Building Deep Learning Models to Predict Mortality in ICU Patients

TL;DR

This paper develops deep learning models to predict ICU patient mortality using electronic health records, addressing challenges with irregular time series data and demonstrating improved predictive performance.

Contribution

Introduces deep learning models for ICU mortality prediction utilizing the SAPS II features and handles irregular time series data effectively.

Findings

Models achieve high precision, recall, and F1 scores.

Demonstrates improved AUC-ROC over traditional methods.

Validates effectiveness on MIMIC-III dataset.

Abstract

Mortality prediction in intensive care units is considered one of the critical steps for efficiently treating patients in serious condition. As a result, various prediction models have been developed to address this problem based on modern electronic healthcare records. However, it becomes increasingly challenging to model such tasks as time series variables because some laboratory test results such as heart rate and blood pressure are sampled with inconsistent time frequencies. In this paper, we propose several deep learning models using the same features as the SAPS II score. To derive insight into the proposed model performance. Several experiments have been conducted based on the well known clinical dataset Medical Information Mart for Intensive Care III. The prediction results demonstrate the proposed model's capability in terms of precision, recall, F1 score, and area under the receiver operating characteristic curve.