Sepsis remains a leading cause of mortality and long-term disability in intensive care, with outcomes varying widely across hospitals despite advances in technology and repeated guideline updates. One persistent barrier is the fragmentation of clinical data, which limits the ability of clinicians, researchers and health systems to learn systematically from real-world practice. The European Health Data Space (EHDS) introduces a federated, privacy-preserving framework designed to connect intensive care units (ICUs) across Europe without transferring patient-level data. Sepsis provides a useful clinical model for this approach because of its heterogeneity, high resource use and ongoing gaps in standardisation and outcomes. By enabling structured analysis of distributed ICU data and supporting complementary use of synthetic data, the EHDS aims to transform how critical care knowledge is generated, shared and applied while maintaining strong ethical and legal safeguards.
Sepsis and the Limits of Isolated ICU Data
Sepsis imposes a substantial clinical and economic burden worldwide and across Europe. It affects tens of millions of individuals each year and contributes to a significant proportion of global deaths. In European health systems, the impact extends beyond acute hospitalisation. Treatment costs vary widely between countries and are compounded by long-term physical, cognitive and psychological impairments among survivors, many of whom require ongoing care. These outcomes underline the need for more effective learning from routine ICU data rather than reliance on isolated registries or time-limited research projects.
Must Read: Cancer Patients’ Views on AI and Data Protection
Clinical heterogeneity is central to the challenge. Sepsis can present as sudden circulatory collapse in a previously healthy person or as a gradual deterioration in older patients with multiple comorbidities. This variability complicates diagnosis, limits the effectiveness of uniform protocols and contributes to persistent differences in outcomes between hospitals. Smaller datasets and single-centre studies struggle to capture this complexity, reducing the generalisability of findings and the reliability of advanced analytics. Artificial intelligence and machine learning offer tools to detect patterns in physiological data, biomarkers and treatment responses, but their value depends on access to large, representative and ethically governed datasets. Without broader connectivity, models risk reflecting local practice rather than wider critical care reality.
Federated Data and the EHDS Model
The EHDS proposes a federated data model that allows hospitals to retain full control over their patient records while enabling standardised analyses to be carried out locally. In this approach, clinical or research questions are translated into common queries that are executed within each institution’s secure environment. Only anonymised and aggregated results are shared, supporting cross-border benchmarking and surveillance while complying with European data protection requirements. National health data access bodies are expected to oversee governance, privacy and ethical compliance, with opt-out mechanisms preserving patient autonomy.
This model marks a shift from manually curated registries towards a long-term learning infrastructure based on routinely collected data. By minimising data transfer and administrative burden, federated systems could enable participation by hospitals with differing resources and digital maturity. Aggregated outputs can be used to monitor trends, compare practices and support quality improvement without exposing identifiable information. The framework is designed to support continuous learning rather than retrospective analysis alone, aligning clinical care, research and education within a shared ecosystem.
A comparison with the United States Medical Information Mart for Intensive Care (MIMIC) database highlights the potential advantages of the EHDS approach. MIMIC has enabled influential research and education but reflects practice from a single US centre with specific case mix, coding systems and organisational structures. European ICUs differ widely in capacity, staffing and admission thresholds, with substantial variation between countries. These differences limit the transferability of single-centre models and reinforce the need for a multi-country framework that preserves variability while making it interpretable through common standards and terminologies.
Synthetic Data, Education and Practical Limits
Synthetic data are presented as a complementary tool within this landscape. Generated statistically from real datasets, synthetic data preserve relationships between variables while containing no identifiable patient information. This makes them suitable for workflow testing, training and early-stage development of analytical tools. Synthetic cases can illustrate rare or complex sepsis scenarios and support discussion of uncertainty, trade-offs and guideline limitations. However, they cannot reproduce the full unpredictability of real patient trajectories and should not be treated as substitutes for real-world data.
Federated data also have practical limitations. Aggregation can obscure rare subgroups, coding practices differ across countries and governance processes for cross-border analyses may be slow. As a result, current applications are strongest in epidemiology, surveillance and benchmarking rather than direct bedside decision support. Educational uses, such as federated grand rounds based on anonymised trends, remain largely conceptual but illustrate how individual cases could be contextualised within broader population patterns.
Implementation challenges remain substantial. Interoperability depends on harmonised standards, consistent coding and sufficient digital maturity across institutions. Adoption of frameworks such as SNOMED CT, openEHR and FHIR varies, and national interpretations of data protection rules influence consent and secondary-use models. Privacy-preserving techniques, including differential privacy and secure multi-party computation, introduce further trade-offs between data utility and protection, particularly for rare events. These factors highlight the need for use-case specific design and transparent governance rather than uniform technical solutions.
Federated and synthetic data approaches offer a realistic pathway towards more connected and learning-oriented intensive care in Europe. For sepsis, where heterogeneity and high mortality persist, the EHDS provides a framework to link ICUs through secure, ethically governed infrastructures that support benchmarking, research and education while preserving local control of data. Federated analytics can strengthen surveillance and quality improvement, while synthetic data can enhance training and system development. Progress will depend on sustained investment in interoperability, standards and governance to ensure that distributed ICU data are translated into shared intelligence that supports better decision-making and patient outcomes in critical care.
Source: Intensive Care Medicine
Image Credit: iStock
