World Health Organization estimates of health care expenditure reveal a global trend of increasing costs, and health care systems need to become more efficient at treating patients to slow this trend. Incentives are in place to develop information-based health care systems, and I claim that using machine learning tools in medicine will lead to improvements in patient care. My work demonstrates new methods to improve collaboration between machine learning experts and clinicians, and new methods for modeling individual responses to treatment. My work in collaboration with clinical experts involves the adaptation of machine learning models to address the challenging task of identifying benign breast cancer biopsies that cannot be definitively diagnosed. I first adapt an inductive logic programming learner to prefer rules that do not misclassify malignant cases, and show promising results that both adhere to the clinical objective and provide insight into the task. I later present a framework for collaboration between clinical and machine learning experts, leveraging clinician expertise to build and refine a model that meets the conservative objective of missing no malignant cases. My work on estimating individual responses to treatment takes lessons from the marketing domain, applying uplift modeling to two primary medical tasks. One task is to identify patients at greater risk of heart attack due to treatment with COX-2 inhibitors, and another is understanding characteristics of in situ breast cancer specific to older women. I first present a statistical relational learner that constructs Bayesian networks to maximize area under the uplift curve (AUU), and show that the learned networks capture clinically-relevant characteristics of indolent, in situ breast cancer. I next present a support vector machine for maximizing AUU and show promising results on both the COX-2 inhibitor and breast cancer tasks, as well as a synthetic marketing task. Finally, I present a collaboration showing strong evidence that machine learning for individualized treatment effect estimation improves upon current methods in multiple ways. Overall, I present multiple works that demonstrate improved clinical collaboration and new methods for modeling individual responses to treatment within machine learning.