SEARCH

SEARCH BY CITATION

Keywords:

  • flow cytometry;
  • WBC differential;
  • complete blood count;
  • immunophenotyping;
  • laboratory instrumentation;
  • hematology;
  • sepsis;
  • principal component analysis;
  • hierarchical clustering;
  • machine learning

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

Background:

We sought to evaluate, on a model of sepsis, the clinical relevance of new parameters obtained on a white blood cell (WBC) differential by flow cytometry, implemented in the routine workflow of our hematology laboratory.

Methods:

A WBC with differential by flow cytometry was done on 459 patients at admission in intensive care unit. They were retrospectively categorized in having (i) infection or not or (ii) a high gravity score (severe sepsis or septic shock) or not. We analyzed by hierarchical clustering, in a multidimensional manner, 50 parameters provided by the flow cytometric platform in place of the standard seven parameters for a standard differential.

Results:

Our approach allows to discriminate on the basis of a WBC differential (i) infected patients at admission based on a 16 parameter signature, with a concordance rate of 72.7% and a specificity of 79.9% and (ii) patients with high gravity score (septic shock or severe sepsis) at admission with a signature of eight parameters, with a concordance rate of 74.7% and a specificity of 75.9%.

Conclusions:

This study shows the clinical relevance of an extended WBC differential to obtain by a flow cytometer integrated into a routine hematology laboratory workflow. Development of such approach implicates the redefinition of the WBC differential by integrating new parameters. © 2012 International Clinical Cytometry Society


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

The complete blood count (CBC) with white blood cell (WBC) differential is one of the most requested tests for the diagnosis or prognosis of a wide spectrum of pathologies, from inflammatory states or sepsis to malignancies. Patient care is governed by decision tree that takes into account, among other parameters, the count of neutrophils, eosinophils, basophils, lymphocytes, and monocytes. Currently, the CBC is provided by an automated cell counter after which, in case of abnormalities, a manual differential review is done on a blood smear stained with May-Grunwald-Giemsa or Romanowsky (1, 2). Manual review consists of a 100–200 nucleated cell count and classification of the different types of leukocytes present. This global scheme remains unchanged since the advent of cell counter in the second half of the 20th century. More recently, this concept has evolved with the description of (i) new clinically relevant circulating cells such as subtypes of monocytes or dendritic cells or (ii) new clinically relevant cell state such as HLA-DR expression on monocytes for septic patient (3–5). To date, analysis of cell subsets (CD4/CD8 lymphocytes, dendritic cells, CD34 progenitors for instance) remains a specialized test, although some of these results may have a direct impact on the patient care if there were available 24 h/day (3).

Multiparameter flow cytometry (FCM) seems to present an attractive alternative to a microscopic differential. Indeed, thousands of cells can be analyzed in a standardized manner giving the opportunity to detect populations usually recognized within a blood smear, as well as other subsets of cells. Recently, we and others have reported the usefulness of the flow cytometric approach in a routine workflow to replace part of the microscopic review and recently, a standardized European Community (CE) labeled test has been proposed for WBC differential by flow cytometry (6–10). The study describes herein an approach applied in the same time frame than for CBC and differential that gives not only five WBC populations but 7 to 9 subsets associated with up to 50 parameters (including positional data) provided by the flow cytometer (11–13). In that way, WBC differential becomes more sophisticated. It is then worth investigating whether these new parameters could better discriminate patients and could thus be integrated in the clinical patient care. However, the challenge is to analyze in high-dimensional datasets, i.e., analyze together many parameters for many samples. Hierarchical clustering is widely used in genomic analysis to explore high-dimensional data and to identify biological pathways. More recently, this approach has been used for flow cytometric data or cytokine expression, emphasizing the usefulness of high-dimensional approaches and the power of machine-learning techniques in improving diagnosis and prognosis (14–18).

In this study, we address the clinical relevance of this multidimensional WBC differential to manage patients from critical care unit. Our results show the potential interest of reporting these new parameters in the context of sepsis and highlight the need for a redefinition of the WBC differential integrating parameters now easily detected by flow cytometry in routine laboratory workflows.

DESIGN AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

Patient Samples and Study Design

All patients aged over 18 years and hospitalized in the medical intensive care unit (ICU) of the Rennes University Hospital between September 15, 2009 and March 15, 2010 were prospectively included. Pregnant women, persons deprived of their liberty by judicial or administrative decision, and patients not requiring blood sampling were excluded. The ethics committee of the hospital previously approved the study protocol. For all patients, the blood sample was drawn for blood count as well as for the usual procedures of the service, i.e., microbiological samples according to the presumed site of infection (blood cultures, lumbar puncture, tracheal aspirate, and urine culture), a C-reactive protein, and clinical chemistry. For each patient, the following data were recorded at admission: age, sex, history, reason for hospitalization, Simplified Acute Physiology Score II (SAPSII), Sepsis-Related Organ Failure Assessment (SOFA), primary diagnosis, length of stay, and mortality in the ICU and the results of all biological samples. Two intensivists and two specialists in infectious diseases retrospectively reviewed medical records and independently determined whether patients had (i) no systemic inflammatory response syndrome (SIRS) nor infection (control group) or had (ii) SIRS alone, (iii) sepsis, (iv) severe sepsis, or (v) septic shock at the time of admission to the ICU according to the consensus definitions (19). A training set was built to define the relevant parameters and evaluate the feasibility of the approach to discriminate group of patients. We randomly selected 24 control, 25 SIRS, and 25 septic shock patients from our database, analyzed jointly with 28 additional healthy blood donors (Table 1).

Table 1. Patients' Characteristics at Admission
Training set
  ICU 
  Healthy donorsControlSIRSSeptic shockP value
  • a

    Patients with coinfection are not listed here.

  • Patients were separated on a training and a validation set of analysis. Results are expressed as median and interquartile except for gender (ratio). NA: not applicable; ns: not significant. Groups were compared using a Kruskal–Wallis test.

Samples (N)28242525NA
Age (years) (IQR)43 (32–61)53 (39–70)51 (39–56)68 (50–77)<0.05
Gender M/F (ratio)15/13 (1.1)16/9 (1.7)16/9 (1.7)14/11 (1.2)ns
WBC (×109/L) median (IQR)7.5 (7–8.1)9.2 (6.1–10.8)11.2 (7.2–15.3)11.8 (5.7–17.9)<0.05
SAPSII score (IQR)NA31 (22–40)36 (28–47)64 (46–79)<0.001
SOFA score (IQR)NA4 (1.5–5)5 (2.5–6)12 (9–14.5)<0.001
Validation set
 ControlSIRSSepsisSevere sepsisSeptic shockP value
Samples (N)88101945250NA
Age (years) (IQR)57 (48–68)57 (45–69)56 (41–71)60 (47–72)65 (52–74)ns
Gender M/F (ratio)55/33 (1.7)65/36 (1.8)66/28 (2.3)33/19 (1.7)27/23 (1.2)<0.05
WBC (×109/L) median (IQR)8.6 (6.3–12.1)14.4 (11.6–17)13.9 (8.7–19.1)12.8 (4.9–20)10.9 (5.3–23)<0.001
SAPSII score (IQR)39.5 (27–52)42.5 (33–61)40.5 (23–55)42 (30–49)57.5 (47–80)<0.001
SOFA score (IQR)6 (4–8)7 (3–10)6 (3–8)5 (2–8)12 (8–14)<0.001
Bacterial infection (N)0059a32a34aNA
Viral/mycosis/parasitic (N)0013a6a3aNA

Remaining cases were included in the validation set. Relevant parameters, defined using the training set, were applied on these patients (Table 1). At admission in ICU, clinicians do not always detect the level of gravity. Moreover, the gravity is also not always correlated with the presence of bacterial infection. Therefore, we assessed in this study the WBC parameters to predict infection or gravity irrespective of the clinical or bacterial status. Patients were all enrolled at admission in ICU, and our aim was to identify “infected” versus “noninfected” patients and “high” versus “low” gravity score (Fig. 1).

thumbnail image

Figure 1. Study flowchart and patient repartition on the training and validation set. Patients from ICU where categorized, after expert review, on five groups (control, SIRS, sepsis, severe sepsis, and septic shock). A training set was built with a random selection of 24 control, 25 SIRS, and 25 septic shock. Remaining patients were analyzed on a validation set. N: number of samples.

Download figure to PowerPoint

Flow Cytometry Analysis

WBC differential is routinely analyzed by flow cytometry in our core facility (Hematoflow, Beckman Coulter, Miami, FL) as previously described (9). Briefly, the Hematoflow platform is built around a cell preparator (FP1000, Beckman Coulter), along with a five-color flow cytometer (FC500, Beckman Coulter). Each WBC-flagged sample was processed with a lyse-no wash protocol and labeled with an antibody cocktail (Cytodiff, Beckman Coulter) including fluorescein isothyocyanate conjugated CD36 (clone F16.152), phycoerythrin (PE) conjugated CD2 (clone 39C1.5), PE conjugated CRTH2 (clone BM16), PE-Texas Red (ECD) conjugated CD19 (clone J4.119), PE-cyanine 5 (PC5) conjugated CD16 (clone 3G8) and PE-cyanine 7 (PC7) conjugated CD45 (clone J.33). For each sample, 20,000 WBC were analyzed after cell debris exclusion. We used an autogating software (Cytodiff CXP, Beckman Coulter), which follows a sequential flow path where populations of interest are isolated at each intermediate step. Data recorded were WBC count from the cell counter (LH755, Beckman Coulter) and percentage of cell subsets from the flow cytometer. Moreover, additional parameters from the flow cytometer, including mean fluorescence intensity and coefficient of variation (CV), were recorded (Supporting Information Table 1).

Neutrophil CD64 expression was measured separately using a Leuko64 kit (Trillium Diagnostics, Brewer, ME) containing fluorescent beads, CD64 and CD163 antibodies analyzed with a FC500 flow cytometer (Beckman Coulter) following manufacturer's recommendations. Leuko64 index was reported on lymphocytes, monocytes, and neutrophils using a lot-specific Leuko64™QuantiCALC automated software (Trillium Diagnostic). The Leuko64 index was calculated using the ratio of the mean fluorescent intensity of the cell populations over that of the beads. Internal negative control (lymphocytes index < 1) and internal positive control (monocytes index > 3) were used for the validation of each sample.

Flow cytometry data were analyzed by a biologist who was blind to the clinical data, and the results were collected at the end of the study period.

Statistical Analysis

Descriptions of continuous variables on patients were expressed as median and interquartile ranges (25th–75th). Groups were compared using a Kruskal–Wallis test. A value of P < 0.05 was considered statistically significant. Statistical analysis was performed using GraphPad 5.0 (Prism Software).

The percentage of cell subsets, the mean, and CV of the fluorescence signal for each marker tested were automatically exported and saved as tabulation delimited files. In an univariate analysis, we used a Kruskal–Walis nonparametric test to compare groups of patients. Differences were defined as statistically significant when P < 0.05 (*), P < 0.01 (**), and P < 0.001 (***). Classification analysis was done using Cluster 3.0, originally written by Michael Eisen (http://bonsai.hgc.jp/∼mdehoon/software/cluster/software.htm) (20). We performed an unsupervised hierarchical clustering with Spearman's rank distance and average linkage. Clustering results were visualized using Treeview (http://jtreeview.sourceforge.net). Principal component analysis (PCA) and predictive modeling was made using Partek Genomics Suite (www.partek.com). Various models (K-Nearest Neighbor, Support Vector Machine and Discriminant Analysis) were tested for the predictive model using a full leave-one-out crossvalidation to increase the prediction accuracy of the tested models. The model giving the best efficiency was reported.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

Patients Characteristics

From September 2009 to March 2010, 505 patients were hospitalized in the medical ICU and prospectively included in the study. Among them, 46 were excluded from our study because CBC was not available at admission (n = 28), technical problems occur in the laboratory (n = 6), and finally, 12 patients did not meet the criteria of classification after expert review (Fig. 1). The characteristics of the entire population are summarized in Table 1. SIRS alone was diagnosed in 126 out of 459 patients (27.5%), sepsis in 94 patients (20.5%), severe sepsis in 52 patients (11.3%), and septic shock in 75 patients (16.3%). Finally, 112 out of 459 patients (24.4%), without SIRS or infection were used as control (Fig. 1, Table 1). Patients were randomly separated on (i) a training set with 28 healthy donors, 25 control, 25 SIRS, and 25 septic shock patients from ICU and (ii) a validation set with 88 control, 101 SIRS alone, 94 sepsis, 52 severe sepsis, and 50 septic shock. On the validation set, the median (interquartile range) of the SAPSII and the SOFA were 57.5 (47–80) and 12 (8–14), respectively.

Relevance on a Training Set of the WBC Differential Signature

Our approach by FCM provides 49 parameters (see Supporting Information Table 1) for each sample: cell count of WBC or subsets gives (11 values) mean and CV of intensity of expression (38 values). In a first attempt to define parameters that best characterize WBC differential and those irrelevant, we used a training set with randomly selected samples including healthy blood donors (n = 28) and patients from an ICU, among which controls (n = 24), SIRS (n = 25), or septic shock (n = 25). Both in an unsupervised analysis, using a Spearman rank algorithm with an average linkage and PCA, septic shock were separated from healthy donors, controls, and SIRS, whereas SIRS and control donor groups were not clearly separated (Figs. 2A and 2B). By comparison, we analyze by PCA and discriminant analysis (with a full leave-one-out crossvalidation) the extended WBC obtained (i) by FCM (Fig. 2B) or (ii) by standard process (Fig. 2C). We found that the standard and FCM review allow, respectively, the correct classification of 70 and 82 out of 102 cases. The ability to discriminate septic shock on this training set give a sensitivity of 83.3%, a specificity of 97.4% with a positive and negative likelihood ratio (LR+ and LR−) of 32.5 and 0.17, respectively, and a positive predictive value (PPV) of 90.9% and a negative predictive value (NPV) of 95% for the FCM. By comparison, for the standard differential, the sensitivity was at 83.3% and the specificity at 92.3% (LR+ of 10.8, LR− of 0.18, PPV of 76.9%, and NPV of 94.7%; data not shown).

thumbnail image

Figure 2. Hierarchical clustering and PCA on the training set. A: Flow cytometric values obtained from the training set were analyzed by an unsupervised hierarchical clustering with Spearman's rank distance and average linkage. Patients were grouped with healthy donors (brown), control (green), SIRS (red) and septic shock (blue); B: PCA using the flow cytometric (FCM) values from the training set; C: PCA on values obtained from the same training set but using results from the standard differential (standard diff).

Download figure to PowerPoint

Using a Kruskal–Wallis nonparametric analysis, we defined parameters significantly modulated among groups of patients. Seven parameters were significantly modulated between control and SIRS, whereas 23 parameters (not all the same) defined the control to septic shock or SIRS to septic shock comparison (Fig. 3).

thumbnail image

Figure 3. Monovariate analysis of each parameter among groups of patients. Statistical difference was analyzed for each of the 49 depicted parameters. We use a Kruskal–Wallis nonparametric test to compare groups of patients. Differences were defined as statistically significant when P < 0.05 (yellow), P < 0.01 (orange), and P < 0.001 (brown). Grey plots represent no statistical difference. Each column represents a comparison between two groups (indicated by x). Nine parameters with no difference among groups are not represented. #: Absolute count; mean: mean fluorescent intensity; CV: coefficient of variation. Cell count was recorded for WBC; neutrophils; eosinophils; basophils; B-; T cytotoxic and NK (Tc/NK-) and T non cytotoxic (Tnc-) lymphocytes; CD16- and CD16+ monocytes; immature granulocytes (Igrans) and blasts. Mean and CV was recorded for forward (FSC) and side scatter (SSC), CD16, CD36, and CD45 intensity of expression, for neutrophils, monocytes CD16- and CD16+ and immature granulocytes. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Download figure to PowerPoint

We conclude from this training set that a signature of 29 out of 49 parameters obtained by FCM on the basis of a regular extended differential allow a better definition of the patients in our study model.

Efficiency on a Validation Set of the Selected Parameters

The 29 parameters as defined on the training set (Fig. 3) were investigated to see if they were suitable to separate infected patients at Day 0 in ICU regardless of their clinical or bacterial status. A cohort of 385 patients was analyzed involving two sets of patients depending of the presence, or not, of an “infected” status; 196 versus 189, respectively (Fig. 1). Among the 196 infected patients, we include 50 with septic shock, 52 with severe sepsis, and 94 with sepsis. The 189 “noninfected” patients consisted of 101 with SIRS alone. The remaining 88 patients were categorized as without SIRS, the so-called “control” (Table 1). As the CD64 expression on polynuclear neutrophils has been proposed as a biomarker of sepsis, we evaluated this marker in combination with the extended differential by flow cytometry in a routine controlled workflow (21). This analysis is standardized and can be setup on a cell counter or a flow cytometer.

Samples were analyzed by discriminant analysis with a full leave-one-out crossvalidation using the Partek software. This model allowed the correct classification of 280 out of 385 cases (72.7%) with a minimal signature of the top 16 ranked parameters (Table 2). Of the 196 “infected” patients, 129 were correctly classified, whereas 151 out of 189 “noninfected” patients were correctly classified, giving a sensitivity of 65.8% and a specificity of 79.9% for this signature. The PPV and NPV were, respectively, of 69.3% and 77.3%, whereas the LR+ and LR− were at 3.3 and 0.4 (Table 3). On the same analysis performed on the 7-part differential, a sensitivity of 57.1%, a specificity of 63.5%, a PPV of 61.9%, and a NPV of 58.8% were found, thus allowing the correct classification of 232 out of 385 samples (60.2%; Table 3).

Table 3. WBC Differential Classifier for Infection Classification
 ValueCI7-part diff
  1. Summary of the classifier efficiency and table of contingency; PPV: positive predictive value; NPV: negative predictive value; CI: 95% confidence interval; 7-part diff: 7-part differential obtained on the conventional CBC; N: number of patients.

Sensitivity (%)65.858.7–72.457.1
Specificity (%)79.973.5–85.463.5
Positive likelihood ratio3.32.4–4.41.6
Negative likelihood ratio0.40.3–0.50.7
PPV (%)77.370.1–83.461.9
NPV (%)69.362.7–75.358.8
  Real 
  PresentAbsenceN
PredictedPositive12938167
Negative67151218
 N196189385

We conclude that a signature of 16 parameters obtained on the basis of a WBC differential, allows the classification of infected patients at admission in ICU with a specificity of 79.9%.

We then addressed the relevance of our approach for the classification of severity score. To this aim, we analyzed 297 patients (Table 1 and Fig. 1) divided in two groups according to severity: (i) patients with a septic shock status (n = 50) and severe sepsis (n = 52), and (ii) patients with sepsis (n = 94) and SIRS (n = 101). Using discriminant analysis with a full leave-one-out crossvalidation, 222 out of 297 cases (74.7%) were correctly classified with a minimal signature of the top eight ranked parameters (Table 2). Out of the 102 “septic shock or severe sepsis,” 74 were correctly classified, whereas 148 out of 195 “SIRS or sepsis” patients were properly assigned, giving a sensitivity of 72.6% and a specificity of 75.9% for this signature. The PPV and NPV were, respectively, of 61.2% and 84.1%, whereas the LR+ and LR− were of 3 and 0.4 (Table 4). By contrast, the same analysis done on the 7-part differential gave a sensitivity of 47.1%, a specificity of 74.4%, a PPV of 49%, and a NPV of 72.9%, allowing the correct classification of 193 out of 297 samples (64.9%; Table 4).

Table 4. WBC Differential Classifier for Severity Classification
 ValueCI7-part diff
  1. Summary of the classifier efficiency and table of contingency; PPV: positive predictive value; NPV: negative predictive value; CI: 95% confidence interval; 7-part diff: 7-part differential obtained on the conventional CBC; N: number of patients.

Sensitivity (%)72.662.8–80.947.1
Specificity (%)75.969.3–81.774.4
Positive likelihood ratio32.3–41.8
Negative likelihood ratio0.40.3–0.50.7
PPV (%)61.251.9.1–69.949
NPV (%)84.177.8–89.272.9
  Real 
  PresentAbsenceN
PredictedPositive7447121
Negative28148176
 N102195297
Table 2. Relevant Parameters from the WBC Signature
Infected versus noninfected patientsSepsis severe and shock versus sepsis and SIRS
ItemsP-valueIn infectedItemsP-valueIn sepsis severe and shock
  1. Items are labeled with (i) #: Absolute count; mean: mean fluorescent intensity; CV: coefficient of variation (except for the CD64 index); (ii) cell type (PN: neutrophils, MonoCD16: monocyte CD16+, MonoCD16-: monocyte CD16-, Igran: immature granulocytes) and (iii) parameter: forward (FSC) and side scatter (SSC), CD16, CD36, and CD45 intensity of expression.

CD64Index8.1E-18UpCD64Index9.8E-11Up
Mean_PN_CD162.4E-8DownMean_PN_CD165.3E-8Down
Mean_MonoCD16_FSC6.6E-6UpCV_MonoCD16-_CD451.9E-5Down
Mean_PN_FSC6.9E-6UpCV_MonoCD16_FSC7.9E-5Down
Mean_MonoCD16-_FSC1.7E-5UpCV_PN_CD368.3E-5Down
CV_MonoCD16-_CD452.6E-4DownCV_Igran_CD451.1E-4Up
Mean_MonoCD16-_CD455.1E-4UpMean_MonoCD16_FSC1.2E-4Up
Mean_PN_SSC6.3E-4DownCV_Igran_CD361.3E-4Up
#Igran0.0011Up   
CV_Igran_CD450.0011Up   
Mean_Igran_CD450.0019Up   
#LyTc0.0028Down   
CV_MonoCD16-_SSC0.0055Up   
#Neutro0.010Up   
CV_PN_CD450.011Down   
#LyTnc0.017Down   

We concluded that a signature of eight parameters obtains on the basis of a WBC differential allows the classification of patient with high gravity score (septic shock or severe sepsis) at admission in ICU with a specificity of 75.9%.

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

To refine the classical WBC differential report, we addressed in this study the clinical relevance of new parameters obtained by standardized FCM on a WBC differential. We used, as a model, patients at admission to the critical care unit that were classified for infection severity according to international guidelines (19). In a previous study, we explored the CD64 index as single biomarker of bacterial infection and found a sensitivity of 63% and a specificity of 89% (22). These values were close from those published in a recent meta-analysis (21). In this study, parameters obtained by flow cytometry were analyzed jointly in a multidimensional manner, to predict infection or disease severity, irrespective of the clinical or bacterial status. Indeed, in clinical practice, early detection of infected or severe patients at admission in ICU is not always obvious. Moreover, there are no consensual biomarkers for the early diagnosis of infection or the prediction of prognosis which depends on early intervention with antibiotic administration (23, 24). This study demonstrates that a WBC signature integrating new parameters from the FCM differential allows the correct classification of infected or high gravity score patients in more than 72% of cases.

On flow cytometry and automated hematology counter, cells are depicted both by count and by positional or structural parameters, i.e., size, granularity, volume, or intensity of fluorescence. These parameters have been used to recognize for instance CLL from reactive lymphocytosis or degranulated neutrophils in myelodysplastic syndrome (11, 12). Regarding sepsis, an increase in neutrophil size, in distribution of breadth, as well as in monocyte volume, have been reported (13, 25–27). The size and granule content of neutrophils or monocytes, as well as the population heterogeneity, are well evaluated by hematologists on a stained slide; however, these parameters are to date not quantifiable and not integrated into a diagnostic algorithm (28).

Of note, all these parameters proven to be clinically relevant have been tested individually but, to our knowledge, they have not yet been studied jointly in a multidimensional analysis, which might offer new insights in WBC differential and for improve the diagnosis. In our analysis, we combined (i) positional parameters (from the flow cytometer), (ii) antigen expression (restricted to monocytes, neutrophils, and immature granulocytes) measured by the flow cytometer, and (iii) absolute count of cells. Altogether, 50 parameters were analyzed for each sample. To analyze this dataset, we used clustering analysis methods and PCA as previously suggested for flow cytometric data (14, 15, 17, 18, 29, 30). In a training set, we obtained a better signature to segregate healthy donors and septic shock using all the parameters available rather than the regular one from the conventional WBC differential. In a validation set, several models were tried and we found, as previously proposed, discriminant function analysis as the best classifier with the minimal misclassification for detection of infection and high gravity score of sepsis (31, 32). As previously described, we found a lower CD16 and a higher CD64 expression on neutrophils, on our classifiers, as the top-ranked parameters (21, 33). Other relevant parameters includes (i) positional parameters from the flow cytometer (Fig. 3), such as size or granularity of neutrophils and monocytes and (ii) cell count such as T lymphocytes that are known to be lower in sepsis (13, 25–27, 34–36).

Using our approach, we found an overall concordance rate of our classifier at 72.7% and 74.7% to separate, respectively, infected from noninfected patients and high gravity score versus others. Using the same set of patients with data obtained from the conventional 7-part differential, we found a concordance rate at, respectively, 60.2% and 64.9%. Although, this efficiency does not seem greater compared with other proposed biomarkers of sepsis (23), our result were obtained under strict methodological conditions, first by comparing, in our validation set, only patients from ICU and avoiding healthy controls. Indeed, sepsis biomarkers are related to inflammation and most case-control studies compare patients from ICU to normal patients or patients coming from other clinical units, although they have a different basic state of inflammation. Moreover, our results were obtained using a full leave-one-out crossvalidation to increase the prediction accuracy of our prediction model, and our results characterized the efficiency of a signature rather than the value of a single parameter.

The FCM method used here is standardized, meaning that our approach is widely applicable in routine laboratories and individual samples can be compared with a controlled reference database. Indeed, the protocol used includes a CE-marked reagent and an autogating software that dramatically reduces the need of an expert review of flow cytometric raw data. Moreover, this system is totally connected to both, the cell counter and the laboratory informatics system allowing for automatic CBC and WBC differential report based on predefined rules.

We described herein a proof of concept study on a model of patients from ICU. The antibody combination used here is not dedicated to inflammatory or infected states, and a definition of a specific combination of antibodies could be more efficient. Thus, no clear difference was found between bacterial and nonbacterial infection status (data not shown). Of note, further work remains necessary to build reference database of cases and to integrate data mining techniques in routine flow cytometric software. Moreover, we restricted this study to positional parameters from the flow cytometer, and of course, it will be interesting to analyze jointly data from the cell counter and from the flow cytometer. The final goal of such approach will be to produce, for a CBC, a predictive score of matching with reference database, for instance, of sepsis or myelodysplastic syndrome.

In summary, we have shown in our model that improvement of the routine workflow with FCM-based WBC differential provides additional relevant parameters. This highlights the need for a redefinition of what a WBC differential should be in the 21st century taking into account new parameters obtained on the basis of a regular standardized CBC with WBC differential and using a high-dimensional analysis approach.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

The authors thank Beckman Coulter France and Trillium Diagnostics, LLC for supplying reagents.

LITERATURE CITED

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. DESIGN AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. LITERATURE CITED
  9. Supporting Information

Additional Supporting Information may be found in the online version of this article.

FilenameFormatSizeDescription
Supplemental_Table-1.doc78KSupporting Information

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.