OUP user menu

Unbiased plasma proteomics for novel diagnostic biomarkers in cardiovascular disease: identification of quiescin Q6 as a candidate biomarker of acutely decompensated heart failure

Alexandre Mebazaa, Griet Vanpoucke, Gregoire Thomas, Katleen Verleysen, Alain Cohen-Solal, Marc Vanderheyden, Jozef Bartunek, Christian Mueller, Jean-Marie Launay, Natalie Van Landuyt, Filip D'hondt, Elisabeth Verschuere, Caroline Vanhaute, Robin Tuytten, Lies Vanneste, Koen De Cremer, Jan Wuyts, Huw Davies, Piet Moerman, Damien Logeart, Corinne Collet, Brice Lortat-Jacob, Miguel Tavares, Wouter Laroy, James L. Januzzi, Jane-Lise Samuel, Koen Kas
DOI: http://dx.doi.org/10.1093/eurheartj/ehs162 2317-2324 First published online: 25 June 2012


Aims Biochemical marker testing has improved the evaluation and management of patients with cardiovascular diseases over the past decade. Natriuretic peptides (NPs), used in clinical practice to assess cardiac dysfunction, exhibit many limitations, however. We used an unbiased proteomics approach for the discovery of novel diagnostic plasma biomarkers of heart failure (HF).

Methods and results A proteomics pipeline adapted for very low-abundant plasma proteins was applied to clinical samples from patients admitted with acute decompensated HF (ADHF). Quiescin Q6 (QSOX1), a protein involved in the formation of disulfide bridges, emerged as the best performing marker for ADHF (with an area under the receiver operator characteristic curve of 0.86, 95% confidence interval: 0.79–0.92), and novel isoforms of NPs were also identified. Diagnostic performance of QSOX1 for ADHF was confirmed in 267 prospectively collected subjects of whom 76 had ADHF. Combining QSOX1 to B-type NP (BNP) significantly improved diagnostic accuracy for ADHF by particularly improving specificity. Using thoracic aortic constriction in rats, QSOX1 was specifically induced within both left atria and ventricles at the time of HF onset.

Conclusion The novel biomarker QSOX1 accurately identifies ADHF, particularly when combined with BNP. Through both clinical and experimental studies we provide lines of evidence for a link between ADHF and cardiovascular production of QSOX1.

  • Diagnosis
  • Acute decompensated heart failure
  • Biomarker
  • Proteomics

See page 2249 for the editorial comment on this article (doi:10.1093/eurheartj/ehs187)


Breathlessness or dyspnoea is the chief complaint of many patients admitted to the hospital, and while symptom-driven management strategies for dyspnoea would be desirable, there is no one single set of treatments that safely fits all cardiovascular or respiratory diagnoses; treatment that may improve one condition, such as diuretics and vasodilators in heart failure (HF), might be detrimental in other conditions such as pulmonary infection.

Heart failure, a major cause of acute dyspnoea, is a current worldwide pandemic with an unacceptable high level of morbidity and mortality.1 The immediate and accurate diagnosis of the mechanism (cardiac and/or non-cardiac origin) of acute dyspnoea may be challenging. Furthermore, cardiac causes of acute dyspnoea, often named acute decompensated HF (ADHF), are diverse including left ventricular (LV) diastolic and/or systolic dysfunction, right ventricular dysfunction and/or acute ischaemia.2 B-type natriuretic peptides (NPs) (NT-proBNP and BNP),2 released from stressed cardiomyocytes, have been recently demonstrated to be highly sensitive to diagnose ADHF but they generally lack specificity, in some conditions, including acute dyspnoea in patients with a history of chronic HF (CHF). The latter represent 60–80% of ADHF patients. Chronic HFs have continuous ventricular overload and high levels of plasma NPs, even in stable conditions. Furthermore, conditions such as pneumonia, heart ischaemia, and/or renal failure, three very common concomitant phenomena in acute dyspnoeic patients, lead to increases in NP levels.3 Thus, high levels of NPs measured in the plasma of CHF patients admitted for dyspnoea do not necessarily indicate an ADHF event; this was named the ‘grey zone’ of NPs (for a review, see Maisel et al.4). Novel biomarkers are needed to complement NPs in establishing a diagnosis of ADHF.

Biomarkers that have shown their merit in clinical practice, such as the NPs, troponins, and prostate-specific antigen, have mainly been the result of serendipitous findings or hypothesis driven efforts and are generally in low abundance in blood. Mass spectrometry (MS) provides a unbiased approach to identifying novel protein biomarkers,5 which required to overcome the large dynamic range as seen in plasma to mine the lower abundance proteome. Using a robust protein discovery platform and a sensitive antibody-free assay, we discovered Quiescin Q6 (QSOX1) as a new candidate diagnostic marker of ADHF.


Patient samples

Study protocols were approved by the local ethics committee, and written informed consent was obtained from all patients and controls. Ten patients hospitalized for acute HF were included in the candidate marker discovery phase. Patients were receiving treatment for HF in accordance with current guidelines at enrolment, and this was continued during the study. Left ventricular ejection fraction (LVEF) was determined by transthoracic echocardiography. Those cases were selected with history of CHF, BNP values >1000 pg/mL, and reduced LVEFs. The control group consisted of 10 subjects with normal LVEF and without signs or symptoms of HF, inflammation, or renal failure (serum creatinine >150 μmol/L; see Supplementary material online, Table S1).

The verification study was a prospective collection conducted in three European centres. Patients >18 years were recruited in the ED with a primary complaint of acute dyspnoea. Final diagnosis of cardiac or non-cardiac origin of dyspnoea was performed after patient hospital discharge, by two physicians including a senior cardiologist. Furthermore, stable CHF patients were recruited from an outpatient setting as well as the healthy controls. At enrolment, blood (10 mL) was collected and ethylenediaminetetraacetic acid-containing plasma was prepared within 2 h according to Tuck et al.6 Both BNP and NT-proBNP were measured after the completion of all samples at one site, using the Triage BNP test (Alere, San Diego, CA, USA) and electrochemiluminescence on the Elecys 2010 Analyzer (Roche-Diagnostics, Rotkreutz, SE, USA), respectively. The clinical trial registration is NCT1374880, URL: ClinicalTrials.gov.

The proteomics pipeline

The experimental strategy using discovery and validation experimental strategy is illustrated in Supplementary material online, Figure S1.

The ‘discovery platform (MASStermind)’ included different MASStermind building blocks (see Supplementary material online, Figure S1A). The major bottleneck in analyzing plasma proteins is their large dynamic range from ng/mL levels to those at mg/mL. The crux of the liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) discovery platform is the upfront selection of a single amino-terminal peptide per protein acting as a signature using the COFRADIC approach.7 This allows to submit only one peptide per protein for MS analysis compared with typical 35 tryptic peptides with conventional approaches.8 Hence, the COFRADIC technology allows mapping of the true in vivo representation of the proteome and provides information about processing events.9 For details, see Supplementary material online. Finally, protein identification was performed using an optimized analysis pipeline including three software packages Mascot (Matrix Science, London, UK), ProteinPilot™ (Applied Biosystems), and SPIDER.10

The ‘verification MASSterclass platform’ was designed to rapidly turn candidate biomarkers into quantitative multiplexed protein assays without the need for antibodies following depletion (see Supplementary material online, Figure S1B).11 For detailed experimental procedure, see Supplementary material online.

Human cardiac tissue samples used herein and the RNA preparation have been described previously.12,13 For QRT information, see Supplementary material online.

Experimental models

The investigation conformed to the National Institutes of Health Guide for Care and Use of Laboratory Animals was allowed by the Inserm Animal Ethics Committee. The study, using thoracic ascending aortic constriction (TAC; from Days 2 to 56 after surgery) and myocardial infarction (6 months after surgery) in rats, was approved by the Local Animal Ethics Committee (CEEA-LV/2010-01-05 and CEEA-LV/2010-02-05).

Experimental surgery, echocardiography, anatomic indexes, and RNA analysis are described in Supplementary material online.

Statistical analysis

All statistical analyses were performed using R computation language14 and MedCalc (MedCalc Software, Belgium).

MASStermind discovery

The LC-MS/MS signals were analysed to reconstruct features corresponding to distinct peptides. The differentially expressed peptides were selected using SAM15 together with the more generic 1R classifier.10 These peptides were put forward for identification using tandem MS.

MASSterclass verification

The relative quantitation produced by the MASSterclass platform was compiled into peptide expression sets. The diagnostic performance of the proteins was estimated using the apparent area under the receiver operating characteristic curve with its 95% confidence interval (CI).16 Receiver operating characteristic curves were compared using the non-parametric method developed by DeLong et al.17 The diagnostic performance for the combination of BNP and QSOX1 biomarkers was estimated using logistic regression. The model was derived from the log-transformed BNP and QSOX1 concentration. The generalization error on the logistic regression model AUC was estimated using bootstrap cross-validation and 0.632+ methods.18 Correlations among continuous variables were assessed by the Spearman rank-correlation coefficient.

Animal model

Results are expressed as the mean ± SEM. Using PASWStatistics (SPSS), the analysis of variance was followed by multiple comparisons using unpaired Student's t-test combined with Bonferonni correction. The Mann–Whitney test was performed when necessary. P < 0.05 was considered as statistically significant. Linear regression and Pearson's correlation coefficient were used for correlation analysis.


Discovery of novel protein biomarker candidates

The experimental design compared the plasma proteome of 10 ADHF patients sampled at admission to the hospital with 10 controls (see Supplementary material online, Table S1). Figure 1 shows an overview of the process leading to the identification of candidate markers. Statistical analysis of microarrays (SAM) yielded 103 differential proteins. Thus, 49 proteins were retained as potential candidate markers for further evaluation in a larger patient cohort. An important proof-of-concept of the platform was the ‘rediscovery’ of NT-proBNP,2,19 the gold standard marker of HF. Using our discovery platform ability to map the true in vivo N-terminal peptide of each protein, three novel isoforms of NT-proBNP were detected in patients with ADHF: N-terminus (−2), (−3), and (−6) amino acids of which the (−2) isoform seemed to be the most prominent one (see Supplementary material online, Figure S2). In addition, other established markers, such as cystatin-C and C-reactive protein, were identified.

Figure 1

Flow diagram illustrating the discovery pipeline applied to acute dyspnoea. The performance of N-terminal pro-B-type natriuretic peptide, as well as our lead candidate marker quiescin Q6, is highlighted. FDR, false discovery rate.

Gene ontology analysis of candidates showed a majority of either structural or cell signalling proteins, the latter being either secreted ligands or transmembrane receptors (Figure 1). Eighteen per cent of the candidates are involved in inflammation or acute phase response, which is not surprising given the comparison of ADHF patients with healthy controls. Several candidate markers could be linked to cardiac physiopathology, such as oxidative stress response, cardiovascular remodelling, and cellular growth.

Validation of novel protein biomarker candidates in acute decompensated heart failure

Accurate quantitative protein assays were developed for 27 of 49 selected candidates, the clinical performance of which was evaluated in 267 prospectively collected subjects including 147 patients admitted with acute dyspnoea from cardiac or non-cardiac origin, 80 stable CHF patients, and 40 healthy controls. Patient characteristics are summarized in Table 1. Patients diagnosed with ADHF (n = 76) were mainly patients with a previous history of CHF (mean LVEF 37 ± 17%), comparable with the stable CHF population (34 ± 14%).

View this table:
Table 1

Summary of patient characteristics of the verification study

Acute HF (n = 76)Non-cardiac dyspnoea (n = 71)Stable CHF (n = 80)
Age (average)72 ± 1262 ± 1961 ± 12
 Males (%)676479
Medical history (%)
 HF history708.5100
 Coronary artery disease30434
 Diabetes mellitus361841
 Chronic kidney disease32724
Physical examination
 Heart rate (bpm)84 (68–107)92 (75–114)70 (60–78)
 Systolic BP (mmHg)135 (107–161)130 (106–145)111 (104–125)
 Diastolic BP (mmHg)74 (61–87)70 (61–80)70 (60–75)
Echographic examination
 LVEF [median (interquartile range)]35 (25–51)65 (59–65)30 (25–40)
Admission labs
 BNP (pg/mL)1006 (470–2027)119.4 (57–297)355 (146–791)
 NT-proBNP (pg/mL)5591 (2453–11 500)670 (289–1939)1295 (552–3435)
 Creatinine (µmol/L)123.2 (89.5–161.5)79 (65–107.5)101 (88–145)
Admission diagnosis (%)
 Pulmonary embolism3.5NA
  • COPD, chronic obstructive pulmonary disease; BP, blood pressure; LVEF, left ventricular ejection fraction; BNP, B-type natriuretic peptide; NT-proBNP, N Terminal-proBNP.

A novel candidate showed similar diagnostic performance in ADHF patients compared with established markers BNP and NT-proBNP (Figure 1). Thus, QSOX1 was identified a potential marker and showed excellent diagnostic performance, remaining at baseline levels in stable CHF patients. Quiescin Q6 was therefore selected for more detailed analysis.

Quiescin Q6: a novel marker for acutely decompensated heart failure

As shown in Figure 2A, the median plasma levels of QSOX1 are 1.5× greater in ADHF patients compared with both non-ADHF dyspnoea and healthy controls. Quiescin Q6 levels were low in stable CHF patients with median levels equalling baseline levels of healthy controls (Figure 2A). In contrast, while BNP and NT-proBNP do show good diagnostic performance in acute dyspnoea patients, with AUCs of 0.90 (95% CI: 0.85–0.96) and 0.88 (95% CI: 0.81–0.94), respectively, their levels are significantly increased in stable CHF patients leading to large overlaps in NP levels between ADHF and stable CHF populations (Figure 2B). Receiver operating characteristic analysis demonstrated QSOX1 to be highly sensitive and specific for diagnosing ADHF in dyspnoea patients, as indicated by an overall median AUC of 0.86 (95% CI: 0.79–0.92; Figure 2C and D). In the present cohort, QSOX-1 performance seemed unaffected by renal function [AUC 0.86 (0.76–0.93) vs. 0.86 (0.75–0.94) with and without renal dysfunction defined as plasma creatinine greater than or less than 150 µmol/L, respectively] conversely to BNP [0.92 (0.83–0.97) vs. 0.77 (0.65–0.88)] or NT-ProBNP [0.89 (0.79–0.95) vs. 0.70 (0.57–0.82)].

Figure 2

Quiescin Q6 diagnostic performance in acute dyspnoea patients. (A) Levels for quiescin Q6 and (B) for B-type natriuretic peptide in acute decompensated heart failure, non-cardiac dyspnoea, stable chronic heart failure, and healthy populations. Receiver operating characteristic curve of quiescin Q6 compared with B-type natriuretic peptide (C) and N-terminal pro-B-type natriuretic peptide (D) for the diagnosis of acute decompensated heart failure cause of dyspnoea. (E) Receiver operating characteristic curve for B-type natriuretic peptide and quiescin Q6 in ‘grey zone’ patients. (F) Impact of combining quiescin Q6 and B-type natriuretic peptide markers on the diagnostic accuracy. Dotted lines represent the rule-out cut-off for B-type natriuretic peptide at 100 pg/mL and the cut-off for quiescin Q6 for maximum combined accuracy. The ‘grey zone’ of B-type natriuretic peptide is between 100 and 400 pg/mL as described previously.4

For each patient, QSOX1 and BNP levels were compared (Figure 2F). Given the modest albeit significant, correlation coefficient (r = 0.38) between the two, we then combined results of QSOX1 with BNP, which resulted in an overall AUC of 0.95 (95% CI: 0.92–0.98), significantly better than the AUC for BNP alone (P = 0.02). The greatest difference between BNP and QSOX1 diagnostic performance was seen when the plasma BNP level was in the borderline zone, ranged from 100 and 400 pg/mL, also named ‘grey zone’.4 In contrast to BNP, which had no diagnostic value (AUC = 0.63; 95% CI: 0.45–0.82) in this ‘grey zone’, QSOX1 reached an AUC of 0.91 (95% CI: 0.83–1) (Figure 2E). When testing QSOX1 and BNP combination, such combination reached 89% diagnostic accuracy. The major impact of adding QSOX1 to BNP was on the specificity. Indeed, despite a high sensitivity (99%), BNP at its rule-out cut-off 100 pg/mL was poorly specific (44%). The combination of BNP and QSOX1 had a sensitivity of 93% and reached a specificity (83%), greater than BNP alone. Hence, QSOX1 can be used to rule out the false positives identified by BNP and the combination of both markers resulted in a nearly clean ADHF population (Figure 2F, upper right quadrant).

Differential expression of quiescin Q6 in pathophysiological conditions

In normal animals, the QSOX1transcription level is low in the skeletal muscle, moderate in the LV and high in the lung and left atria (LA). To further define the conditions leading to altered QSOX1 expression, we used TAC, a pressure-overload model20 that rapidly progresses from compensated cardiac hypertrophy to ADHF (see Supplementary material online, Table S2 and Figure S3). QSOX1 mRNA levels were up-regulated in the TAC-LV compared with sham-operated at 12 days post-surgery, coinciding with the time of HF, as illustrated by the depressed shortening fraction. Besides, BNP mRNA levels show an earlier rise, at 4.5 days post pressure overload (Figure 3A and B). During the 28 days following surgery, the LV QSOX1 mRNA level was positively correlated with the degree of LV hypertrophy (Figure 3D). At the ADHF time, LV QSOX1 expression is negatively correlated with the LVSF (Figure 3C). The up-regulation of QSOX1 mRNAs was also evident in the LA, 56 days after surgery, when LA hypertrophy is maximal and ADHF is truly developed (see Supplementary material online, Figure S4). A weak but significant correlation of LA QSOX1 levels with the degree of lung congestion is observed (see Supplementary material online, Figure S4). Accordingly, cardiac QSOX1 up-regulation is related to the ADHF development. Quiescin Q6 transcription was unaltered in other studied organs in TAC animals (see Supplementary material online, Figure S5). A similar result was observed in human tissue with a higher QSOX1 expression in LA compared with LV in end-terminal HF patients (see Supplementary material online, Figure S6).

Figure 3

Quiescin Q6 and B-type natriuretic peptide mRNA levels in a rat thoracic ascending aortic constriction model. Quiescin Q6 (A) and B-type natriuretic peptide (B) levels were quantified by quantitative reverse transcription–polymerase chain reaction in thoracic ascending aortic constriction and sham-operated left ventricular at different time-points. The left ventricular performance is indicated by the shortening fraction. (C) Correlation between left ventricular quiescin Q6 mRNAs and left ventricular shortening fraction at Day 56 post-surgery. (D) Quiescin Q6 mRNA levels in the left ventricular correlate with the degree of hypertrophy from Days 0 to 28 after thoracic ascending aortic constriction. *P < 0.05, n = 5–10/groups.

The prime QSOX1 induction during ADHF was also assessed in an ischaemia-induced chronic HF model.20 In comparison with the TAC, this CHF model with reduced LV contraction only shows moderate lung congestion and LA hypertrophy and stable LV QSOX1 transcription, while the BNP level increased by 3-fold (Figure 4).

Figure 4

Quiescin Q6 (A) and B-type natriuretic peptide (B) mRNA levels in a rat chronic heart failure model. Transcripts were quantified in the rat left ventricular 6 months after myocardial infarction (ischaemic heart failure) and in sham-operated animals. EF, left ventricular ejection fraction. *P < 0.05; n = 8/groups.


We here describe the hypothesis-free discovery of known and novel low-abundant plasma protein biomarkers for the diagnosis of ADHF, using a novel clinical proteomics biomarker discovery and a verification platform. Using this technique, we found a novel biomarker QSOX1 to be highly sensitive and specific for ADHF diagnosis in patients with acute dyspnoea with a performance equalling the gold standard biomarkers BNP and NT-proBNP. We further demonstrate that the combination of QSOX1 and BNP markedly reduces false positives and exerts the best specificity for ADHF diagnosis in patients with dyspnoea.

Quiescin Q6 would never have been selected from a hypothesis-driven candidate selection process, as current literature does not link QSOX1 to heart disease. The main physiological role of QSOX1 is the formation of disulfide bridges. While natural substrates for the enzyme remain to be identified,21 it has not escaped our attention that BNP requires disulfide bridge formation in order to possess its crucial ring structure. Importantly, QSOX1 has been linked with the regulation of cellular growth,22 angiogenesis,23 MMP activity,24 and in the cell protection against oxidative stress-induced apoptosis.25 Thus, the potential link between QSOX1 and myocardial injury, cell death, and remodelling is not without significance.

Plasma from ADHF patients shows increased levels of soluble QSOX1, while all other groups tested (including patients with stable CHF) have marker levels similar to baseline (healthy volunteers). This is corroborated in the animal models showing QSOX1 cardiac induction in TAC rats with ADHF, while QSOX1 expression was unaltered in heart of ischaemia-induced CHF. This is in line with a recent publication showing an increased LV QSOX1 levels using the microarray technology in a TAC model in mice.26 Furthermore, all other major organs so far tested expressed a similar level of QSOX1 in sham and TAC animals.

Our findings may be of a high clinical value for ADHF diagnosis in an emergency setting as it would omit the requirement to know patients ‘dry weight’ marker levels;4 NPs are flawed in this context for diagnostic use because of baseline elevation in CHF patients as well as due to cardiac stress from other conditions such as pneumonia even when no important cardiac disease is present.3 Importantly, many diagnoses that confound BNP testing do not seem to affect QSOX1; there is one report on elevated levels of plasma QSOX1 in pancreas cancer patients27; however, the impact of this on diagnostic accuracy in HF patients is expected to be limited.

The success of the proteomics pipeline largely depends on the following aspects: reproducibly and quantitatively tapping into the low ng/mL proteome. By rediscovering NT-proBNP, the gold standard marker for ADHF, we provide evidence that we are examining the right pool of proteins. Furthermore, our approach highlighted new isoforms of NT-proBNP, and we believe that the (−2) form is the main circulating form of the peptide, which could correspond to consecutive dipeptidyl peptidase IV cleavage events, comparable with what has been reported for BNP.28 As current immunoassays for NT-proBNP are based on antibody directed against its starting amino acids (see Supplementary material online, Figure S2), they may not adequately detect the now discovered N-terminally truncated peptides, causing underestimation of the actual NT-proBNP amount. Furthermore, the different processed forms may represent different aspects of HF pathology.

The present study suffers from some limitations. The validation of the new biomarker QSOX1 was performed in a small cohort of patients. It had, however, enough power to qualify QSOX1 as a good performing biomarker to discriminate ADHF from all other causes of acute dyspnoea and even from stable chronic HF. Confirmatory prospective human studies are, however, necessary to further establish QSOX1 as a novel marker for ADHF in acute dyspnoeic patients and to assess the exact impact of different comorbidities on QSOX1 diagnostic performance; this project is ongoing using commercially available standard immune-based assays. The precise pathophysiological role of QSOX1 in ADHF remains unclear. Though we provide evidence that the heart is the source of QSOX1 in ADHF, further studies will decipher the role of QSOX-1.

In summary, unbiased proteomics discovery strategy allowed the discovery of a novel HF biomarker QSOX1, showing a particular value for the clinical evaluation of dyspnoea. We found that QSOX1 is unaffected by many of the factors that weaken the value of BNP, and the combination of BNP plus QSOX1 provided the best sensitivity and specificity for ADHF in patients with dyspnoea.

Author contributions

A.M., A.C.S., M.V., J.B., C.M., K.K., G.V., H.D., P.M., J.-L.S., K.V., and W.L. designed research; G.T., J.W., G.V., A.M., and J.-L.S. analyzed data; N.L., F.D., E.V., C.V., R.T., L.V., K.C., J.-L.S., and J.-M.L. performed research; A.M., M.V., J.B., M.T., B.L.-J., C.C., and D.L. contributed clinical samples; A.M., J.-L.S., K.K., J.L.J., and G.V. drafted the paper and all coauthors edited the paper.


This work was supported by the Inserm (AAP 2009/RRC), AP-HP (CI/JLS), and the Agency for Innovation through Science and Technology in Flanders (IWT, Belgium).

Conflict of interest: A.M. and A.C.S. received speakers' fees from Alere and BRAHMS; A.M. received consulting fees from Pronota and Bayer Pharma. C.M. received research support from Pronota, Brahms, Alere, Abbott, and Roche and speakers' honoraria from Brahms, Alere, Abbott, and Roche. A financial collaboration exists between Pronota and U942 and Cardiovascular-Center-Aalst, respectively. No funding bodies had any role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.


We acknowledge C. Delcayre C, P.U. Milliez and M. Sadoune S. Laribi, N. Deye, Pr B Cristani, N Cigna (UMR700 Inserm- Paris-7), for their contribution as well as K. Sandra, A. Schoonjans, S. Degroeve, D. Vlieghe, M. Moshir, B. Ruttens, R. Colman, K. Huijben and S. Eyckerman. We further acknowledge all members of the GREAT (Global REsearch on Acute conditions Team) network.


View Abstract