Federated learning is a type of machine learning that trains algorithms without the need to share data itself. It’s an emerging AI model that is “privacy-friendly” since it does not require multiple partners with data to share or centralize it. What this means is that STRONG AYA can gain insight from the data held at our participating centres whilst adhering to local and pan-European data privacy regulations like GDPR.

The particular paradigm STRONG AYA is following is something called the Personal Health Train. The Personal Health Train allows the STRONG AYA consortium to ask a research question that an algorithm can execute (our “train” in the metaphor) which is sent along the “tracks” connecting each of the centres. Each of the centres have FAIR (findable, accessible, interoperable, reusable) data “stations”. These stations are secure so only a permitted type of question or train sent by a permitted or trusted user can use the tracks and stop at the station, and each station has rules that dictate the type of question ‘trains’ that can stop. Control over these rules (and by extension the data) remains with each of the participating centres. As the ‘train visits each “station”, it gathers aggregate data insight from pseudonymised data. The passengers on our information train are NOT data, but rather insights from the data.

You can check out a video on the personal health train for more information

In this way, STRONG AYA can answer important research questions relevant to AYAs with cancer in a way that is privacy informed.