JMIR Publications

ABSTRACT

Background:

Recent advances in hard- and software enabled the use of Artificial Intelligence (AI) algorithms for analysis of complex data in a wide range of daily-life use cases. We aim to explore the benefits of applying AI to a specific medical use from nephrology: post kidney transplantation risk prevention. On the one hand, the development of accurate AI models for nephrology requires access to high-quality real-world nephrology data. On the other hand, these data are highly sensitive and require specific legal and technical protection measures.

Objective:

The German-Canadian NephroCAGE consortium aims to develop and evaluate specific processes, software tools, and methods to (a) combine transplant data of 8k+ cases over the past decades from leading nephrology centers in Germany and Canada, (b) implement specific measures to protect sensitive transplant data, and (c) use multi-national data as foundation for developing high-quality AI models for prediction of post-transplant outcomes.

Methods:

To protect sensitive transplant data addressing (a) and (b), we aim to implement a decentralized NephroCAGE Federated Learn- ing Infrastructure (FLI) upon a private blockchain. The NehproCAGE FLI enables a switch of paradigms: Instead of pooling sensitive data into a central database for analysis, our NephroCAGE FLI enables the transfer of Clinical Prediction Models (CPMs) to clinical sites for local data anal- ysis. Thus, sensitive transplant data resides protected in their original sites whilst the comparable small algorithms are exchanged instead. For (c), we will compare the performance of selected AI algorithms, e.g. random forest and extreme gradient boosting, as the basis for building CPMs for specific post-transplant outcomes. The CPMs will be trained on donor and recipient data from ret- rospective cohorts of kidney transplant patients to predict patient-specific risks for developing severe short- and long-term post-transplant endpoints, e.g. graft failure or mortality.

Results:

For the first time, we will (a) combine kidney transplant data from nephrology centers in Germany and Canada, (b) implement federated learning as a foundation to use such real-world transplant data as basis for training of CPMs in a privacy-preserving way, and (c) develop a learning software system to investigate population specifics, e.g. to understand population heterogeneity, treatment specificities, and individual impact on selected post-transplant outcomes.

Conclusions:

Existing longitudinal data of historic transplantations offer a great opportunity for development of high-quality AI-based CPMs for severe clinical endpoints. The NephroCAGE consortium is the first of its kind, who targets the use of transplant data from multiple nephrology centers across Germany and Canada. The foundation for a secure and privacy-preserving way of handling of such sensitive data will be the NephroCAGE FLI, which will be designed for this dedicated use case. If the NephroCAGE demonstrator is successful, the project findings and developed infrastructure components can open up a new way of medical collaboration assisted by modern AI-based data analytics tools also for future projects.