William Osler noted in 1893 that “If it were not for the great variability between individuals, medicine might as well be a science, not an art”. In contrast, this thesis is based on the scientific paradigm that prediction models have the potential to guide medical decisions by exploiting identifiable heterogeneity across individual patients. Prediction research focuses on the development of well performing prediction models and on the assessment of their generalizability and applicability. Several methods to measure prediction model performance across clusters of patients are proposed in PART I of this thesis. PART II contains novel methods for development and validation of models that incorporate heterogeneity of treatment effect across patients. In PART III, methods for development and validation of prediction models are applied to several case studies in cardiovascular medicine, oncology, and public health.