Advancements in medical research have increasingly relied on robust data analytics to support diagnostic and treatment decisions. However, data analysis still faces challenges when investigating datasets with severe class imbalance, often stemming from the rarity of certain conditions and uneven disease distributions. To address this issue, we propose the Noise-Aware Undersampling with Subsampling (NAUS) algorithm. NAUS integrates clustering, noise removal, and Tomek-link identification techniques to create refined subsamples that assess the significance of individual observations, while systematically removing redundant and noisy data. The proposed approach was evaluated on datasets related to chronic kidney disease, liver disease, heart disease and its performance was compared to that of traditional oversampling methods (e.g., SMOTE, ADASYN, LoRAS) and undersampling techniques (e.g., random undersampling, Tomek-links). Our experimental results, based on machine learning classifiers (e.g. Random Forest, LightGBM, and Multilayer Perceptron). Data visualization further confirmed that NAUS effectively mitigates class imbalance, making it a promising tool for enhancing the reliability of medical data analysis.

Background: In an earlier murine model of myocardial infarction (MI), we showed that CD8 cells and myeloid dendritic cells (mDCs) infiltrate the infarcted myocardium within the first week. However, in humans, the spatial interplay between CD8+ T cells and dendritic cells in the spatial context of human myocardial infarction remains underexplored. Objective: In the present study, we applied spatial transcriptomics and functional assays to characterize immune–stromal dynamics in infarcted myocardium and peripheral blood. Methods & Results: Spatial transcriptomics analysis of infarcted human myocardium at days 2 and 6 post-MI, combined with peripheral blood flow cytometry and EPC colony-forming assays, was performed. Cell composition, pathway enrichment, and cell-to-cell communication analyses were conducted to map immune–stromal cells’ dynamics across time points. Spatial mapping identified dynamic shifts in immune, fibroblast, and endothelial populations, with fibroblasts and endothelial cells remaining abundant throughout. CD8+ T cells accumulated in ischemic regions while their circulating levels declined. Gene Ontology and pathway analyses of CD8A+ transcripts revealed enrichment of proinflammatory and NF-κB survival programs. ITGAX/CD33/THBD+ APCs progressively increased within infarct zones, activating antigen-presentation and leukocyte chemotaxis pathways. Early (day 2) APC–endothelial crosstalk showed the strongest predicted recruitment signals for CD8+ T cells, which diminished by day 6. Finally, EPC colony-forming capacity showed a tendency for reduction in MI patients and inversely correlated with coronary lesion burden, indicating impaired vascular repair potential. Conclusions: This integrative spatial and functional study demonstrates that APC-driven CD8+ recruitment and EPC dysfunction are key features of human MI. Immune–endothelial niches facilitate early cytotoxic T-cell infiltration, while progenitor depletion limits vascular regeneration. These findings provide mechanistic insight into immune–vascular imbalance during infarct healing and highlight potential therapeutic targets to modulate inflammation and restore vascular repair.

INTRODUCTION:Clinical text de-identification is essential for secondary healthcare data use but remains challenging due to heterogeneous documentation. METHODS:We evaluated LLM-based de-identification of synthetic German discharge letters (GraSCCo) using four prompting strategies with GPT-4o and an additional GPT-OSS comparison. Performance was assessed via precision, recall, F1-score, false positives, and relative text reduction. RESULTS:The baseline setup was not suitable due to excessive text loss. F1-scores reached 0.81 (session-isolated), 0.79 (structured-input), 0.93 (optimized prompt, GPT-4o), and 0.90 (GPT-OSS). DISCUSSION:Results indicate that prompt refinement has a stronger impact on de-identification quality than structural preprocessing. LLM-based de-identification is feasible but requires careful prompt design and validation on real-world data.

The generation of synthetic clinical trial data offers a promising approach to mitigating privacy concerns and data accessibility limitations in medical research. However, ensuring that synthetic datasets maintain high fidelity, utility, and adherence to domain-specific constraints remains a key challenge. While hyperparameter optimization (HPO) improves generative model performance, the effectiveness of different optimization strategies for synthetic clinical data remains unclear. This study systematically evaluates four HPO objectives across nine generative models, comparing single-metric to compound metric optimization. Our results demonstrate that HPO consistently improves synthetic data quality, with Tab DDPM achieving the largest relative gains, followed by TVAE (60%), CTGAN (39%), and CTAB-GAN+ (38%). Compound metric optimization outperformed single-metric objectives, producing more generalizable synthetic datasets. Despite improving overall quality, HPO alone fails to prevent violations of essential clinical survival constraints. Preprocessing and postprocessing played a crucial role in reducing these violations, as models lacking robust processing steps produced invalid data in up to 61% of cases. These findings underscore the necessity of integrating explicit domain knowledge alongside HPO to generate high-quality synthetic datasets. Our study provides actionable recommendations for improving synthetic data generation, with future work needed to refine metric selection and validate findings on larger datasets.

Background/Objectives: Solid organ transplantation remains a critical life-saving treatment for end-stage organ failure, yet it faces persistent challenges, such as organ scarcity, graft rejection, and postoperative complications. Artificial intelligence (AI) has the potential to address these challenges by revolutionizing transplantation practices. Methods: This review article explores the diverse applications of AI in solid organ transplantation, focusing on its impact on diagnostics, treatment, and the evolving market landscape. We discuss how machine learning, deep learning, and generative AI are harnessing vast datasets to predict transplant outcomes, personalized immunosuppressive regimens, and optimize patient selection. Additionally, we examine the ethical implications of AI in transplantation and highlight promising AI-driven innovations nearing FDA evaluation. Results: AI improves organ allocation processes, refines predictions for transplant outcomes, and enables tailored immunosuppressive regimens. These advancements contribute to better patient selection and enhance overall transplant success rates. Conclusions: By bridging the gap in organ availability and improving long-term transplant success, AI holds promise to significantly advance the field of solid organ transplantation.

INTRODUCTION:The German Medical Informatics Initiative (MII) promotes the use of routine clinical data for research, supported by the broad consent framework to ensure patient engagement. This work proposes a data management process and reference infrastructure to improve transparency by enabling patients to track their consent history and data use in research. METHODS:We analyzed the data provision process at the University Hospital Dresden (UKD) to identify roles and data flows relevant to secondary data use under broad consent. Established MII tools in use at UKD were evaluated for their suitability in enabling secure data access. RESULTS:We developed a structured data access process and implemented a reference infrastructure that lays the groundwork for a potential patient-facing application providing secure access to consent and study details. CONCLUSION:The reference infrastructure demonstrates how existing MII tools can be repurposed to offer patient-centric transparency in secondary data use. Future work will address scalability, access control, and ethical considerations, such as patient expectations and the clarity of information.

The Cancer Genome Atlas (TCGA) provides comprehensive genomic data across various cancer types. However, complex file naming conventions and the necessity of linking disparate data types to individual case IDs can be challenging for first-time users. While other tools have been introduced to facilitate TCGA data handling, they lack a straightforward combination of all required steps. To address this, we developed a streamlined pipeline using the Genomic Data Commons (GDC) portal’s cart system for file selection and the GDC Data Transfer Tool for data downloads. We use the Sample Sheet provided by the GDC portal to replace the default 36-character opaque file IDs and filenames with human-readable case IDs. We developed a pipeline integrating customizable Python scripts in a Jupyter Notebook and a Snakemake pipeline for ID mapping along with automating data preprocessing tasks (https://github.com/alex-baumann-ur/TCGADownloadHelper). Our pipeline simplifies the data download process by modifying manifest files to focus on specific subsets, facilitating the handling of multimodal data sets related to single patients. The pipeline essentially reduced the effort required to preprocess data. Overall, this pipeline enables researchers to efficiently navigate the complexities of TCGA data extraction and preprocessing. By establishing a clear step-by-step approach, we provide a streamlined methodology that minimizes errors, enhances data usability, and supports the broader utilization of TCGA data in cancer research. It is particularly beneficial for researchers new to genomic data analysis, offering them a practical framework prior to conducting their TCGA studies.

Risk stratification in acute myeloid leukemia (AML) is driven by genetics, yet patient age substantially influences therapeutic decisions. To evaluate how age alters the prognostic impact of genetic mutations, we pooled data from 3062 pediatric and adult AML patients from multiple cohorts. Signaling pathway mutations dominated in younger patients, while mutations in epigenetic regulators, spliceosome genes, and TP53 alterations became more frequent with increasing age. Machine learning models were trained to identify prognostic variables and predict complete remission and 2-year overall survival, achieving area-under-the-curve scores of 0.801 and 0.791, respectively. Using Shapley (SHAP) values, we quantified the contribution of each variable to model decisions and traced their impact across six age groups: infants, children, adolescents/young adults, adults, seniors, and elderly. The highest contributions to model decisions among genetic variables were found for alterations of NPM1, CEBPA, inv(16), and t(8;21) conferring favorable risk and alterations of TP53, RUNX1, ASXL1, del(5q), -7, and -17 conferring adverse risk, while FLT3-ITD had an ambiguous role conferring favorable treatment responses yet poor overall survival. Age significantly modified the prognostic value of genetic alterations, with no single alteration consistently predicting outcomes across all age groups. Specific alterations associated with aging such as TP53, ASXL1, or del(5q) posed a disproportionately higher risk in younger patients. These results challenge uniform risk stratification models and highlight the need for context-sensitive AML treatment strategies.

Genomic and epigenomic instability are defining features of cancer, driving tumor progression, heterogeneity, and therapeutic resistance. Central to this process are epigenetic echoes, persistent and dynamic modifications in DNA methylation, histone modifications, non-coding RNA regulation, and chromatin remodeling that mirror underlying genomic chaos and actively influence cancer cell behavior. This review delves into the complex relationship between genomic instability and these epigenetic echoes, illustrating how they collectively shape the cancer genome, affect DNA repair mechanisms, and contribute to tumor evolution. However, the dynamic, context-dependent nature of epigenetic changes presents scientific and ethical challenges, particularly concerning privacy and clinical applicability. Focusing on lung cancer, we examine how specific epigenetic patterns function as biomarkers for distinguishing cancer subtypes and monitoring disease progression and relapse.

Angiogenesis is crucial in myocardial healing after myocardial infarction (MI). The αvβ3-integrin, a key regulator of angiogenesis, is targeted by RGD-based PET tracers like [68Ga]Ga-NODAGA-RGD. Yet, angiogenesis imaging using RGD-based tracers is seriously hampered by the lack of true specificity of the αvβ3-integrin for angiogenic cells. Therefore, our study aimed to identify the cell type with the highest αvβ3-integrin expression in the process of myocardial healing in order to determine the actual value of the PET tracer [68Ga]Ga-NODAGA-RGD for imaging post-MI angiogenesis. Cardiac magnetic resonance imaging (CMR) was used to assess cardiac function and morphology after 28 days in two groups: permanent ligation (PL) of the left anterior descending coronary artery and transient occlusion for 30 min (I/R). Following these measurements, hearts were excised for histological and immunohistological examinations to evaluate scar formation, capillary density, and cellular composition. PET imaging with [68Ga]Ga-NODAGA-RGD was conducted on day 5 and day 7 post-MI. Single-nucleus transcriptomics were performed to identify cell clusters expressing αvβ3-integrin. Both infarct models induced scar formation, with the PL group developing large infarcts accompanied by massive left ventricular dilation and hypertrophy of cardiomyocytes, while the I/R group exhibited small intramural scars without significant changes in LV geometry or function. PET imaging revealed significantly higher tracer accumulation in the infarct area of the PL group compared to the I/R group. Single-nucleus transcriptomics performed 5 days post-MI revealed that angiogenesis markers were enriched in the I/R group, while the highest αvβ3-integrin mRNA expression was identified in the fibroblast cluster, indicating an activated phenotype. Activated fibroblasts are the primary target cells of [68Ga]Ga-NODAGA-RGD, rather than angiogenic cells. In this regard, [68Ga]Ga-NODAGA-RGD is most probably not a valid tracer for imaging angiogenesis during the first days post-MI.

Integrating Patient-Reported Outcome Measures (PROMs) into Molecular Tumor Boards (MTBs) remains challenging due to the complexity of data visualization and integration into clinical workflows. This work, as part of the German PM4Onco project, aims to identify visualization requirements for PROMs and develop a prototype for PROMs integration in cBioPortal, facilitating broader application within oncology care. We employed a qualitative research approach, including developing personas for MTB stakeholders, conducting a literature-based requirements analysis, organizing a co-design workshop to create low-fidelity prototypes, and evaluating the highest-rated prototype variant through an online survey distributed to MTB physicians across Germany. Seven specialist groups were identified, with key needs including intuitive visualization, clear axis labeling, and longitudinal symptom tracking. The resulting mid-fidelity mockup incorporated PROMs data within cBioPortal’s timeline view and a detailed PROMs tab, featuring trend indicators, line graphs, and customizable health displays. Usability evaluation by MTB members yielded a SUS score of 67, indicating an initial indicator of acceptable usability, with suggestions for improvements like threshold scores and deeper clinical data integration. While PROMs offer critical patient insights, they remain underused in MTBs. Our early-stage prototype demonstrates potential for addressing this gap, with future work focusing on implementation, broader testing, and international validation.

Subset selection has been widely studied but remains underexplored for synthetic tabular data, particularly in data sharing contexts that require high quality data. While generative models can produce large volumes of synthetic data, directly sampling and releasing such data risks including low quality or unrepresentative samples, which can reduce data utility. An alternative approach is to generate more data than needed and subsequently select a subset that better meets specific quality criteria. This paper introduces a genetic algorithm (GA)-based method for optimizing such subset selection. The proposed GA is independent of any specific fitness function, enabling adaptation to diverse evaluation metrics, their combinations, or varying use case requirements. We benchmarked the method on five medical datasets, each synthesized by multiple generative architectures, and consistently found that the GA selected subsets outperformed both the initial synthetic datasets and a random subset selection baseline. Notably, initializing the GA with systematically generated synthetic subsets led to nearly twice the improvement over the baselines compared to random initialization, emphasizing the importance of more informed starting solutions. The proposed GA-based method proved especially beneficial for smaller datasets, which are frequently encountered in clinical domains, such as rare disease research. While performance gains diminished for larger datasets due to combinatorial complexity, this work highlights the potential of GA-driven optimization as a foundation for future research into scalable and adaptive subset selection methods for synthetic data sharing.

The journey from laboratory research to clinical practice is marked by significant advancements in the fields of single-cell technologies and non-coding RNA (ncRNA) research. This convergence may reshape our approach to personalized medicine, offering groundbreaking insights and treatments in various clinical settings. This chapter discusses advancements in (nc)RNAs in the clinics, innovations in single-cell technologies and algorithms, and the impact on actual precision medicine, showing the integration of single-cell and ncRNA research can have a tangible impact on precision medicine. Case studies in Oncology, Immunology, and other fields demonstrate how these technologies can guide treatment decisions, tailor therapies to individual patients, and improve outcomes. This approach is particularly potent in addressing diseases with high inter- and intra-tumor heterogeneity. The final sections address standardization, data integration, and analysis challenges because the complexity and volume of data generated by single-cell and ncRNA research poses significant challenges. Medical Informatics is not just a support tool but could be seen as a pivotal component in advancing clinical applications of single-cell and ncRNA research by bridging the gap between bench and bedside. The future of personalized medicine depends on our ability to harness the power of these technologies, and Medical Informatics in combination with ncRNA and single-cell technologies may stand at the forefront of this endeavor.

Background:Accurate prediction of kidney graft failure at different phases post-transplantation is critical for timely intervention and long-term allograft preservation. Traditional survival models offer limited capacity for dynamic, time-specific risk estimation. Machine learning (ML) approaches, with their ability to model complex patterns, present a promising alternative. Methods:This study developed and dynamically evaluated phase-specific ML models to predict kidney graft failure across five post-transplant intervals: 0-3 months, 3-9 months, 9-15 months, 15-39 months, and 39-72 months. Clinically relevant retrospective data from deceased donor kidney transplant recipients were used for training and internal validation, with performance further confirmed on a blinded external validation cohort. Predictive performance was assessed using ROC AUC, F1 score, and G-mean. Results:The ML models demonstrated varying performance across time intervals. Short-term predictions in the 0-3 month and 3-9 month intervals yielded moderate accuracy (ROC AUC = 0.73 ± 0.07 and 0.72 ± 0.04, respectively). The highest predictive accuracy observed in mid-term or the 9-15-month window (ROC AUC = 0.92 ± 0.02; F1 score = 0.85 ± 0.03), followed by the 15-39-month period (ROC AUC = 0.84 ± 0.04; F1 score = 0.76 ± 0.04). Long-term prediction from 39 to 72 months was more challenging (ROC AUC = 0.70 ± 0.07; F1 score = 0.65 ± 0.06). Conclusion:Phase-specific ML models offer robust predictive performance for kidney graft failure, particularly in mid-term periods, supporting their integration into dynamic post-transplant surveillance strategies. These models can aid clinicians in identifying high-risk patients and tailoring follow-up protocols to optimize long-term transplant outcomes.

The emergence of collaborations, which standardize and combine multiple clinical databases across different regions, provide a wealthy source of data, which is fundamental for clinical prediction models, such as patient-level predictions. With the aid of such large data pools, researchers are able to develop clinical prediction models for improved disease classification, risk assessment, and beyond. To fully utilize this potential, Machine Learning (ML) methods are commonly required to process these large amounts of data on disease-specific patient cohorts. As a consequence, the Observational Health Data Sciences and Informatics (OHDSI) collaborative develops a framework to facilitate the application of ML models for these standardized patient datasets by using the Observational Medical Outcomes Partnership (OMOP) common data model (CDM). In this study, we compare the feasibility of current web-based OHDSI approaches, namely ATLAS and “Patient-level Prediction” (PLP), against a native solution (R based) to conduct such ML-based patient-level prediction analyses in OMOP. This will enable potential users to select the most suitable approach for their investigation. Each of the applied ML solutions was individually utilized to solve the same patient-level prediction task. Both approaches went through an exemplary benchmarking analysis to assess the weaknesses and strengths of the PLP R-Package. In this work, the performance of this package was subsequently compared versus the commonly used native R-package called Machine Learning in R 3 (mlr3), and its sub-packages. The approaches were evaluated on performance, execution time, and ease of model implementation. The results show that the PLP package has shorter execution times, which indicates great scalability, as well as intuitive code implementation, and numerous possibilities for visualization. However, limitations in comparison to native packages were depicted in the implementation of specific ML classifiers (e.g., Lasso), which may result in a decreased performance for real-world prediction problems. The findings here contribute to the overall effort of developing ML-based prediction models on a clinical scale and provide a snapshot for future studies that explicitly aim to develop patient-level prediction models in OMOP CDM.

BackgroundThe COVID-19 pandemic revealed a need for better collaboration among research, care, and management in Germany as well as globally. Initially, there was a high demand for broad data collection across Germany, but as the pandemic evolved, localized data became increasingly necessary. Customized dashboards and tools were rapidly developed to provide timely and accurate information. In Saxony, the DISPENSE project was created to predict short-term hospital bed capacity demands, and while it was successful, continuous adjustments and the initial monolithic system architecture of the application made it difficult to customize and scale.MethodsTo analyze the current state of the DISPENSE tool, we conducted an in-depth analysis of the data processing steps and identified data flows underlying users' metrics and dashboards. We also conducted a workshop to understand the different views and constraints of specific user groups, and brought together and clustered the information according to content-related service areas to determine functionality-related service groups. Based on this analysis, we developed a concept for the system architecture, modularized the main services by assigning specialized applications and integrated them into the existing system, allowing for self-service reporting and evaluation of the expert groups' needs.ResultsWe analyzed the applications' dataflow and identified specific user groups. The functionalities of the monolithic application were divided into specific service groups for data processing, data storage, predictions, content visualization, and user management. After composition and implementation, we evaluated the new system architecture against the initial requirements by enabling self-service reporting to the users.DiscussionBy modularizing the monolithic application and creating a more flexible system, the challenges of rapidly changing requirements, growing need for information, and high administrative efforts were addressed.ConclusionWe demonstrated an improved adaptation towards the needs of various user groups, increased efficiency, and reduced burden on administrators, while also enabling self-service functionalities and specialization of single applications on individual service groups.

The integration of artificial intelligence (AI) algorithms into clinical practice holds immense potential to improve patient care, but widespread adoption still faces significant challenges, including interoperability issues. We propose a concept for the agile development of an IT platform to integrate AI-based applications into clinical workflows for a use case in ophthalmology.