Powered by Centre Hospitalier Universitaire de Liège
TRUMPET is an European research and innovation action activity that focuses on TRUstworthy Multi-site Privacy Enhancing Technologies. Its consortium includes partners from Belgium, France, Italy, Spain, and the United Kingdom. Among these is EHTEL member, the Centre Hospitalier Universitaire de Liège (CHU Liège) which is a Belgian university hospital.
Federated learning provides a number of security guarantees towards compliance with the application of the 2016 General Data Protection Regulation (GDPR). However, at least two major challenges have been identified.
First, federated learning was originally positioned as a major privacy-preserving innovation. However, later research has cast doubt on the strength of privacy protection it provides. One of federated learning’s main threats involves inference attacks that aim to re-identify data subjects as a result of parameter updates embedded in its model of artificial intelligence (AI).
Second, in the age of AI and big data, research is often limited by the lack of availability and accessibility of high quality datasets. AI-grade data quality includes many dimensions that range from its sheer quantity to its semantic interoperability. Several others include: its data cleanliness, conformance to findable, accessible, interoperable, and re-usable (FAIR) principles, uniformity of statistical distribution, richness of attributes, and coverage of population(s) in specific domains/applications. (Datasets of adequate quality are often not available from a single source. Instead, they need to be assembled from subsets owned by different organisations, which may have a variety of access policies and often grant no access at all to entities outside of themselves.) High quality datasets may therefore apply and implement access policies that are far more stringent than those required by the actual GDPR.
Goal and aims
The goal of the TRUMPET project is twofold. It is to research and develop novel privacy enhancement methods for federated learning, and to deliver a federated AI service platform for researchers. The platform will enable the analysis of siloed, multi-site, cross-domain, cross-border European datasets with privacy guarantees that exceed the requirements of the GDPR.
The TRUMPET project aims to:
- collaborate with relevant initiatives and agencies such as the European Union Agency for Cybersecurity (ENISA) or the European Cybersecurity Competence Centre (ECCC),
- engage with the relevant H2020 and Horizon Europe initiatives, and
- develop a strategy to promote the European adoption of the TRUMPET privacy metric for the certification of GDPR compliance of federated learning implementations.
CHU Liege’s role in TRUMPET
TRUMPET is a ‘dual-sided’ platform that reflects the needs of both supply side and demand side actors. On the demand side, the key end users are researchers; on the supply side, they are the data owners.
CHU Liège represents both types of end users that use the platform: the hospital is both a health data provider and also includes clinical researchers. As a health data provider, CHU Liège has longstanding experience in this field through its integration in regional health hubs, the Trinetx network (a global health research network that shares real world data), patient-facing-apps (e.g., Andaman7), and with other research programmes.
In the local TRUMPET node, CHU Liège will register de-identified curated datasets that cover two use cases in the field of radiotherapy:
- Risk ratios of dose volume histograms towards late side-effects of intensity-modulated radiotherapy in head and neck cancer patients.
- Survival prediction for the eligibility to stereotactic body radiotherapy, a radiotherapy method that delivers accurate, high radiation doses in just one or few treatments.
As an academic hospital, CHU Liège’s bio-statistics unit provides support for statistical methodology and analysis both to randomised clinical trials and retrospective cohort studies. In the TRUMPET project, CHU Liège will provide domain expertise in radiotherapy cancer treatment and bio-statistics for the specification and validation of the cohorts’ datasets, the statistical models, and data owner and researcher dashboards.
What’s in TRUMPET for EHTEL members?
Today, there is an increased interest in the use of real world data to support the continuum of evidence generation for innovative medicines.
It is expected that real world data should play three roles in innovative medicines: it should enable the generation of additional evidence after the medicine's launch, support dynamic price-setting in relation to the value of medicines, and optimise appropriate use of medicines in daily practice.
The collection of real world data has many benefits that advance further than conventional randomised control trials. During the market usage phase of innovative medicines, real world data enables the provision of evidence on real world usage of medicines and treatments as well as on longer-term clinical benefits and harm. Before drug development, real world data will help identify those patients who will benefit most from treatments as well as those who will not benefit. Thus, it provides a large number of opportunities to compare multiple alternative interventions.
In the context of the proposed upcoming European Health Data Space Regulation, TRUMPET is developing both a privacy-enhanced federated platform and a privacy metric that could act as enablers for the uptake of federated learning in the clinical sector and as a secure processing environment for real world data.
In TRUMPET, it is expected that such privacy-related guarantees will address barriers for the adoption of federated learning both by natural persons and by health organisations (dependent on trust), and will foster collaboration between researchers and data custodians.
EHTEL members will be able to understand how improved federated learning can overcome current barriers and can play a role within the future European Health Data Space.
The TRUMPET project has written a paper for an upcoming conference that describes its work in detail.
The TRUMPET project has received funding from a Research and Innovation action activity under Horizon Europe Framework with Grant Agreement No.101070038.