We describe a predictive modeling and decision support framework that integrates machine learning and optimization for personalized clinical decision support. We pilot the approach on data from a congestive heart failure patient cohort, and demonstrate the ability to predict and optimize readmission risk in a clinically meaningful manner.