The COVID-19 pandemic highlighted the complexities of diagnosing and managing acute Respiratory Failure (RF). Early prediction of RF remains a key challenge, with no established tools currently available. This study developed a machine learning model to predict RF in hospitalised COVID-19 patients, using structured data (demographic and clinical variables) and clinical reports processed through Natural Language Processing. Early results show an AUC-ROC of 0.856 and an accuracy of 76.5∖% with a Random Forest model, demonstrating the potential of AI to enhance early prediction of patient outcomes in the context of RF.
Fair Digital Objects (FDOs) are designed to manage complex scientific and technological data through open standards and metadata enhancing transparency and traceability. A conceptual framework has been developed to be reliably describe, locate and use any digital entity. This paper aimed to compile and analyze the literature related to FDOs applied in biomedical research, presenting a real case: the OHDO (Open Health research supports Digital Objects life-cycle) tool. Using the PRISMA-ScR methodology, a comprehensive review was conducted to identify gaps and highlight the potential of FDOs, demonstrating OHDO’s ability to leverage existing FAIR frameworks. The review showed that while FDOs improve data management and reproducibility in biomedical research, they face challenges such as technical barriers and the need for better interoperability.
Suicide is currently the leading cause of preventable death among young people, yet early identification of reattempt risk remains limited by its multifactorial nature, stigma, fragmented data, reliance on subjective assessments, and heavy workloads. The PROSODIC project developed an interpretable machine learning model to predict first suicide reattempts among youths, using routinely collected clinical and sociodemographic data from 1,031 patients in Seville, Spain, aged 5–25 years. Logistic Regression achieved the best performance (AUC-ROC = 0.65), balancing accuracy and interpretability. A three-level graded risk scale was derived to guide clinical action. Results highlight social vulnerability, behavioral dysregulation, and care discontinuity as key predictors, supporting the model’s clinical relevance and feasibility for integration into real-world suicide secondary prevention pathways. Limitations include the use of single-site data and moderate discriminative performance; however, future work will extend validation across additional regions to enhance generalizability and robustness.