|
Abstract:
|
In kidney paired donation (KPD), transplant candidates, paired with their incompatible donor, are grouped together in an effort to uncover transplant opportunities via donor exchange. The aim is typically to maximize the number of transplants. Exchanges, represented by the largest cycle-packing in the underlying KPD network, are selected in a greedy manner for transplantation. However, KPD is inherently temporal: nodes (pairs) and edges (potential transplants) are added and removed over time. Thus, the system may be sub-optimal; certain pairs may be disadvantaged as others join KPD and form exchanges quickly. We propose to better account for the dynamics of KPD via a tensor of mode-3, representing the evolution of the underlying KPD network over time. We use non-negative decomposition techniques to uncover latent factors that allow nodes to be clustered based on their propensity to form exchanges. We update the latent factors in an online manner as nodes and edges are added or removed. Simulations suggest adequate separation of pairs based on their propensity to form exchanges, which can inform an allocation system that better emphasizes balance between utility and equity in KPD.
|
