A clinical language model trained on unstructured clinical notes from the electronic health record enhances prediction of clinical and operational events.