A machine learning model for predicting short-term outcomes after rapid response system activation

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Machine Learning in Healthcare

Background:

• Maintaining the quality of rapid response team (RRT) interventions is challenging.
• Predictive models for short-term prognosis following rapid response system (RRS) activation are limited.
• RRS activation is crucial for patient safety, necessitating better prognostic tools.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting short-term outcomes after RRS activation.
• To compare the performance of machine learning models against established scoring systems like NEWS and MEWS.
• To identify key predictors of adverse outcomes in patients receiving RRS support.

Main Methods:

• A retrospective cohort study utilizing data from the In-Hospital Emergency Registry in Japan.
• Development of logistic regression (LR), Random Forest (RF), and eXtreme Gradient Boosted Tree Classifier (XGB) models.
• Comparison of model performance using receiver-operating area under the curve (AUC), with benchmarking against NEWS and MEWS.

Main Results:

• The study included 5414 cases, with an outcome event rate of 28.4%.
• The XGB model achieved the highest AUC (0.798), outperforming RF (0.796), LR (0.785), NEWS (0.696), and MEWS (0.660).
• Key predictors identified by the XGB model included doctor activation, hypotension, and oxygen usage.

Conclusions:

• The developed XGB model represents the first machine learning approach for predicting short-term prognosis post-RRS activation.
• This model shows potential to significantly aid RRT decision-making and improve patient care.
• The findings highlight the utility of machine learning in enhancing RRS effectiveness.