Search for a command to run...
PETs are one of the Top Ten Emerging Technologies of 2024, as announced in the report published by the World Economic Forum in collabora on with Fron ers (World Economic Forum, 2024) 1 .The research topic explores the poten als of PETs to foster data u liza on across ins tu ons and na ons, to promote precision medicine and personalized healthcare. Their objec ve is to safeguard privacy and security while enabling large-scale data u liza on, processing, and analysis between different par es. By facilita ng the u liza on of vast datasets across ins tu ons and na ons, PETs are set to revolu onize mul ple sectors, including healthcare, mobility, and energy. We refer to [a] for a classifica on and defini on of PETs in light of this editorial.To protect sensi ve private data in healthcare, the step towards sta s cal tools is rather small as it closely aligns with known exper se and so ware used in the domain and they offer a fair level of data protec on.The six papers in our research topic nicely illustrate this approach and show how the different PETs relate to important developments in healthcare. This editorial highlights the synergy between the six papers and future challenges to apply PETs in the context of healthcare.If mul ple medical ins tu ons, each having their own data source, want to combine their data for e.g. survival analysis, a federated and privacy-preserving variant of machine learning avoids centralizing all data. In [3] they implement and evaluate a distribu ve protocol for Cox regression based on federated learning which shows that collabora on increases accuracy.For protec ng the privacy of a single medical data source, and overcoming data scarcity, a way to go is synthesizing this data, i.e. ar ficially genera ng data that mimic the sta s cal proper es of real data without revealing sensi ve informa on. The use of synthe c data is further inves gated for the purpose of rare disease research [6]. The poten al of the technique becomes clear and supports ensuring compliance with regula ons like GDPR and HIPAA. This is confirmed by a literature review [1] for the use of medical imaging, although there are some technical limita ons for synthe cally genera ng likely images. They also examine the role of synthe c medical image data within the European Health Data Space (EHDS), a policy ini a ve aimed at enabling secure access to health data across the EU. They provide an overview of European regulatory instruments to apply PETs, related to the ongoing debates on how synthe c imaging data can support a privacy-preserving, data-driven healthcare ecosystem in Europe.Sta s cal techniques for genera ng synthe c data resemble the ones used in differen al privacy: a method to add ar ficial noise to data records for hiding iden fiable a ributes. [5] Assesses the use of differen al privacy in tabular data and finds that it is hard to measure its added value. They consider five genera ve models for crea ng synthe c data, a key technology in healthcare where many a ributes are sensi ve. Synthe c data also facilitates the training of AI models, simula on of clinical trials and cross border coopera on [6].In [2] they suggest adap ng Differen al Privacy, such that the mechanism can account for temporal context and correla ons, making it suitable for me-con nuous data, such as electrocardiographic (ECG) records. Such an anonymiza on mechanism is necessary in the context of European Health Data Space. Synthe c data aligns with regula ons as EHDS but should avoid reiden fica on risks and conflicts with GDPR's protec on of individual rights and freedom (6). Although synthe c data can replicate real world data, the ques on remains to what extent quality can be maintained because not all nuances for high research value may not be captured in the generated data set. This means that hybrid models and a mixed method approach combining real world data with synthe c data are relevant for reasons of validity and clinical values [6].Challenges for using PETs are maintaining data quality, mi ga on bias and overcoming computa onal demands [6], and alignment with regula ons to avoid reiden fica on. Transparency of algorithms, data sta s cs, Third Party reviews and domain experts' evalua ons [6] can ensure an adequate and valid reuse of the data sets.A closer look at EHDS, and how PETs could and should support this, is presented in [4]. They use semistructured interviews to explore risks, challenges and gaps in the implementa on of PETs within EHDS, leading to a research agenda that addresses important future research ques ons. PETs should be considered directly from start, following privacy by design principles, to ensure data protec on is embedded into the development of tech-systems, PETs are no a erthought [4].The use of PETs is hindered by technological and legal aspects, the so-called LEFT aspects [b], legal, ethical, financial and technology aspects of implementa on of novel or emergent technologies in healthcare. In [4] the socio-technical and legal concerns are iden fied to apply PETs, and solu ons are discussed from the perspec ve of governance and policy and technology. The technology aspects are meaningful to ensure data quality, to enable federated systems, homomorphic encryp on, and to improve access and data minimaliza on. To address the cri cal issue of data accuracy in the EHDS, [4] proposes the integra on of zero-knowledge proofs (ZKPs) for the verifica on of AI models. The societal aspects demonstrate the risk of reinforcing dispari es rather than democra zing access to data, and the lack of clear defini ons (data holder, informed consent etc.) that can lead to misinterpreta ons across EU-member states.Investments in cloud-based pla orms to support compu ng resources for reducing computa onal overhead and emerging technologies such as Genera ve Adversarial Networks (GANs) and Varia onal Autoencoders (VAEs) should be leveraged [6].Applying PETs demand for adequate business model to understand cost/benefits in healthcare prac ces and required resources and capaci es to perform PETs [c]. A mul stakeholder collabora on between PET-experts, policymakers, industry leaders, and healthcare professionals is needed to develop standards for applying PETs, regarding quality, governance, building trust (public, healthcare) and compliance with frameworks for be er and smarter data reuse (based on all six papers).The future research agenda should explore legal, ethical financial and technology aspects that foster or hinder the implementa on of PETs in Healthcare across boundaries (ins tu ons, borders) to develop a mul -stakeholder approach for secure and safe data collabora on to develop adequate business models for sustainable implementa on of PETs [d]. Such a business model is essen al to be er understand the cost, benefits and ROI for stakeholders that are involved in the development and implementa on of PETs (ref to scoping review).Addi onal references within the editorial: Accepted papers in this research topic:1. Synthe c data in medical imaging within the EHDS: a path forward for ethics, regula on, and standards 2. Challenges of iden fica on and anonymity in me con nuous data from medical environments 3. Horizontal Federated Learning and Assessment of Cox Models 4. Secondary use under the European Health Data Space: Se ng the Scene and towards a research agenda on Privacy-Enhancing Technologies 5. Comprehensive Evalua on Framework for Synthe c Tabular Data in Health: Fidelity, U lity and Privacy Analysis of Genera ve Models with and without Privacy Guarantees 6. Synthe c data genera on: a privacy-preserving approach to accelerate rare disease research