Machine Learning-Based Multimodal Prediction of In-Hospital Cardiac Arrest in the ICU (Preprint)

Abstract

Early identification of impending in-hospital cardiac arrest (IHCA) improves clinical outcomes but remains elusive for practicing clinicians.

We aimed to develop a multimodal machine learning algorithm based on ensemble techniques to predict the occurrence of IHCA.

Our model was developed by the MIMIC-IV database and validated in the eICU-CRD database. Baseline features consisting of patient demographics, presenting illness, and comorbidities were collected to train a Random Forest (RF) model. Next, vital signs were extracted to train a long short-term memory (LSTM) model. A Support Vector Machine (SVM) algorithm then stacked the results to form the final prediction model.

Of 23,909 patients in the MIMIC-IV database and 10,049 patients in the eICU database, 452 and 85 patients had incident IHCA. Up to 13 hours in advance of an IHCA event, our algorithm maintained an area under the ROC curve above 0.78. Satisfactory results were also seen in validation from two external databases and comparison to existing warning systems.

Using only vital signs and information available in the electronic medical record, our model demonstrates it is possible to detect a trajectory of clinical deterioration up to 13 hours in advance. This predictive tool, which has undergone external validation, could forewarn and help clinicians identify patients in need of assessment to improve their overall prognosis.