Large electronic health record- or population-based datasets form the basis for many diagnostic error studies. This article raises the issue of data-driven feedback loop failures which occur when disease incidence, presentation, and risk factors are misunderstood in research and, therefore, future medical practice. For example, men presenting with "classic" symptoms of heart attack are more frequently targeted for evaluation than women with "atypical" symptoms, thereby resulting in underdiagnoses of heart attack in women and underrepresentation in the evidence base.