Aligning Data-Driven Predictors with Allocation: A Decision-Focused Approach to Survival Analysis
Researchers have developed a new decision-focused learning approach for survival analysis, aiming to better align predictive models with their downstream allocation tasks. This method optimizes for Normalized Discounted Cumulative Gain (NDCG) instead of traditional metrics like the C-index, which can lead to suboptimal outcomes in high-stakes scenarios such as organ allocation. By applying this framework to historical heart transplant data, the approach significantly improved NDCG scores, potentially leading to substantial gains in life years annually. AI
IMPACT This new framework could improve decision-making in critical allocation systems by better aligning predictive models with real-world outcomes.