Predicting the Need For Vasopressors in the Intensive Care Unit Using an Attention Based Deep Learning Model

Area of Science:

• Critical care medicine
• Biomedical informatics
• Machine learning in healthcare

Background:

• Existing shock prediction models often focus on septic shock and necessitate laboratory data.
• This study addresses the need for early vasopressor requirement prediction in critically ill patients.

Purpose of the Study:

• To develop and validate a deep learning model for predicting vasopressor requirement in critically ill patients within 24 hours of intensive care unit (ICU) admission.
• To utilize only routinely collected vital signs, excluding laboratory results, for prediction.

Main Methods:

• Utilized data from the Medical Information Mart for Intensive Care III (MIMIC-III) and eICU Collaborative Research Database.
• Employed data preprocessing techniques including cohort matching, oversampling, and imputation.
• Developed a Bidirectional Long Short-Term Memory (Bi-LSTM) multivariate time series model to predict vasopressor need using serial physiological data.

Main Results:

• The Bi-LSTM model achieved an area under the curve (AUC) of 0.96 for initial prediction and 0.83 after matching for class imbalance.
• The model demonstrated strong predictive performance for vasopressor therapy needs within the first 24 hours of ICU admission.
• Key physiological determinants identified were heart rate, respiratory rate, and mean arterial pressure.

Conclusions:

• A Bi-LSTM model effectively predicts vasopressor requirements in critically ill patients during the initial 24 hours of ICU stay.
• The attention mechanism highlighted respiratory rate, mean arterial pressure, and heart rate as critical sequential predictors.
• This approach offers a promising method for early identification of patients needing vasopressor support using non-invasive vital signs.