Improving the performance of predictive models induced from medical data can help clinicians in making more accurate predictions that in turn may lead to better patient outcomes, greater safety and reduced healthcare costs. Most predictive modeling techniques applied to the medical domain induce population-wide models that are expected to perform well on average on all future patient cases. We hypothesize that learning patient-specific models from data that are optimized to predict well for a given patient case can yield better predictive performance. We discuss a lazy classification algorithm developed by the machine learning community as an example of a patient-specific algorithm and our proposed extensions to it.