We investigate the social value of medical interventions at the end of life that tend to have a high cost-benefit ratio compared to measures based on the value of a statistical life-year. We model the optimal allocation of health resources across a continuum of diseases that differ by severity and treatment options, and extend it to allow for learning spillovers between treatments. We calibrate our model to admissions to intensive care units in Switzerland. We show that cancer treatments associated with learning spillovers that decrease mortality for non-cancer patients by 1 percentage point justify a cost-benefit ratio per additional life-year of 1.78.