Data from the electronic health record (EHR) are frequently used to generate predictive models, or as the basis of artificial intelligence tools, to support medical decision making and health care operations. There are fundamental limitations based on the quality of the underlying data. Missingness and errors are unlikely to be random, and shouldn’t be ignored. This is true both for developing the models and then for using these tools to make decisions at the individual patient level. In this session, recorded at Translational Science 2019, the presenter covers the principal challenges, how errors propagate through modeling processes, and how they translate to sub-optimal decision making. Potential solutions and opportunities for research are discussed.
Speaker: Jareen Meinzen-Derr, Cincinnati Children’s Hospital Medical Center