Decentralised approaches to infectious disease modelling could support real-time epidemic and pandemic response in a secure and equitable way, a Nature Medicine review article argues.
According to researchers at PSI, the University of Oxford and several institutions worldwide, federated approaches, which do not require movement of raw data, comply with international laws and ethical standards while enabling sustainable surveillance efforts.
Study co-author Professor Moritz Kraemer (PSI, Oxford Martin School and Department of Biology) commented: “Global health security will depend on the successful adoption of federated AI infrastructures for disease surveillance.”
Currently, access to high-quality data is usually restricted due to privacy or proprietary reasons, or may be disincentivised due to data sovereignty concerns.
Within federated approaches, in contrast, the data never leaves its local server. What is shared is meaningful data analytics, minimising risks related to data privacy, ownership and security.
“Valid concerns around privacy are growing and are not going away. They are a reality we must accept and work within”, said co-author Professor Samir Bhatt of Imperial College London and the University of Copenhagen.
“Federated learning is key to enabling data sharing and collaboration within the limits of what is possible across jurisdictions. Ultimately, it will help us detect outbreaks faster and fulfil our collective promise to strengthen preparedness.”
Leveraging a wide range of data across locations is crucial to understanding how infectious outbreaks emerge and spread. Unlike federated approaches, centralised approaches consolidate and host data from various sources in a single place, and are subject to limitations. Mobility data, for example, is often held by private companies, with little or no margin for data sharing; and sharing healthcare data carries risks related to patient identification and stigma.
Such difficulties come at the cost of rapid analysis, and have historically delayed responses to outbreaks such as COVID-19, mpox, Ebola, and Marburg. Federated approaches could enable secure use of multiple types of data, helping researchers address key epidemiological questions during outbreaks. Beyond tracking disease progression, these approaches could help assess the effectiveness of vaccines and public health interventions.
Commenting on the key role federated learning can play in global health, co-author Professor Samuel V. Scarpino of Northeastern University in Boston said: “Providing global access to key datasets related to mobility and genomics can power a more effective, equitable approach global biosecurity.”
For federated approaches to be fully leveraged, the authors argue, global data sharing ecosystems will need to be strengthened, ensuring data interoperability, attribution and transparency. Existing initiatives, such as the Observational Health Data Sciences and Informatics (OHDSI) programme provide frameworks that can be adapted and expanded to facilitate data standardisation across different protocols.