Developing and evaluating prognostic models using data from electronic health records
Benjamin French, PhD Radiation Effects Research Foundation
Clinical research across a range of specialties has focused on identifying prognostic models to predict future morbidity and mortality. Accurate models could be used to counsel patients more effectively and to guide personalized treatment strategies over time. As interest in individualized prediction has grown, so too has the availability of large-scale clinical information systems. Detailed data from electronic health records facilitate the development and evaluation of prognostic models, but comprehensive analyses based on these data require consideration of several statistical challenges, such as outcome misclassification and competing risks. In this talk, I will provide an overview of statistical approaches for developing and evaluating prognostic models that maximize the utility of electronic health records data. I will focus on methods that address prospective data in which multiple events occur sequentially for each patient, such as a series of hospital admissions for heart failure, and incorporate event severities that are linked to the times at which patients experience an event, such as length of hospital stay among admitted patients. Such approaches are expected to improve the accuracy and clinical relevance of prognostic models..