BMC Medical Informatics and Decision Making最新文献

{"title":"D3MI: an efficient and powerful federated imputation method for bias reduction in the analysis of distributed incomplete data by accounting for within-site correlation and between-site heterogeneity.","authors":"Yi Lian, Xiaoqian Jiang, Qi Long","doi":"10.1186/s12911-026-03663-2","DOIUrl":"10.1186/s12911-026-03663-2","url":null,"abstract":"<p><strong>Background: </strong>Electronic health records (EHRs) collected from diverse healthcare institutions offer a rich and representative data source for clinical research. Federated learning enables analysis of these distributed data without sharing sensitive patient-level information, preserving privacy. However, missing data remain a major challenge and can introduce substantial bias if not properly addressed. Very few distributed imputation methods currently exist, and they fail to account for two critical aspects of EHR data: correlation within sites and variability across sites. We aim to fill this important methodological gap.</p><p><strong>Methods: </strong>We propose Distributed Mixed Model-based Multiple Imputation (D3MI), a novel federated imputation method designed to reduce bias in distributed EHRs. D3MI integrates the strengths from federated learning techniques, statistical learning methods for correlated data, and multilevel imputation algorithms to explicitly account for both within-site correlation and between-site heterogeneity using site-specific random effects. It operates under the Missing at Random (MAR) assumption. It preserves privacy by avoiding sharing raw data and features communication and computational efficiency.</p><p><strong>Results: </strong>Through extensive simulation studies, we demonstrate that D3MI outperforms SOTA distributed imputation methods in both accuracy and consistency. We further demonstrate the use of D3MI in a real-world EHR case study involving incomplete and clustered data from participating hospitals in the Georgia Coverdell Acute Stroke Registry.</p><p><strong>Conclusion: </strong>By explicitly modeling the complex structure of distributed EHR data, D3MI addresses key limitations of existing approaches. It provides a powerful and efficient solution for handling missing data in distributed and privacy-sensitive settings and enhances the rigor and reproducibility of collaborative clinical research.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-10-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13647586/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149126596","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Using interpretable machine learning models to predict the occurrence of severe influenza in hospitalized children: a retrospective cohort study.","authors":"Jinhuan Wu, Wenxian Qian, Shuqin Xie, Yi Yan, Gang Chen, Ye Wang, Xing Liu, Xiuchang Ma, Yuchen Du, Yaowen Liang, Shuanglin Gu, Wei Li, Apeng Chen, Changhua Yi, Man Tian","doi":"10.1186/s12911-026-03664-1","DOIUrl":"https://doi.org/10.1186/s12911-026-03664-1","url":null,"abstract":"<p><strong>Background: </strong>Influenza, a prevalent disease, significantly threatens public health. Accurately predicting severe influenza occurrences is crucial for developing personalized prevention strategies and treatment plans.</p><p><strong>Aims: </strong>This study aimed to construct a highly interpretable model to assess the risk of severe influenza in hospitalized children, using the SHapley Additive exPlanation (SHAP) method to interpret the Random Forest (RF) model and identify risk factors for severe influenza.</p><p><strong>Methods: </strong>A retrospective cohort study was conducted, collecting hospitalization records of influenza patients from the Electronic Medical Records (EMR) system for model development and performance evaluation. Data were collected within the first 24 h after patient admission and the dataset was randomly divided into two parts: 70% for model training and 30% for model accuracy verification.</p><p><strong>Results: </strong>A total of 436 eligible influenza patients were included in the study analysis. The RF model demonstrated the highest predictive accuracy among the six models compared, with an area under the receiver-operating-characteristiccurve(AUROC) of 0.88. The decision curve analysis (DCA) indicated that the net benefit of the RF model surpassed that of other machine learning models. According to the importance ranking by the SHAP method, blood glucose was identified as the most significant predictive variable.</p><p><strong>Conclusion: </strong>The study established a RF model capable of predicting severe influenza occurrence in hospitalized children. The model, interpreted through the SHAP method, assists physicians in providing better treatment plans.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13637386/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149104480","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CTG-FRAME: development and validation of a multi‑modal deep learning for intrapartum escalation support and severe fetal acidemia prediction.","authors":"Dian Tjondronegoro, Elizabeth Irenne Yuwono, Fabricio Da Silva Costa","doi":"10.1186/s12911-026-03651-6","DOIUrl":"10.1186/s12911-026-03651-6","url":null,"abstract":"<p><strong>Background: </strong>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.</p><p><strong>Methods: </strong>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.</p><p><strong>Results: </strong>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.</p><p><strong>Conclusions: </strong>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.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13617899/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149073268","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Interpretable machine learning models for predicting tocilizumab response in rheumatoid arthritis using clinical data.","authors":"Mengsi Ma, Xinya Chen, Xue Wu, Chen Chen, Zhuoyu Li, Qianxi Xu, Ziyang Zhang, JinChao Chen, Zhenyang Guan, Xinming Feng, Xiangcao Liang, Wentao Qin, Lijun Wu","doi":"10.1186/s12911-026-03486-1","DOIUrl":"10.1186/s12911-026-03486-1","url":null,"abstract":"<p><strong>Background: </strong>Approximately 30% of patients with rheumatoid arthritis (RA) do not respond to tocilizumab (TCZ). This study aimed to predict TCZ response using machine learning models trained on clinical data.</p><p><strong>Methods: </strong>Baseline and follow-up data from patients with RA treated with TCZ were collected. Seven different machine learning models (logistic regression, K-nearest neighbor algorithm, random forest, support vector machine, decision tree, gradient boosting, and boosting algorithm) were trained to predict the treatment response of patients with RA across different imaging stages after three to six months of therapy. The area under the receiver operating curve (AUC) was the main performance evaluation feature to screen the best model, and the relative importance of each variable in the model was ranked using the Shapley Additive Explanation (SHAP) value.</p><p><strong>Results: </strong>A total of 245 RA patients treated with TCZ were included. The logistic regression model demonstrated superior prediction performance across imaging stages, achieving AUC scores of 0.78 for patients without imaging changes, 0.73 for stage I, and 0.82 for stages II, III, and IV. Key features influencing predictions varied by imaging stage. The most important features were D-dimer concentration for patients without imaging changes, DAS28 grade for patients in stage I, and the physician's overall disease score (EGA) for patients in stages II, III, and IV, with SHAP values of 0.5180, 0.6661, and 1.3978, respectively.</p><p><strong>Conclusions: </strong>This study demonstrates the potential of machine learning to predict TCZ treatment outcomes and identify stage-specific features in patients with RA.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13621592/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149073290","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Cardiovascular disease prediction and mortality risk assessment using a QSAO‑optimized BP neural network.","authors":"Mengmeng Chen, Yani Yan, Yige Xue, Wenyi Tu","doi":"10.1186/s12911-026-03865-8","DOIUrl":"https://doi.org/10.1186/s12911-026-03865-8","url":null,"abstract":"<p><strong>Background: </strong>Cardiovascular disease remains a leading cause of mortality worldwide, imposing a substantial burden on global healthcare systems. Early detection and accurate mortality risk assessment are critical for timely intervention and improved patient outcomes. However, existing predictive models often suffer from limited clinical applicability due to their restriction to single diagnostic tasks, insufficient prediction accuracy, and a tendency toward overfitting.</p><p><strong>Methods: </strong>This study proposes a novel hybrid intelligent model, termed QSAO-BP, which integrates a quartile based Snow Ablation Optimizer (QSAO) with a Back Propagation (BP) neural network for simultaneous cardiovascular disease (CVD) diagnosis and mortality risk assessment. The QSAO enhances the SAO by incorporating Gaussian and Lévy random walks through a quartile based strategy, effectively mitigating premature convergence and local optima entrapment of SAO. The QSAO optimizes the hyperparameters of the BP neural network to improve its predictive accuracy and generalization capability. A dual validation framework is established using the Cleveland Heart Disease Dataset and the Heart Failure Clinical Records Dataset. Extensive experiments compare QSAO-BP against nine optimizer based BP variants and four classical machine learning algorithms across ten evaluation metrics.</p><p><strong>Results: </strong>QSAO-BP achieves superior performance on both datasets, attaining an accuracy of 0.873 on Dataset 1 and 0.854 on Dataset 2, representing improvements of 8.178% and 9.347% over SAO-BP, respectively. The model also achieves the highest accuracy, precision, recall, F1-score, and MCC while maintaining competitive specificity and NPV. SHAP analysis reveals that the key predictors identified by QSAO-BP are consistent with established clinical knowledge.</p><p><strong>Conclusions: </strong>QSAO-BP delivers robust and accurate predictions across two distinct clinical tasks with satisfactory interpretability, positioning it as a promising tool to support clinical decision making in CVD.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149227298","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CleanSurvival: automated data preprocessing for time-to-event models using reinforcement learning.","authors":"Yousef Koka, David Selby, Gerrit Großmann, Kathan Pandya, Sebastian Vollmer","doi":"10.1186/s12911-026-03786-6","DOIUrl":"10.1186/s12911-026-03786-6","url":null,"abstract":"<p><strong>Background: </strong>Data preprocessing is often paid little attention in machine learning, despite its potentially significant impact on model performance. While automated machine learning pipelines are starting to recognise and integrate data preprocessing into their solutions for classification and regression tasks, this integration is lacking for more specialised tasks like time-to-event models for censored data. As a result, survival analysis not only faces the general challenges of data preprocessing but also suffers from the lack of tailored, automated solutions in this area.</p><p><strong>Method: </strong>To address this gap, this paper presents CleanSurvival, a reinforcement-learning-based solution for optimizing preprocessing pipelines, extended specifically for survival analysis. The framework can handle continuous and categorical variables. It builds upon Learn2Clean's [Formula: see text]-learning to select which combination of data imputation, outlier detection and feature extraction techniques achieves optimal performance for a Cox, random forest, neural network or user-supplied time-to-event model. The Python package is available on GitHub: https://github.com/datasciapps/CleanSurvival .</p><p><strong>Results: </strong>Experimental benchmarks on real-world datasets show that the [Formula: see text]-learning-based data preprocessing can improve predictive performance relative to simple baselines, while runtime behaviour is condition-dependent and most clearly interpretable in the best-covered benchmark cells. Furthermore, a simulation study demonstrates effectiveness across different types and levels of missingness and noise.</p><p><strong>Conclusion: </strong>With an increase in the use of machine learning, it becomes important to generalise AutoML pipelines to a variety of models now present, including survival analysis. Tools like CleanSurvival, which integrate preprocessing for survival analysis, can make survival studies faster and easier to perform, while also yielding more robust results.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13613583/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149036386","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evaluation of short- and long-term mortality prediction in patients undergoing hip fracture surgery using a biomarker-based interpretable predictive modeling approach: a retrospective cohort analysis.","authors":"Sait Fatih Öner, Sevim Şenol Karataş, Oğuz Kağan Bulut, Ferhat Karataş","doi":"10.1186/s12911-026-03797-3","DOIUrl":"10.1186/s12911-026-03797-3","url":null,"abstract":"<p><strong>Purpose: </strong>The aim of this study was to develop an interpretable predictive modeling approach to predict short-term (three-month) and long-term (one-year) mortality in patients undergoing hip fracture surgery, and to evaluate the prognostic value of routinely available clinical and laboratory variables, including age, sex, lactate-to-albumin ratio (LAR), red cell distribution width-standard deviation (RDW-SD), and neutrophil-to-lymphocyte ratio (NLR).</p><p><strong>Methods: </strong>This retrospective cohort study included 1,000 patients who underwent hip fracture surgery. Demographic characteristics and preoperative laboratory parameters (LAR, RDW-SD, and NLR) were recorded. Three-month and one-year mortality were defined as the primary outcomes. Independent predictors of mortality were assessed using a multivariable logistic regression model implemented within an interpretable predictive modeling framework. Model performance was evaluated using receiver operating characteristic (ROC) curves, area under the curve (AUC), precision-recall AUC (PR-AUC), and Brier score. Calibration was assessed using calibration plots, and optimal probability thresholds were determined using the Youden J index.</p><p><strong>Results: </strong>Among the included patients, 17.4% died within three months and 24.6% within one year following surgery. Non-survivors were older and had higher LAR, RDW-SD, and NLR values compared with survivors (p < 0.001). In multivariable analyses, advanced age and elevated NLR were identified as independent predictors of both short-term and long-term mortality (p < 0.001). LAR was independently associated with one-year mortality (odds ratio: 1.035; 95% confidence interval: 1.008-1.063; p = 0.010). The model demonstrated acceptable discriminative performance, with an AUC of 0.807 for three-month mortality and 0.810 for one-year mortality. At the optimal thresholds, sensitivity and specificity were 80% and 73% for three-month mortality, and 77% and 71% for one-year mortality, respectively.</p><p><strong>Conclusion: </strong>The biomarker-based interpretable predictive modeling approach developed in this study demonstrated acceptable discriminative performance for estimating short- and long-term mortality after hip fracture surgery. Advanced age, elevated NLR, and increased LAR emerged as important predictors. As these variables are routinely available in clinical practice, the proposed framework may support perioperative risk awareness and assist clinical evaluation when interpreted together with comprehensive clinical assessment.</p><p><strong>Clinical trial number: </strong>Not applicable.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13602654/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149029953","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"FAIR by design (TRACE): A Trusted Research Access & Collaboration Environment.","authors":"Marcel Müller, Benjamin P Geisler, Samar Shamas, Nikolaus V Bomhard, Eva Hoster, Markus Pfirrmann, Ulrich Mansmann","doi":"10.1186/s12911-026-03840-3","DOIUrl":"10.1186/s12911-026-03840-3","url":null,"abstract":"<p><strong>Background: </strong>A tension exists between the open science imperative for data transparency through FAIR (Findable, Accessible, Interoperable, Reusable) data sharing and the stringent requirements of current data protection regulations for health data. Researchers lack efficient yet affordable tools to share sensitive data responsibly. Existing solutions are often either difficult to use, expensive, or lack features like variable-level access control and auditable and reproducible workflows. Our objective was to develop a solution that lowers the threshold for researchers to share sensitive data responsibly and that can be operated by existing institutional staff.</p><p><strong>Results: </strong>TRACE (Trusted Research Access & Collaboration Environment) was designed to support several data-sharing elements highlighted in the 2025 SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) and CONSORT (Consolidated Standards of Reporting Trials) statements by providing controlled and auditable workflows and supporting reproducibility at the level of data and analysis code. Its features for data governance, sharing, reproducibility and security allow institutions to manage study data access and to establish consistent, transparent standard operating procedures through granular variable selection, integrated study visualisation, and provisioning into a sealed analysis environment. TRACE provides an automated workflow for selected data-access steps that had previously relied on manual curation and ad-hoc sharing. A core feature versions datasets together with the corresponding analysis scripts, supporting reproducibility at the level of data and code; the computational environment is not captured. TRACE has supported data governance across seven clinical studies and processed 56 data access applications. Across three studies, an uncontrolled before-and-after comparison found 42.4 manual interface exports per 30 days before TRACE and 4.7 after implementation; causality cannot be inferred.</p><p><strong>Conclusions: </strong>Platforms like TRACE may help shift resource-intensive, expert-dependent data governance towards automated processes. Operated by existing institutional staff, TRACE may make controlled data sharing feasible for institutions with limited resources. By automating previously manual steps, TRACE is intended to reduce the administrative burden of governed data sharing; whether it shortens the interval from data request to analysis has not been measured. TRACE is used as a training environment for preparing and using data in sharing processes.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13576130/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148976009","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A framework for extraction of clinical information from radiological mammography reports using large language models and retrieval augmented generation.","authors":"Eduardo Godoy, Joaquín De Ferrari, Sofia Lazo, Catalina Pacheco, Diego Mellado, Alex Saez, Steren Chabert, Rodrigo Salas","doi":"10.1186/s12911-026-03633-8","DOIUrl":"10.1186/s12911-026-03633-8","url":null,"abstract":"<p><strong>Background: </strong>Extracting structured information from free-text radiology reports is essential for downstream clinical analysis and decision support. In mammography, this challenge is amplified by heterogeneous writing styles, the absence of standardized terminology across institutions, and limited availability of annotated Spanish datasets. Large language models (LLMs) offer a compelling alternative by enabling few-shot generalization without task-specific fine-tuning.</p><p><strong>Methods: </strong>We propose a structured Retrieval-Augmented Generation (RAG) framework that leverages annotated mammography reports as in-context demonstrations, enabling joint named entity and relation extraction without task-specific fine-tuning. The framework is evaluated on Spanish BI-RADS-style mammography reports and achieves performance comparable to fine-tuned BETO models in low-resource settings, while significantly reducing training and deployment overhead.</p><p><strong>Results: </strong>In NER, GPT-based few-shot models achieved competitive performance (0.89-0.94 F[Formula: see text]) relative to the fine-tuned BETO baseline (0.97 F[Formula: see text]), despite requiring no task-specific training. In RE, LLMs showed moderate performance (up to 0.78 F[Formula: see text]), remaining below the supervised BETO model (0.99 F[Formula: see text]), reflecting the greater sensitivity of RE to boundary errors and cross-sentence context. Inference costs were low (fractions of a cent per report) and latency remained within seconds, enabling practical deployment scenarios. Local open-weight models preserved privacy and runtime efficiency but exhibited substantially lower accuracy.</p><p><strong>Conclusion: </strong>Few-shot RAG combined with modern LLMs provides a viable, data-efficient alternative for structuring Spanish mammography reports, particularly in low-resource or rapid-deployment settings. While fine-tuned encoders remain preferable for high-accuracy RE, the proposed framework offers a practical balance between performance, operational cost, and accessibility. A remaining limitation is the need for an initial annotated subset to populate the RAG store; future work will explore weak supervision, multimodal extensions, and cross-site generalization.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13560278/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148928630","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Machine learning-assisted mRNA vaccine pharmacovigilance: a systematic review of multi-source real-world data.","authors":"YuLong He, Yan Mao, XinYu Wang","doi":"10.1186/s12911-026-03832-3","DOIUrl":"10.1186/s12911-026-03832-3","url":null,"abstract":"<p><strong>Background: </strong>The rapid deployment of mRNA vaccines during the COVID-19 pandemic exposed limitations in traditional pharmacovigilance systems, including delayed reporting, high underreporting rates, and inability to calculate true incidence. Machine learning (ML) offers new pathways to overcome these challenges by integrating multi-source real-world data.</p><p><strong>Methods: </strong>We systematically reviewed English-language studies from database inception to June 2026. Searches were performed in PubMed, Embase, and Web of Science. Two reviewers independently screened records. Given substantial heterogeneity across ML tasks (signal detection, text extraction, risk prediction, prognosis stratification), algorithms, data sources, and metrics, we performed narrative synthesis. Risk of bias was assessed using adapted QUADAS-2.</p><p><strong>Results: </strong>We identified 43 studies. For adverse-event prediction, tree-based models reported AUCs of 0.85-0.87, though estimates derive from heterogeneous settings. NLP reduced redundant signals by 17% in vaccine reporting systems. For myocarditis, ML models reached AUCs up to 0.899 in cardiovascular cohorts, but direct mRNA vaccine applications remain limited and retrospective. Emerging platforms (self-amplifying and tumor mRNA vaccines) lack post-marketing data, rendering ML applications largely conceptual.</p><p><strong>Conclusion: </strong>ML-assisted pharmacovigilance enables a shift from passive to active, intelligent monitoring. Despite challenges in data quality, model interpretability, and regulatory approval, intelligent pharmacovigilance systems will become essential infrastructure for safeguarding public health.</p>","PeriodicalId":9340,"journal":{"name":"BMC Medical Informatics and Decision Making","volume":"26 1","pages":""},"PeriodicalIF":5.5,"publicationDate":"2026-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13555883/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148918262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}