Development and validation of a machine learning-based model for post-sepsis frailty.

Area of Science:

• Critical Care Medicine
• Gerontology
• Artificial Intelligence in Healthcare

Background:

• Post-sepsis frailty is a prevalent and significant clinical challenge.
• Predicting the development of frailty after sepsis remains difficult.

Purpose of the Study:

• To develop and validate a machine learning model for predicting frailty in sepsis survivors.
• To identify key clinical variables for predicting post-sepsis frailty.

Main Methods:

• A deep learning model was developed using data from a Korean national multicenter prospective observational cohort (September 2019–December 2021).
• Ten routinely collected sepsis variables were used to train six machine learning models, including conventional and neural network approaches.
• Model performance was evaluated using cross-validation and temporal validation, including external validation with a COVID-19 dataset.

Main Results:

• 8518 sepsis patients were analyzed; 64.1% were identified as frail at discharge.
• The Extreme Gradient Boosting (XGB) model demonstrated the highest performance with an AUC of 0.8175 and accuracy of 0.7414.
• External validation confirmed the XGB model's generalizability, achieving an AUC of 0.7668.

Conclusions:

• A machine learning-based model effectively predicts frailty after sepsis.
• High predictive performance was achieved using a limited set of baseline clinical parameters.