This study demonstrated that NE-WY was extremely useful for detecting bacteremia and allowed for the establishment of specific cutoff values. Furthermore, results learned and validated by machine learning supported this, with NE-WY showing superiority over other parameters. To our knowledge, this is the first report demonstrating that NE-WY can help detect bacteremia by using machine learning to analyze CPD.
In this study, we compared the diagnostic values of NE-WY with other concurrently obtained parameters (WBC, HB, PLT, and CRP) in adult patients with clinically suspected bacteremia and submitted blood cultures. Previous CPD studies have tested healthy subjects as controls or patients with acute bacterial infections, but this study was conducted in a manner that is most similar to actual clinical practice. Additionally, optimal matching was performed at a 1:1 ratio of fever, gender, and age, eliminating other factors that could cause differences between groups. In NE-WY, the greatest difference between the bacteremia and nonbacteremia groups was observed, with a cutoff value of 686.5 (no units) based on the ROC curve and an AUC of 0.7268. This was a major difference compared to CRP, and was the next highest.
The fluorescence distribution of neutrophil area (NE-WY) of CPDs produced by the SYSMEX XN series hematology analyzer correlates with fluorescence intensity and nucleic acid content, reflecting the immaturity associated with neutrophil activation. In bacteremia, various cytokines first cause changes in the organelles of the cell. Furthermore, neutrophils within blood vessels gather around the bacteria, while immature neutrophils with high nucleic acid content are recruited from the bone marrow to peripheral blood. After reacting with the bacteria, neutrophils eventually undergo cell death, and the cytokines produced during this process are reflected in the nucleic acids, increasing NE-WY levels. Therefore, in severe infections, NE-WY is higher.
Although some biomarkers are said to predict bacteremia, the biomarkers used are highly sensitive and specific, and can be measured in hospitals and are inexpensive to measure17. Currently, the NE-WY is being measured simultaneously with the blood count and photographs of the SYSMEX XN series hematology analyzer. This allows you to quickly deliver results along with blood counts and photos without the need for additional equipment or costs in the measurement process (there are data within the device, but they are not usually displayed).
NE-WY meets these criteria and can be easily accessible biomarkers for predicting bacteremia. Additionally, biomarker combinations have been investigated to establish more accurate and efficient diagnostic procedures for bacterial infections. Effective combination of NE-WY with traditional biomarkers is important18,19.
To validate the validity of these results and make them easy to apply in clinical practice, we conducted 2:1 training and validation using machine learning. We confirmed that the first branch of the decision tree is always NE-WY, regardless of the number of training sessions, and that this is the most important parameter for diagnosing bacteremia. Furthermore, the node boundary of the decision tree, NE-WY = 689.5, was roughly the same as the cutoff value obtained from the ROC curve above. CRP was sorted in a similar way to NE-WY, so it was thought that the next branch of NE-WY was determined by WBC. Results do not necessarily indicate a higher NE-WY or WBC level, the higher the rate of bacteremia. However, there is no doubt that there is a trend in that direction. Also, this was thought to be within the error range, as the minimum number of branches was set to n = 20. The ROC curves created based on this decision tree show balanced AUC and high AUC in both training and validation, indicating that no overtraining occurred.
Among the cases examined in this study, some cases did not show bacteremia despite high NE-WY. However, most of these cases were from patients with very poor general conditions and were on the verge of death. Therefore, it is possible that a cytokine storm that occurred when death was imminent could have affected the increase in NE-WY. Conversely, there were some cases where NE-WY was low, but bacteremia was present. Specifically, the majority of cases were involved in infectious endocarditis and catheter infection. This reason is thought to indicate that CRP responds to both systemic and local infections, whereas NE-WY increases primarily in response to systemic infections, slowing the fluctuations in local infections.
Additionally, training and validation was performed by increasing the number of variables using a support vector machine (SVM). CRP is currently the most clinically used marker of infectious diseases and its usefulness is well established, so it was used for comparative purposes. The results showed that performance improved as the number of variables increased, but the difference was not significant compared to NE-WY alone. This indicates that NE-WY has a significant impact on the diagnosis of bacteremia.
In this study, we looked at NE-WY in CPD, but there are many other parameters besides NE-WY.20 (Supplementary Table S2). Analyzing a single parameter or combination of multiple parameters is very complicated to show significant differences between groups. However, machine learning allows for accurate learning and verification in a short amount of time. Machine learning is extremely versatile in CPD analysis, and has achieved results that can be quickly applied in clinical practice.
Recently, several papers have been published attempting to create predictive models of bacteremia using machine learning based on various parameters, including CPD.21,22,23,24. This study not only developed predictive models, but also aimed to first focus on the NE-WY of CPD, demonstrating its usefulness in itself, and then develop it into a NE-WY-centered bacteremia diagnosis. Boerman et al. He said it was important to find parameters that are more positively correlated with bacteremia.twenty four; NE-WY is the perfect example of this. This is the first report demonstrating that NE-WY can help detect bacteremia by analyzing CPD using machine learning. In this study, patients were matched by age and gender to demonstrate the utility of NE-WY. However, both elderly and male patients are at increased risk for bacteremia. Therefore, combining these factors can help construct a more useful risk score for bacteremia.
This study was related to several limitations. First, this could involve selection bias as it was a single-centered retrospective observational study. Second, the sample size may be small, limiting the reproducibility of the results. However, this study was conducted using machine learning with optimal 1:1 matching, and the conclusions were not significantly affected.
In conclusion, this study demonstrated that NE-WY is extremely useful in detecting bacteremia in adult patients who received blood cultures due to clinically suspected bacteremia. Considering the versatility in clinical practice, we propose a cutoff value of NE-WY = 690 (no units). There are two major advantages. The first is providing results more rapidly than other biomarkers. Second, the measurement process does not require additional equipment or involves additional costs. Therefore, NE-WY is expected to become a new biomarker in the treatment of infectious diseases. Furthermore, CPD has not yet been fully analyzed, so machine learning should be applied to further explore its relationship with pathology and disease.
