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

TL;DR

AttDMM is a novel attentive deep Markov model that improves real-time ICU risk scoring by jointly modeling disease dynamics and states, outperforming existing methods in early warning accuracy.

Contribution

This paper introduces AttDMM, the first ICU prediction model to combine attention-based long-term disease dynamics with latent disease states.

Findings

Achieved AUROC of 0.876, 2.2% better than previous methods.

Provided earlier risk warnings, enabling timely interventions.

Validated on MIMIC-III dataset with 53,423 ICU stays.

Abstract

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