Abstract
Introduction Imatinib reduced 90-day mortality in hospitalised COVID-19 patients in a recent clinical trial, but the biological effects that cause improved clinical outcomes are unknown. We aimed to determine the biological changes elicited by imatinib in patients with COVID-19, and what baseline biological profile moderates the effect of imatinib.
Methods Secondary analysis of a randomised, double-blind, placebo-controlled trial of oral imatinib in hospitalised, hypoxemic COVID-19 patients. Mediating effects of changes in plasma concentration of 25 plasma host response biomarkers on the association between randomisation group and 90-day mortality were studied by combining linear mixed-effect modelling and joint modelling. Moderation of baseline biomarker concentrations was evaluated by Cox regression modelling. We identified subphenotypes using Ward's method clustering and evaluated moderation of these subphenotypes using the above-described method.
Results 332 out of 385 participants had plasma samples available. Imatinib increased the concentration of surfactant protein D (SP-D), and decreased the concentration of interleukin-6, procalcitonin, angiopoietin 2 to 1 ratio, E-selectin, tumour necrosis factor (TNF)α, and TNF receptor I. The effect of imatinib on 90-day mortality was fully mediated by changes in these biomarkers.
Cluster analysis revealed three host response subphenotypes. Mortality benefit of imatinib was only present in the subphenotype characterised by alveolar epithelial injury indicated by increased SP-D levels in the context of systemic inflammation and endothelial dysfunction (HR 0.29, 95%-CI: 0.10–0.92).
Conclusions The effect of imatinib on mortality in hospitalised COVID-19 patients is mediated through modulation of innate immune responses and reversal of endothelial dysfunction, and possibly moderated by biological subphenotypes.
Introduction
Acute hypoxemic respiratory failure is the most common reason for hospitalisation in patients infected with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1]. The introduction of different treatment strategies, including steroids, interleukin (IL)-6 inhibitors and therapeutic anticoagulation [2–4], has resulted in improved clinical outcomes in hospitalised non-critically ill COVID-19 patients. However, with the best standard of care, mortality in hospitalised patients remains substantial with a rate of around 6.5% [5].
Observational studies have linked unfavourable outcomes in COVID-19 patients to dynamic changes in the plasma concentrations of biomarkers reflecting modulation of the innate immune response, endothelial barrier protection and epithelial injury [6, 7]. Imatinib, an ABL tyrosine kinase inhibitor, has been shown to improve the endothelial barrier by reversing the loss of cell-matrix adhesion and adherens junctions in vitro and in animals in vivo [8, 9], and in addition has immunomodulatory effects [10]. A randomised controlled trial of imatinib in hospitalised, hypoxemic patients with COVID-19 did not show statistical improvement in the primary endpoint (i.e. duration of oxygen therapy), but revealed a large decrease in 28-day mortality [11]. Imbalances in baseline characteristics between treatment arms were suggested to drive part of the protective effect, which could have led to a type I error. In an extended follow-up study, the survival benefit of imatinib at day 90 remained statistically significant in both the unadjusted and adjusted analysis, increasing the likelihood of a true protective effect [12].
Our current understanding of how pharmacological interventions improve outcomes in COVID-19 is limited. There is a general conception that immunomodulation of the innate immune response and endothelial protection are important in the treatment of severe COVID-19 [13]. It is therefore assumed 1) that a change in these biological processes is required for the drug to work and 2) that patients with higher baseline activation of these pathways are more likely to respond [14–16]. However, this assumption has not been formally tested. Mediation analysis can be used to evaluate if a drug only results in improved outcomes when it elicits specific biological effects, thus testing if an intermediate response is required for the drug to work. Moderation analysis can be used to study whether the relationship between two variables is dependent on the value of a third variable, e.g. a baseline biomarker concentration.
In this study, we aimed to describe what biological changes are elicited by imatinib and how these changes relate to clinical outcomes. We hypothesized that the effect of imatinib on 90-day mortality was mediated by reversal of endothelial dysfunction and modulation of innate immune responses. We also postulated that the baseline biological profile of a patient moderated the effect of imatinib on 90-day mortality.
Methods
Study design and patient selection
This is a pre-specified secondary analysis of clinical data and biological material obtained from a randomised, double-blind, placebo-controlled, clinical trial that was done at 13 hospitals in the Netherlands. Details on study design and patient selection are described elsewhere [11]. In short, patients were eligible for inclusion if they were aged 18 years or older, had been admitted to the hospital with a SARS-CoV-2 infection (confirmed with a RT-PCR test), and required supplemental oxygen to maintain a peripheral oxygen saturation of greater than 94%.
The trial was approved by the Medical Ethics Committee of Amsterdam UMC (location VUmc, Amsterdam, Netherlands), and was done in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki. All patients provided written informed consent before randomisation.
Study procedures
Patients were randomly assigned (1:1) to either placebo or oral imatinib treatment. After randomisation, patients in the imatinib group received a loading dose of 800 mg imatinib on day 0, followed by 400 mg once daily on days 1–9. Patients in the placebo group received placebo tablets in a similar dosing scheme. Heparin anticoagulated blood was collected right before first study drug administration (baseline), and at day 2, day 3 and day 5 thereafter. Plasma was harvested and stored at −80°C within 4 h after blood draw. Plasma was obtained from 20 healthy volunteers to obtain reference normal values.
Data collection
Measurements were done in heparin anticoagulated plasma. Twenty-five biomarkers were measured by Luminex multiplex assay (R&D Systems Inc., Minneapolis, United States), using the Bio-Plex 200 System (Bio-Rad Laboratories Inc., California, United States) in one batch at the end of the study (table S1). Ferritin and C-reactive protein (CRP) were not measured, due to differences in dilution. The data quality assessment is described in the supplementary methods.
Endpoints
For this secondary analysis focusing on the biological effects of imatinib, we used 90-day mortality as the primary endpoint and 28-day mortality as secondary endpoint.
Statistical analysis
All statistical analyses were performed in R (version 4.1.0) using RStudio (version 1.4.766).
Mediation analysis
All biomarkers were log10 transformed to better approach a normal distribution. The association between randomisation to imatinib and the longitudinal biomarker values were estimated using linear mixed-effects models (with the lme4 package) [17]. The randomisation group, the measurement day and their interaction were included as fixed effects. Random intercepts were given to each subject. The effect of imatinib over time was determined by evaluating the interaction term and the 95% confidence interval (95% CI) of this term was calculated using bootstrapping. In a sensitivity analysis, baseline differences (i.e. age, body mass index (BMI), diabetes, and cardiovascular disease) were included as covariates.
Biomarkers that were statistically significant in the above described linear mixed-effects models were subsequently studied using Baron and Kenny's approach for mediation [18]. First, the abovementioned model was used to describe the effect of treatment on biomarker concentration (nlme package) [19]. Second, a joint model that combines a linear mixed-effects model and a Cox proportional hazards model was used to describe the effect of a change in biomarker concentration on mortality (using the survival and JM package) [20, 21]. Third, a Cox proportional hazard model was used to describe the association between randomisation group and mortality. Next, the isolated effect of imatinib on mortality (i.e. the effect explained if no change in biomarker was observed) was calculated. As a sensitivity analysis, a mediation analysis using natural effects as described in the medflex package was performed [22]. All model assumptions are described in the supplementary methods.
Moderation analysis
To estimate the moderation of the baseline biological profile on the association between randomisation group and outcome, we performed Cox regression modelling (with the survival package) [20]. Randomisation group, baseline biomarker concentration and its interaction term were used as independent variables and 90-day mortality as time-to-event variable. Resulting p values were corrected for multiple testing using the Benjamini-Hochberg false discovery rate. A significant interaction term indicates that the effect between imatinib and mortality is influenced by the baseline biomarker concentration. In a secondary analysis, the baseline biomarker concentrations were dichotomised by maximally selected rank statistics (survminer package) [23]. This dichotomised variable was included in the above-described Cox models.
Lastly, we identified subphenotypes of patients with a similar baseline biological status using Ward's method clustering. For this, baseline (pre-treatment) host response biomarker concentrations were used. IL-10 and IL-17 were excluded from this analysis since these disproportionally affected the clustering due to a high proportion of values below the lower limit of quantification (table S2). The optimal number of clusters was determined using a majority ruling as described in the NbClust package [24]. This approach has been used previously to identify and validate subphenotypes of acute respiratory distress syndrome (ARDS) [25]. To evaluate the effect of imatinib treatment on mortality within each cluster subgroup, a Cox regression model with randomisation group as only covariate was performed for patients within each cluster. A Cox regression model with randomisation group, age, BMI, diabetes and cardiovascular disease as covariates was performed as a sensitivity analysis.
Role of funding source
The funders of this study had no role in study design, data collection, data analysis, data interpretation or writing of the report.
Results
Patients
Between March 2020 and January 2021, 385 patients were included in the final analysis population, of which 197 patients were randomised to the imatinib group and 188 patients to the placebo group (figure 1). Baseline biomarker data were available for 154 imatinib patients (78%) and 142 placebo patients (76%). 169 imatinib patients (86%) and 163 placebo patients (87%) had at least one measurement during the study period (table S3). Patients included in the secondary analysis were comparable in terms of age and sex (table 1). Patients in the placebo group more often had obesity, diabetes mellitus and cardiovascular disease. Baseline routine laboratory values and medical treatments, both chronic medication use and medication initiated at hospital admission, in particular dexamethasone, were comparable between the groups. In line with the analysis of the total population [12], 90-day mortality was significantly lower in the imatinib group, with an unadjusted HR of 0.49 (0.26–0.92) and an adjusted HR of 0.47 (0.24–0.94). The primary endpoint of the clinical trial (i.e. time to discontinuation of ventilation and supplemental oxygen for more than 48 consecutive hours, while being alive during a 28-day period after randomisation) was also comparable to the analysis of the total population, with an unadjusted HR of 0.99 (0.77–1.26). The comparison between patients included in the secondary analysis versus patients excluded in the secondary analysis did not demonstrate meaningful differences between the two groups, indicating that the patients in the secondary analysis are a representative reflection of the full study cohort (table S4).
Flowchart of patient selection.
Clinical characteristics of patients included in the cohort that was used for the presented secondary analyses
Effect imatinib on biomarkers
In the linear mixed-effect models, imatinib was found to increase the log10 transformed concentration of the epithelial injury marker SP-D with 0.028 (95% CI: 0.003–0.054) per day (figure 2b; figure S1). Imatinib caused a decrease in the log10 transformed concentration of pro-inflammatory markers: IL-6 with 0.044 (0.015–0.073) per day, procalcitonin with 0.037 (0.016–0.058) per day, TNFα with 0.018 (0.006–0.030) per day, and TNFRI with 0.011 (0.001–0.021) per day. Imatinib also resulted in a decrease of endothelial markers: angiopoietin 2 to 1 ratio (Ang-2/Ang-1) with 0.025 (0.001–0.049) per day, and E-selectin with 0.018 (0.005–0.032) per day. The concentrations of other biomarkers were not affected by imatinib treatment (figure S2A). In a sensitivity analysis, corrected for age, BMI, cardiovascular disease and diabetes, the estimated effects remained the same (figure S2B), confirming that the results were not caused by baseline differences in these variables.
Visualization of mediation analysis. 2B: The effect of imatinib on the biomarker concentration over time, when compared to placebo. 2C: The effect of an increased biomarker concentration over time on 90-day mortality. 2D: The effect of imatinib on 90-day mortality, when the effect of the biomarkers is left out. The effect of imatinib on 90-day mortality is completely mediated by changes in TNFRI, TNFα, Eselectin, Ang-2/Ang-1, procalcitonin and IL-6. Abbreviations: SP-D=surfactant protein D, TNFRI=tumour necrosis factor receptor I, TNFα=tumour necrosis factor alpha, Ang-2/Ang-1=angiopoietin 2 to 1 ratio, IL-6=interleukin-6.
Mediation analysis
Mediation analysis was performed by estimating direct and indirect effects of imatinib on mortality (figure 2a). For the above-described significant biomarkers, the association between the change in biomarker concentration and 90-day mortality was estimated. Higher concentrations of TNFRI, TNFα, E-selectin, Ang-2/Ang-1, procalcitonin and IL-6 and lower concentrations of SP-D were associated with a higher mortality (figure 2c). Incorporation of these models in mediation analysis showed that imatinib was not directly related to 90-day mortality when accounting for the indirect effect via IL-6, procalcitonin, Ang-2/Ang-1, E-selectin, TNFα, TNFRI and SP-D (figure 2d). In a sensitivity analysis using natural effects mediation, complete mediation via IL-6 and TNFRI was confirmed (figure S3). The above-described findings were replicated by a sensitivity analysis modelling 28-day mortality (figure S4).
Moderation analysis
Baseline characteristics of patients included in the moderation analysis (i.e. patients with baseline biomarker data) were comparable to those of patients in the mediation analysis (table S5). The baseline biomarker concentrations in this cohort were comparable between the randomisation groups (table S6). After correction for multiple testing, moderation analysis showed no significant interaction between baseline concentration of any single biomarker and treatment with imatinib on 90-day mortality. In a secondary analysis where patients were categorised by having a high or low baseline biomarker level, no moderation was found either (figure S5).
As none of the biomarkers could capture the biological complexity observed in the included patients, hierarchical clustering was used to group patients into biologically similar groups based on baseline plasma biomarker levels. Majority rules showed that three clusters (10 out of 22 classifications) best explained the variation based on 22 plasma biomarkers (figure 3). 90-day mortality was highest in cluster 1 and 3 (17.6% and 13.9%, respectively), and lowest in cluster 2 (5.7%) (table S7). Plasma concentrations of all biomarkers were generally highest in cluster 1 and lowest in cluster 2, with plasma concentrations of patients in cluster 3 in between (figure S6). Patients in cluster 3 were distinct in their higher concentration of SP-D in plasma, indicative of alveolar epithelial injury. Longitudinal plasma concentrations of patients within each cluster are visualised in figures S7–S9. Only in patients assigned to cluster 3, imatinib resulted in a 90-day mortality reduction (HR 0.30 (0.10–0.92)) (figure 3). There was no mortality difference in patients in cluster 1 or cluster 2. These results remained the same in a sensitivity analysis, correcting for baseline imbalances (table S8).
Heatmap of subphenotypes, based on baseline biological profile. Rows represent patients, columns represent biomarkers. First column: three clusters; yellow is cluster 1, green is cluster 2, orange is cluster 3. Second column: patients that deceased within 90 days are indicated with black, surviving patients with grey. Third column: patients who received imatinib therapy are indicated with gold, placebo patients with grey. Heatmap: a higher concentration in comparison to the other included patients is indicated with red, while a lower concentration is indicated by blue. Ang-2/Ang-1=angiopoietin 2 to 1 ratio, IL =interleukin, IFNγ=interferon gamma, TNFα=tumour necrosis factor alpha, ICAM-1=intracellular adhesion molecule 1, VCAM-1=vascular cell adhesion molecule 1, TNFRI=tumour necrosis factor receptor I, PDGFAB=platelet-derived growth factor AB, SP-D=surfactant protein D, RAGE=receptor for advanced glycation end products, vWF=Von Willebrand factor. 3B: Baseline plasma concentrations of three biomarkers reflective of cluster analysis, stratified according to subphenotype. Data is depicted as box and whisker plots. Dotted lines indicate median values obtained in healthy controls. Asterisks indicates statistical significance by analysis of variance (ANOVA) tests. **** p<0.0001. 3C: Kaplan Meier curves and risk tables for imatinib depicted in gold and placebo shown in grey stratified per biological subphenotype identified by cluster analysis shown in panel A. A Cox proportional hazards model was used to provide p values.
Discussion
In this study, we aimed to describe the biological changes elicited by imatinib and the relationship between these changes and clinical outcomes in hospitalised COVID-19 patients. Our results suggest that the benefit of imatinib is mediated through modulation of innate immune responses and reversal of endothelial dysfunction. None of the individual baseline biomarker concentrations showed evidence for predictive enrichment of patients benefitting from imatinib. Classification of patients into three subphenotypes suggested that a subgroup of patients with profound alveolar injury combined with systemic inflammation and endothelial dysfunction were selectively profiting from imatinib treatment. This information could aid in providing insight in the mechanism of action of COVID-19 related therapies, and in the relationship between biomarkers and a clinical intervention in general.
This is the first human study to assess the effect of imatinib on the host response, and – as far as we know – the first study that links data from a randomised controlled trial to detailed biological profiles. In vitro and in vivo studies showed that imatinib reinforces the endothelial barrier and mitigates alveolar inflammatory responses through nuclear factor kappa B mediated chemotaxis, resulting in lower IL-6 concentrations [10, 26, 27]. Anti-inflammatory effects and endothelial barrier protection were therefore a priori likely to mediate imatinib effectiveness in COVID-19. Although previous studies demonstrated that COVID-19 is not specifically associated with a strong cytokine release syndrome [28], therapy strategies targeting the release of cytokines (e.g. steroids [2], IL-6 inhibitors [3], IL-1 receptor antagonists [29], Janus kinase 1/2 inhibitors [30], and granulocyte macrophage-colony stimulating factor inhibitors [31]) have also shown to be effective in reducing COVID-19 related mortality. The mortality mediating mechanisms of these therapeutics remain uncertain, since only clinical outcomes and no biological data was collected in those studies.
The mediation analysis presented here strongly suggested that a reduction in IL-6 concentration completely mediated the mortality reduction of imatinib. In other words, mortality was only reduced when the plasma concentration of IL-6 decreased after imatinib treatment. Yet, baseline plasma IL-6 concentration did not moderate the effect of imatinib on outcomes, so we dispute the hypothesis that patients with a more pro-inflammatory starting position have more benefit. So why would patients with severe COVID-19 only benefit when systemic anti-inflammatory effects are seen, given that there is little evidence for innate immune responses compatible with cytokine release syndrome? A possible explanation is that imatinib primarily restores the endothelial barrier function, compatible with the observed changes in Ang-2/Ang-1 seen in this study, which might have resulted in cytokine leakage from the alveolar compartment to the systemic compartment [32]. It has indeed been suggested that COVID-19 is characterised by an alveolar cytokine storm instead of a systemic cytokine storm [33, 34].
Given that biological complexity is insufficiently captured by single biomarkers, we used an established clustering method to identify three biological subphenotypes in hospitalised COVID-19 patients [25]. Separation into two biological subphenotypes has been described in ARDS [35] and in COVID-19 [36, 37]. Most of these studies relied on clinical data [38] or a combination of clinical data and biomarkers [35], while we used biological data alone to identify subphenotypes. Furthermore, we focused on patients admitted to the ward, while previous studies were restricted to a critically ill population admitted to the ICU. The inclusion of a comprehensive set of biomarkers provided separation by plasma concentration of SP-D within the subset of patients with an inflammatory profile and endothelial dysfunction (subphenotype 1 versus 3). SP-D is a biomarker of alveolar injury and is increased in patients developing ARDS [39], and an increase in plasma concentration is indicative of alveolar permeability [40]. Imatinib only decreased mortality in the subgroup with this biological profile, suggesting that a certain amount of alveolar permeability in the context of systemic inflammation and endothelial dysfunction needs to be present in order for imatinib to have a protective effect. When validated independently, patients with more alveolar injury in the setting of an inflammatory state could therefore preferentially be selected for imatinib treatment. It remains to be explained how a further increase in SP-D mediated the protective effect imatinib on mortality, as this is counter intuitive in light of the moderation analysis.
Our study has important strengths and some limitations. The use of randomised group allocation eliminates most forms of bias and therefore provides the best possible estimate of a causal treatment effect. Although mortality was a secondary endpoint and the protective effect of imatinib attenuated after correction for baseline differences, the long-term analysis at day 90 demonstrated a persistent survival benefit of imatinib, even after adjusting for baseline imbalances. Furthermore, in our study, the pre-treatment biomarker concentrations did not show any differences between the groups, confirming a comparable baseline biological profile and limiting the explanation that baseline differences were responsible for the observed mediating effects. Because the data was collected systematically with the performed analyses in mind, we obtained biomarker data of a large share of the study population. All patients in the biomarker cohort were alive at the time of the second measurement, excluding immortal time bias as explanation for our findings. Patients without biomarker data had similar baseline characteristics, but had fewer days of oxygen therapy. We assume that no bias occurred in the selection of patients for whom biomarker data was available. We selected 25 biomarkers representative of host response pathways implicated in COVID-19 and the mechanism of action of imatinib; nonetheless, we could have missed an important mediator. The absence of commonly measured biomarkers (e.g. ferritin or CRP) in our biomarker panel limits the ability to compare our dataset to other studies. Second, only the systemic host response was evaluated and the alveolar environment was not sampled nor studied because obtaining alveolar samples in non-intubated patients is infeasible. Last but not least, the study is likely underpowered to detect heterogeneity of treatment effects via moderation analysis and a larger sample might have yielded different results [41]. When examining the moderating effects of the three subphenotypes by an interaction term instead of a stratified analysis, the HRs were comparable but confidence intervals were wide resulting in p values above 0.05, as expected. Therefore, future prospective testing is required to validate our results.
The findings of this study extend our biological understanding of how mortality can be reduced in patients with severe COVID-19. Changes in innate immune responses and endothelial barrier protection appear to mediate the reduction in mortality observed in the imatinib group. We speculate that this may translate to other immunomodulatory treatments as well. Furthermore, we illustrate that we should not assume that patients who have a high concentration of a single biomarker that is considered to be reflective of activation of the pathway that is targeted by the drug results in predictive enrichment within the context of severe COVID-19 pneumonia. Rather, identification of subphenotypes by comprehensive analysis of multiple pathways provided three clusters that responded differently to the tested intervention. This is in line with studies in ARDS [35, 42, 43], and shows that biological profiles should be used for predictive enrichment rather than single biomarker values. The subphenotype that responded favourably to imatinib treatment had a similar severity of illness compared to a subphenotype that did not have survival benefit from imatinib. This is in contrast to previous studies, and might suggest that this subphenotype could be used for predictive enrichment rather than prognostic enrichment.
To conclude, we here show that imatinib works as an effective therapy against severe COVID-19 only when circulating biomarkers confirm decreased systemic innate immune response and improved endothelial barrier function after treatment. Changes in these biomarker concentrations may be used as a surrogate endpoint when validated as mediators for therapy-related survival in other randomised controlled trials. Three biological subphenotypes were identified and only patients classified as having alveolar injury by increased levels of SP-D in the context of systemic inflammatory response and endothelial dysfunction benefitted from imatinib treatment.
Acknowledgements
The authors would like to thank all the patients and their families who participated in this study. We also thank Barbara Smids-Dierdorp and Tamara Dekker for their expert technical assistance in the performance of the Luminex multiplex assays.
Footnotes
Ethics approval: The ethical boards of the participating hospitals approved the collection of data for the study purposes.
Data sharing: De-identified participant data with data dictionary can be shared after approval of a proposal with a signed data access agreement and always in collaboration with the study group.
Collaborators: The CounterCOVID study group: Jurjan Amana, Sara Azhanga,b, Imke H Bartelinkd, Ahmed A Bayoumya,f, Pierre M Betd, Wim Boersmag, Harm J Bogaarda, Peter I Bontah, Karin A T BoomarsI, Lieuwe D J Bosh,j, Liza Botrosa, Job J M H van Bragth, Gert-Jan Braunstahll, Lucas R Celanta, Erik Duijvelaara, Katrien A B Egerh, J J Miranda Geelhoedm, Yurika L Evan Glabbeeka, Hans P Grotjohann, Laura A Hagensj, Chris M Happea, Boaz D Hazesa, Leo M A Heunkse, Michel van den Heuvelo, Wouter Hoefslooto, Rianne J A Hoeka, Romke Hoekstrap, Herman M A Hofsteec, Nicole P Juffermansj,q, E Marleen Kemperk, Azar Kianzada, Renate Kosh, Peter W A Kunstr, Ariana Lammersh, Ivo van der Leet, E Laurien van der Leea, Anke-Hilse Maitland-van der Zeeh, Frances S de Mana, Pearl F M Mau Asamh, Adinda Mierasa, Mirte Mulleru, Elisabeth C W Neefjesu, Anton Vonk Noordegraafa, Esther J Nossenta, Laurien M A Oswaldl, Maria J Overbeekb, Carolina C Pamplonaa, Nienke Paternotteg, Niels Pronkv, Michiel A de Raafa, Bas F M van Raaijm, Merlijn Reijrinka, Job R Schippersa, Marcus J Schultzj, Ary Serpa Netow, Elise M A Slobh, Patrick J Smeelea, Frank W J M Smeenku, Marry R Smitj, A Josien Smitsa, Janneke E Stalenhoefs, Pieter R Tuinmane, Arthur L E M Vanhovea, Jeroen N Wesselsa, Jessie C C van Wezenbeeka
aDepartment of Pulmonary Medicine, Amsterdam Cardiovascular Sciences, Amsterdam UMC, VUmc, Vrije Universiteit Amsterdam, Amsterdam, Netherlands
bDepartment of Pulmonology, Haaglanden Medisch Centrum, The Hague, Netherlands
cDepartment of Internal Medicine, Haaglanden Medisch Centrum, The Hague, Netherlands
dDepartment of Pharmacy, Amsterdam UMC, VUmc, Amsterdam, Netherlands
eDepartment of Intensive Care, Amsterdam UMC, VUmc, Amsterdam, Netherlands
fDepartment of Pulmonology, Chest Unit, Suez Canal University, Suez, Egypt
gDepartment of Pulmonology, Noordwest Ziekenhuisgroep, Alkmaar, Netherlands
hDepartment of Respiratory Medicine, Amsterdam UMC, AMC, University of Amsterdam, Amsterdam, Netherlands
iDepartment of Pulmonology, Erasmus Medisch Centrum, Rotterdam, Netherlands
jDepartment of Intensive Care, Amsterdam UMC, AMC, Amsterdam, Netherlands
kHospital Pharmacy, Amsterdam UMC, AMC, Amsterdam, Netherlands
lDepartment of Pulmonology, Sint Franciscus Ziekenhuis, Rotterdam, Netherlands
mDepartment of Pulmonology Leiden University Medical Center, Leiden, Netherlands
nDepartment of Pulmonology, Isala Ziekenhuizen, Zwolle, Netherlands
oDepartment of Pulmonology, Radboud UMC, Nijmegen, Netherlands
pDepartment of Pulmonology, Antonius Ziekenhuis, Sneek, Netherlands
qDepartment of Intensive Care, Onze Lieve Vrouwe Gasthuis, Amsterdam, Netherlands
rDepartment of Pulmonology, Onze Lieve Vrouwe Gasthuis, Amsterdam, Netherlands
sDepartment of Internal Medicine, Onze Lieve Vrouwe Gasthuis, Amsterdam, Netherlands
tDepartment of Pulmonology, Spaarne Gasthuis, Haarlem, Netherlands
uDepartment of Pulmonology, Catharina Ziekenhuis, Eindhoven, Netherlands
vDepartment of Pulmonology, Gelre Ziekenhuis, Apeldoorn, Netherlands
wDepartment of Critical Care Medicine and Institute of Education and Research, Hospital Israelita Albert Einstein, São Paulo, Brazil
Author contributions: The authors designed the study together and were involved in collecting the data with the help of the study collaborators. JdB and LDJB had access to the raw data, did the analyses and drafted the manuscript. The other authors revised the initial draft. All authors approved the final version of the manuscript.
Funder: This project was funded by an unrestricted grand from the Amsterdam Medical Center Foundation and a bottom-up grant from Nederlandse Organisatie voor Wetenschappelijk Onderzoek/ZonMW (10430 01 201 0007). In addition, this project received funding from the Innovative Medicines Initiative 2 Joint Undertaking (grant agreement number 101005142). Amsterdam Medical Center Foundation; Innovative Medicines Initiative; DOI: http://dx.doi.org/10.13039/501100010767; Grant: 101005142; ZonMw; DOI: http://dx.doi.org/10.13039/501100001826; Grant: 10430 01 201 0007.
Conflict of interest: LDJB reports grants from the Dutch Lung Foundation, grants from the Dutch Lung Foundation and Health Holland (Public-Private Partnership grant), grants from the Dutch Lung Foundation (Dirkje Postma Award), grants from the IMI COVID19 initiative, and grants from Amsterdam UMC fellowship, outside the submitted work. JA is inventor on a patent (WO2012150857A1; 2011) covering protection against endothelial barrier dysfunction through inhibition of the tyrosine kinase abl-related gene (arg). JA reports serving as a non-compensated scientific advisor for Exvastat. All other authors declare no competing interests.
- Received April 13, 2022.
- Accepted June 23, 2022.
- Copyright ©The authors 2022.
This version is distributed under the terms of the Creative Commons Attribution Non-Commercial Licence 4.0. For commercial reproduction rights and permissions contact permissions{at}ersnet.org
References
