Self-Supervised Predictive Coding with Multimodal Fusion for Patient Deterioration Prediction in Fine-grained Time Resolution

TL;DR

This paper introduces a self-supervised predictive coding approach with multimodal data fusion to improve hourly prediction of patient deterioration, enhancing accuracy especially for long-term forecasts in critical care settings.

Contribution

It presents a novel hourly prediction framework using self-supervised learning and multimodal fusion, significantly improving performance over existing methods in urgent clinical scenarios.

Findings

Significant performance gains from multimodal fusion and self-supervised regularization.

High AUROC scores achieved for mortality and vasopressor need prediction.

Notable improvements in far-future prediction accuracy.

Abstract

Accurate time prediction of patients' critical events is crucial in urgent scenarios where timely decision-making is important. Though many studies have proposed automatic prediction methods using Electronic Health Records (EHR), their coarse-grained time resolutions limit their practical usage in urgent environments such as the emergency department (ED) and intensive care unit (ICU). Therefore, in this study, we propose an hourly prediction method based on self-supervised predictive coding and multi-modal fusion for two critical tasks: mortality and vasopressor need prediction. Through extensive experiments, we prove significant performance gains from both multi-modal fusion and self-supervised predictive regularization, most notably in far-future prediction, which becomes especially important in practice. Our uni-modal/bi-modal/bi-modal self-supervision scored 0.846/0.877/0.897 (0.824/0.855/0.886) and 0.817/0.820/0.858 (0.807/0.81/0.855) with mortality (far-future mortality) and with vasopressor need (far-future vasopressor need) prediction data in AUROC, respectively.