The advance of mobile devices and communication technologies enable patients to communicate with their doctors in a more convenient way. We have developed an App that allows patients to record their symptoms and submit them to their doctors. Physicians can keep track of patients’ conditions by looking at the self-report messages. Nevertheless, physicians are usually busy and may be overwhelmed by the large amount of incoming messages. As a result, critical messages may not receive immediate attentions, and patient care is compromised. It is imperative to identify the messages that require physicians’ attention, called noteworthy messages. In this research, we propose an approach that applies text-mining technologies to identify medical symptoms conveyed in the messages and their associated sentiment orientation, as well as other factors. Noteworthy messages are subsequently characterized by symptom sentiment and symptom change features. We then construct a prediction model to identify messages that are noteworthy to the physicians. We show from our experiments using data collected from a teaching hospital in Taiwan that the different features have different degrees of impact on the performance of the prediction model, and our proposed approach can effectively identify noteworthy messages.