Machine learning identification of specific changes in myeloid cell phenotype during bloodstream infections

Abstract

The early identification of bacteremia is critical for ensuring appropriate treatment of nosocomial infections in intensive care unit (ICU) patients. The aim of this study was to use flow cytometric data of myeloid cells as a biomarker of bloodstream infection (BSI). An eight-color antibody panel was used to identify seven monocyte and two dendritic cell subsets. In the learning cohort, immunophenotyping was applied to (1) control subjects, (2) postoperative heart surgery patients, as a model of noninfectious inflammatory responses, and (3) blood culture-positive patients. Of the complex changes in the myeloid cell phenotype, a decrease in myeloid and plasmacytoid dendritic cell numbers, increase in CD14+CD16+ inflammatory monocyte numbers, and upregulation of neutrophils CD64 and CD123 expression were prominent in BSI patients. An extreme gradient boosting (XGBoost) algorithm called the “infection detection and ranging score” (iDAR), ranging from 0 to 100, was developed to identify infection-specific changes in 101 phenotypic variables related to neutrophils, monocytes and dendritic cells. The tenfold cross-validation achieved an area under the receiver operating characteristic (AUROC) of 0.988 (95% CI 0.985–1) for the detection of bacteremic patients. In an out-of-sample, in-house validation, iDAR achieved an AUROC of 0.85 (95% CI 0.71–0.98) in differentiating localized from bloodstream infection and 0.95 (95% CI 0.89–1) in discriminating infected from noninfected ICU patients. In conclusion, a machine learning approach was used to translate the changes in myeloid cell phenotype in response to infection into a score that could identify bacteremia with high specificity in ICU patients.

A reactive monocyte subset characterized by low expression of CD91 is expanded during sterile and septic inflammation

“…A series of studies have shown that downregulated HLA-DR expression is associated with acute illness-induced immune suppression and poor outcome [17][18][19]. Depressed HLA-DR expression by peripheral blood monocytes during BSI was also observed by our laboratory in a recent study [16]. CD14+CD91low monocytes express the lowest levels of HLA-DR on their cell surface compared to all other monocyte subsets.…”

“…CD91low monocyte levels were not correlated with severity scores SOFA or SAPS II. We previously showed that the discrimination between infected and non-infected intensive care unit patients cannot be achieved with a single monocyte marker but requires the combination of multiple phenotypic changes in a composite score [16]. Longitudinal observation of cardiac surgery patients as a model of inflammation highlights the early increase in CD14+CD91low monocytes while the rise of CRP is witnessed 24 h later, reflecting their inverse correlation.…”

“…While this is the case for monocytes in steady-state, it has not been addressed so far whether CD91 is also expressed by all monocytes in inflammatory states. In studying the dynamics of monocyte subset redistribution in pathological conditions, we recently uncovered a significant fraction of monocytes expressing low to negligible amounts of CD91 [16].…”

Speaking of sepsis: semantics, syntax, and slang

“…Neither the current concept of sepsis nor laboratory confirmation of bloodstream infection offer much help to the physician who has to make a call on presumptive antimicrobial therapy during the early stages of bloodstream infection. In teaching hospitals, where subject matter expertise reaches critical mass, alignment between the patient journey and the clinical laboratory workflow can be fine tuned with the help of near-at-hand decision-support tools including clinical laboratory support and emerging techniques such as machine learning algorithms and immunophenotyping ( 6 ). There is a recent observation from the United Kingdom that laboratory-enhanced surveillance of Gram negative bloodstream infections can predict associated in-hospital mortality ( 7 ).…”

“…Some early attempts at using machine learning to find a way out of the sepsis bind were a little clunky, and were hampered by under-coding when benchmarked by the current International Classification of Diseases ( 13 ). More recent attempts to achieve specific types of clinical decision support, such as prediction of positive blood culture results and immunophenotyping ( 6 , 14 ), are starting to look useful. A valuable feature of supervised machine learning is its ability to prevent the user from jumping to conclusions before the data has been acquired, curated, classified and visualised.…”