key: cord-0682791-rnuf8pum
authors: Herold, Tobias; Jurinovic, Vindi; Arnreich, Chiara; Lipworth, Brian J.; Hellmuth, Johannes C.; von Bergwelt-Baildon, Michael; Klein, Matthias; Weinberger, Tobias
title: Elevated levels of interleukin-6 and CRP predict the need for mechanical ventilation in COVID-19
date: 2020-05-18
journal: J Allergy Clin Immunol
DOI: 10.1016/j.jaci.2020.05.008
sha: 385de938da146d8666c1200b146925e1f346c0a4
doc_id: 682791
cord_uid: rnuf8pum

Abstract Background COVID-19 can manifest as a viral induced hyperinflammation with multi-organ involvement. Such patients often experience rapid deterioration and need for mechanical ventilation. Currently, no prospectively validated biomarker of impending respiratory failure is available. Objective We aimed to identify and prospectively validate biomarkers that allow the identification of patients in need of impending mechanical ventilation. Methods Patients with COVID-19 hospitalized from February 29th to April 09th, 2020 were analyzed for baseline clinical and laboratory findings at admission and during the disease. Data from 89 evaluable patients were available for the purpose of analysis comprising an initial evaluation cohort (n=40) followed by a temporally separated validation cohort (n=49). Results We identified markers of inflammation, LDH and creatinine as most predictive variables of respiratory failure in the evaluation cohort. Maximal interleukin-6 (IL-6) levels before intubation showed the strongest association with the need of mechanical ventilation followed by maximal CRP. Respective AUC values for IL-6 and CRP in the evaluation cohort were 0.97 and 0.86 and similar in the validation cohort 0.90 and 0.83. The calculated optimal cutoff values in the course of disease from the evaluation cohort (IL-6> 80 pg/ml and CRP> 97 mg/l) both correctly classified 80% of patients in the validation cohort regarding their risk of respiratory failure. Conclusion Maximal levels of IL-6 followed by CRP were highly predictive of the need for mechanical ventilation. This suggests the possibility of using IL-6 or CRP levels to guide escalation of treatment in patients with COVID-19 related hyperinflammatory syndrome.

The other authors declare no conflict of interest. 49

The pandemic Coronavirus-disease 19 (COVID-19) is characterized by a highly variable 94 course. While most patients experience only mild symptoms, a relevant proportion develops 95 severe disease progression up to respiratory failure. Interestingly, many patients do not show 96 signs of respiratory distress, despite severe hypoxemia in blood gas analysis 1 . About 5% of 97 patients require intensive care including mechanical ventilation 2, 3 . 98

Recently published large retrospective analyses provide a detailed characterization of 99 COVID-19 and identify variables associated with disease severity and high mortality 4, 5 . One 100 of the largest studies so far shows that age, quick sequential organ failure assessment score 101 (qSOFA score) and D-Dimer correlate with in-hospital death in a multivariate analysis 2 

Another group showed a correlation of obesity and increased inflammatory markers in the 103 blood with respiratory failure 6 . 104

In many aspects, severe COVID-19 may be regarded as a viral induced hyperinflammatory 105 condition with multi-organ involvement due to a cytokine cascade 7 . Of these various 106 cytokines, the presence of raised circulating levels of interleukin-6 (IL-6) appears to be key 107 and is closely connected to disease severity not only in COVID-19 8 but also in avian-origin 108 H7N9 influenza infections 9 and the common seasonal H1N1 influenza A 10 . 109

While these studies identify the correlation of parameters with disease severity, prospective 110 factors predicting impending deterioration of patients are not yet established. The broad 111

All patients with PCR proven COVID-19 hospitalized at our institution from February 29 th to 119

April 09 th , 2020 (n=115) were screened and analyzed for baseline clinical and laboratory 120 findings. In total, 26 patients were excluded from the study and the depicted cohort consisted 121 of 89 patients (Table 1) . Patients with palliative treatment (n=3) or hospitalization due to other 122 medical reasons and nosocomial Sars-CoV2-infection on the ward (n=13) were excluded 123 from this study. Additionally, patients already mechanically ventilated at admission (n=8) and 124 those receiving anti-IL-6 antibody treatment (n=2) were excluded ( Figure 1) . 125

Of the 89 evaluable patients, 40 were part of an initial evaluation cohort hospitalized from 126

February 29 th to March 27 th , 2020 (Supplementary Table E1 ). This cohort was used to 127 identify predictive markers of respiratory failure. 128

Following an interim analysis of the initial evaluation cohort 11 , we performed a power analysis 129 to estimate the number of patients needed to validate our findings. Assuming the need of 130 mechanical ventilation to be 20% in the validation cohort and the risks for mechanical 131 ventilation to be 70% and 20% in the high-risk and the low-risk group, respectively, the total 132 sample size for a two-sided test was determined to be 40. We defined an additional safety 133 margin of 10%. This subsequent validation cohort consisted of patients hospitalized from 134

March 27 th to April 09 th , 2020 (n=49) (Supplementary Table E2 

The fully automated Elecsys® system on a cobas e801 platform (Roche Diagnostics, 147

Switzerland) was used to measure single levels of IL-6, as described previously 13, 14 . The 148

Elecsys® IL-6 immunoassay has been standardized against the NIBSC 1st IS 89/548 149 Standard. CRP values were measured on a cobas c702 platform using the Tina-quant® C-150

Reactive Protein assay (Roche Diagnostics, Switzerland). 151

All variables with less than 50% of missing data in the initial cohort were tested for the 153 association with respiratory failure. Categorical variables were tested with the χ 2 test, and 154 numerical variables with the Mann-Whitney U test. When appropriate, a paired test was 155 performed. All tests were two-sided. The p-values were adjusted for multiple testing with the 156

Benjamini-Hochberg-method to avoid inflating the alpha error. An adjusted p-value (q-value) 157 of ≤0.05 was considered significant. We constructed receiver operating characteristic (ROC) 158 curves and calculated the area under the curve (AUC) to compare the predictive ability of 159 continuous variables. The AUC can be interpreted as the probability that the predictor's value 160 for a randomly chosen patient requiring intubation will be higher than its value for a randomly 161 chosen patient not requiring intubation. The optimal cut off was defined as the one 162 maximizing the Youden's Index 15 . Statistical analyses were performed using the R software 163 package (version 3.6.2). Figures were drawn using Graphpad Prism® (Version 6.0).

To initially evaluate predictors of respiratory failure, 40 patients with confirmed COVID-19 169 were recruited from February 29 th to March 27 th , 2020 and served as an evaluation cohort 170 ( Figure 1 ). Thirteen (32.5%) patients deteriorated during hospitalization and required 171 mechanical ventilation. The time from hospital admission to intubation varied from less than 172 two hours to 9 days (median 2 days). Patients requiring mechanical ventilation did not differ 173 in age, comorbidities, radiological findings, respiratory rate or qSofa score (Supplementary 174 Table E1) . 175

Heart rate, markers of inflammation, LDH and creatinine at admission were significantly 176 associated with respiratory failure (Supplementary Table E1 ). Elevated IL-6 showed the 177 strongest association with the need for mechanical ventilation (Figure 2A , p=1.2x10 -5 ). 178

In addition to values at first assessment, follow-up data were available for laboratory 179 variables. These follow-up data were used to test if there are critical laboratory values that 180 are associated with respiratory failure once they have been reached during disease course. 181

For each patient, we assessed the maximum level of each parameter during disease (for 182 patients requiring ventilation, only values before intubation were used). The maximal values 183 were correlated with respiratory failure (Table 2) April 09 th , 2020, of which 19 (39%) required mechanical ventilation. As in the initial cohort, 194 creatinine, LDH, and several markers of inflammation were significantly elevated in patients 195 requiring intubation (Table 2 and Supplementary Table E2 ). Again, IL-6 at assessment was 196 strongly associated with respiratory failure (Figure 2B ), and maximal IL-6 was the best 197 predictor of future respiratory failure among all parameters ( Figure 2D To validate our findings from the initial cohort, we analyzed the number of patients correctly 203 classified regarding their need of mechanical respiratory support by the determined cutoffs of 204 IL-6 and CRP at presentation and in the course of disease (Table 3) . At presentation, IL-6 205 >35 pg/ml as well as CRP >32.5 mg/l showed high sensitivity to detect patients at risk for 206 respiratory failure (84% and 95%) with moderate specificity (63% for both parameters). 207

Measuring IL-6 and CRP values in the course of disease (cutoffs 80 pg/ml and 97 mg/l) 208 increased the specificity for both parameters (83% and 77%) accompanied with a decrease 209 in sensitivity (74% vs. 84%). In detail, nineteen (39%) patients exceeded the calculated 210 maximal IL-6 cutoff (>80 pg/ml) in the validation cohort, compared to 23 (47%) patients 211 exceeding the CRP cutoff (>97mg/l). Of these patients, 74% and 70% were correctly 212 classified by IL-6 and CRP, respectively, as being at risk for respiratory failure (positive 213 predictive value). Of the 30 patients with values below the IL-6 cutoff, 83% did not require 214 mechanical ventilation, while this was the case for 88% of the 26 patients remaining below 215 the CRP cutoff of 97 mg/l (negative predictive value). In total, the calculated cutoffs for 216 maximal IL-6 and CRP both correctly classified 80% of patients regarding their risk of 217 respiratory failure (Table 3) , while values at assessment show poorer predictor properties gained when examining values in the course of disease. The risk ratios for the cutoffs of IL-6 221 and CRP were 4.4 and 6.0 in the validation cohort, with corresponding p-values of 0.00022 222 and 0.00011. The optimal cut point in the validation cohort was slightly lower for IL-6 (60 223 pg/ml) and identical for CRP (97 mg/l). 224

To further evaluate positive and negative predictive values (PPV/NPV) of IL-6 and CRP we 226 combined the two cohorts (Table 1) . We calculated predictive values across the range of all 227 possible cutoffs. The PPV of CRP was consistently lower compared to IL-6 in the overall 228 study cohort ( Figure 4 ). In other words, increased CRP misclassified more patients as being 229 at risk for respiratory failure than IL-6. However, the predictive values strongly depend on the 230 selected cutoff (Figure 4 ). For cutoffs <50 pg/ml for IL-6 and <40 mg/l for CRP (dotted line), 231 the risk of intubation for patients with sub-threshold levels is roughly zero, while patients with 232 levels above these values show a dramatic increase in the risk of respiratory failure. The risk 233 for respiratory failure in patients with IL-6 levels exceeding 210 pg/ml was 100% (dashed 234 line). The NPV of IL-6 and CRP parameters was comparable. In the combined cohort, the 235 optimal threshold value (maximal Youden index 15 ) is highest at 65 pg/ml for IL-6 and for CRP 236 at 97 mg/l (corresponding risk ratio of 18.1 and 6.9). 237 Furthermore, we analyzed the time lag from reaching the cutoff values to intubation in the 238 combined cohort. Patients reached the cutoff of IL-6 (>65 ng/ml) and CRP (>97 mg/l) at a 239 median of 23.2 and 15.7 hours before intubation, resulting in a significant time difference 240 between the two values of 7.5 hours in favor for IL-6 ( Figure 5 ; p=0.014).

circulating levels of IL-6 as well as CRP were highly predictive of the need for invasive 246 ventilation, with corresponding AUC values of 0.97 and 0.90 for IL-6 and 0.86 and 0.83 for 247 CRP in the first and the second cohorts, respectively. Secondly, we defined cutoffs for IL-6 248 (at presentation >35 pg/ml; maximal value >80 pg/ml) and CRP (at presentation >32.5 mg/l; 249 maximal value >97 mg/l) in the evaluation cohort. Cutoff values at assessment correctly 250 classified 71% (for IL-6) and 76% (for CRP) of patients in the validation cohort with a further 251 increase when measuring maximal values in the course of disease (80% for both 252 parameters). Thirdly, elevated IL-6 levels in the course of disease predicted respiratory 253 failure significantly earlier than CRP (23.2 vs. 15.7 hours). Therefore, IL-6 and CRP are 254 useful markers that predict impending respiratory failure with high accuracy and can help 255 physicians correctly allocate patients who might benefit from early treatment escalation, for 256 example using anti-cytokine strategies. We believe that having these data reproduced across 257 the two separate cohorts enhances the strength of our conclusions. It is important to note 258 that the commercial diagnostic IL-6 assay used in our study allows the measurement of Il-6 259 in a comparable time scale as CRP. Since it uses the broadly available Cobas platform it can 260 be implemented in most laboratories. 261

Our study also has several limitations. It is still unclear whether elevated inflammatory 262 markers merely represent an epiphenomenon or a causal pathogenic element of severe 263 COVID-19 16 . It is likely that elevated IL-6 reflects the cytokine mediated hyperinflammatory 264 state as evidenced by the similarly predictive values for CRP. Further, even though IL-6 and 265 CRP levels are significantly elevated in patients requiring ventilation, they are relatively low 266 compared to levels observed in patients with septic shock 17 . However, earlier studies in 267 severe acute respiratory syndrome (SARS) or H7N9 influenza patients show that 268 inflammatory cytokines are highly expressed in lung tissues. Autopsy reports from SARS 269 patients showed a high amount of inflammatory cytokines in cells expressing angiotensin-converting enzyme 2 18 , the functional receptor for SARS-CoV and in even higher affinity for higher concentrations of different cytokines including IL-6 compared to plasma levels, hinting 273 towards a massively increased local concentration of inflammatory cytokines in the diseased 274 lung 9 . Recent preprints provide detailed single cell RNA-sequencing data from immune cells 275

in peripheral blood as well as BAL from COVID-19 patients. The authors report that 276 peripheral monocytes did not substantially express proinflammatory cytokines 20 , while there 277 was high expression in monocyte derived macrophages in BAL 21 . Taken together, these 278 data possibly suggest that circulating levels of IL-6 might be a putative surrogate for the 279 burden of lung tissue damage and provide a "window" into the lung 9 . 280 IL-6 and CRP have been associated with severity of COVID-19 (in most cases defined by the 281

Chinese National Health Commission) and mortality before 22-24 . However, to our knowledge 282 our study is the first to demonstrate a prospective prediction of the end point "mechanical 283 ventilation", which is of high clinical relevance not only for patient treatment but also for 284 resource planning. Very recent publications provide additional data that strengthen the role of 285 IL-6 and CRP in COVID-19 as predictive markers 22, 23 . Unfortunately, these studies did not 286 include a prospective validation cohort and sometimes did not mention analysis platforms 22 . 287 A further difference between our and other studies is the dramatic discrepancy in mortality of 288 severely diseased patients. We are not able to analyze mortality as an end point because 289 only two patients had died until April 12 th . This number has only increased by one until May 290 6 th (overall mortality 3.4%). While still some patients are in critical condition and the mortality 291 rate in our cohort is likely to increase in the next weeks it will be significantly below those 292 reported. We can only speculate about the reasons for this huge difference but argue that 293 overwhelmed hospitals and patient selection might have contributed to the increased 294 mortality observed in other studies. As we did not perform sequential CT-scans after 24-48 295 hours in our patients due to radiation hygiene, we are not able to precisely calculate severity 296 of COVID-19 according the Chinese National Health Commission classification to compare 297 our patient cohort to the cohorts of the mentioned studies. However, our validation cohort at least exists of 63% of severe patients due to the available parameters (2% with mild and Since the start of the pandemic, hundreds of research articles on COVID-19 have been 301 published 25 . To our knowledge, we report the first predictive marker for respiratory failure 302 that was prospectively validated in an independent cohort. Although our sample sizes were 303 small, the large difference in risk for respiratory failure between the high-risk and the low-risk 304 group made it possible to successfully validate our findings. Interestingly, a study of 134 305 patients with avian-origin H7N9 influenza in 2013 also showed a strong correlation of IL-6 306 and disease severity. In analogy to our findings, this study reports that IL-6 plasma levels 307 >80 pg/ml were found in all patients with lethal outcome compared to only 8.3% in surviving 308 patients 9 . The combined cohort (n=89) produced an only slightly lower cutoff for IL-6 (65 309 pg/ml) while the cutoff for CRP levels remained the same at 97 mg/l when calculated from 310 the combined cohort. However, even the combined sample size is probably too small to 311 determine an optimal cutoff value. Furthermore, the acceptable proportion of falsely identified 312 low-risk patients, and therefore the set threshold, is largely dictated by the availability of 313 health care resources. Future prospective studies with larger sample sizes are needed to 314 formally address this issue. We want to stress that IL-6 and CRP should be used as a 315

predictor not an indication for invasive respiratory support, as mechanical ventilation per se 316 has several unintended adverse consequences and may support inflammation of distal 317 airways in COVID-19 patients. 318

Immunologically, CRP and IL-6 are closely intertwined. IL-6 is known to induce gene 319 expression and release of CRP from the liver 26, 27 and also from immune cells 28 . A functional 320 connection has been shown in different trials using IL-6 inhibition, in which CRP-levels 321 rapidly normalized after blocking IL-6 29 . In analogy, we found that IL-6 levels predicted 322 respiratory failure significantly earlier than CRP-levels, which is essential for a predictive 323 marker. While inhibition of inflammatory pathways represents a promising approach to treat response to virus-induced pneumonias 30, 31 . Thus, our study does not facilitate any 326 recommendations for or against IL-6 inhibition. Ongoing randomized controlled clinical trials 327 of IL-6-antibodies in the treatment of COVID-19 will shed light on this question (e.g. 328 NCT04320615 and NCT04331795). More importantly, in times of missing established 329 therapeutic options, best supportive care is essential 32 . 330

In summary, we were able to validate our finding that IL-6 and CRP levels serve as strong 

Procalcitonin as a 386 diagnostic marker and IL-6 as a prognostic marker for sepsis

Index for rating diagnostic tests

COVID-19: a new virus, but an old cytokine release 390 syndrome

Assessment of Clinical Criteria for Sepsis: For the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3)

Defining community acquired pneumonia severity on presentation to hospital: an international derivation and validation study

Clinical Features Predicting Mortality Risk in Patients With Viral Pneumonia: The MuLBSTA Score