Combination Biomarkers to Diagnose Sepsisin the Critically Ill Patient

Sebastien Gibot1,2, Marie C. Bene3, Robin Noel4, Frederic Massin5, Julien Guy6, Aurelie Cravoisy1,Damien Barraud1, Marcelo De Carvalho Bittencourt3, Jean-Pierre Quenot4, Pierre-Edouard Bollaert1,Gilbert Faure3, and Pierre-Emmanuel Charles4

1Reanimation Medicale, Hopital Central, Nancy, France; 2Contrat Avenir INSERM, Groupe Choc, Faculte de Medecine, Nancy Universite, Nancy,

France; 3Laboratoire d’Immunologie, CHU and Nancy Universite, Vandoeuvre-les-Nancy, France; 4Reanimation Medicale, Hopital du Bocage, Dijon,

France; 5Laboratoire d’Immunologie, CHU Brabois, Vandoeuvre-les-Nancy, France; and 6Laboratoire d’Hematologie, Hopital du Bocage, Dijon,

France

Rationale: Although the outcome of sepsis benefits from the promptadministration of appropriate antibiotics on correct diagnosis, theassessment of infection in critically ill patients is oftena challenge forclinicians. In this setting, simple biomarkers, especiallywhen used incombination, could prove useful.Objectives: To determine the usefulness of combination biomarkersto diagnose sepsis.Methods: Three hundred consecutive patients were enrolled to con-struct a biologic score that was next validated in an independentprospective cohort of 79 critically ill patients from another center.Measurement andMain Results: Plasma concentrations of soluble trig-gering receptor expressed onmyeloid cells-1 (sTREM-1) and procal-citonin (PCT) were assayed, and the expression of the high-affinityimmunoglobulin-Fc fragment receptor I (FcgRI) CD64 on neutro-phils (polymorphonuclear [PMN] CD64 index) in flow cytometrywas measured. A “bioscore” combining these biomarkers was con-structed. Serum concentrations of PCT and sTREM-1 and the PMNCD64 index were higher in patients with sepsis compared with allothers (P, 0.001 for the three markers). These biomarkers were allindependent predictors of infection, the best receiver-operatingcharacteristic curve being obtained for the PMN CD64 index. Theperformance of the bioscore, better than that of each individualbiomarker, was externally confirmed in the validation cohort.Conclusions: This prospective study, including inceptive and valida-tioncohortsofunselected intensive careunitpatients, demonstratesthe high performance of a bioscore combining the PMNCD64 indextogether with PCT and sTREM-1 serum levels in diagnosing sepsis inthe critically ill patient.

Keywords: sepsis; diagnostic biomarkers; procalcitonin; CD64; sTREM-1

Among critically ill patients admitted to intensive care units (ICU),at least one-third to half of them is ultimately diagnosed with sep-sis, which is a leading cause of mortality and morbidity (1). The

outcome of sepsis has been shown to benefit from the promptadministration of appropriate antibiotics on correct diagnosis(2). This latter issue is critical to limit the selection of antibior-esistant bacterial species, whereas early recognition of sepsisremains a matter of concern given the low diagnostic accuracyof such usual tools as clinical signs, fever, or elevated whiteblood cell counts (3, 4). In this context, the development ofnew biomarkers is desirable. Among the most recent ones, cu-mulative published evidence supports the use of procalcitonin(PCT) measurement. The soluble triggering receptor expressedon myeloid cells-1 (sTREM-1) and the intensity of expressionof the high-affinity immunoglobulin-Fc fragment receptor I(FcgRI) CD64 on polymorphonuclear (PMN) cells are also ofpotential interest (3, 5–9). Indeed, several studies have sug-gested the usefulness of these markers in the early stages ofsepsis, before a positive microbiologic identification can beobtained. However several limitations should be mentioned.PCT elevation also occurs in nonseptic inflammatory states,such as postoperative conditions, resuscitated cardiac arrest,or cardiogenic shock. In addition, low PCT levels may be seenin patients with localized infections or early sepsis (10). Al-though less published data exist regarding sTREM-1, a recentmetaanalysis concluded that it was of clinical interest (8). It is,however, worth noting that it may lack specificity becausesTREM-1 elevation has been observed during noninfectiouslung injury conditions, such as trauma or aspiration (11, 12). Ad-ditional markers are therefore required to improve PCT diag-nosis accuracy.

The increased expression level of CD64 on PMN, indicativeof these cells’ activation, was published as highly indicative ofsepsis in small cohorts (13–17). Although promising, these find-ings should be validated in larger studies including adults. Moreover,in contrast to previous biomarkers, the specificity and sensitivity

(Received in original form January 8, 2012; accepted in final form April 14, 2012)

Supported by a grant from the French Ministry of Health interregional Program

Hospitalier de Recherche Clinique (PHRC 2009–2011) to S.G. Trillium Inc. freely

provided the Leuko64 tests.

Author Contributions: S.G., M.C.B., and P.-E.C. designed and conducted the study;

R.N., A.C., D.B., and J.-P.Q. collected data and managed the patients; F.M., J.G.,

M.d.C.B., and G.F. analyzed data; and S.G., M.C.B., P.-E.B., and P.-E.C. interpreted

data and prepared the manuscript. All authors read and approved the manuscript.

S.G. had full access to all of the data in the study and takes responsibility for the

integrity of the data and the accuracy of the data analysis.

Correspondence and requests for reprints should be addressed to Sebastien

Gibot, M.D., Ph.D., Hopital Central, Service de Reanimation Medicale, 29 avenue

du Marechal de Lattre de Tassigny, 54035 Nancy Cedex, France. E-mail:

s.gibot@chu-nancy.fr

This article has an online supplement, which is accessible from this issue’s table of

contents at www.atsjournals.org

Am J Respir Crit Care Med Vol 186, Iss. 1, pp 65–71, Jul 1, 2012

Copyright ª 2012 by the American Thoracic Society

Originally Published in Press as DOI: 10.1164/rccm.201201-0037OC on April 26, 2012

Internet address: www.atsjournals.org

AT A GLANCE COMMENTARY

Scientific Knowledge on the Subject

Rapid diagnosis of sepsis remains a challenge to clinicians.Measurement of individual biomarkers is often of marginalusefulness.

What This Study Adds to the Field

This prospective study, including inceptive and validationcohorts of unselected intensive care unit patients, demon-strates the good performance of a bioscore combining thepolymorphonuclear CD64 index with procalcitonin andsoluble triggering receptor expressed on myeloid cells-1 se-rum levels in diagnosing sepsis in the critically ill patient.

of the assay of CD64 expression by PMN measurement has tobe better assessed. Altogether, these data suggest that giventheir respective predictive values, a diagnosis approach seldomused (18, 19) combining these three biomarkers could improvesepsis recognition.

Here we performed a systematic study aimed at evaluatingthe individual and combined diagnostic accuracy of PCT,sTREM-1, and CD64 expression on PMN for the differential di-agnosis of sepsis in critically ill patients at ICU admission. Ina derivation cohort 300 consecutive patients were enrolled toconstruct a simple biologic score combining the three biomarkersand was next validated in an independent prospective cohort of79 critically ill patients from another center.

METHODS

Study Population

All consecutive patients newly hospitalized in the ICU of the UniversityHospital of Nancy, France, were prospectively enrolled in the study, ex-cept those admitted on weekends. There were no exclusion criteria.These 300 patients constituted the inception cohort, which was usedto determine the performance of the parameters examined in this study.

A distinct population of 79 patients was prospectively recruited inthe ICU of the University Hospital of Dijon, France, with the same cri-teria and during a 4-month period. This population constituted the val-idation cohort.

Approval of the institutional review board and informed consentwere obtained before inclusion.

Data Collection

On admission to the ICU, the following items were recorded for eachpatient: age, sex, severity of underlying medical condition stratifiedaccording to the criteria ofMcCabe and Jackson, SimplifiedAcute Phys-iology Score II (20), Sepsis-related Organ Failure Assessment score(21), and reason for admission into the ICU. The following baselinevariables were also recorded at inclusion: body temperature, leukocytecount, presence of shock, and use of previous antimicrobial therapy.The length of ICU stay and ICU mortality were also recorded.

When infection at admission was suspected by the attending physi-cian, microbiologic tests and antimicrobial therapy were prescribedaccording to the usual practice of the ICU without interference bythe research team or the knowledge of biomarker measurement results.Two intensivists retrospectively reviewed all medical records pertainingto each patient and independently classified the diagnosis as no infec-tion, sepsis, severe sepsis, or septic shock at the time of admission,according to established consensus definitions (4). Agreement concern-ing the diagnosis was achieved in all cases. Both intensivists weremasked to the results of the biomarkers studied.

Measurements of PCT, sTREM-1, and PMN CD64 Index

Within 12 hours after admission blood samples were drawn to assay PCTand sTREM-1 and to evaluate CD64 expression on PMN. Samplingwas repeated on Day 2. PCT concentrations were measured in a serumsample, using an immunoassay with a sandwich technique and a chemi-luminescent detection system (LumiTest; Brahms Diagnostica, Berlin,Germany). Serum concentrations of soluble TREM-1 were measured induplicate by sandwich ELISA using the Quantikine kit assay (R&DSystems, Minneapolis, MN). Interassay and intraassay coefficients ofvariation were lower than 7%. The expression of CD64 on neutrophilswas measured by quantitative flow cytometry within 12 hours afterblood sampling using the Leuko64 assay (Trillium Diagnostics, LLC,Brewer, ME) as previously described (13–17).

Statistical Analyses

Descriptive results of continuous variables were expressed as mean(6 SD) or median (interquartile range) depending of the normalityof their distribution. Variables were tested for their association withthe diagnosis using Pearson chi-square test for categorical data and

Mann-Whitney U test for numerical data. Receiver-operating charac-teristic (ROC) curves were constructed to illustrate cut-off values ofPCT, sTREM-1, and PMN CD64 index. Biomarker values were trans-formed into quartiles based on their distribution and their accuracy todiagnose sepsis was evaluated using a multiple stepwise logistic regres-sion model. Any covariate with univariate significance of P less than0.10 was eligible for inclusion in the model. Having determined thatthese biomarkers were independently associated with the diagnosis, wegenerated a scoring system or “bioscore” attributing one point perbiomarker with a value above the optimal cut-off point. This bioscorewas next tested for its association with diagnosis through logisticregression analysis and calibration of the model performed throughHosmer-Lemeshow testing.

The modeling process was performed on the inception cohort ofpatients and applied then to the validation cohort.

RESULTS

Characteristics of the Inception Cohort Patients

The baseline characteristics of the inception cohort are shownin Table 1. Among the 300 patients enrolled in this study, 154(51.3%) were diagnosed with sepsis. The source of infection andmicroorganisms involved are shown in Table E1 in the onlinesupplement. At admission, PCT, sTREM-1, and PMN CD64index were higher in patients with sepsis compared with allothers (P , 0.001 for the three markers) (Table 1; see FigureE1). Among patients with sepsis, PCT and sTREM-1 concen-trations were higher in case of gram-negative as compared withgram-positive infections, respectively 30.09 (4.55–76.01) versus1.33 (1.04–4.98) ng/ml for PCT, and 943 (518–1339) versus 676(390–1,010) pg/ml for sTREM-1 (P , 0.0001 for both).

Accuracy of Baseline PCT, sTREM-1, and PMN CD4 Index

in Diagnosing Sepsis

As shown in Figure 1, the PMN CD64 index yielded the highestdiscriminative value with an area under the ROC curve (AUC)of 0.95 (95% confidence interval [CI], 0.92–0.97; P , 0.001),followed by PCT (AUC 0.91 [0.87–0.94]; P , 0.001), andsTREM-1 (AUC 0.73 [0.67–0.79]; P , 0.001). Table 2 summa-rizes the performances of each of these biomarkers in diagnos-ing sepsis. PMN CD64 index proved to be the best individualmarker in terms of specificity (95.2%) and sensitivity (84.4%).

In multiple logistic regression, PCT, sTREM-1, and PMNCD64 index were all found to be independent predictors of sepsis(Table 3).

Combination of PCT, sTREM-1, and PMN CD64 Index

in a Bioscore

To determine whether the combination of these three biomarkersinto a single score could improve the diagnostic performance, in-dividual data were scored as 0 or 1 whether they were below orabove the threshold previously determined with the ROC curvesas shown in Table 2. This constituted the bioscore, which there-fore ranged between 0 (all three markers below their respectivethresholds) and 3 (all three markers above threshold). We choseto not weigh individual biomarkers based on their respectiveWald coefficients (Table 3) because results did not differ fromthose presented below, and from a clinical standpoint the sim-plest tests are likely to be the most applicable.

The probability of sepsis increased together with the bioscorewith rates of infection of 3.8% for a bioscore of 0–100% fora bioscore of 3 (Figure 2).

When the bioscore was entered into the multiple logistic re-gression model (Table 3) its performance was shown to be farbetter than that of each individual biomarker taken individually.The Hosmer-Lemeshow goodness-of-fit test showed that the

66 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 186 2012

model was well calibrated with P equal to 0.74. The AUC was0.97 (95% CI, 0.95–0.99) and 90.9% of patients were correctlyclassified by use of the model.

When one of the biomarkers was omitted from the bioscore(this score thus ranging from 0–2), diagnostic performances weremodified. The best odds ratio was observed for the combinationPMN CD64 index plus sTREM-1 at 26.50 (95% CI, 13.47–52.14), followed by PMN CD64 index plus PCT at 18.16 (95%CI, 10.15–32.50), and sTREM-1 plus PCT at 11.48 (95% CI,6.80–19.38).

To further test the performance of the bioscore on sepsis pre-diction for clinical relevance, the entire inception cohort of 300patients was classified according to baseline bioscore. A bioscoreof 0 (35% of patients) excluded the presence of sepsis for all but4 of the 105 patients (96.2%) in this category (two suffering fromherpes simplex virus-1 encephalitis, and two with a mild acuteexacerbation of chronic obstructive pulmonary disease). By con-trast, all but 5 of the 134 patients (96.3%) with a bioscore of 2 or3 (44.6% of the cohort) were septic. Between these two groups,the 61 patients with a bioscore of 1 (20.3% of the cohort) werebalanced between infected (34.4%) and not infected (65.6%)(Figure 3).

Recalculation of the bioscore for these 61 patients with a scoreof 1, using values obtained 24 hours after admission, showed im-proved data with 26.2% of the patients down to 0 and absence ofsepsis, and 14.8% with a bioscore increased up to 2 or 3 and thepresence of sepsis. For the 36 patients (59%) who did not changeand remained with a bioscore of 1 the sepsis rate still was 33.3%.Thus, from a clinical point of view, the bioscore was overall use-ful in 88% of patients over the first 24 hours after admission.

Usefulness of PCT, sTREM-1, PMN CD64 Index, and Bioscore

among Patients with an Initial Suspicion of Infection

We next evaluated the usefulness of the three biomarkers andthe bioscore among the 228 patients presenting with a clinicalsuspicion of infection. All of them initially received antibiotics.

Performances of individual or combinedmarkers were very closeto those obtained in the entire cohort of patients (see Tables E2and E3, and Figure E2). Interestingly, the optimal cut-offs asdetermined from ROC curves were similar to those derivedfrom the entire population (see Table E2).

External Validation of the Bioscore

The characteristics of the 79 patients included in the validationcohort are presented in Table E4. Compared with patients

TABLE 1. BASELINE CHARACTERISTICS OF THE INCEPTION COHORT

Characteristic All Patients (n ¼ 300) Patients without Sepsis (n ¼ 146) Patients with Sepsis (n ¼ 154) P Value†

Age, yr* 61 (46–73) 60 (44.5–71) 61 (47.5–74) 0.26

Sex, n (%)

Male 171 (57) 86 (59) 85 (63) 0.32

Female 129 (43) 60 (41) 69 (45)

McCabe score 1 (0.5–2) 1 (0.5–2) 1 (0.5–2) 0.98

History of immunodepression, n (%) 24 (8) 11 (7.5) 13 (8.4) 0.67

SAPS II score 46 (34–60) 42 (29–53) 50 (37–67) ,0.001

SOFA score 6 (3–10) 4 (2–6) 9 (5–14) ,0.001

Reason for admission, n (%)

Acute respiratory failure 55 (18.3) 33 (22.6) 22 (14.3) 0.18

Neurologic 86 (28.7) 67 (45.9) 19 (12.3) ,0.001

Shock 104 (34.7) 23 (15.7) 81 (52.6) ,0.001

Miscellaneous 55 (18.3) 23 (15.7) 32 (20.8) 0.43

Previous antiotherapy within 24 h, n (%) 45 (15) 10 (7) 35 (23) ,0.001

Mechanical ventilation, n (%) 193 (64.3) 82 (56.2) 111 (72.1) 0.038

Vasopressors, n (%) 128 (42.6) 26 (17.8) 102 (65.6) ,0.001

Body temperature, 8C 37.3 (36.7–37.9) 37.2 (36.7–37.7) 37.3 (36.6–38) 0.21

Leukocyte count, cells/mm3 11,700 (8,100–16,600) 11,000 (8,300–13,600) 13,400 (7,900–19,850) 0.002

Procalcitonin, ng/ml 1.55 (0.21–13.56) 0.24 (0.12–0.87) 10.38 (2.47–36.71) ,0.001

sTREM-1, pg/ml 540 (346–894) 403 (278–602) 773 (426–1,185) ,0.001

Neutrophils CD64 Index 1.51 (0.98–3.05) 0.99 (0.84–1.26) 2.99 (2.04–4.79) ,0.001

Length of ICU stay, d 3 (2–7) 2 (1–4) 5 (3–11) ,0.001

Mortality, n (%) 54 (18) 14 (10) 40 (26) ,0.001

Definition of abbreviations: ICU ¼ intensive care unit; SAPS II ¼ Simplified Acute Physiologic Score II; SOFA ¼ Sepsis-related Organ Failure Assessment; sTREM-1 ¼soluble triggering receptor expressed on myeloid cells-1.

* Unless specified, values are given as median (25th percentile to 75th percentile).y P values are comparisons between nonseptic and septic groups.

Figure 1. Receiver-operating characteristic curves for various cut-off

levels of PCT, sTREM-1, and PMN CD64 index in differentiating be-

tween the presence and absence of sepsis at admission. Areas underthe receiver-operating characteristic curves for PMN CD64 index (0.95

[95% CI, 0.92–0.97]); PCT (0.91 [95% CI, 0.87–0.94]); and sTREM-1

(0.73 [95% CI, 0.67–0.79]). CI ¼ confidence interval; PCT ¼ procalci-

tonin; PMN ¼ polymorphonuclear; sTREM-1 ¼ soluble triggering re-ceptor expressed on myeloid cells-1.

Gibot, Bene, Noel, et al.: Combined Biomarkers in Sepsis 67

enrolled in the inception cohort, they were older, more oftenimmunosuppressed, received more antibiotics before ICU ad-mission, and required vasopressors more frequently. Amongthese 79 patients, 36 (45.6%) were diagnosed with sepsis. Thecharacteristics of patients with and without sepsis are shownin Table E5.

The diagnostic accuracy of PCT, sTREM-1, PMN CD64 in-dex, and bioscore was similar to what had been observed inthe inception cohort with respective AUC at 0.88 (95% CI,0.79–0.97), 0.73 (0.67–0.79), 0.93 (0.87–0.99), and 0.95 (0.89–0.99) (all P , 0.0001).

In logistic regression analysis, the diagnostic odds ratios of thebaseline bioscore in this smaller cohort was close to that for theinception cohort (Table 4). A similar usefulness of the bioscoreat baseline and on Day 1 was observed (see Figure E3), with

a definitive diagnosis ultimately predicted for 86.1% of thepatients over the first 24 hours.

In this validation cohort, the bioscore also performed wellin the subgroup of patients with a clinical suspicion of infection(n ¼ 40). Respective odds ratios were 3.3 (95% CI, 1.3–40.2),40 (2–794.5), and 90 (4.7–1,709) for bioscore 1, 2, and 3 (all P ,0.05).

DISCUSSION

This study reports, in two independent cohorts of patients, a pro-spective evaluation of the diagnostic value of three sepsis bio-markers considered respectively as individual parameters or asa combined bioscore. We show here that PCT, sTREM-1, andPMN CD64 index were useful to diagnose sepsis, the best being

TABLE 2. CLINICAL PERFORMANCE OF BIOMARKERS IN DIAGNOSING SEPSIS

Biomarker Cut-off* Sensitivity (%)† Specificity (%)† PLR NLR PPV (%) NPV (%)

PCT 1.55 ng/ml 83.1 (76.2–88.6) 84.9 (78.1–90.3) 5.5 0.20 85.3 82.7

sTREM-1 755 pg/ml 53.2 (45.1–61.3) 86.3 (79.6–91.4) 3.9 0.54 80.4 63.6

PMN CD64 Index 1.62 84.4 (77.7–89.7) 95.2 (90.3–98.1) 17.6 0.16 94.9 85.3

Definition of abbreviations: NLR ¼ negative likelihood ratio; NPV ¼ negative predictive value; PCT ¼ procalcitonin; PLR ¼positive likelihood ratio; PMN ¼ polymorphonuclear; PPV ¼ positive predictive value; sTREM-1 ¼ soluble triggering recep-

tor expressed on myeloid cells-1.

* Cut-offs were determined using the Youden index (J ¼ max[sens 1 spec 2 1]).y Presented with 95% confidence intervals.

TABLE 3. MULTIPLE LOGISTIC-REGRESSION ANALYSIS OF FACTORS USED FOR DIFFERENTIATINGBETWEEN PATIENTS WITH AND THOSE WITHOUT SEPSIS IN THE INCEPTION COHORT

Variable Coefficient Standard Error x2 P Value Odds Ratio (95% CI)

Model 1*

SAPS II score 0.07 0.01 0.3 0.57 0.9 (0.9–1.1)

SOFA score 0.48 0.23 1.7 0.14 1.9 (0.8–13.6)

Vasopressor use 0.38 0.46 0.7 0.41 1.5 (0.6–3.6)

Mechanical ventilation 1.15 0.67 1.1 0.11 2.6 (0.8–10.9)

Previous antibiotherapy within 24 h 0.88 0.61 2.1 0.15 2.4 (0.7–7.9)

White blood cells count 0.32 0.20 2.4 0.12 1.4 (0.9–2.1)

PMN CD64 Index

Q2 0.29 0.65 0.2 0.64 1.3 (0.3–4.8)

Q3 4.05 0.72 31.4 ,0.001 57.4 (13.9–236.9)

Q4 5.99 1.23 23.5 ,0.001 400 (35.5–4509.6)

PCT

Q2 0.41 0.67 0.4 0.53 1.5 (0.4–5.6)

Q3 2.22 0.99 5 0.02 9.2 (1.3–65.3)

Q4 2.29 0.70 10.4 0.001 9.9 (2.4–39.6)

sTREM-1

Q2 0.38 0.65 0.3 0.55 1.4 (0.4–5.3)

Q3 0.66 0.66 1 0.31 1.9 (0.5–7.1)

Q4 2.09 0.76 7.5 0.006 8.1 (1.8–36.4)

SAPS II score 0.15 0.06 0.4 0.63 0.9 (0.9–1.1)

SOFA score 0.63 0.42 1.1 0.17 1.7 (0.6–15.8)

Vasopressor use 0.49 0.62 0.6 0.43 1.6 (0.5–5.5)

Mechanical ventilation 1.24 0.65 2.6 0.12 2.4 (0.8–12.6)

Previous antibiotherapy within 24 h 0.85 0.61 1.9 0.16 2.3 (0.7–7.8)

White blood cells count 0.24 0.25 0.9 0.34 1.3 (0.8–2.1)

Model 2†

BIOSCORE

1 2.72 0.59 20.9 ,0.001 15.2 (4.7–48.8)

2 6.05 0.75 65.6 ,0.001 425.7 (98.4–1842.2)

3 7.76 1.21 41.4 ,0.001 2355.2 (221.2–25079.9

Definition of abbreviations: CI ¼ confidence interval; PCT ¼ procalcitonin; PMN ¼ polymorphonuclear; Q ¼ quartile;

SAPS II ¼ Simplified Acute Physiologic Score II; SOFA ¼ Sepsis-related Organ Failure Assessment; sTREM-1 ¼ soluble

triggering receptor expressed on myeloid cells-1.

Model 1: individual biomarkers were entered into the model.

Model 2: biomarkers were combined into the bioscore that was then entered into the model.

* Pseudo R2 (Cox & Snell) 0.663.y Pseudo R2 (Cox & Snell) 0.669.

68 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 186 2012

assessment of CD64 expression on PMNs. We then combinedthese markers in a simple score, called “bioscore,” which turnedout to be associated with an impressive diagnostic accuracy andproved really useful from a clinical standpoint in rapidly assigningmore than 80% of the patients to having or not having sepsis.

PCT has proved useful for the diagnosis of sepsis in criticallyill patients, although it may lack specificity as attested by the highrate of elevated PCT in clinical conditions unrelated to sepsis(5–7, 22–24). Accordingly, several studies have provided lessconvincing data making controversial the use of this biomarker(25, 26). Here, we observed satisfactorily discriminating sensitiv-ity and specificity values reaching 83% and 84%, respectively.

By contrast, sTREM-1 showed a diagnostic accuracy lowerthan expected in both these cohorts. Our group and others havereported that the test was sensitive and specific in the ICU setting(8, 27). The sensitivity of the assay was poorer here, possibly ex-plained by a lower prevalence of infection than in publishedreports. Nevertheless, the performance of the bioscore decreasedwhen this marker was omitted, and the best odds ratio was ob-served for the combination PMN CD64 index plus sTREM-1.

CD64 expression up-regulation on the cell surface of PMNand monocytes is considered to be a very early step of the im-mune host response to bacterial infection (28). Conversely,CD64 expression seems unchanged in patients with inflamma-tory states from noninfectious origin, especially on neutrophils(16). Accordingly, previously published clinical data have sug-gested the high diagnostic value of the PMN CD64 index mea-surement (13–15, 29–31). However, these were small series ofpatients including mainly pediatric subjects. In a recently pub-lished work performed in an ICU the CD64 index was found tohave a lower sensitivity (63%) for the CD64 index than previ-ously reported, although its specificity was excellent (32). Ourfindings are also in contradiction with this study. This may beexplained by the fact that Gros and coworkers (32) only in-cluded patients with a documented infection, thus missing manyinfected patients without microbial documentation, leading toa higher CD64 cutoff. The second explanation may be technical:although using the same commercial kit than in our study, Grosand coworkers (32) measured CD64 index “within 36 hours”after blood sampling rather than within 12 hours. This may haveled to a progressive decrease of CD64 expression and a highnumber of false-negative patients. Despite these small discrep-ancies, the PMN CD64 index was confirmed here to be the mostperforming parameter. In addition, our work illustrates the highfeasibility of the CD64 index measurement in daily practice.The test is performed as an individual parameter, and can berealized at any time, provided a flow cytometry facility is avail-able. It is a whole-blood lysis no-wash approach unlikely to

artifactually modify any of the intrinsic parameters of the cellstested through the analytic pathway.

Beyond the individual performance of each biomarker, theircombination within a bioscore additionally seemed to be a prag-matic and efficient way to differentiate between ICU patientswith and without sepsis. Therefore, with at least two of the threemarkers above their respective threshold (bioscore 2 or 3) atbaseline, more than 90% of patients were found to be infected.

After the clinical work-up is achieved, and before getting mi-crobiologic data, the diagnosis of sepsis remains in many casesa plausible yet not demonstrated option. With one single bio-marker, incertitude often remains and the best therapeuticschedule is difficult to establish, keeping in mind the consequen-ces of unnecessary antibiotherapy. Some authors have proposedto combine biomarkers to improve the diagnostic accuracy indi-vidually provided by each one (19, 33). The pragmatic approachapplied here relies on a simple bedside bioscore of 0–3, witha strong positive predictive value for infection if above 1. Thisthreshold is reliable at baseline for scores 0, 2, and 3, and rein-forced by Day 1 data for most patients with a baseline bioscoreof 1. Although the bioscore’s calculation implies the measure-ment of three biomarkers, it clearly provides relevant informa-tion likely to strengthen the physician’s decision in addition toclinical work-up.

This study is especially interesting because consecutive ICUpatients were enrolled, therefore without necessarily a clinicalsuspicion of sepsis. This had the clear advantage of not usinghealthy subjects as controls subjects, but to place the tests inthe “real life” conditions of the ICU. It may be argued thatfor some of these patients, the dilemma of “sepsis or not sepsis”did not exist and that in routine conditions the indication for

Figure 2. Percentage of patients with sepsis according to the bioscore

in the inception cohort.

Figure 3. Classification of the patients according to their bioscore (in-ception cohort). NS ¼ nonseptic; S ¼ septic.

Gibot, Bene, Noel, et al.: Combined Biomarkers in Sepsis 69

prescribing either test or the entire bioscore could be morelimited, and this can be accepted. Nevertheless, when analyseswere restricted only to patients presenting with a clinical suspi-cion of sepsis, performances of the biomarkers and the bioscoreseemed to be very close to those obtained in the entire cohort ofpatients. Such an absence of selection criteria may also explainthe higher thresholds established for all three parameters stud-ied than previously reported.

The combination of several biomarkers to improve diagnosisor predict outcome is used in various disorders, such as liver fi-brosis, breast cancer, and cardiovascular diseases (34–36), al-though this strategy has not yet been validated for sepsis. Thebiologic scoring system we propose here implies the measure-ment of three different markers. At first glance, this raises fea-sibility issues. Indeed, if PCT measurement may be performedeasily, sTREM-1 determination requires an ELISA assay whereasCD64 assessment requires access to a flow cytometer. These fa-cilities may not be routinely available in all hospitals. This mayconstitute an obstacle to the widespread use of such a strategy,although point-of-care technologies are rapidly spreading.

Finally, the scoring strategy we propose must be evaluatedthrough medicoeconomic studies that should balance the weightof useless antibiotic therapies toward unacceptable delays in theadministration of appropriate antimicrobial agents in any patientwith actual but unconfirmed sepsis.

In conclusion, this prospective, two-part study, including in-ceptive and validation cohorts of unselected ICU patients in twodifferent settings strengthens the high diagnostic value of thePMN CD64 index, and the even higher value of a bioscore com-bining this parameter with PCT and sTREM-1 serum levels.Whether any of these should routinely be implemented in ICUs,and on which screening criteria, remains to be established byconsensus confrontations. This work, however, provides robustthresholds for all three parameters as a strong basis for furthermulticenter studies.

Author disclosures are available with the text of this article at www.atsjournals.org.

Acknowledgment: The authors are grateful to Arthur Lefebvre and the techniciansin the Immunology and Haematology laboratories in Nancy and Dijon, who per-formed the tests.

References

1. Vincent JL, Rello J, Marshall J, Silva E, Anzueto A, Martin CD, Moreno

R, Lipman J, Gomersall C, Sakr Y, et al. EPIC II Group of Inves-

tigators. International study of the prevalence and outcomes of in-

fection in intensive care units. JAMA 2009;302:2323–2329.

2. Kumar A, Roberts D, Wood KE, Light B, Parrillo JE, Sharma S, Suppes

R, Feinstein D, Zanotti S, Taiberg L, et al. Duration of hypotension

before initiation of effective antimicrobial therapy is the critical de-

terminant of survival in human septic shock. Crit Care Med 2006;34:

1589–1596.

3. O’Grady NP, Barie PS, Bartlett JG, Bleck T, Carroll K, Kalil AC,

Linden P, Maki DG, Nierman D, Pasculle W, et al; American College

of Critical Care Medicine. Infectious Diseases Society of America.

Guidelines for evaluation of new fever in critically ill adult patients:

2008 update from the American College of Critical Care Medicine

and the Infectious Diseases Society of America. Crit Care Med 2008;

36:1330–1349.

4. Levy MM, Fink MP, Marshall JC, Abraham E, Angus D, Cook D,

Cohen J, Opal SM, Vincent JL, Ramsay G; SCCM/ESICM/ACCP/

ATS/SIS. 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis

Definitions Conference. Crit Care Med 2003;31:1250–1256.

5. Harbarth S, Holeckova K, Froidevaux C, Pittet D, Ricou B, Grau GE,

Vadas L, Pugin J; Geneva Sepsis Network. Diagnostic value of pro-

calcitonin, interleukin-6, and interleukin-8 in critically ill patients admit-

ted with suspected sepsis. Am J Respir Crit Care Med 2001;164:396–402.

6. Muller B, Becker KL, Schachinger H, Rickenbacher PR, Huber PR,

Zimmerli W, Ritz R. Calcitonin precursors are reliable markers of

sepsis in a medical intensive care unit. Crit Care Med 2000;28:977–983.

7. Uzzan B, Cohen R, Nicolas P, Cucherat M, Perret GY. Procalcitonin as

a diagnostic test for sepsis in critically ill adults and after surgery or

trauma: a systematic review and meta-analysis. Crit Care Med 2006;

34:1996–2003.

8. Jiyong J, Tiancha H, Wei C, Huahao S. Diagnostic value of the soluble

triggering receptor expressed on myeloid cells-1 in bacterial infection:

a meta-analysis. Intensive Care Med 2009;35:587–595.

9. Hoffmann JJ. Neutrophil CD64: a diagnostic marker for infection and

sepsis. Clin Chem Lab Med 2009;47:903–916.

10. Hausfater P, Juillien G, Madonna-Py B, Haroche J, Bernard M, Riou B.

Serum procalcitonin measurement as diagnostic and prognostic

marker in febrile adult patients presenting to the emergency depart-

ment. Crit Care 2007;11:R60.

11. Bingold TM, Pullmann B, Sartorius S, Geiger EV, Marzi I, Zacharowski

K, Wissing H, Scheller B. Soluble triggering receptor expressed on

myeloid cells 1 is expressed in the course of non-infectious inflam-

mation after traumatic lung contusion: a prospective cohort study.

Crit Care 2011;15:R115.

12. El Sohl AA, Akinusi ME, Peter M, Berim I, Schultz MJ, Pineda M.

Triggering receptors expressed on myeloid cells in pulmonary aspi-

ration syndromes. Intensive Care Med 2008;34:1012–1019.

13. Groselj-Grenc M, Ihan A, Pavcnik-Arnol M, Kopitar AN, Gmeiner-

Stopar T, Derganc M. Neutrophil and monocyte CD64 indexes,

lipopolysaccharide-binding protein, procalcitonin and C-reactive

protein in sepsis of critically ill neonates and children. Intensive Care

Med 2009;35:1950–1958.

14. Cardelli P, Ferraironi M, Amodeo R, Tabacco F, De Blasi RA, Nicoletti

M, Sessa R, Petrucca A, Costante A, Cipriani P. Evaluation of neu-

trophil CD64 expression and procalcitonin as useful markers in early

diagnosis of sepsis. Int J Immunopathol Pharmacol 2008;21:43–49.

15. Livaditi O, Kotanidou A, Psarra A, Dimopoulou I, Sotiropoulou C,

Augustatou K, Papasteriades C, Armaganidis A, Roussos C, Orfanos

SE, et al. Neutrophil CD64 expression and serum IL-8: sensitive early

markers of severity and outcome in sepsis. Cytokine 2006;36:283–290.

16. Matsui T, Ohsumi K, Ozawa N, Shimada K, Sumitomo S, Shimane K,

Kawakami M, Nakayama H, Sugii S, Ozawa Y, et al. CD64 on neu-

trophils is a sensitive and specific marker for detection of infection in

patients with rheumatoid arthritis. J Rheumatol 2006;33:2416–2424.

TABLE 4. MULTIPLE LOGISTIC-REGRESSION ANALYSIS OFFACTORS USED FOR DIFFERENTIATING BETWEEN PATIENTS WITHAND THOSE WITHOUT SEPSIS IN THE VALIDATION COHORT

Variable Coefficient

Standard

Error x2 P Value

Odds Ratio

(95% CI)

Model 1*

PMN CD64

Index

Q2 21.17 1.32 0.8 0.37 0.3 (0.1–4.1)

Q3 3.11 1.10 7.92 0.004 22.6 (2.6–197.6)

Q4 4.00 1.12 12.7 ,0.001 55 (6.1–498)

PCT

Q2 0.32 0.90 0.1 0.68 1.4 (0.2–9.1)

Q3 2.85 1.07 7.1 0.007 17.3 (2.1–141.7)

Q4 4.30 1.37 9.8 0.001 73.9 (5–1093.7)

sTREM-1

Q2 0.77 0.70 1.21 0.27 2.1 (0.5–8.6)

Q3 0.92 0.28 4.3 0.02 2.8 (1.3–10.7)

Q4 1.68 0.86 3.8 0.04 5.4 (1.1–29)

Model 2†

BIOSCORE

1 1.93 1.18 2.7 0.10 6.9 (0.7–69.8)

2 5.02 1.48 11.6 ,0.001 152 (8.4–2741.9)

3 5.83 1.45 16.1 ,0.001 342 (19.8–5889.2)

Definition of abbreviations: CI ¼ confidence interval; PCT ¼ procalcitonin; PMN ¼polymorphonuclear; Q ¼ quartile; sTREM-1 ¼ soluble triggering receptor expressed

on myeloid cells-1.

Model 1: individual biomarkers were entered into the model.

Model 2: biomarkers were combined into the bioscore that was then entered

into the model.

* Pseudo R2 (Cox & Snell) 0.548.y Pseudo R2 (Cox & Snell) 0.609.

70 AMERICAN JOURNAL OF RESPIRATORY AND CRITICAL CARE MEDICINE VOL 186 2012

17. Barth E, Fischer G, Schneider EM, Wollmeyer J, Georgieff M, Weiss M.

Differences in the expression of CD64 and mCD14 on polymorpho-

nuclear cells and on monocytes in patients with septic shock. Cytokine

2001;14:299–302.

18. Gibot S, Cravoisy A, Dupays R, Barraud D, Nace L, Levy B, Bollaert PE.

Combined measurement of procalcitonin and soluble TREM-1 in the

diagnosis of nosocomial sepsis. Scand J Infect Dis 2007;39:604–608.

19. Kofoed K, Andersen O, Kronborg G, Tvede M, Petersen J, Eugen-Olsen

J, Larsen K. Use of plasma C-reactive protein, procalcitonin, neu-

trophils, macrophage migration inhibitory factor, soluble urokinase-

type plasminogen activator receptor, and soluble triggering receptor

expressed on myeloid cells-1 in combination to diagnose infections:

a prospective study. Crit Care 2007;11:R38.

20. Le Gall JR, Lemeshow S, Saulnier F. A new Simplified Acute Physiology

Score (SAPS II) based on a European/North American multicenter

study. JAMA 1993;270:2957–2963.

21. Vincent JL, Moreno R, Takala J, Willatts S, De Mendonca A, Bruining H,

Reinhart CK, Suter PM, Thijs LG. The SOFA (Sepsis-related Organ

Failure Assessment) score to describe organ dysfunction/failure. On be-

half of the Working Group on Sepsis-Related Problems of the European

Society of Intensive Care Medicine. Intensive Care Med 1996;22:707–710.

22. Clec’h C, Ferriere F, Karoubi P, Fosse JP, Cupa M, Hoang P, Cohen Y.

Diagnostic and prognostic value of procalcitonin in patients with

septic shock. Crit Care Med 2004;32:1166–1169.

23. Charles PE, Kus E, Aho S, Prin S, Doise JM, Olsson NO, Blettery B,

Quenot JP. Serum procalcitonin for the early recognition of nosoco-

mial infection in the critically ill patients: a preliminary report. BMC

Infect Dis 2009;9:49.

24. Charles PE, Ladoire S, Snauwaert A, Prin S, Aho S, Pechinot A, Olsson

NO, Blettery B, Doise JM, Quenot JP. Impact of previous sepsis on

the accuracy of procalcitonin for the early diagnosis of blood stream

infection in critically ill patients. BMC Infect Dis 2008;8:163.

25. Ugarte H, Silva E, Mercan D, De Mendonca A, Vincent JL. Procalci-

tonin used as a marker of infection in the intensive care unit. Crit Care

Med 1999;27:498–504.

26. Tang BM, Eslick GD, Craig JC, McLean AS. Accuracy of procalcitonin

for sepsis diagnosis in critically ill patients: systematic review and

meta-analysis. Lancet Infect Dis 2007;7:210–217.

27. Gibot S, Kolopp-Sarda MN, Bene MC, Cravoisy A, Levy B, Faure GC,

Bollaert PE. Plasma level of a triggering receptor expressed on

myeloid cells-1: its diagnostic accuracy in patients with suspected

sepsis. Ann Intern Med 2004;141:9–15.

28. van der Meer W, Pickkers P, Scott CS, van der Hoeven JG, Gunnewiek

JK. Hematological indices, inflammatory markers and neutrophil

CD64 expression: comparative trends during experimental human

endotoxemia. J Endotoxin Res 2007;13:94–100.

29. Lewis SM, Treacher DF, Bergmeier L, Brain SD, Chambers DJ, Pearson

JD, Brown KA. Plasma from patients with sepsis up-regulates the

expression of CD49d and CD64 on blood neutrophils. Am J Respir

Cell Mol Biol 2009;40:724–732.

30. Bhandari V, Wang C, Rinder C, Rinder H. Hematologic profile of sepsis

in neonates: neutrophil CD64 as a diagnostic marker. Pediatrics 2008;

121:129–134.

31. Icardi M, Erickson Y, Kilborn S, Stewart B, Grief B, Scharnweber G.

CD64 index provides simple and predictive testing for detection and

monitoring of sepsis and bacterial infection in hospital patients. J Clin

Microbiol 2009;47:3914–3919.

32. Gros A, Roussel M, Sauvadet E, Gacouin A, Marque S, Chimot L,

Lavoue S, Camus C, Fest T, Le Tulzo Y. The sensitivity of neutrophil

CD64 expression as a biomarker of bacterial infection is low in crit-

ically ill patients. Intensive Care Med 2012;38:445–452.

33. Pierrakos C, Vincent JL. Sepsis biomarkers: a review. Crit Care 2010;

14:R15.

34. Maor Y, Cale S, Bashari D, Kenet G, Lubetsky A, Luboshitz J, Schapiro

JM, Naranda GPE, Bar-Meir S, Martinowitz U, et al. Improving es-

timation of liver fibrosis using combination and newer noninvasive

biomarker scoring systems in hepatitis C-infected haemophilia pa-

tients. Haemophilia 2007;13:722–729.

35. Evans DGR, Lalloo F, Cramer A, Jones EA, Knox F, Amir E, Howell

A. Addition of pathology and biomarker information significantly

improves the performance of the Manchester scoring system for

BRCA1 and BRCA2 testing. J Med Genet 2009;46:811–817.

36. Zethelius B, Berglund L, Sundstrom J, Ingelsson E, Basu S, Larsson A,

Venge P, Arnlov J. Use of multiple biomarkers to improve the pre-

diction of death from cardiovascular causes. N Engl J Med 2008;358:

2107–2116.

Gibot, Bene, Noel, et al.: Combined Biomarkers in Sepsis 71

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.