Development of the STRONG-AYA federated learning-based adolescent and young adult cancer information portal: a co-creation process with adolescents and young adults with lived experience of cancer (Preprint)

BACKGROUND Adolescent and young adult (AYA) cancer represents a unique sub-domain in oncology, as AYAs experience cancer in a starkly different way from children, and older adults. The diverse and complex needs of AYAs result in an increasing impact and burden upon individuals and society. Nevertheless, widely accessible information and data insights are lacking for AYA stakeholders. In part, this is because widespread dissemination of such information is difficult and confined to local regions. Federated learning (FL) enables large-scale multi-centre data analysis without individual-level information disclosure. The STRONG-AYA Initiative combines precisely these components by setting up a FL-ecosystem including a vast amount of AYA cancer relevant data. Such developments present promising opportunities to tackle issues such as data availability and timeliness, creating a rich information landscape of AYA cancer for its diverse stakeholders. OBJECTIVE To harness the opportunities introduced by FL we designed a FL-based AYA cancer information portal, aiming to transform AYA health information dissemination across the AYA landscape. METHODS We described the portal’s setting and variety of target audiences (stakeholder groups) for an AYA cancer portal, based on a consortium-wide consultation. With representatives of these stakeholder groups, we conducted unstructured interviews to obtain an initial concept of the requirements of this portal. The portal was designed to meet their requirements in bi-weekly iterative co-creation events between an AYA with lived experience and a researcher. The design was then evaluated using a think-aloud method, by a larger group of AYAs with lived experience and healthcare professionals. Finally, we developed a simplified proof-of-concept of our portal. RESULTS Key stakeholders, including AYAs, and healthcare professionals, expressed a strong interest in an AYA cancer information portal. They acknowledged the benefits of harnessing FL insights and the specific opportunities provided by our FL methods. Our initial designs were well received but improvements were highlighted in information findability, usage guidance, and textual understandability. In our final version, reviewers felt their voices were heard and evaluated the resulting portal favourably. They did however raise remaining concerns about the need for extensive continuous quantitative user experience testing and associated adaptations and updates required throughout the portal’s lifetime. Our proof-of-concept lastly demonstrated it was possible to: extract relevant information from the STRONG-AYA FL-infrastructure without sacrificing data privacy, retrieve up-to-date information in an automated yet curated manner, and display analytical findings in a user-friendly manner. CONCLUSIONS By means of co-creation, we designed an AYA-specific cancer information portal, that was welcomed by stakeholders, and regarded as user-friendly, balanced in its design, and utilising the strengths of STRONG-AYA’s FL-ecosystem. Simultaneously, we demonstrated that the realisation of such a FL-based portal is feasible using existing infrastructure.