The Machine Learning Support for organ Donation Matches (MLSDM) project aims at developing machine learning-based technological solution to improve matchmaking between donors and recipients of organ transplantation and to broaden the pool of potential donors with more successful DCD (determined circulatory death) attempts. More generally, the ultimate objective of the solution is to provide better survivability and wellbeing to organ recipients. Getting access to large datasets on organ donations and transplantations is paramount to build and train ML models that can meet the overall objective of the solution. As the United States are renowned for the extensive historical data on the health of their population, BI Expertise has reached out to the SRTR (United States’ (US) Department of Health and Human Services’ Scientific Registry of Transplant Recipients) to access relevant data. SRTR datasets are also commonly used in research, which is helpful benchmark the performance of new models or solutions.
The solution was implemented entirely in Microsoft Azure public cloud leveraging Azure Machine Learning for the core components and data management. Azure Security Center and Azure Sentinel were also activated to secure the end-to-end solution and very sensitive medical information. Ethical challenges were also addressed in relation with predictive modeling and potential biases. As in many ethical challenges, there is no one straightforward best answer and participation of multiple stakeholders during prototyping helped build consensus around a good enough course of action for fairness and social acceptability of the solutions brought to these challenges.