Complete blood count and C-reactive protein to predict positive blood culture among neonates using machine learning algorithms

Abstract

Purpose:  The authors aimed to develop a Machine-Learning (ML) algorithm that can predict positive blood culture in the neonatal intensive care unit, using complete blood count and C-reactive protein values.

Methods:  The study was based on patients’ electronic health records at a tertiary neonatal intensive care unit in São Paulo, Brazil. All blood cultures that had paired complete blood count and C-reactive protein measurements taken at the same time were included. To evaluate the machine learning model’s performance, the authors used accuracy, Area Under the Receiver Operating Characteristics (AUROC), recall, precision, and F1-score.

Results:  The dataset included 1181 blood cultures with paired complete blood count plus c-reactive protein and 1911 blood cultures with paired complete blood count only. The f1-score ranged from 0.14 to 0.43, recall ranged from 0.08 to 0.59, precision ranged from 0.29 to 1.00, and accuracy ranged from 0.688 to 0.864.

Conclusion:  Complete blood count parameters and C-reactive protein levels cannot be used in ML models to predict bacteremia in newborns.

Keywords: Critical care; Neonatology; Artificial intelligence; Machine learning; Sepsis

HIGHLIGHTS

It can take days to get the result of blood culture.

CBC and CRP are readily available exams and could be used to predict blood culture.

ML algorithms based on CBC and CRP couldn’t predict neonatal blood culture positivity.