Background
Sepsis remains a leading cause of death worldwide, especially in the intensive care unit (ICU) setting [
1]. It is currently accepted that improving the outcome of critically ill patients with sepsis relies mainly on the adequacy and the timeliness of key interventions such as administering appropriate antibiotics and sufficient amounts of fluid, especially the sickest ones [
2].
It is therefore mandatory to accurately assess the severity of the acute illness in such patients. Severity scores based on the assessment of underlying disease and organ failure have been derived from large studies [
3]. However, these large cohorts included patients without sepsis. In addition, the interest of repeated clinical assessments has not been validated with such scoring systems, and one should consider only the worst values of the physiological and biological parameters within the first 24 h following ICU admission. As a result, the Simplified Acute Physiology Score (SAPS) II is not theoretically available before day 2 and finally of limited value in clinical practice. In contrast, the Sequential Organ Failure Assessment (SOFA) score, which is easier to calculate since it relies solely on daily organ dysfunction assessment, could be more suitable. In addition, it was first evaluated in septic patients [
4,
5]. However, as organ failure is the end-stage complication of sepsis, it would be useful to predict it before it becomes clinically obvious, in order to prevent or at least to attenuate it whenever possible. In addition, clinical judgment may lack objectivity, thus leading to wrong evaluations and potentially inappropriate interventions. Moreover, the administration of innovative therapies is thought to provide the greatest benefit if given early to the potentially sickest septic patients.
In addition to the clinical evaluation, biomarkers provide a unique but only theoretical opportunity to predict the risk of bad outcomes reliably and promptly in patients with sepsis. Since the host inflammatory response is of paramount importance, measuring some of its most relevant mediators as well as surrogates within various body fluids including plasma has been proposed as a promising way to improve the management of such patients. Among these biomarkers, procalcitonin (PCT) and the soluble Triggering Receptor Expressed by Myeloid cells 1 (sTREM-1) have been shown to exhibit good diagnostic accuracy for bacterial sepsis [
6‐
8]. More recently, we showed that the CD64 leucocyte index measured upon ICU admission was even more accurate [
9].
The prognostic value of these biomarkers, however, remains to be clearly established and compared with relevant clinical scores. Actually, although it is tempting to believe that the same biomarker could be both a reliable diagnosis tool for sepsis and a powerful outcome predictor, none of those mentioned above has demonstrated these abilities within the same cohort of patients.
We therefore assessed the predictive value of PCT, sTREM-1 and the PMN CD64 index, with regard to the risk of a bad outcome in a large cohort of ICU septic patients included in a prospective observational study that aimed primarily to evaluate their diagnostic accuracy.
Methods
The methodology has already been extensively described elsewhere [
9].
Study population
Briefly, the approval of the institutional review board and written informed consent were obtained before inclusion. All consecutive patients newly hospitalized in two French medical intensive care units (Nancy and Dijon) were prospectively enrolled in the study. There were no exclusion criteria.
Data collection
On admission to the ICU, the following items were recorded for each patient: age; sex; severity of underlying medical condition stratified according to the criteria of McCabe and Jackson; SAPS II score [
10]; Sepsis-related Organ Failure Assessment (SOFA) score (range, 0 to 24, with scores for each organ system [respiration, coagulation, liver, cardiovascular system, central nervous system, and kidney] ranging from 0 [normal] to 4 [most abnormal]) [
4]; and the reason for admission to the ICU. The following baseline variables were also recorded at inclusion: body temperature; leucocyte count; ratio of the partial pressure of arterial oxygen to the fraction of inspired oxygen (Pao
2/Fio
2); presence of shock, defined as systolic arterial pressure lower than 90 mmHg with signs of peripheral hypoperfusion or need for continuous infusion of vasopressor or inotropic agents; and the use of previous antimicrobial therapy. The length of the ICU stay and ICU deaths were also recorded.
Two intensivists retrospectively reviewed all of the medical records pertaining to each patient and independently classified the diagnosis as no infection, sepsis, severe sepsis, or septic shock at the time of admission, according to established consensus definitions [
11]. Only patients with sepsis, severe sepsis, or septic shock were kept for the present study.
Measurement of Neutrophil CD64 index, and plasma levels of Procalcitonin, and sTREM-1
Within 12 h after admission and enrolment in the study, 5 mL of whole heparinized blood was drawn. Sampling was repeated on days (D) 2,3,5,7,10,14,21, and 28, provided the patient was still in the ICU. The expression of CD64 on neutrophils and monocytes was measured by quantitative flow cytometry using the Leuko64TM assay (Trillium Diagnostics, LLC, Brewer, ME). The sample preparation and flow cytometer setup were based on the manufacturer’s instructions. Index calculations were performed using Leuko64 QuantiCalc software (Trillium Diagnostics, LLC, Brewer, ME) [
12]. Flow cytometry was performed within 12 h after blood sampling. The whole procedure took less than 2 h. The reproducibility of measurements was excellent with a coefficient of variation lower than 5 %.
Procalcitonin concentrations were measured using an immunoassay with a sandwich technique and a chemiluminescent detection system, according to the manufacturer’s protocol (LumiTest, Brahms Diagnostica, Berlin, Germany).
Plasma concentrations of sTREM-1 were measured by ELISA using the Quantikine kit assay (RnD Systems, MN, USA) according to the manufacturer’s recommendations. All analyses were performed in duplicate. Inter and intra-assay coefficients of variation were lower than 7 %.
Clinical endpoints
The outcome of the included patients was assessed according to the two following endpoints: all-cause death in the ICU and an increasing SOFA score between day 1 and day 3 as early surrogates for deteriorating organ function. Given the fact that late mortality may be caused rather by secondary infections or comorbidities than by sepsis itself, biomarkers’ predictive value regarding the risk of death before day-14 was also evaluated.
Statistical analysis
Descriptive results for continuous variables were expressed as means (±SD) or medians (IQR) depending on the normality of their distribution as assessed by the Kolmogorov-Smirnov test. Variables were tested for their association with the outcome (i.e., death in the ICU or increasing SOFA scores between D1 and D3) using the Pearson χ2 test for categorical data and the Mann-Whitney U test for numerical data. Receiver-operating-characteristic (ROC) curves were constructed to illustrate various cut-off values of soluble TREM-1 and PCT. The sensitivity, specificity, positive and negative likelihood ratios and their confidence intervals were calculated (19). These values were calculated for the cut-off that represented the best discrimination as derived from the Youden index (J = max [sensitivity + specificity-1]).
To account for wide distributions of data and potential nonlinear associations with outcomes, if mean baseline values were found to be significantly different between survivors and non-survivors, biomarker values were transformed into quartiles based on their distribution. The corresponding Kaplan-Meier curves were then constructed in order to compare ICU survival between the different quartiles through a time-dependent analysis, using the log-rank test.
We also evaluated the value of PCT and sTREM-1 in predicting the outcome as independent factors using a Cox model. Any covariate with univariate significance of p < 0.10 was eligible for inclusion in the model.
Finally, we compared PCT and sTREM-1 with the clinical scoring systems usually used in critically ill patients upon ICU admission (i.e., SAPS II and SOFA) and with serum lactate concentration, regarding their ability to predict the outcome. Several combinations were also tested.
Subgroup analyses were conducted in patients with proven infection (i.e., microbiologically documented) and in those who were not given antibiotics within the 48 h preceding ICU admission.
Statview software (Abacus Concepts, Berkeley CA) and Prism (Graphpad®) were used for the analyses. A two-tailed p < 0.05 was considered significant.
Discussion
Accurately evaluating the severity of acute illness in septic patients on admission to an ICU is challenging. It usually relies on a combination of various clinical and biological data, including the assessment of organ failure, as well as knowledge of underlying disease(s). Although of paramount importance, clinical judgment might be biased given the need for prompt decision-making. Indeed, having an objective overview of a patient’s risk of death remains a matter of concern. Calculating clinical scores is fastidious and could remain subjective. Measuring biomarkers could overcome these drawbacks.
The main findings of the present study were the following: (i) the PMN CD64 index had no prognostic value despite its promising accuracy regarding the diagnosis of sepsis; (ii) the predictive value of sTREM-1 regarding the risk of death was greater than that of PCT, especially when considering early mortality. In addition, among the biomarkers tested, sTREM-1 was an independent predictor of death and sTREM-1 kinetics looked different between survivors and non survivors. Nonetheless, both biomarkers were outperformed by clinical scores; (iii) the combination of sTREM-1 and SAPS II offered the best accuracy for predicting ICU survival in our cohort; (iv) none of the indicators tested were valuable tools for reliably predicting clinical worsening within the first 48 h of the ICU stay.
Several studies have aimed to evaluate biomarkers including sTREM-1 as prognostic factors in critically ill patients with sepsis. However, despite interesting findings, it remains unclear whether they are of any value or not (Table
8).
Table 8
Summary of the main clinical studies evaluating sTREM-1 prognostic value
Nb. of Patients | 63 | 90 | 52 | 63 | 130 (100 with sepsis) | 102 | 40 with cancer | 190 |
Median Age | 61 | Unknown | Unknown | 63.7 | 58.9 | 63 | 67 | 60.6 |
Unit | ICU | ICU | ICU | ER (and then ICU) | ICU | ICU | ICU | ICU |
Severity of sepsis | Septic shock (53 %) | Severe sepsis + Septic shock (70 %) | Severe sepsis + Septic shock (71.1 %) | Severe sepsis + Septic shock (100 %) | Severe sepsis + Septic shock (64 %) | Sepsis, Severe sepsis and Septic shock | Severe sepsis + Septic shock (100 %) | Septic shock (68.4 %) |
Site of Infection | Miscellaneous | VAP | Miscellaneous | Miscellaneous | Miscellaneous | Miscellaneous | Miscellaneous | Miscellaneous |
Overall Mortality | 33 % (ICU) | 36.7 % | 30.7 % (D28) | 25.4 % (D28) | 43 % (D28) | 41.2 % (D28) | In cancer patients: | 25.8 % (ICU) |
34.7 % (ICU) |
40 % (D28) |
SAPS II | 53 (21) | Unknown | Unknown | Unknown | Unknown | Unknown | Unknown | 52.6 (21.5) |
APACHE 2 | Unknown | Unknown | Unknown | Unknown | 13.4 (6.1) | Unknown | In cancer patients 19.9 (5.1) | Unknown |
SOFA D0 | 936 (3.1) | Unknown | Unknown | Unknown | 7.8 (4.4) | Unknown | 6.2 (2.7) | 9 (5.2) |
Type of sample | - sTREM1 | - Serum | - Serum | - Serum | - Serum | - Serum | - Serum | - Serum |
- sTREM1 | - sTREM1 | - sTREM1 | - sTREM1 | - sTREM1 | - sTREM1 | - sTREM1 |
- ELISA | - ELISA | - ELISA | - ELISA | - ELISA | - ELISA | - ELISA |
Median sTREM1 level on admission (pg/ml) | | | | | | | In Cancer Patients | |
Survivors | 154 | Unknown | 193.4 | 182.4 | Unknown | 161.95 | 848 | 671 |
Non survivors | 94 (p = 0.02) | Unknown | 240.2 (NS) | 514.1 (p = 0.001) | Unknown | 320 (p < 0.001) | 558 (NS) | 1148 (p < 0.01) |
sTREM1 cut-off value (pg/ml) | 180 (baseline) | | | | | 252.05 (baseline) | | 954.4 (baseline) |
Sensitivity | 86 % | | | | | 85.7 % | | 54.5 % |
Specificity | 70 % | | | | | 75.7 % | | 78 % |
AUROCC | 0.74 | | | | | 0.856 | | 0.64 |
PPV | | | | | | 70.6 % | | 49.2 % |
NPV | | | | | | 88.2 % | | 81.5 % |
Best Relevant Prognostic Predictor (associated with sTREM1) | -sTREM1 baseline | - sTREM1/IL-6 baseline ratio | None (sTREM1 increase between D1 and D14 was NS) | - Log (sTREM1) baseline | None | sTREM1 (baseline > 252.05) | -sTREM1 value on D2 for ICU mortality. AUROCC = 0.69 | sTREM1 (baseline value > 954.4) + SAPS II (>67.5) |
Other relevant Predictors | - SOFA baseline | - TNF alpha baseline | - SOFA score baseline and evolution | - ScVO2 baseline | | | - sTREM1 value on D1 for D28 mortality. AUROCC = 0.75 | |
- IL-6 baseline | - SAPS II | - SOFA | - PCT (baseline > 10.6 ng/ml) | - Days of MV | - SAPS II |
- IL-10/IL-6 baseline ratio | - sCD163 | - SOFA baseline > 6.5 | - Use of corticosteroids | - Core Temperature |
TREM-1 is a member of the immunoglobulin superfamily of receptors that is specifically expressed on the surface of neutrophils and monocytes. The primary role of all TREM is both the tuning and integration of multiple signals rather than the direct initiation of an inflammatory response. Soluble TREM-1 is the soluble form of TREM-1, which is up-regulated when the host innate immune system is exposed to infectious invaders. Any sustained increase in the sTREM-1 level indicates that the overall expression of TREM-1 is continuously rising, along with the release of larger amounts of pro-inflammatory mediators. Thereafter, any further increase in sTREM-1 suggests a protracted inflammatory response generally related to a poor clinical outcome.
To our knowledge, this study is the largest to date to evaluate the prognostic interest of measuring sTREM-1 in septic patients. Concerning the predictive value of baseline sTREM-1 levels, our findings are in accordance with those obtained by Li et al., who showed in 102 ICU patients that day-1 sTREM-1 concentrations yielded an AUROCC of 0.85 regarding the risk of death at day-28 [
13]. In this study, PCT was also showed a good predictive value. The far higher mortality rate reported in the Li et al. cohort than in ours (41.2 vs. 25.8 %, respectively) may account for the greater accuracy in predicting death. Moreover, the number of patients with antibiotic exposure prior to biomarker measurement was quite high in our study, as was the number of patients with negative cultures. These characteristics probably diminished the predictive value of sTREM-1 [
14]. Finally, Jeong et al. also found that the sTREM-1 concentration on admission was the best biomarker regarding the short-term prognosis in patients presenting with severe sepsis, outperforming baseline blood lactate levels [
15].
Altogether, however, these results are apparently conflicting if compared with the data previously obtained by our group in 63 septic patients, among whom more than half presented with shock, that the lower the sTREM-1 baseline level, the poorer the outcome [
16]. Similar findings were obtained recently in a small cohort of cancer patients. Indeed, Ravetti et al. showed that levels of sTREM-1 over time were higher in survivors than in non-survivors [
17]. This may be subsequent to the dual significance of plasma levels of sTREM-1. Basically, it may reflect the overall expression of TREM-1, including the membrane anchored as well as the soluble form. Then, high levels might be deleterious for the host since they could result in an overwhelming inflammatory response. Conversely, the release of large amounts of sTREM-1 could be protective through the neutralization of yet unknown TREM-1 ligands likely to amplify the host inflammatory response [
18]. An anti-inflammatory effect of sTREM-1 could then be expected as suggested by findings made in a small cohort of patients with sepsis related to ventilator-associated pneumonia [
19,
20]. Moreover, as suggested by previous works, elevated sTREM-1 does not necessarily reflect TREM-1 gene expression [
21]. In addition, one could speculate that depending on the immunoassay, either one or more isoforms of sTREM-1, with or without its shed form, were detected, thus accounting for such a discrepancy. Finally, TREM-1 expression on immune cell surfaces is known to be highly time-dependent, but also variable according to the pathogen involved and the source of infection [
14]. Differences may exist regarding these points in the different cohorts mentioned above and account for these apparently conflicting findings. However, the present findings may be more representative since a larger number of patients were included in two distinct ICUs.
In addition, we should admit that several other studies failed to demonstrate any interest of sTREM-1 as a predictor of prognosis. For example, Phua et al. found that baseline sTREM-1 was a poor predictor of death in the ICU, as did Zhang et al. [
22,
23]. Procalcitonin performed even better in the former study. One larger study published by Su et al. showed similar results, although the sTREM-1 level was probably helpful in diagnosing sepsis, and in differentiating between sepsis, severe sepsis and septic shock [
24].
To overcome the above-described issues regarding the interpretation of a single sTREM-1 value, serial measurements are of potential interest. As previously reported, we showed herein that an early decrease in sTREM-1 levels was associated with a better outcome in the ICU [
15,
16,
23,
25]. The high rate of previous exposure to antibiotics (around 30 %) in our cohort may account for the lack of correlation between PCT time-course and the outcome. Actually, we and others have previously shown in large cohorts of critically ill patients that a decrease in PCT levels within the first 72 h of sepsis management was closely related to the outcome [
26,
27].
Finally, we compared biomarkers with clinical scores since they are still considered the “gold-standard” for predicting the outcome of critically ill patients [
4,
10,
28]. It is worth noting that in our study the AUROCC achieved with both SOFA on admission and SAPS II was found to be greater than that for sTREM-1. However, adding sTREM-1 to SAPS II improved its specificity, since it reached 98 %, and showed a positive likelihood ratio of 27.3. In addition, the baseline sTREM-1 level remained an independent predictor of death in the ICU as did the SAPS II score, after adjustment for potential confounders, whereas neither lactate levels nor the SOFA score did. We could then consider that measuring sTREM-1 upon admission to the ICU, provides relevant information regarding the severity of sepsis in addition to clinical data.
There are, however, some limitations. First, our cohort was small, thus precluding the external validity of our findings. Second, sTREM-1 levels were not routinely measured since it relies on one ELISA assay that so far has not been automated. Finally, one can argue that given the large proportion of patients with negative bacterial cultures, non-septic inflammatory states were included in our study. However, this could be easily explained by the fact that prior exposure to antibiotics before ICU admission was quite frequent, thus reflecting “real life” conditions.
Acknowledgments
We are indebted to Philip Bastable for English expression revisions.