JMIR Publications

ABSTRACT

Background:

The early identification of serious bacterial infections is important in clinical practice.

Objective:

This study aimed to establish and validate a clinically applicable model to identify serious bacterial infections in patients with infective fever.

Methods:

Clinical data from a 2200-bed teaching hospital were retrospectively obtained between January 2013 and December 2021. Two models (an early admission assessment model and a risk assessment model within 24 hours of admission) were built. We employed machine learning artificial intelligence (AI) algorithms (Boruta, Lasso and recursive feature elimination (RFE)) to filter the features. Logistic regression (LR), random forest (RF) and extreme gradient boosting (XGboost) were used to build the models. Five-fold cross-validation was used to evaluate all of the datasets.

Results:

In total, 945 patients clinically considered infective fever were included in the analysis. There were 661 patients with serious bacterial infections. In the early admission assessment model, the RF exhibited the highest accuracy. In the risk assessment model within 24 hours of admission, LR exhibited highest accuracy. A nomogram was developed to provide the exact SBI risk for each patient.

Conclusions:

The constructed models exhibit a good effect on predicting serious bacterial infections in adults suspected of infective fever, and the models will thus have a good auxiliary decision-making function. Further prospective studies on clinical application scenarios are required to determine the clinical utility of this AI model.