Serious adverse drug reactions are an important cause of hospitalization and can result in the withdrawal of licensed drugs. Genetic variation has been shown to influence adverse drug reaction susceptibility, and predictive genetic tests have been developed for a limited number of adverse drug reactions. The identification of patients with adverse drug reactions, obtaining samples for genetic analysis and rigorous evaluation of clinical test effectiveness represent significant challenges to predictive genetic test development. Using the example of serious drug-induced liver injury, we illustrate how a database of routinely collected electronic health records (EHRs) could be used to overcome these barriers by facilitating rapid recruitment to genome-wide association studies and supporting efficient randomized controlled trials of predictive genetic test effectiveness.