A Structure-Based Scoring Policy for Patient Stratification on the U.S. Heart Transplant National Waiting List

Nowadays, mainstream organ transplantation services across the world are facing the challenge of patient stratification. Most policies rely on either specific patient attributes or if-else questions to form “scores” or “types”, respectively. The rigidity often deviates the whole system from optimality as the medical indices of patients evolve. Consequently, policymakers are forced to change the criteria every few years. In this paper, we develop a structure-based scoring policy for patient stratification with heart transplant history from the United Networks for Organ Sharing (UNOS) data. Specifically, our approach adopts a joint learning and optimization framework. It utilizes both the background knowledge of the national waiting list and massive patients' autonomous choices of transplant centers. Our results from thoracic transplantation data show high consistency with the current criteria of UNOS. Though not decreasing the complexity to perform, the algorithm merging learning and optimization has the potential to maintain both causality and precision. Most importantly, it is adaptive to regional disparity, policy variance, and possible medical developments.