Abstract: In the era of EHRs, it is possible to examine the outcomes of decisions made by doctors during clinical practice to identify patterns of care that are optimal—generating evidence based on the collective practice of experts. We will discuss methods that transform unstructured patient notes into a de-identified, temporally ordered, patient-feature matrix. With the resulting high-throughput data, it is possible to monitor for adverse drug events, profile specific drugs, identify off-label drug usage, uncover ‘natural experiments’ and generate practice-based evidence for difficult-to-test clinical hypotheses. We will review four use-cases to illustrate the usefulness of learning practice-based evidence from unstructured clinical notes. By examining the frequency and co-frequency of drug and disease mentions, we can detect associations among drugs and their adverse events about 2 years before an alert is issued as well as learn the prevalence of known drug-drug interactions. Using the patient feature matrix as well as prior knowledge about drugs, diseases, and known usage, we identify novel off-label uses; ranked on the basis of drug safety and cost. We uncover a natural experiment—a subset of CHF patients who were prescribed Cilostazol despite its black box warning—and profile its safety in this high-risk group of patients. We will discuss a use case of testing a clinical hypothesis about a possible association between allergic conditions and the complication of chronic uveitis in juvenile idiopathic arthritis (JIA) patients.