As much as 80% of valuable data in EHR systems exist in an unstructured textual form, such as clinical notes. Although challenging to exploit, the data has great potential for generating new knowledge and improving patient care and cost efficiencies.
Clinical data is, however, sensitive by nature, so ensuring privacy is paramount not only for the protection of individuals but also for maintaining trust in health systems.
Moreover, processing clinical text in one language is challenging enough; doing so across multiple languages multiplies the complexity, especially when aiming for consistent and accurate results.
Federated network approach
The Federated Health project is funded by the Nordic Ministerial Council for Digitalization, representing the collaborative digital effort across Nordic countries and beyond.
Led by the Norwegian Centre for E-Health Research and with DNV as a project partner, the project features two key strategies to create a scalable, privacy-centred platform that fosters collaboration across Denmark, Estonia, Finland, Norway, and Sweden.
Instead of centralizing data, the project will use federated learning, which allows each dataset to remain in its original location. Machine learning models are trained locally on each dataset, and only the model updates, not the raw data, are shared. This method ensures data privacy while benefiting from collective learning.
The project leverages state-of-the-art multilingual language models to process clinical texts across various languages, breaking down linguistic barriers and ensuring consistent and accurate analysis of health data across the region.
The project incorporates two ‘demonstrator’ use cases, where one aims to identify patients with implants from clinical text, as this is crucial to prevent harm to patients undergoing MRI scans, and the other aims to automatically analyse clinical text to quickly identify serious adverse drug reactions.
The benefits
With more data at their disposal, healthcare professionals can make decisions based on patterns, trends and evidence, rather than solely on individual experiences or intuition.
Processing both structured and unstructured clinical data can lead to more accurate diagnoses and personalized treatment plans and so improve patient care.
By pooling data insights from various regions and languages, researchers can identify broader health trends, leading to new medical discoveries. Access to diverse datasets enhances the robustness of research, ensuring findings are more transferable.
The data security element ensures privacy and helps institutions to comply with data protection rules.
Market potential
The federated network and its security aspect could allow for the creation of a Privacy-Preserving Virtual Data Warehouse for varied user-level data analysis and research. Users could include SMEs, hospitals, universities and similar stakeholders as well as policymakers and politicians involved in developing healthcare infrastructure.