AttDMM: An Attentive Deep Markov Model for Risk Scoring in Intensive Care Units

Clinical practice in intensive care units (ICUs) requires early warnings when\na patient's condition is about to deteriorate so that preventive measures can\nbe undertaken. To this end, prediction algorithms have been developed that\nestimate the risk of mortality in ICUs. In this work, we propose a novel\ngenerative deep probabilistic model for real-time risk scoring in ICUs.\nSpecifically, we develop an attentive deep Markov model called AttDMM. To the\nbest of our knowledge, AttDMM is the first ICU prediction model that jointly\nlearns both long-term disease dynamics (via attention) and different disease\nstates in health trajectory (via a latent variable model). Our evaluations were\nbased on an established baseline dataset (MIMIC-III) with 53,423 ICU stays. The\nresults confirm that compared to state-of-the-art baselines, our AttDMM was\nsuperior: AttDMM achieved an area under the receiver operating characteristic\ncurve (AUROC) of 0.876, which yielded an improvement over the state-of-the-art\nmethod by 2.2%. In addition, the risk score from the AttDMM provided warnings\nseveral hours earlier. Thereby, our model shows a path towards identifying\npatients at risk so that health practitioners can intervene early and save\npatient lives.\n