Prognostic performance of endothelial biomarkers to early predict clinical deterioration of patients with suspected bacterial infection and sepsis admitted to the emergency department

Background The objective of this study was to evaluate the ability of endothelial biomarkers to early predict clinical deterioration of patients admitted to the emergency department (ED) with a suspected sepsis. This was a prospective, multicentre, international study conducted in EDs. Adult patients with suspected acute bacterial infection and sepsis were enrolled but only those with confirmed infection were analysed. The kinetics of biomarkers and organ dysfunction were collected at T0, T6 and T24 hours after ED admission to assess prognostic performances of sVEGFR2, suPAR and procalcitonin (PCT). The primary outcome was the deterioration within 72 h and was defined as a composite of relevant outcomes such as death, intensive care unit admission and/or SOFA score increase validated by an independent adjudication committee. Results After adjudication of 602 patients, 462 were analysed including 124 who deteriorated (27%). On admission, those who deteriorated were significantly older (73 [60–82] vs 63 [45–78] y-o, p < 0.001) and presented significantly higher SOFA scores (2.15 ± 1.61 vs 1.56 ± 1.40, p = 0.003). At T0, sVEGFR2 (5794 [5026–6788] vs 6681 [5516–8059], p < 0.0001), suPAR (6.04 [4.42–8.85] vs 4.68 [3.50–6.43], p < 0.0001) and PCT (7.8 ± 25.0 vs 5.4 ± 17.9 ng/mL, p = 0.001) were associated with clinical deterioration. In multivariate analysis, low sVEGFR2 expression and high suPAR and PCT levels were significantly associated with early deterioration, independently of confounding parameters (sVEGFR2, OR = 1.53 [1.07–2.23], p < 0.001; suPAR, OR = 1.57 [1.21–2.07], p = 0.003; PCT, OR = 1.10 [1.04–1.17], p = 0.0019). Combination of sVEGFR2 and suPAR had the best prognostic performance (AUC = 0.7 [0.65–0.75]) compared to clinical or biological variables. Conclusions sVEGFR2, either alone or combined with suPAR, seems of interest to predict deterioration of patients with suspected bacterial acute infection upon ED admission and could help front-line physicians in the triage process.

present in 6% of adult hospitalizations [2]. Over the last decade, a decrease in the mortality rate has been observed [3] in particular thanks to improved management, and more appropriate intervention approaches in the emergency department (ED) [4]. Although the recently proposed qSOFA score [5] aims to help frontline clinicians detecting severe patients with a higher risk of mortality [6], it fails to get decisive support for discharge decision, especially in patients without initial organ dysfunction [7,8], that could help to reduce ED crowding and cost.
Even if widely used as an infection biomarker and diagnostic of severity, procalcitonin (PCT) has not been fully validated for deterioration assessment, and no other biological marker has yet been validated to accurately early predict clinical deterioration in unselected patients admitted to the ED with infection or sepsis [9][10][11]. Asymptomatic endothelial injury participates in the development of organ failure with poor outcome [12,13]. Endothelial biomarkers have been presented as predictors of death and/or organ dysfunction during sepsis [14][15][16][17][18][19][20][21][22]. Of those, soluble vascular endothelial growth factor receptors 2 (sVEGFR2, growth factor for vascular endothelial cells) and soluble urokinase plasminogen activator receptor (suPAR, pro-inflammatory activation of the immune system) were proposed. VEGFR2, which is selectively expressed in the endothelium, mediates endothelial growth, proliferation and permeability and pathological angiogenesis, and bound to VEGF increases microvascular permeability resulting in oedema and hypotension [23]. The uPAR receptor is expressed on different cell types including vascular endothelial cells [24]. After cleavage from the cell surface, the soluble receptor, suPAR, can be found in the blood and other organic fluids. Increased activation of the immune system caused by different types of infections results in increased suPAR concentrations. These biomarkers have been shown to be associated with initial severity and subsequent clinical worsening [25][26][27][28][29][30][31][32], but their ability to early predict deterioration on ED admission remains to be determined.
This study aimed to evaluate the ability of sVEGFR2 and suPAR biomarkers to early predict the clinical deterioration of patients with infection upon ED admission and compare them to conventional clinical and biological parameters (qSOFA and SOFA score, lactates, PCT, CRP). Second, we assessed the prognostic performance of biomarkers according to the presence of sepsis or not in accordance with the new definitions of Sepsis-3.

Population
We conducted a prospective, multicentre, international study in 14 EDs from 2015 to 2018. Inclusion criteria included adult patients (age ≥ 18 years) with an acute suspected bacterial community-acquired infection (≤ 3 days), evolution time window being checked with the patient and/or relatives, associated with at least two systemic inflammatory response syndrome (SIRS) criteria [33], which were currently the most sensitive criteria for sepsis [34,35]. All patients admitted to the ED with a suspected infection, based on fever and/or any other infectious symptom reported by referral practitioner were screened 24/7 by emergency physicians for eligibility and treated following the Surviving Sepsis Campaign guidelines [36]. Exclusion criteria were patients with septic shock (based on ACCP/SCCM criteria), patients with a healthcare-associated infection, immunosuppression (e.g. human immunodeficiency virus (HIV), transplant, ongoing chemotherapy, steroid treatment > 20 mg/ day of prednisone or equivalent for more than a week), non-infectious diseases potentially associated with SIRS (cancer), patients with a prior episode of infection within the 30 days before ED admission and onset of symptoms greater than 72 h and absence of consent. The protocol was recorded on ClinicalTrials.gov (N°: NCT02739152), approved by the Ethics Committee for Clinical Research (CPP SOOM IV: CPP15-004).

Endpoints
The primary endpoint was the occurrence of early clinical deterioration within 72 h following ED admission. Deterioration was determined by an independent adjudication committee (including one experienced emergency physician and two intensive care physicians) who were blinded to biomarker results, and followed a pre-defined adjudication charter. Patients were then classified according to their initial course during the first 72 h of hospitalization, as exhibiting an early deterioration defined by a composite endpoint (increase SOFA score of at least 1 point, or ICU admission directly related to the initial infectious disease because of documented sustained hypotension requiring vasopressors or ventilation support requirement, or death) or not. The same adjudication committee also confirmed the bacterial origin of infection according to available clinical, biological and microbiological data and based on pre-defined criteria for every different type of infection [37]. Patients without confirmed infection were excluded from the analysis.

Study design and measured variables
All the patients were included and received their first care and blood collection in ED. Clinical criteria, biological data, lactates [38], qSOFA score, SOFA score [39] and studied biomarkers were measured at three time points: the first within emergency room (T0) and the others at 6 ± 2 h (T6) and 24 ± 2 h (T24) after ED admission. The following data were prospectively collected by the study team blinded to biomarker results during ED stage: demographics, Charlson score, site of infection, antimicrobial therapy and initiation time, traditional biological parameters (leukocytes, CRP, platelets) and orientation after ED discharge. Pathogens, length of stay and mortality at day 28 were collected during hospitalization or at the end of follow-up.

Biomarker measurements
Serum sVEGFR2 concentrations (soluble vascular endothelial growth factor receptors 2) were measured using the enzyme-linked fluorescent assay (ELFA) technique. The results were automatically analysed by VIDAS ® and expressed in relative fluorescence intensity or RFV (relative fluorescent value). Plasma suPAR (soluble urokinase plasminogen activator receptor) levels were analysed using the commercially available CE/IVDlabelled suPARnostic ® AUTO Flex ELISA kit, according to the manufacturer's instructions (Virogates, Birkeroed, Denmark). For sUPAR ELISA test, the inter-assay coefficient of variation (CV) given by the manufacturer is below 6%. For sVEGFR2, the inter-assay coefficient of variation was calculated at 3.09%. Serum PCT (procalcitonin) levels were measured using VIDAS BRAHMS PCT assay (Biomerieux, Marcy l'Etoile, France) according to the manufacturer's instructions.

Analysis
The prognostic performance of studied biomarkers was evaluated in the entire cohort at T0. Following the current definitions and to analyse the prognostic performance of biomarkers according to the severity, the Sepsis-3 criteria were applied to define 2 groups: infected patients (SOFA score < 2) and septic patients (SOFA score ≥ 2) [5]. A model of risk of early deterioration based on the value of biomarkers on ED admission was proposed on non-septic patients. Data were censored after deterioration. No patient was lost to follow-up until T24.

Statistics
Data are presented either as means ± SD, median with interquartile range or as box and whisker plots with representation of the median value, 25th, 75th and 90th percentiles, and outliers. Parameters and biomarkers were compared between the two groups of patients according to their initial course (i.e., early deterioration or not), using the nonparametric Mann-Whitney U test for continuous variables, while categorical variables were compared with the Pearson χ 2 test or the Fisher's exact test when appropriate. The level of significance was set at 5% and results of regression analyses were presented with their 95% CI. All analyses were computed using the R version 3.4.0.
Logistic regressions were fit using single or both biomarkers. Association with clinical variables was independently evaluated, and clinical parameters with a p value below 0.1 in the univariate analyses were selected as adjustment covariates for multivariate analyses. Among significant clinical parameters, a selection was made to avoid collinearity and limit the number of variables introduced in multivariate models. Strength of association was reported using inter-quartile range (IQR) adjusted odds ratios (OR). Areas under the ROC curve and their 95% confidence interval (CI) were computed and compared using the DeLong's method. Predictive performances were calculated/evaluated under constraint of a sensitivity higher than 90% (rule-out test). In a complementary approach, a decision tree was built. The thresholds used to partition the data were chosen to optimize sensitivity (> 90%).

Biomarkers: predictive performances in the global cohort
At T0 in univariate analysis, the age, the Charlson score, the qSOFA score and the SOFA score were associated with the early clinical deterioration, but not traditional biological markers. Biomarker association with clinical deterioration was also observed in secondary excluded patients (Additional file 1: Figure 1). Levels of sVEGFR2, PCT and suPAR at inclusion were significantly associated with the degree of organ dysfunction, as reflected by the SOFA score on ED admission (Fig. 2).

Biomarkers: predictive performances in infected patients without sepsis (SOFA < 2)
Among the 233 patients considered with infection but no sepsis at enrolment (SOFA < 2 at admission), 48 of them (21%) deteriorated within 72 h (Fig. 1b). The clinical characteristics of this derivative cohort at admission were not different from those of the overall cohort (  Table 4). At T6, only sVEGFR2 and sUPAR were found significantly associated with worsening (p < 0.01) (Additional file 1: Figure 2b.). As data were censored after deterioration, the low number of patients deteriorating after T6 and T24 did not allow further analysis. The prognostic value of biomarkers on septic patients (SOFA ≥ 2) is presented in Additional file 1: Table 1.

Proposal of a stratification model in non-severe patients on ED admission
The best prognostic model including sVEGFR2 and suPAR combination and using cut-off values optimized to yield a high sensitivity allowed identifying distinct levels of risk (i.e. low and high) for deterioration. When comparing risk groups, we found that the low-risk group had a 15-fold lower risk of worsening than the high-risk group (OR = 14.50 p < 0.05) (Fig. 3).

Discussion
In this ED-based multicentre study, the endothelial biomarker sVEGFR2, used either alone or combined with suPAR, proved the best early predictor of patient deterioration, independently of potential confounders. High medical value biomarkers are those that are able to predict outcomes even before any clinical evidence of deterioration to help front-line physicians to better anticipate the complicated course. Since early prediction of patient deterioration is crucial to allow safe rule-out, over-triage reduction and better allocation of hospital resources, the high negative value of these potential endothelial biomarkers appears particularly useful in the ED settings with inherent peaks of activity and overcrowded units.
In the present study, 27% of the entire cohort deteriorated within 72 h of ED admission. Importantly, 21% of non-sepsis patients, without any severity criteria on ED admission, did deteriorate within the first 72 h of hospitalization. Intrinsically, these patients presented to the ED with low SOFA (i.e., SOFA < 2) and non-qualifying qSOFA score. This proportion is similar to that reported in previous studies [18][19][20]. About 20 to 25% of patients progressed to severe sepsis whereas they had no sign of seriousness at first medical contact [7]. Saeed et al. reported that early clinical deterioration occurred in more than 16% of patients presenting to the ED with sepsis, even when patients were non-severe with low lactate level (< 2 mmol/L) or low clinical score (qSOFA < 2) [40]. Recently, Cleek et al. confirmed that in predicting 28-day in-hospital mortality among infected ED patients, qSOFA did not outperform or improve physician judgment [41]. Overall, the clinical deterioration occurred very early after ED admission since two-third of the patients deteriorated within the first 6 h of inclusion. Although information on the delay between ED admission and deterioration are scare, some authors have reported that it may occur within 48 h [42], even within the first 12 h following ED arrival [43].
Due to various presentations of infected patients on ED admission, determining the severity early in the disease course remains challenging since clinical scoring systems have limited prognostic accuracy [44,45]. Many  conventional biomarkers reflecting end-organ compromise are not informative until significant clinical deterioration has occurred [46,47]. The Sepsis-3 definition underlines organ dysfunction as the mainstay of sepsis and the value of the SOFA score to identify patients with a higher risk of subsequent death [48]. However, 21% of our patients with a SOFA score < 2, i.e. non-sepsis according to Sepsis-3, deteriorated within 72 h after ED arrival. In these circumstances, assessment of endothelial injury could be a good predictor of deterioration [49]. Likewise Fang et al. [20] described a relationship between endothelial biomarkers and variations of the SOFA score during the first week after admission. Liu et al. [50] also showed an association between the presence of endothelial injury on admission and severity of sepsis. More recently, Henning et al. also confirmed that biomarkers of endothelial activation and inflammation in combination with emergency department physician judgment improved prediction in hospital mortality [51,52]. These observations are concordant with our findings showing that the level of sVEGFR2 and suPAR is associated with that of the SOFA score. Importantly, we have shown that sVEGFR2 alone or combined with suPAR is the best predictor of patient deterioration, independently of potential confounding factors. If confirmed, this result could allow safe rule out of patients who have low risk of deterioration, hence leading to a decrease in hospital admissions.
This prospective, multicentre, international, observational study presents several strengths, such as (i) a biological collection of biomarkers combining with the evolving clinical criteria/in line with the requirements of the new Sepsis-3 definition of sepsis, and (ii) the appointment and careful evaluation by an independent adjudication committee. We also demonstrated that circulating markers of endothelial activation, at the earliest time in ED, have a potential of risk stratification and could help emergency physicians better manage patients with sepsis.
Our study, however, has several limitations, the first one being the limited possibility to fully investigate the heterogeneity of the different subtypes of infections. Indeed, patients with pneumonia may differ from patients with abdo-pelvic infections. The population may be biased against deterioration, as it requires decompensation from a less ill state. Half of the cohort having a SOFA ≥ 2 at ED baseline may have been already quite ill. Therefore, the study may have been stronger if focused on a light-/middle-severity sepsis cohort, using new definition of sepsis if it had been available when designing the study. Also, the design and purpose of the study did not allow analysis of patients with septic shock, while they could have been used as a control group of severity. In addition, as the number of patients with non-confirmed or viral infection was low, no prediction analysis was done. The entire analysis has somewhere a modest sample size (n = 462 patients, of which 127 experienced deterioration) and do not support strong conclusions but serve as a robust early basis for future validation. Finally, we did not perform health economics and outcome research that could have brought useful information on the potential cost savings for hospitals.

Conclusion
The current findings highlight the potential interest of the sVEGFR2 protein, alone or in combination with suPAR, to diagnose initial endothelium stress and to predict/anticipate subsequent organ dysfunction. Such tool, suitable for routine test measurement, with timeto-results within 1 h and only one-time measurement required, could be used together with other laboratory findings and clinical assessments, to help in early prediction of the risk of deterioration and safely ruling out infected patients after ED admission.