INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Severe sepsis is a systemic inflammatory response to infection, which involves a characteristic spectrum of pathologic changes in many host systems. It is one of the most devastating problems for patients who are critically ill. On infection, the pathophysiologic sequence for sepsis involves cytokine release, and endothelial and neutrophil activation, initiating a cascade of leukocyte-endothelium interactions and adhesions. This is followed by transendothelial migration and subsequent microvascular and tissue injury, leading ultimately to multiple organ failure (MOF) (1). Several authors have reported that endothelial and tissue damage is correlated with the intensity of the inflammatory response and with leukocyte sequestration within tissues (2).

Adhesion of neutrophils to the endothelium is regulated by at least three adhesion molecule families (selectins, integrins, and the immunoglobulin superfamily) and by chemotactic signals (3, 4). The expression of endothelial leukocyte adhesion molecule (E-selectin, CD62E) is restricted to stimulated endothelial cells (5), and it plays a key role in the initial phase of adhesion, known as rolling. Rolling is a prerequisite for eventual firm adhesion, which is mediated by interaction of neutrophil ₂-integrin with endothelial intercellular adhesion molecule 1 (ICAM-1, CD54). ICAM-1 is constitutively expressed at very low levels on the endothelium and is highly inducible on cell activation (3). Both E-selectin and ICAM-1-inducible adhesion molecules are found in the plasma, in soluble forms that increase in concentration during inflammation (6- 13). Another protein synthesized by endothelial cells, von Willebrand factor antigen (vWf:Ag), is stored in endothelial cell-specific Weibel-Palade bodies and may be constitutively secreted in the plasma (14). High levels of circulating vWf:Ag have been reported in several inflammatory diseases and in acute lung injury in patients with nonpulmonary sepsis in which endothelial damage is thought to be important (15, 16).

Neutrophil production and the release of toxic granular components increase on stimulation, with beneficial antimicrobial activity but potentially deleterious effects on host tissues (2). Myeloperoxidase (MPO), which is a major component of primary granules, and lactoferrin (Lacto-f  ), which is restricted to secondary granules, are secreted into the extracellular environment following neutrophil stimulation and their plasma levels may reflect neutrophil activation (17, 18).

Until now, the direct assessment of microvascular inflammatory phenomena in vivo has not been possible in clinical practice. We thought that levels of the soluble forms of E-selectin and ICAM-1 (sE-selectin, sICAM-1) might be correlated with the intensity of leukocyte-endothelium interactions during inflammation whereas levels of vWf:Ag might be correlated with the intensity of endothelial damage. Therefore we performed a prospective study of patients with sepsis to define the sensitivity and specificity of concentrations of sICAM-1, sE-selectin, vWf:Ag, MPO, and Lacto-f as predictive plasma markers of disease severity and eventual final outcome, in addition to commonly monitored clinical variables.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Design and Patients

This study was conducted in the medical intensive care unit (ICU) of the Boucicaut University-Medical Hospital, Paris. Our procedures were reviewed and approved by the Medical Ethics Committee of the Société de Réanimation de Langue Française and, whenever possible, verbal consent was obtained from every patient and control and from healthy volunteers. Twenty-five patients hospitalized in our ICU were enrolled prospectively in the study, and met clinical criteria of severe sepsis, or septic shock, as defined subsequently. Such criteria were met either on admission or during hospitalization within the ICU. Entry into the study was defined as Day 0. The first clinical evaluation and blood sampling were performed on Day 0 as soon as possible after diagnosis (usually within 1 h). The putative onset of sepsis was defined as the beginning of symptoms, as defined subsequently. During the follow-up period (from Day 0 to discharge from the ICU) and for each patient, the assessment of clinical data was based on computation of the simplified acute physiological score (SAPS) (19), daily acute lung injury score (ALI) according to the criteria of Murray and coworkers (20), and the number of organ failures (MOF score) according to the criteria of Tran and colleagues (21).

Levels of the five plasma proteins were also studied in a nonseptic control group, consisting of seven patients from the ICU. Blood samples were taken within 24 h of diagnosis except for control patients ventilated for nonpulmonary pathology, from whom blood samples were taken at least 72 h after the beginning of ventilatory support. All nonseptic control patients were followed for at least 48 h and there was no evidence of sepsis symptoms. Their clinical characteristics are reported in RESULTS.

Definitions of Clinical Severity Scoring, Infection, and Outcome

The criteria of sepsis, severe sepsis, and septic shock used were those established by a consensus conference (22). The systemic inflammatory response syndrome (SIRS) was defined by at least two of the following criteria: temperature > 38° C or < 36° C; heart rate (HR) > 90 beats/min; respiratory rate > 20 cycles/min or Pa_CO2 < 32 mm Hg; leukocytosis > 12,000 or < 4,000 elements/mm³. SIRS was treated as sepsis when it was associated with infection. Classification of patients was determined as follows:

1. Severe sepsis: Sepsis associated with at least one of the following conditions: organ dysfunction; signs of hypoperfusion (oliguria, encephalopathy, lactic acidosis, etc.); hypotension (systolic blood pressure [SBP] < 90 mm Hg or a decrease of 40 mm Hg from initial SBP) controlled by vascular expansion.

2. Septic shock: Severe sepsis associated with hypotension that cannot be controlled by vascular expansion and therefore requires vasopressive agents (dopamine,  7.5 mg/kg/min; epinephrine or norepinephrine at any dose) to maintain SBP > 100 mm Hg.

Multiple organ failure (MOF score) was assessed according to the criteria of Tran and coworkers (21). A score of 1 was assigned to each identified organ failure, and the MOF score was calculated every day by summing the number of involved organs. The cumulative MOF was the highest daily score calculated during the follow-up period. Organ failure included at least one symptom from the relevant category:

1. Cardiovascular: Mean arterial pressure (MAP) < 50 mm Hg, requirement for vasopressive drugs to maintain SBP above 100 mm Hg, sustained ventricular tachycardia, ventricular fibrillation, or reversible cardiovascular arrest, heart rate  50 beats/min, myocardial necrosis.

2. Respiratory: Spontaneous respiratory rate  50 or  5 cycles/ min, mechanical ventilation for more than 3 d, FI_O2 > 40% and/or PEEP > 5 cm H₂O (FI_O2, fraction of inspired oxygen; PEEP, positive end expiratory pressure).

3. Renal: Serum creatinine concentration > 280 µmol/L, requirement for extracorporeal exchange (dialysis or hemofiltration).

4. Neurological: Glasgow coma score  6 in the absence of sedation and neuromuscular blocking agents.

5. Gastrointestinal: Stress ulcer with bleeding requiring at least two erythrocyte units per 24 h, necrotizing pancreatitis defined as serum amylase level higher than twice the normal upper limit, acalculous cholecystitis, bowel ischemia or perforation.

6. Hepatic: Total bilirubin > 50 µmol/L (in the absence of hemolysis), serum glutamate pyruvate transaminase (SGPT) higher than twice the upper normal limit, hepatic encephalopathy with asterixis.

7. Hematological: Leukocytosis < 3,000 elements/mm³, disseminated intravascular coagulation, hematocrit < 20%.

The ALI was also assessed every day according to the published criteria of Murray and coworkers (20), based on a four-point scoring system: (1) Pa_O2/FI_O2 ratio, (2) pulmonary infiltrates, (3) static pulmonary compliance (when patients are ventilated), and (4) PEEP level. The cumulative ALI score was the highest daily score calculated during the follow-up period.

Organ infection was defined clinically and microbiologically as follows:

1. Pulmonary: New infiltrates visible on chest X-ray and purulent sputum or aspiration, altered polymorphonuclear cells or bacteria observed on a protected lung specimen (brush or minilavage), positive culture of  10³ colony-forming units/ml.

2. Abdominal: Pain and rigidity of the abdominal wall and scan evidence of a local abscess, polymorphonuclear cell-enriched peritoneal liquid.

3. Kidney: Clinical signs of cystitis or urethritis and leukocyturia with more than 5 elements/field or presence of bacteria after Gram staining, radiographic image evidence of dilation of the high urinary tract.

4. Neurologic: Nuchal rigidity and leukocyte-enriched cerebrospinal fluid (CSF) with high protein and low glucose concentrations.

5. Skin: Local erythema and purulent drainage of a postsurgical wound or peripheral or central catheter insertion site.

Infection was demonstrated microbiologically when the growth of at least one microorganism was observed in a biological fluid that should have been sterile (blood, pleural or peritoneal liquid, urine, CSF). Staphylococcus epidermidis bacteremia was regarded as true bacteremia when the same strain was obtained from at least two positive blood cultures.

Patients were classified as survivors if they were alive 30 d after inclusion in the study, even if they had already been discharged from the ICU.

Blood Samples

Serum samples were collected in nonheparinized tubes for measurement of sE-selectin and sICAM-1 concentrations and in EDTA tubes for that of vWf:Ag, Lacto-f, and MPO. All samples were kept at 4° C, centrifuged (2,500 × g for 15 min at 4° C) within 30 min of collection, aliquoted, and stored at 80° C until the time of assay. Blood specimens were obtained from arterial catheters or through peripheral venipuncture. The values obtained from arterial and vein blood samples collected simultaneously from the same patient were not significantly different.

Assay for the Five Plasma Proteins

Levels of sE-selectin, sICAM-1, vWf:Ag, MPO, and Lacto-f were measured by sandwich-type enzyme-linked immunosorbent assays (ELISAs).

A group consisting of nine normal healthy volunteers from the ICU personnel (three men and six women) with a mean age of 26 ± 3.6 yr (range, 22-33 yr) formed the basis for determination of normal values for sE-selectin (40.5 ± 4.5 ng/ml), sICAM-1 (208 ± 20.5 ng/ml), vWf:Ag (115 ± 26.7%), Lacto-f (557 ± 181.6 ng/ml), and MPO (49 ± 15.1 ng/ml).

Data Analysis

All values are given as mean ± standard error of the mean (SEM). Nonparametric statistics (Kruskall-Wallis, Mann-Whitney) were used to compare continuous or ordinal (MOF score, ALI score, SAPS) variables and Spearman rank nonparametric correlation coefficients were calculated.

We used several statistical methods to determine the relationships between outcome and various patient characteristics evaluated on Day 0. Receiver-operator characteristics (ROC) analysis was used to identify the Day 0 cutoff values for sE-selectin, sICAM-1, vWf:Ag, MPO, and Lacto-f that could be used as early outcome predictors. True Positive (TP) = value higher than cutoff when outcome was "nonsurvivor"; True Negative (TN) = value lower than cutoff when outcome was "survivor"; False Negative (FN) = value lower than cutoff when outcome was "nonsurvivor"; False Positive (FP) = value higher than cutoff when outcome was "survivor." Sensitivity (Se), specificity (Sp), positive predicted value (PPV) and negative predicted value (NPV) were calculated as follows: Se = TP/(TP + FN); Sp = TN/(FP + TN); PPV = TP/(TP + FP); NPV = TN/(FN + TN). The selected cutoff value was associated with the highest combination of sensitivity and specificity. The dichotomous prognostic criteria (higher or lower than cutoff) for the clinical and biological features studied were then analyzed by a stepwise logistic regression using SAS and Logxact statistical software (23). We compared the mean values of survivors and nonsurvivors over time using analysis of variance for repeated measures to test whether persistently high levels of circulating adhesion molecules were related to death. Differences were considered significant for p < 0.05, and 95% confidence intervals (CI 95%) were given for relative risks.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patient and Control Group Characteristics

The clinical characteristics of each patient with sepsis are given in Table 1A. The 25 patients studied (18 men and seven women) had a mean age of 57.2 ± 3.9 yr (range, 19 to 89 yr), mean SAPS of 14.7 ± 1.05, mean ALI score of 1.3 ± 0.23, and mean MOF score of 2.3 ± 0.38. None of the patients received corticosteroids or other immunosuppressive treatments for at least 2 mo before being enrolled in the study. Within the whole study population, 11 had septic shock (44%) and 20 were under mechanical ventilation. Sites of infection thought to be responsible for severe sepsis were identified for all of the patients and microbiological identification of the source of infection was obtained in 22 patients (88%). Gram-positive bacteria (n = 14) (Streptococcus [n = 6], Staphylococcus aureus [n = 7], Listeria monocytogenes [n = 1]), gram-negative bacteria (n = 8) (Escherichia coli [n = 3]; Klebsiella pneumoniae [n = 2], Pseudomonas aeruginosa [n = 1], Proteus mirabilis [n = 1], and Serratia marcescens [n = 1]), and in one case Plasmodium falciparum were the microorganisms responsible for severe sepsis. The most frequent sites of infection were lung (n = 10), urinary tract (n = 6), and skin (n = 7). One patient had a documented P. falciparum infection with neurologic complications. Positive results were obtained for blood cultures in 60% of patients.

The control group consisted of seven patients (four men and three women) hospitalized in the ICU for nonsepsis pathology. Their clinical characteristics are given in Table 1B. They had a mean age of 68.6 ± 4.5 yr (range, 44 to 80 yr), mean SAPS of 11.14 ± 2.16, and mean MOF score of 0.86 ± 0.4. Three patients had cardiogenic shock with no signs of acute ischemic process (end stage of a nonobstructive cardiomyopathy [n = 2], degenerative mitral valve rupture [n = 1]), two patients had hemodynamic pulmonary edema without cardiogenic shock, and two had strokes and required ventilatory support for more than 48 h. Three patients were under mechanical ventilation. There were no significant differences in terms of age, creatininemia, proteinemia, leukocytosis, hemoglobin, and MAP (62.8 ± 2.90 versus 75.3 ± 7.8 mm Hg) between the septic and control groups. As expected, temperature (39.2 ± 0.21 versus 37.0 ± 0.12° C; p < 0.001) and MOF score (2.3 ± 0.38 versus 0.86 ± 0.4; p < 0.05) were significantly higher in patients with infection.

Plasma Markers on Day 0

Soluble adhesion molecules (sE-selectin and sICAM-1), vWf:Ag, MPO, and Lacto-f mean plasma levels of patients with severe infection, determined on Day 0, were significantly higher than those of the noninfected control group of patients (Table 2).

To test whether any of the endothelium-derived (sE-selectin, sICAM-1, vWf:Ag) and leukocyte-derived (MPO and Lacto-f  ) proteins obtained on Day 0 could predict the value of another variable, a nonparametric Spearman rank correlation test was performed. This showed a strong correlation between levels of sE-selectin, sICAM-1, and vWf:Ag, but MPO and Lacto-f levels were correlated only with those of vWf:Ag.

We then compared each of the levels of the soluble plasma proteins with dichotomous clinical characteristics. Patients with septic shock (44%) had significantly higher levels of sE-selectin (295 ± 50 versus 181 ± 62 ng/ml; p < 0.05), sICAM-1 (1,133 ± 200 versus 660 ± 159 ng/ml; p < 0.05), and vWf:Ag (811 ± 74 versus 491 ± 64%; p < 0.01) than those without septic shock. There was no significant difference for MPO (309 ± 143 versus 58 ± 8 ng/ml) and for Lacto-f (2,189 ± 482 versus 1,047 ± 323 ng/ml) levels.

Patients with bacteremia (60%) had significantly higher levels of sE-selectin (301 ± 60 versus 126 ± 35; p < 0.05) and sICAM-1 (1,039 ± 170 versus 611 ± 188; p < 0.05) only.

Production of inflammatory plasma proteins is usually delayed after the initiation of inflammation and its kinetics might change rapidly with time, giving a transient peak (24), so we compared two subgroups differing in terms of lag time between the putative onset of sepsis and the first blood sample obtained on Day 0. On the basis of endothelial pathophysiology, and the known kinetics of sE-selectin expression after endotoxin challenge in vitro as well as in vivo (24), we chose 6 h as a discriminating lag time: < 6 h (n = 9); > 6 h (n = 16). sICAM-1 and MPO levels were slightly higher and sE-selectin slightly lower in the second subgroup (Table 3), but the differences were not significant, probably because survivors and nonsurvivors were evenly distributed throughout the study population.

Differences between Nonsurvivors and Survivors on Day 0

Comparison of the clinical features of survivors (n = 15) and nonsurvivors (n = 10) on Day 0 showed significant differences for MAP, before any vasopressive drug treatment, SAPS, MOF score, and platelet count, but not for age, temperature, creatininemia, proteinemia, leukocytosis, and ALI score (Table 4).

Significantly higher mean concentrations of sE-selectin (286 ± 51.4 versus 195 ± 60.2 ng/ml; p < 0.05), sICAM-1 (1,219 ± 200 versus 634 ± 150 ng/ml; p < 0.01), and vWf:Ag (846 ± 78.2 versus 490 ± 57.0%; p < 0.01) were present in nonsurvivors than in survivors. There was no significant difference for MPO (328 ± 157.0 versus 61 ± 7.8 ng/ml) or for Lacto-f (2,014 ± 534 versus 1,227 ± 334 ng/ml). Individual and mean values for the noninfected control group and the two survivor and nonsurvivor subgroups of the 25 patients with sepsis are shown in Figure 1.

View larger version (21K): [in this window] [in a new window]   Figure 1.   Individual and mean values (boldface bars) of the nonsepsis control group (C ), survivor (S ), and nonsurvivor (NS ) subgroups of patients with sepsis for sE-selectin, sICAM-1, vWf:Ag, myeloperoxidase (MPO), and lactoferrin (Lacto-f  ). Values are plotted on a logarithmic scale. Analysis of variance was significant for sE-selectin, sICAM-1, and vWf:Ag (p < 0.001 for each) and for Lacto-f (p < 0.01), and borderline for MPO (p = 0.06). p Values of the nonparametric comparison between S and NS groups are reported in the figure.

Relationship to Clinical Severity

We assessed the possibility that one or several of the five studied plasma proteins were related to clinical severity of illness by a nonparametric Spearman rank correlation test with SAPS, Day 0, and cumulative MOF and ALI scores. Unlike MPO and Lacto-f levels, sE-selectin, sICAM-1, and vWf:Ag levels correlated significantly with SAPS and Day 0 MOF score, but not with Day 0 or cumulative ALI score. Cumulative MOF for the entire sepsis study population was significantly correlated with levels of endothelial cell activation proteins (E-Selectin, ICAM-1, and vWf:Ag; p < 0.01 for each) (Figure 2) but not with neutrophil activation markers.

View larger version (19K): [in this window] [in a new window]   Figure 2.   Correlation of sE-selectin, sICAM-1, and vWf:Ag with cumulative MOF score determined according to the criteria of Tran and colleagues. The scatter diagrams were plotted on a logarithmic scale. Spearman rank coefficient (rho) values are reported with their respective p values.

Determination of Cutoff Values and Prediction of Survival Outcome

The distribution of endothelial and leukocyte markers for survivors and nonsurvivors overlapped (Figure 1). We performed ROC analysis to estimate the value of those best predicting survival outcome. The cutoff values were those that gave the fewest misclassified patients. True positives were defined as patients who died, with values higher or equal to the cutoff for each of the five markers. This analysis gave the following values: sE-selectin,  128 ng/ml; sICAM-1,  715 ng/ml; vWf:Ag,  717%; MPO,  97 ng/ml; Lacto-f,  797 ng/ml. A similar analysis was performed to determine cutoff values for clinical scores: SAPS,  15; MOF score on Day 0,  3 organ failures. The presence or absence of septic shock was also used as a dichotomous prognostic clinical variable. For dichotomous clinical and biological variables, treated as independent factors, we calculated the relative risk of death and its 95% confidence interval (CI 95%) (Table 5). Sensitivity, specificity, and negative and positive predictive values were also recorded and show that survival outcome was predicted with high sensitivity and specificity by markers of endothelial cell activation (sE-selectin, sICAM-1, and vWf:Ag) but not by markers of neutrophil activation (MPO and Lacto-f  ).

We then performed a multivariate analysis, using two distinct logistic stepwise algorithms: one for clinical variables and the other for soluble plasma proteins. We identified the Day 0 MOF score  3 organ failures as an independent clinical variable (p < 0.0001) with no statistical gain if the SAPS and the presence of septic shock were added into the model. Of the five proteins studied, only sICAM-1 (p < 0.0016) and vWf:Ag (p < 0.002) were independent prognostic variables. When we tested by a global logistic stepwise model the relationship of clinical variables and soluble proteins to survival status, the criterion of MOF  3 organ failures was the only independent prognostic variable. In our population, MOF  3 gave perfect discrimination between survivors and nonsurvivors and, if we omitted it from the regression model, two independent prognostic variables predictive of survival were then significant: septic shock (p < 0.002) and sICAM-1  715 ng/ml (p < 0.05).

Trends in Soluble E-Selectin and Soluble ICAM-1 Levels during Sepsis

Trends in the profiles of sE-selectin and sICAM-1 of survivors and nonsurvivors were studied by repeated analysis of variance, showing the following:

View larger version (25K): [in this window] [in a new window]   Figure 3.   Changes in sE-selectin and sICAM-1 levels for survivors (open squares) and nonsurvivors (closed circles) during disease progression. For statistical analysis see RESULTS.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Our results show that levels of the circulating adhesion molecules, sE-selectin and sICAM-1, and of vWf:Ag are higher in patients with sepsis and are correlated with clinical illness severity scores as assessed by SAPS and the number of organ failures. They were highest when severe sepsis was associated with septic shock or bacteremia. High levels of sICAM-1, either continuously high or increasing during the follow-up period, were associated with a lower probability of survival in our population. Such results were consistent with other studies of critically ill patients. Boldt and colleagues, studying the trends over 5 d of adhesion molecule (sE-selectin, sICAM-1, and soluble vascular cell adhesion molecule 1 [sVCAM-1]) levels in patients experiencing polytrauma found higher levels in nonsurvivors (10). Cowley and coworkers reported that organ dysfunction and death in a population of 40 patients with SIRS was best predicted by sE-selectin levels at admission rather than by sICAM-1 levels (11). In contrast, Donnelly and colleagues found that mean levels of sE-selectin were not correlated with outcome (12). However, most of their study population were patients with polytrauma, and E-selectin is widely distributed on vascular endothelium in experimental animals with septic shock but not traumatic or hypovolemic shock (25).

Endothelial upregulation of E-selectin on in vitro stimulation by endotoxin or cytokine (tumor necrosis factor  [TNF-] and interleukin 1 [IL-1]) is time dependent with maximum expression at 6 h and a return to baseline after 24 h. In contrast, the expression of ICAM-1 reaches a maximum between 12 and 24 h and remains high for at least 48 h (4). The role of membrane-anchored E-selectin and ICAM-1 in pathology is reasonably well understood, but that of the soluble forms is unknown. In the ICAM-1 mouse knockout model, the inflammatory reaction is abnormal despite normal production of cytokines such as TNF- and IL-1 (26, 27). The authors suggested that in this model, the survival of lethal endotoxin or exotoxin challenge by mice was at least partly due to a decrease in leukocyte-endothelium interactions and subsequent influx of polymorphonuclear leukocytes (PMNs) into tissue. Studies in animals, focusing on other adhesion molecules, have led to similar conclusions, supporting the importance of leukocyte-endothelium interactions in tissue injury and in determining outcome (28).

Soluble E-selectin and ICAM-1 are present at very low levels in healthy humans. Their concentration increases markedly in various inflammatory pathologies (6), including severe infection in animals (25) or humans (10). The mechanisms by which the concentrations of soluble adhesion molecules increase in the plasma during inflammation are unclear but they probably involve the proteolytic cleavage of transmembrane proteins (6). Soluble E-selectin levels in vivo are endotoxin dose dependent, hence the interest of this soluble molecule as a quantitative marker of inflammation-induced endothelial activation (24). As in other studies (7), we found that sICAM-1 and sE-selectin levels were significantly higher in cases of bacteremia. In our population, sE-selectin levels were higher in patients who did not survive, suggesting that the endothelium could be more intensely activated in these patients. In another study focusing on sICAM-1 levels in a population of 25 patients with sepsis, 12 of whom had septic shock, Sessler and co-workers (13) reported a strong correlation between the severity of shock (evaluated by the requirement of vasoactive drugs to maintain systolic blood arterial pressure above 100 mm Hg) and the level of sICAM-1. They demonstrated, using a logistic regression model, that an sICAM-1 cutoff level 3.4 times higher than normal values predicts nonsurvival. In our study, a plasma sICAM-1 concentration  715 ng/ml (3.43 times higher than normal values) gave a high NPV (92%) and PPV (75%) for mortality, whereas sE-selectin concentrations  128 ng/ml (3.1 times above normal values) were only associated with an NPV of 85% (Table ).

The increase in expression of endothelial membrane-bound E-selectin and ICAM-1 is correlated with an increase in levels of their soluble forms without necessarily being associated with endothelial injury (9). We attempted to determine whether shedding of soluble adhesion molecules following sepsis was associated with endothelial injury, by simultaneously measuring the release of circulating vWf:Ag. We found very high levels of vWf:Ag, strongly correlated with levels of sE-selectin and sICAM-1, suggesting that these three molecules may be part of the same phenomenon at the early onset of severe sepsis. High levels of vWf:Ag in plasma have been described in inflammatory diseases in which endothelial cells may be activated, but large increases generally reflect endothelial damage (15, 16). Also, some authors have suggested that high levels of vWf:Ag production may also be due to new growing cells, which may be present in inflammatory infiltrates. Endothelial cells produce more vWf:Ag in vitro in the active growth phase than when they are confluent (29). We think that this occurs at later stages rather than at the start of sepsis. However, very high levels of vWf:Ag, as found in our patients, are often associated with severe disease. When used to predict the outcome, a plasma level of vWf:Ag  717% (6.2 times higher than normal) gives a sensitivity of 80%, a specificity of 87%, and a positive predictive value of 80%, slightly higher than those obtained with sICAM-1. Interestingly, Rubin and coworkers (16) reported similar results for sensitivity, specificity, and positive predictive value for identifying patients who were not likely to survive. Thus, the vWf:Ag concentration appears to be a useful biochemical marker for endothelial injury with prognostic value. The logistic regression analysis, including clinical and biological variables, retained Day 0 MOF  3 organ failures as an independent prognostic factor, as expected, because it gave perfect discrimination between survivors and nonsurvivors (Table ). However, when this variable was removed from the model, ICAM-1 was the best plasma marker in combination with septic shock. A plasma level of vWf:Ag  717% was not retained as an independent prognosis variable in the global logistic regression model, probably because of our small population size.

High levels of MPO and Lacto-f are detected in vitro and in vivo after neutrophil activation (18, 24). When PMNs are activated and attached to the endothelial surface, they may release MPO, which chlorinates proteins and could trigger lipid peroxidation and subsequent endothelial injury (30). Mean levels of MPO and Lacto-f were significantly higher in patients with sepsis than in healthy volunteers and noninfected control patients, but their widespread occurrence makes them less sensitive and less specific than endothelium-derived proteins for predicting illness severity or survival outcome. In humans, the increase in Lacto-f after intravenous administration of endotoxin coincides with PMN margination (24). The lack of sensitivity of Lacto-f in predicting illness severity and survival outcome may be partly due to its uptake by liver endothelial and Kupffer cells (31).

It has been demonstrated in critically ill patients that clinical scores such as the SAPS (19) as well as the number of organ failures (21) are correlated with survival outcome. Complication of severe sepsis by septic shock favors poor survival outcomes. Septic shock occurred in 80% of our nonsurvivor subgroup. However, when performing univariate analysis, we found that the SAPS, MOF score, and septic shock were correlated with sE-selectin, sICAM-1, and vWf:Ag levels, whereas the ALI score was not. Widespread organ damage after the onset of sepsis leads to "panendothelial" activation in which circulating plasma markers cannot provide information on organ-specific alterations. Nevertheless, 10 subjects of the study population already had a primary lung infection associated with ALI on Day 0 (Table ). The Pa_O2/FI_O2 ratio and the initial ALI scores were significantly different in patients with and without pulmonary sepsis. Thus, the predictive value of plasma markers for the development of ALI should be assessed only in patients without primary pulmonary sepsis, and only six of this subgroup of patients (40%) developed secondary ALI. Rubin and coworkers, in a prospective study including 45 patients with nonpulmonary sepsis (16), found that an elevated plasma vWf:Ag level at the time of entry was 87% sensitive and 77% specific for predicting the subsequent development of lung injury.

Several studies performed in patients with sepsis to determine whether circulating cytokines could predict survival outcome have given conflicting results (32). Under septic conditions, most cytokines peak early and briefly. Such kinetics account for the poor reproducibility of results because the onset of sepsis often cannot be determined. The levels of sE-selectin in endotoxin-challenged healthy volunteers remained high for up to 24 h. Increases in TNF-, soluble TNF- receptor (sTNF-R), IL-1, IL-6, IL-8, IL-1ra, and granulocyte colony-stimulating factor (G-CSF), if they occur, are detected for only a few hours after the challenge (24). Despite ill-defined clearance mechanisms, plasma sE-selectin and sICAM-1 levels for a given patient vary slowly and progressively. It has also been reported that plasma vWf:Ag levels increase 3 h after intravenous administration of endotoxin to normal subjects and remain high at 24 h (35). We did not find significant differences for sE-selectin, sICAM-1, and vWf:Ag levels before and after 6 h from the putative onset of sepsis. Moreover, plasma levels of sE-selectin and sICAM-1 increased rapidly after the onset of sepsis and continued to increase for 3 or 4 d, especially in patients who did not survive. However, after this period, sE-selectin returned to levels similar to those of patients who survived whereas sICAM-1 remained high or continued to increase in patients who did not survive. Thus, sICAM-1 may provide an interesting and useful plasma marker during the late stages of sepsis.

In conclusion, unlike neutrophil-derived proteins (MPO and Lacto-f  ), circulating sE-selectin, sICAM-1, and vWf:Ag are useful plasma markers of endothelial hyperactivation in severely infected patients. Our data suggest that these three plasma endothelium-derived proteins may be useful early predictors of disease severity. During the late stage of sepsis, levels of sICAM-1 reflect more accurately than those of sE-selectin the extent and intensity of inflammation and tissue injury. They may provide a valuable means of monitoring therapy with new drugs. A larger and more homogeneous population of patients with septic shock is required to confirm these results.