,805 patients from the eICU-CRD database were enrolled in this study. After variable selection by LASSO regression, eight predictors were included in the nomogram. The model was compared with conventional ICU scoring systems, including SAPS II, SOFA, and CCI. In the training cohort, the AUROC values were 0.802 for the nomogram, 0.725 for SAPS II, 0.709 for SOFA, and 0.509 for CCI. In the external validation cohort, the corresponding AUROC values were 0.733, 0.596, 0.701, and 0.517, respectively. The Hosmer-Lemeshow test yielded P-values of 0.976 in the training set and 0.905 in the validation set, indicating favorable goodness-of-fit and satisfactory calibration. Moreover, DCA demonstrated that the nomogram provided greater net benefit compared with traditional scoring systems. The established nomogram for predicting ICU mortality in patients with CA demonstrated acceptable predictive performance and calibration. As a quantitative risk stratification tool, it may provide clinicians with useful individualized prognostic information for this vulnerable patient population.
Extra-articular bony impingement may contribute to posture-specific groin pain after total hip arthroplasty (THA), but the relative roles of bony morphology and functional pelvic tilt remain unclear. CT reconstructions from 100 cadavers (200 hips) were used to build three-dimensional bone-on-bone collision detection models to quantify impingement-free range of motion (RoM). Femora were rotated in internal/external rotation and abduction/adduction from 30° hyperextension to 120° flexion in 5° flexion increments. Simulations were repeated with pelvic tilt ranging ±30° from neutral as extreme boundaries. The contribution of bony anatomy to early impingement was evaluated using Pearson correlations and two-stage regression, with nested cross-validation used for supplementary internal validation. Mean external rotation to extra-articular bony impingement in extension was 46.6° ± 14.8°, and mean internal rotation to impingement at 90° flexion was 34.4° ± 14.1°. In standing, 30° posterior pelvic tilt reduced external rotation clearance by 13.2° ± 10.2° (p < 0.001); while sitting, 30° anterior pelvic tilt reduced internal rotation clearance by 33.0° ± 12.1° (p < 0.001). Standing impingement was most frequently between the ischial tuberosity and posterior intertrochanteric crest or lesser trochanter, whereas seated impingement predominantly involved the anterior inferior iliac spine (AIIS) and intertrochanteric crest. Multivariable models showed strong in-sample fit and retained good held-out performance on nested cross-validation. High-risk morphologies combined with adverse pelvic tilt produced markedly earlier impingement, with the greatest reduction observed in the seated high-risk subgroup under extreme anterior tilt. These findings quantify posture-dependent extra-articular bony constraints and identify morphology-based phenotypes associated with reduced clearance, providing a biomechanical framework for future patient-specific and implant-based studies after THA.
Head and neck squamous cell carcinoma (HNSCC) is frequently diagnosed at advanced stages, when outcomes are poor. Biomarkers for HNSCC detection remain needed. We performed serum metabolomics profiling in 938 participants, including 457 patients with HNSCC, 451 controls, and 30 patients with oral potentially malignant disorders (OPMDs). Discovery and external validation cohorts were analysed using LC-MS/MS. Machine learning-based feature selection was used to develop an 8-metabolite diagnostic model. Disease specificity was assessed in an independent third-centre cohort. Multi-omics analyses integrating TCGA transcriptomics and spatial transcriptomics data were used to explore molecular heterogeneity. ChiCTR-BRC-17014040. Serum metabolomics profiling showed altered arginine metabolism, characterised by arginine decrease and ornithine accumulation. The 8-metabolite model achieved an internal validation AUC of 0.990, with 91.3% sensitivity and 97.1% specificity. In the multicentre external validation cohort, the model achieved an AUC of 0.901, with 91.1% sensitivity and 80.2% specificity, and maintained performance in Stage I disease, with an AUC of 0.927. In the independent disease-control cohort, the model distinguished HNSCC from healthy controls with an AUC of 0.980 and showed moderate discrimination between HNSCC and OPMDs, with an AUC of 0.710. Multi-omics analyses identified two candidate molecular subtypes with distinct immune features, prognosis, TP53 mutation status, and predicted treatment sensitivities. Spatial transcriptomics supported an inverse association between metabolic score and immune infiltration. This multicentre study identifies a serum metabolomics signature associated with HNSCC detection and molecular stratification. The findings support further prospective validation in clinically representative populations. National Natural Science Foundation of China; Shanghai Committee of Science and Technology; Innovative Research Team of High-level Local Universities in Shanghai; Shanghai Jiao Tong University Trans-med Awards Research; The Explorers Program of Shanghai; National Major Science and Technology Infrastructure for Translational Medicine, Shanghai; Beijing Municipal Natural Science Foundation; Beijing High-Level Innovation and Entrepreneurship Talent Support Program Young Top Talent Projects.
To investigate the epidemiological and forensic characteristics of rotator cuff injuries caused by road traffic accidents, and to explore the timing and influencing factors of disability assessment of rotator cuff injuries. A retrospective analysis was conducted on 173 cases of rotator cuff injury caused by road traffic accidents. Factors analyzed included the time of post-injury assessment, the degree of functional loss of shoulder joint movement in six directions, the treatment methods, the presence or absence of combined shoulder injuries and the type of combined shoulder injuries, and the type of rotator cuff injury. The Kruskal-Wallis test and Wilcoxon rank sum test were used to analyze the differences in functional loss among the six directions of shoulder joint movement. The t-test was used to compare the differences in functional loss between the conservative treatment and the surgical treatment, as well as between cases with and without combined shoulder injuries. One-way analysis of variance (ANOVA) and Student-Newman-Keuls (SNK) test were used to analyze the differences in functional loss of shoulder joint movement among the injury types of contusion, mild tear, moderate tear and severe tear. One-way ANOVA and least significant difference (LSD) test were used to analyze the differences according to the number of involved rotator cuff tendons (1, 2, 3, or 4 tendons) and post-injury assessment time (3-<6 months, 6-<9 months, 9-<12 months, and ≥12 months). The median assessment time after injury was 290.00 days. Among the six directions of shoulder joint movement, statistically significant differences in the degree of functional loss were observed between abduction up, extension back and external rotation compared with adduction, as well as between external rotation and forward flexion up (P<0.05). There was no statistically significant difference in the degree of functional loss between conservative treatment and surgical treatment, or among different post-injury assessment times (P>0.05). However, there were significant differences between the non-combined shoulder injury and combined shoulder injury, different types of rotator cuff injuries, and the different numbers of involved rotator cuff tendons (P<0.05). Treatment methods had no significant effect on the degree of functional loss of shoulder joint movement, whereas combined shoulder injuries, more severe rotator cuff injuries, and involvement of a greater number of rotator cuff tendons were associated with more severe functional loss of shoulder joint movement. Rotator cuff injuries caused by road traffic accidents exhibit the epidemiological and forensic characteristics including delayed diagnosis after injury, prolonged post-injury assessment time, and the rotator cuff injuries being mainly mild to moderate tears (≤3 cm) and involving 1 or 2 tendons. Combined shoulder injuries are mostly characterized by greater tuberosity fractures of the humerus, long head of the biceps tendon injuries and shoulder dislocations. The recommended timing for disability assessment in cases involving rotator cuff tears and/or surgical treatment should be appropriately extended. The presence or absence of combined shoulder injuries, the types of rotator cuff injury, and the number of involved rotator cuff tendons should be emphasized in forensic clinical identification. 目的: 分析道路交通事故致肩袖损伤的流行病学及法医学特征，探索肩袖损伤的残疾等级评定时机和影响因素。方法: 从伤后鉴定时间、肩关节6个方向活动功能丧失程度、治疗方式、有无肩部复合伤及肩部复合伤类型、肩袖损伤类型等方面，对173例道路交通事故致肩袖损伤的案件进行回顾性分析。采用Kruskal-Wallis检验和Wilcoxon秩和检验分析肩关节6个方向活动功能丧失程度的差异性;采用t检验分析保守治疗和手术治疗、无肩部复合伤和肩部复合伤的肩关节活动功能丧失程度的差异性;采用单因素方差分析和Student-Newman-Keuls（SNK）检验对损伤类型（挫伤、轻度撕裂、中度撕裂、重度撕裂）的肩关节活动功能丧失程度进行差异性分析;采用单因素方差分析和最小显著性差异（least significant difference，LSD）检验分别对肩袖损伤累及肌腱根数（1、2、3、4根）和伤后鉴定时间（3~<6个月、6~<9个月、9~<12个月、≥12个月）肩关节活动功能丧失程度进行差异性分析。结果: 伤后鉴定时间的中位数为290.00 d。肩关节6个方向中，外展上举、后伸、外旋较内收方向，外旋较前屈上举方向的活动功能丧失程度差异具有统计学意义（P<0.05）。肩关节活动功能丧失程度在保守治疗和手术治疗、伤后不同鉴定时间之间的差异无统计学意义（P>0.05），无肩部复合伤和肩部复合伤、肩袖损伤不同类型之间、肩袖损伤累及不同肌腱根数之间的差异均有统计学意义（P<0.05）。治疗方式对肩关节活动功能丧失程度无影响，肩部复合伤、肩袖损伤重、累及肩袖肌腱根数多可加重肩关节活动功能丧失程度。结论: 道路交通事故致肩袖损伤具有伤后确诊时间滞后，伤后鉴定时间长，肩袖损伤以轻、中度撕裂（≤3 cm）和累及肌腱以1、2根为主，肩部复合伤以肱骨大结节骨折、肱二头肌长头肌腱损伤和肩关节脱位居多等流行病学和法医学损伤特点，涉及肩袖撕裂和（或）手术治疗案件的残疾等级评定时机建议适当延长，有无肩部复合伤、肩袖损伤类型、肩袖损伤累及肌腱根数可作为法医临床鉴定时重点考量的方面。.
Neoadjuvant chemotherapy (NAC) is widely used in the management of breast cancer, as it can downstage tumors and increase the rate of breast-conserving surgery (BCS). Accurate preoperative prediction of BCS eligibility after NAC is essential for optimizing treatment planning and surgical decision-making. However, reliable noninvasive tools for evaluating BCS eligibility remain limited. This retrospective multicenter study included 315 patients with pathologically confirmed breast cancer who underwent NAC between January 2018 and December 2024. Patients from Center 1 (n = 227) were randomly divided into a training cohort (n = 181) and an internal validation cohort (n = 46), while data from Center 2 (n = 88) served as an external validation cohort. Habitat radiomics features were extracted from dynamic contrast-enhanced MRI (DCE-MRI) to construct a habitat model. Clinical and radiological variables associated with BCS feasibility were identified using univariate and multivariate logistic regression to establish a clinical-radiological model. A combined model integrating radiomic and clinical predictors was subsequently developed. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC). The overall BCS rate was 73.65% (232/315). The habitat model achieved AUCs of 0.886, 0.778, and 0.725 in the training, internal validation, and external validation cohorts, respectively. The clinical-radiological model yielded AUCs of 0.742, 0.733, and 0.707. The combined model demonstrated improved performance with AUCs of 0.910, 0.839, and 0.755 across the 3 cohorts. The combined model integrating habitat radiomics and clinical-radiological variables demonstrated favorable performance for predicting BCS feasibility after NAC. This noninvasive approach may assist clinicians in preoperative surgical planning and facilitate individualized breast-conserving treatment strategies.
Postoperative delirium is a common and serious complication after general anesthesia; its accurate prediction remains a substantial challenge in perioperative medicine. Existing models primarily rely on clinical variables and may have limited predictive accuracy. This study aimed to evaluate the added value of heart rate variability parameters in predicting postoperative delirium and construct an interpretable multimodal predictive model. In this prospective observational study, 1418 patients undergoing general anesthesia were included. Seventy-three features, including electrocardiogram abnormalities and heart rate variability time-, frequency-, and nonlinear-domain indicators, were extracted from electrocardiogram data. Postoperative delirium was assessed using the Chinese version of the 3-Minute Diagnostic Interview for Delirium within 3 days postoperatively. Feature selection was conducted by combining least absolute shrinkage and selection operator (LASSO) regression, the Boruta algorithm, and random forests, and 10 machine learning models were developed. Model performance was evaluated through receiver operating characteristic curves and decision curve analysis, with interpretability assessed via Shapley additive explanations. Clinical prediction tools were derived from key features. We used an external validation set to further evaluate the generalization ability of the models. Postoperative delirium occurred in 255 (18%) patients. Seventeen key predictors were identified in total. The combined clinical-electrocardiogram-heart rate variability model demonstrated the highest predictive performance (area under the curve = 0.728), outperforming clinical-only (area under the curve = 0.673) and electrocardiogram-only models (area under the curve = 0.679). Logistic regression showed the highest discrimination. In the external validation set, the model maintained robust performance with an area under the curve value of 0.836. Shapley additive explanations highlighted seven core predictors: atrial or ventricular arrhythmia, operative time, ST-segment abnormalities, age, American Society of Anesthesiologists classification, heart rate variability entropy, and overall electrocardiogram abnormalities. A nomogram and online platform enabled personalized risk assessment. Our results indicate that integrating heart rate variability with clinical and electrocardiogram features significantly enhances the personalized predictive efficacy of postoperative delirium.
Congenital heart disease (CHD) is the most common congenital anomaly, with lifelong implications as survival into adulthood becomes the norm. Despite advances in prenatal detection, surgical care and long-term follow-up, the aetiology of CHD remains incompletely understood. This narrative review synthesises current evidence across genetic, epigenetic and environmental domains, with particular attention to their interplay.Growing evidence indicates that genetic susceptibility in combination with environmental exposures may shape CHD risk. While chromosomal anomalies and single-gene defects explain a minority of cases, polygenic contributions and emerging evidence on epigenetic programming suggest additional layers of complexity in CHD aetiology. Maternal health conditions, such as diabetes, autoimmune disease and infections together with medication use, play critical roles in determining fetal cardiac development and contribute to modifiable risk pathways. Lifestyle, reproductive and external environmental factors, including smoking, assisted reproductive technology and air pollution, further underscore the need for proactive counselling and early risk mitigation. Emerging evidence also points to gene-environment interactions as a key mechanism through which genetic susceptibility modulates the impact of external exposures, offering new avenues for precision prevention.Looking ahead, progress will depend on integrating multi-omic data with longitudinal cohorts, multinational registries and interoperable data infrastructures. Such efforts must be coupled with implementation research to translate mechanistic insights into predictive tools and scalable interventions. Framing CHD within this bench-to-bedside-to-policy continuum positions it as both a model for precision prevention and lifelong, multidisciplinary care and a test case for approaches that could be extended to other multifactorial conditions with major population-health impact.
We aimed to develop and validate a practical deep learning model integrating commonly collected clinical data and knee radiographs to predict the need for knee arthroplasty (total or partial) in patients with, or at risk of, knee osteoarthritis, as well as the time to surgery. Data from the Multicentre Osteoarthritis Study (MOST) and the Osteoarthritis Initiative (OAI) were used. The MOST dataset, comprising 3,026 patients, was the primary training and testing cohort, while the OAI dataset provided external validation. The final architecture was based on DenseNet-201, with a head that combined outputs from the radiological analysis with commonly collected clinical data. Model evaluation used the area under the receiver operating characteristic curve (AUROC) and precision-recall curve (AUPRC). The integration of clinical and radiological data significantly improved predictive accuracy. The combined model achieved an AUROC of 0.85 and AUPRC of 0.62, outperforming models using either data source alone. External validation with the OAI dataset yielded an AUROC of 0.79, confirming the model's generalizability. The AUROC and AUPRC for surgical interventions within 40 months was 0.83 and 0.26, respectively, on the validation dataset, demonstrating higher predictive accuracy for earlier surgical needs. This study highlights the potential of deep learning models, which integrate clinical and radiological data, to predict the need for knee arthroplasty. The robust performance and generalizability of the developed model could streamline clinical pathways and predict local demand for surgery during the next three years. This will facilitate resource planning for providers and accurate and timely access to surgical interventions for patients.
Timely detection of circulatory and respiratory instability (CRI) is critical in intensive care units (ICUs), yet existing early warning systems often rely on single-parameter indices that underutilize continuous vital-sign data or on delayed, difficult-to-interpret multimodal clinical data. Leveraging routinely collected high-frequency vital-sign monitoring, we developed an interpretable, expert-augmented early warning system based on 1-second-resolution heart rate, blood pressure, respiratory rate, and oxygen saturation data. Machine‑learning models were trained on 627,958 h of continuous vital‑sign data from 1702 ICU patients at the First Affiliated Hospital of Sun Yat‑sen University and externally validated in the MIMIC‑III cohort. Models incorporating trend-based, reference, and statistical features derived from vital-sign trajectories achieved strong predictive performance (AUROC > 0.8) in both internal and external validation, outperforming conventional single-parameter indices and achieving performance comparable to models incorporating laboratory and demographic variables. Increasing temporal resolution improved predictive accuracy, with trend-based features contributing most strongly to model predictions. To improve clinical interpretability, tree-based models were transformed into physiologically meaningful decision rules and refined through expert-augmented learning, resulting in the Expert-Augmented Early Warning System (EAEWS). EAEWS generated accurate, low-frequency alerts with transparent explanations aligned with bedside monitoring, and may provide a scalable framework for real-time CRI detection in ICUs.
Declining NAPLEX pass rates have raised concerns, prompting pharmacy programs to administer readiness exams that guide student self-preparation. The objective of this research is to examine the association between self-regulated learning behavior clusters and performance on a NAPLEX readiness examination during Advanced Pharmacy Practice Experiences (APPEs). This retrospective observational cohort study surveyed final-year PharmD students enrolled in a required NAPLEX readiness course during APPEs (August 2022-May 2024). Students completed an anonymous survey assessing demographics, external factors, and self-regulated learning behaviors using a modified Self-Regulation Strategy Inventory-Self-Report. NAPLEX Readiness and NAPLEX readiness remediation examination performance were collected. Individual study behaviors and conceptually grouped behavioral clusters were evaluated for associations with readiness exam performance using correlation analyses and exploratory regression models. Forty-eight of 105 eligible students completed the survey. Higher pre-APPE GPA moderately correlated with readiness exam performance. External factors, including work hours, commuting time, caregiving responsibilities, and total reported study time, were not associated with exam success. Several individual self-regulated learning behaviors were associated with higher readiness exam scores including environmental control, planning, persistence, help-seeking, and avoidance of last-minute studying. When evaluated as integrated clusters such as planning, organization, environmental control, persistence, help-seeking, and avoidance of last-minute studying, they demonstrated substantially stronger associations than individual behaviors alone. Clusters of self-regulated learning behaviors show stronger associations with NAPLEX readiness exam performance than isolated study strategies. These findings suggest that academic coaching should emphasize bundled self-regulation strategies rather than isolated techniques to support licensure readiness.
The study aimed to develop and validate a multimodal radiomics model that integrates radiologist-informed feature augmentation leveraging expert-selected suspicious lymph nodes (LNs) based on ESGAR criteria to improve the accuracy of preoperative lymph node metastasis (LNM) prediction in patients with rectal cancer (RC). This retrospective study included 563 eligible patients with RC. From each patient's high-resolution T2-weighted imaging (HRT2WI) and diffusion-weighted imaging (DWI) sequences, we extracted radiomic features from three distinct regions: the primary tumor, the entire mesorectal nodal region, and suspicious mesorectal nodes identified by radiologists. Clinical factors associated with LNM were identified through univariate and multivariate logistic regression analyses to establish a clinical prediction model. Finally, we constructed an integrated predictive model by combining these clinical factors with multimodal radiomic features, followed by a comprehensive comparison and evaluation of the predictive performance across all developed models. The integrated model, incorporating radiomic features derived from DWI sequences of the entire mesorectal nodal region and radiologist-annotated suspicious LNs, along with clinical factors, achieved optimal performance in predicting LNM. It yielded an area under the curve of 0.87 (95% confidence interval [CI]: 0.83-0.90) in the internal validation cohort and 0.83 (95% CI: 0.78-0.89) in the external validation cohort. Our findings show that the multimodal radiomics model integrating radiologists' prior knowledge offers potential for improving preoperative LNM assessment in RC, particularly in internal validation, and may provide supportive information for personalized treatment strategies in clinical practice. However, the incremental benefit of the radiologist-informed component was not consistently demonstrated in external validation, and further multi-center prospective studies are warranted.