Impact of incomplete patient records on predictive models that use electronic health record (EHR) data

Electronic health records (EHR) contain a plethora of information that might be used to improve the clinical and operational decision-making processes. However, the databases that contain EHR data are typically characterized by incomplete patient information. In this paper, we investigate the impact of incomplete and/or missing EHR data on predictive modeling and the clinical recommendation. Using a multi-model approach, we show that a medical recommendation might be made even based on incomplete information, but its quality is a function of the characteristics of the patient population and of resource availability.