Background: Cardiovascular diseases are a significant health burden with the prevalence increasing worldwide. Thus, a highly accurate assessment and prediction of death risk are crucial to meet the clinical demand. This study sought to develop and validate a model to predict in‐hospital mortality among patients with the acute coronary syndrome (ACS) using nonlinear algorithms.Methods: A total of 2414 ACS patients were enrolled in this study. All samples were divided into five groups for cross‐validation. The logistic regression (LR) model and XGboost model were applied to predict in‐hospital mortality. The results of two models were compared between the variable set by the global registry of acute coronary events (GRACE) score and the selected variable set.Results: The in‐hospital mortality rate was 3.5% in the dataset. Model performance on the selected variable set was better than that on GRACE variables: a 3% increase in area under the receiver operating characteristic (ROC) curve (AUC) for LR and 1.3% for XGBoost. The AUC of XGBoost is 0.913 (95% confidence interval [CI]: 0.910–0.916), demonstrating a better discrimination ability than LR (AUC = 0.904, 95% CI: 0.902–0.905) on the selected variable set. Almost perfect calibration was found in XGBoost (slope of predicted to observed events, 1.08; intercept, −0.103; p < .001).Conclusions: XGboost modeling, an advanced machine learning algorithm, identifies new variables and provides high accuracy for the prediction of in‐hospital mortality in ACS patients.

The data generated by the structured electronic medical records is helpful for mining and extracting medical data, and it is an effective way to make effective use of valuable data resources. However, the hospitals have accumulated a large number of unstructured data in electronic medical records, which cannot be effectively searched, resulting in serious waste of resources. In this paper, we study the problem of extracting attribute values from the unstructured text in electronic medical records. By observing intestinal cancer diagnostic texts, our attributes have two categories - discriminative attributes and extractive attributes, which use the text classification and the sequence labeling to tackle attribute values extraction problems. For discriminative attributes, we firstly divide the text into sentences/segments as instances. Secondly, we fine-tune the pre-trained word embedding to capture domain-specific semantics/knowledge. Thirdly, we also use an attention mechanism to select the most important instance for different attribute extractors. Finally, multi-tasking learning is used to share useful information to get better experimental results. For extractive attributes, we propose a novel model to get attribute values, including the BiLSTM layer, the CNN layer and the CRF layer. In particular, we use BiLSTM and CNN to learn text features and CRF as the last layer of the model. Experiments have shown that our method is superior to several competitive baseline methods.

Background: Accurate staging of esophageal cancer is crucial for determining prognosis and guiding treatment strategies, but manual interpretation of radiology reports by clinicians is prone to variability and limited accuracy, resulting in reduced staging accuracy. Recent advances in large language models (LLMs) have shown promise in medical applications, but their utility in esophageal cancer staging remains underexplored. Objective: This study aims to compare the performance of 3 locally deployed LLMs (INF-72B, Qwen2.5-72B, and LLaMA3.1-70B) and clinicians in preoperative esophageal cancer staging using free-text radiology reports. Methods: This retrospective study included 200 patients from Shanghai Chest Hospital who underwent esophageal cancer surgery from May to December 2024. The dataset consisted of 1134 Chinese free-text radiology reports. The reference standard was derived from postoperative pathological staging. A total of 3 LLMs determined tumor classification (T1-T4), node classification (N0-N3), and overall staging (I-IV) using 3 prompting strategies (zero-shot, chain-of-thought, and a proposed interpretable reasoning [IR] method). The McNemar test and Pearson chi-square test were used for comparisons. Results: INF-72B+IR achieved a superior overall staging accuracy of 61.5% and an F₁-score of 0.60, substantially higher than the clinicians’ accuracy of 39.5% and F₁-score of 0.39 (all P<.001). Qwen2.5-72B+IR also demonstrated an advantage, achieving an overall staging accuracy of 46% and an F₁-score of 0.51, which was better than the clinicians’ performance (P<.001). LLaMA3.1-70B showed no statistically significant difference in overall staging performance compared to clinicians (all P>0.5) Conclusions: This study demonstrates that LLMs, particularly when guided by the proposed IR strategy, can accurately and reliably perform esophageal cancer staging from free-text radiology reports. This approach not only provides high-performance predictions but also offers a transparent and verifiable reasoning process, highlighting its potential as a valuable decision-support tool to augment human expertise in complex clinical diagnostic tasks.

Background: The application of big data resources and the development of medical collaborative networks (MCNs) boost each other. However, MCNs are often assumed to be exogenous. How big data resources affect the emergence, development, and evolution of endogenous MCNs has not been well explained. Objective: This study aimed to explore and understand the influence of the mechanism of a wide range of shared and private big data resources on the transaction efficiency of medical services to reveal the impact of big data resources on the emergence and development of endogenous MCNs. Methods: This study was conducted by administering a survey questionnaire to information technology staff and medical staff from 132 medical institutions in China. Data from information technology staff and medical staff were integrated. Structural equation modeling was used to test the direct impact of big data resources on transaction efficiency of medical services. For those big data resources that had no direct impact, we analyzed their indirect impact. Results: Sharing of diagnosis and treatment data (β=.222; P=.03) and sharing of medical research data (β=.289; P=.04) at the network level (as big data itself) positively directly affected the transaction efficiency of medical services. Network protection of the external link systems (β=.271; P=.008) at the level of medical institutions (as big data technology) positively directly affected the transaction efficiency of medical services. Encryption security of web-based data (as big data technology) at the level of medical institutions, medical service capacity available for external use, real-time data of diagnosis and treatment services (as big data itself) at the level of medical institutions, and policies and regulations at the network level indirectly affected the transaction efficiency through network protection of the external link systems at the level of medical institutions. Conclusions: This study found that big data technology, big data itself, and policy at the network and organizational levels interact with, and influence, each other to form the transaction efficiency of medical services. On the basis of the theory of neoclassical economics, the study highlighted the implications of big data resources for the emergence and development of endogenous MCNs.

Background: Radiofrequency ablation (RFA) for atrial fibrillation (AF) is associated with a risk of complications. This study aimed to develop and validate risk models for predicting complications after radiofrequency ablation of atrial fibrillation patients. Methods: This retrospective cohort study included 3365 procedures on 3187 patients with atrial fibrillation at a single medical center from 2018 to 2021. The outcome was the occurrence of postoperative procedural complications during hospitalization. Logistic regression, decision tree, random forest, gradient boosting machine, and extreme gradient boosting were used to develop risk models for any postoperative complications, cardiac effusion/tamponade, and hemorrhage, respectively. Patients’ demographic characteristics, medical history, signs, symptoms at presentation, electrocardiographic features, procedural characteristics, laboratory values, and postoperative complications were collected from the medical record. The prediction results were evaluated by performance metrics (i.e., the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, F score, and Brier score) with repeated fivefold cross-validation. Results: Of the 3365 RFA procedures, there were 62 procedural complications with a rate of 1.84% in the entire cohort. The most common complications were cardiac effusion/tamponade (28 cases, 0.83%), and hemorrhage (21 cases, 0.80%). There was no procedure-related mortality. The machine learning algorithms of random forest (RF) outperformed other models for any complication (AUC 0.721 vs 0.627 to 0.707), and hemorrhage (AUC 0.839 vs 0.649 to 0.794). The extreme gradient boosting (XGBoost) model outperformed other models for cardiac effusion/tamponade (AUC 0.696 vs 0.606 to 0.662). Conclusions: The developed risk models using machine learning algorithms showed good performance in predicting complications after RFA of AF patients. These models help identify patients at high risk of complications and guiding clinical decision-making.