key: cord-334801-p5mxc694
authors: Van Singer, Mathias; Brahier, Thomas; Ngai, Michelle; Wright, Julie; Weckman, Andrea M.; Erice, Clara; Meuwly, Jean-Yves; Hugli, Olivier; Kain, Kevin C.; Boillat-Blanco, Noémie
title: COVID-19 risk stratification algorithms based on sTREM-1 and IL-6 in emergency department
date: 2020-10-09
journal: J Allergy Clin Immunol
DOI: 10.1016/j.jaci.2020.10.001
sha: 
doc_id: 334801
cord_uid: p5mxc694

Background The COVID-19 pandemic has led to surges of patients presenting to emergency departments (ED) and potentially overwhelming health systems. Objective This study aimed to assess the predictive accuracy of host biomarkers at clinical presentation to the ED for adverse outcome. Methods Prospective observational study of PCR-confirmed COVID-19 patients in the ED of a Swiss hospital. Concentrations of inflammatory and endothelial dysfunction biomarkers were determined at clinical presentation. We evaluated the accuracy of clinical signs and these biomarkers in predicting 30-day intubation/mortality, and oxygen requirement by calculating the area under the receiver operating characteristic curve (AUROC) and by classification and regression tree analysis. Results Of 76 COVID-19 patients included, 24 were outpatients or hospitalized without oxygen requirement, 35 hospitalized with oxygen requirement and 17 intubated/died. We found that soluble triggering receptor expressed on myeloid cells (sTREM-1) had the best prognostic accuracy for 30-day intubation/mortality (AUROC 0.86; 95% CI 0.77-0.95) and interleukin-6 (IL-6) measured at presentation to the ED had the best accuracy for 30-day oxygen requirement (AUROC 0.84; 95% CI 0.74-0.94) .An algorithm based on respiratory rate and sTREM-1 predicted 30-day intubation/mortality with 94% sensitivity and 0.1 NLR. An IL-6-based algorithm had 98% sensitivity and 0.04 negative likelihood ratio (NLR) for 30-day oxygen requirement. Conclusion sTREM-1 and IL-6 concentrations in COVID-19 in the ED have good predictive accuracy for intubation/mortality and oxygen requirement. sTREM-1- and IL-6-based algorithms are highly sensitive to identify patients with adverse outcome and could serve as early triage tools.

sTREM-1 and IL-6 concentrations in COVID-19 in the ED have good predictive accuracy for 53 intubation/mortality and oxygen requirement. sTREM-1-and IL-6-based algorithms are 54 highly sensitive to identify patients with adverse outcome and could serve as early triage The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic has led to 84 surges of patients that can overwhelm health systems (1) (2) (3) (4) . Medical resource constraints and 85 scarcity are a new reality including lack of hospital beds, oxygen concentrators, and 86 ventilators in both high-and low-income countries (5, 6) . There is an urgent need to have 87 guidance allowing the rational allocation of scarce medical equipment and resources. In this 88 context, it is critical to have triage tools in place permitting the early recognition of patients at 89 risk of adverse outcomes, and allow for optimal resource allocation (7-9). Current prediction 90 models to support clinical decision making for Coronavirus disease (COVID-19) patients 91 were developed based on demographics, clinical signs and symptoms, imaging techniques, 92 biomarkers or a combination of these variables, however most are poorly validated and at risk 93 of bias (10). 94 Cytokine dysregulation during COVID-19 is suspected to contribute to disease severity and 95 multiple clinical trials are underway to test the efficacy of immuno-modulatory therapies (11-96 13). A recent study identified different inflammatory cytokine profiles according to the stage 97 of the disease, suggesting their usefulness for risk stratification in COVID-19 patients (13). 98 Interleukin 6 (IL-6), interleukin 8 (IL-8) and C-reactive protein (CRP) have been proposed for 99 monitoring and prognosis purpose in the context of COVID-19 (14-16). Triggering receptor 8 inflammatory response (23, 24) . In support of this contention, several studies have reported an 108 association between markers of a pro-coagulable state (e.g. D-dimers and fibrinogen) and 109 patients outcome (25) (26) (27) . In a French study, level of Angiopoietin-2 (Ang-2) as marker of 110 endothelial activation predicted with a medium accuracy the ICU admission for COVID-19 111 (28). 119 This prospective observational cohort study of COVID-19 patients was conducted in the ED 120 of Lausanne University Hospital (CHUV), a tertiary care center in Switzerland. We 121 prospectively screened all patients (age ≥ 18 years) upon arrival at the ED with symptoms of 122 an acute lower respiratory tract infection (cough, sputum, dyspnea or chest pain for <21 days) 123 between February 6 th 2020 and April 3rd 2020 (29). Patients were included in this study if 124 COVID-19 was confirmed by real time polymerase chain reaction (RT-PCR) for SARS-CoV-125 2 from a nasopharyngeal swab.

Patients demographics, comorbidities, symptoms, vital signs and laboratory results performed 127 during routine care were recorded in the ED using a standardized electronic case report form 128 in REDCap (Research Electronic Data Capture). Easy-to-measure bedside clinical scores to 129 identify patients at risk of poor outcome were calculated at inclusion: (i) Quick Sequential A plasma sample was collected in the ED at enrollment and stored at -80°C without freeze-142 thaw until analysis. Plasma concentrations of endothelial and immune mediators (Ang-2, IL-143 6, IL-8, and sTREM-1) were analyzed head-to-head on a multi-analyte Ella TM platform with 144 custom-developed reagents from Protein Simple (California, USA) as described (32). C-145 reactive protein (CRP) was quantified by enzyme-linked immunosorbent assays (R&D 146 DuoSet, Minneapolis, MN).

Statistical analyses 148 Differences between the three groups were evaluated by one-way ANOVA, Kruskal-Wallis or 149 Chi-squared, as appropriate. A bilateral p value <0.05 was considered indicative of statistical 150 significance.

The primary outcome was 30-day intubation/mortality. Secondary outcomes were 7-day 152 intubation/mortality, 7-and 30-day oxygen requirement (all patients hospitalized with oxygen 153 requirement, including those who were intubated or died, as they all received oxygen).

These outcomes are important for pragmatic clinical decision-making in the ED. They are 155 objective and therefore more reproducible in other settings than clinician decision to admit a 156 patient to hospital or to the intensive care unit. We hypothesized first, that patients intubated the dependent variable was assessed via the Box-Tidwell procedure and by inspecting the 166 partial residuals. Non-linear variables were transformed for multivariate logistic regression.

The predictive validity of a combinatorial model adding top biomarkers to vital signs or 168 clinical severity scores was measured using logistic regression, and the predicted probabilities 169 were used to generate AUROC. The combinatorial models were compared using the DeLong 186 All included patients signed an informed consent form. Table 1 shows patient demographics, clinical characteristics and laboratory results. While sex 207 distribution, prevalence of co-morbidities and duration of symptoms were similar between the 208 three severity groups, patients in the intubation/mortality group were significantly older than Table 2 ). The need for 30-day 233 oxygen supplementation was best predicted by IL-6 (AUROC 0.84, 95% CI 0.75-0.94) and 234 IL-8 (AUROC 0.82, 95% CI 0.72-0.92) ( Figure 2B ). sTREM-1 and CRP showed an 235 acceptable performance, which was not statistically lower than IL-6. Ang-2 performed 236 significantly worse than IL-6, IL-8 and sTREM-1 (p<0.005). Table 2 ). The combination of respiratory rate 248 with sTREM-1 performed significantly better than the respiratory rate alone (p = 0.024), but 249 not than sTREM-1 alone. We found similar results when combining clinical scores and 250 sTREM-1 (Table 2) . 

After having evaluated the magnitude of the association between clinical signs, scores, 261 biomarkers and adverse outcome, we performed a CRT analysis to find optimal variables and 262 cut-points suitable to generate simple algorithms. The resulting classification tree represents 263 visual decision making. Since it is also important to identify patients who require immediate medical attention, we 275 also assessed the performance of this algorithm to predict 7-day intubation/death. The 276 sensitivity was maintained at 93%, the specificity at 59% and the negative and positive 277 likelihood ratio were 0.17 and 2.3 respectively.

Since we could use this triage tool to identify patients requiring oxygen, we tested this 279 algorithm to predict 7-and 30-day oxygen requirement. While it was highly specific (100%), 280 its sensitivity was low (25%) in this setting. 282 The CRT analysis performed with all clinical signs, severity scores and biomarkers to predict Since we could also use this triage tool to identify patients at high risk of poor outcomes, we 290 tested this algorithm to predict 7-and 30-day intubation/death. It had an excellent sensitivity 291 (100%) but a poor specificity (21%) for 7-day intubation/death and showed a similar 292 performance for 30-day intubation/death (sensitivity at 100% and specificity at 22%). sensitivity. However, the performance of this algorithm to identify patients with a lower level 320 of severity (those requiring oxygen) was not optimal with a low sensitivity. Our data do not 321 support the use of a respiratory rate/sTREM-1 triage algorithm for oxygen requirement. Nevertheless, our data support sTREM-1 and IL-6 as potential candidates for POCT, which 363 could be used as a triage tool at presentation to predict disease severity in COVID-19. In this study both algorithms were very sensitive with a low negative likelihood ratio assuring 376 safe management of patients. Ultimately biomarkers-based algorithms such as these, could 377 enhance clinical decision making and resource allocation; however, this will require further 378 prospective trials to confirm their risk-stratification performance in actual practice.

Acknowledgements 380 We thank all the patients who accepted to participate and make this study possible. We thank 381 Professor Carron, head of the emergency department, who supported the study. We thank all 382 healthcare workers of the emergency department, internal medicine ward, infectious disease 

Nonparametric ROC curves were generated and AUROC were plotted to illustrate the ability 560 of these markers to discriminate between patient groups. Each AUROC was compared to 561 other using DeLong method.

AUROCs for the outcome of each biomarker are presented to the right of its respective forest 563 plot, with 95% CIs in parentheses.

* angiopoietin-2 performed significantly worse than sTREM-1, IL-6 and IL-8 (p<0.05) to 565 predict 30-day oxygen requirement.

No other comparison reached a statistically significant difference (p<0.05). 

For all algorithms, the cost of misclassifying a patient who was intubated or died was 578 designated as 10 times the cost of misclassifying a patient that survived without intubation. AUROCs were calculated from the predictive probabilities of logistic regression models to 30-day mortality/intubation. AUROC of clinical parameters alone and combined with sTREM-1 or IL-6 are presented. Differences in AUROCs were assessed using the DeLong method. * p value < 0.05 comparing the clinical parameter AUROC versus the combined clinical parameter with sTREM-1 or IL-6 AUROC. 

Covid-19: how doctors 399 and healthcare systems are tackling coronavirus worldwide

Critical Care Utilization for the COVID-19 Outbreak 403 in Lombardy, Italy: Early Experience and Forecast During an Emergency Response

COVID-19 in Europe: the Italian lesson -The Lancet

Lessons Learned From the Coronavirus Health Crisis in Madrid, Spain: How 409 COVID-19 Has Changed Our Lives in the Last 2 Weeks

Fair 413 Allocation of Scarce Medical Resources in the Time of Covid-19

Clinical 416 ethics recommendations for the allocation of intensive care treatments in exceptional

Critical Care

Impact of non-pharmaceutical interventions (NPIs) to reduce COVID19 mortality and 421 healthcare demand

Imperial College London

Elevated levels of IL-6 and CRP predict the need for mechanical ventilation in 444 COVID-19

The role of interleukin-6 in 447 monitoring severe case of coronavirus disease 2019

Clinical and pathological investigation of 451 patients with severe COVID-19

Metalloproteinases Shed TREM-1 Ectodomain from

Lipopolysaccharide-Stimulated Human Monocytes. The Journal of Immunology

Cutting Edge: Inflammatory Responses Can Be 458 Triggered by TREM-1, a Novel Receptor Expressed on Neutrophils and Monocytes. The 459 Journal of Immunology

Pathological inflammation in patients with COVID-19: a key role 461 for monocytes and macrophages

Heart rate, bpm; Median (IQR)

Kruskal-Wallis or Chi-squared, as appropriate. # BMI > 30 kg/m 2 . * Heart failure, coronary disease. † Stroke, dementia, parkinson. ‡ Stade III-V according to CKD classification. ® Autoimmune or chronic inflammatory disease. Missing values: obesity 7, duration of symptoms 8, fever 1, cough 1, vital signs 5, blood count 1, chest radiograph 6