Predicting Gram-negative bloodstream infection in elderly patients after isolation of GNB from non-blood specimens: a machine learning-based tool

1

A machine learning model was developed to identify elderly inpatients at high risk of Gram-negative bloodstream infection after detecting Gram-negative bacteria in non-blood samples.
2

The study included 9,646 elderly inpatients from three centers, using LASSO regression and Boruta algorithm for variable selection.
3

The XGBoost model outperformed others with an AUC of 0.816, while logistic regression provided a simpler alternative with acceptable accuracy.
4

Seven predictive variables were identified, with maximum procalcitonin level, length of hospital stay, and max neutrophil percentage being the most important.
5

External validation in the MIMIC-IV database confirmed the model's generalizability for predicting Gram-negative bloodstream infections in elderly patients.

1