The temporal dynamics of the patient journeys are important to make better clinical decisions as well as predict the outcomes. Longitudinal health records record time-stamped progressions of diagnoses, treatment, and other clinical happenings that show disease progression patterns and a reaction to treatment. The aim of the present paper is to design a deep learning methodology to capture temporal properties of patient journeys using longitudinal electronic health records (EHRs). Based on the recurrent neural networks (RNNs), temporal convolutional networks (TCNs), and transformer-based structures, we could obtain the temporal dependencies and predictive features of disease onset, hospital readmission, and mortality. To assess our models we consider a publicly available healthcare dataset, which shows better results at predicting and sequence representation compared to conventional models. We found that deep learning temporal models present an effective solution to proactive and personalized health care.
Veerendra Nath Jasthi (Fri,) studied this question.