Algorithmic Fairness in AI Surrogates for End-of-Life Decision-Making