Machine learning algorithm for predicting 30-day mortality in patients receiving rapid response system activation: A...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Clinical Prediction Models

Background:

• Rapid Response System (RRS) activation is crucial for in-hospital critical events.
• Accurate prediction of mortality following RRS activation can improve patient outcomes.
• Existing prediction models may not fully leverage advanced machine learning techniques.

Purpose of the Study:

• To evaluate the accuracy of machine learning algorithms in predicting 30-day mortality in patients receiving RRS activation.
• To identify key variables contributing to mortality prediction.
• To compare the performance of different machine learning classifiers.

Main Methods:

• Retrospective cohort study using nationwide data from the In-Hospital Emergency Registry in Japan.
• Multiple imputation techniques for handling missing data.
• Development and comparison of four machine learning models: LightGBM, XGBoost, random forest, and neural network.
• Analysis of 52 variables including patient characteristics, RRS details, and hospital capacity.

Main Results:

• The LightGBM algorithm achieved the highest predictive accuracy for 30-day mortality (AUC = 0.860).
• Key predictors identified included hospital capacity, site of incidence, code status, and abnormal vital signs within 24 hours.
• The study analyzed data from 4,997 patients across 34 hospitals.

Conclusions:

• Machine learning, particularly LightGBM, offers a highly accurate approach for predicting 30-day mortality post-RRS activation.
• Hospital capacity and patient clinical status are significant factors in mortality prediction.
• These findings can inform clinical decision-making and resource allocation for RRS interventions.