Exposure and Outcome in practice: a retrospective cohort study between fibrinolytic suppression and hypercoagulability, the severity of hypoxemia, and mortality in COVID-19 patients

Abstract

Background:

COVID-19 causes hypercoagulability, but the association between coagulopathy and hypoxemia in critically ill patients has not been thoroughly explored. We hypothesized that severity of coagulopathy would be associated with ARDS severity, major thrombotic events, and mortality in patients requiring ICU-level care.

Methods:

Viscoelastic testing by ROTEM and coagulation factor biomarker analyses were performed in this prospective observational cohort study of critically ill COVID-19 patients from April 2020 to October 2020. Statistical analyses were performed to identify significant coagulopathic biomarkers such as fibrinolysis-inhibiting plasminogen activator inhibitor-1 (PAI-1) and their associations with clinical outcomes such as mortality, extracorporeal membrane oxygenation (ECMO) requirement, occurrence of major thrombotic events, and severity of hypoxemia (PaO2/FiO2 categorized into mild, moderate, and severe per the Berlin Criteria).

Results:

In total, 53/55 (96%) of the cohort required mechanical ventilation and 9/55 (16%) required ECMO. ECMO-naïve patients demonstrated Lysis Indices at 30 minutes indicative of fibrinolytic suppression on ROTEM. Survivors demonstrated less procoagulate acute phase reactants such as MP-Tissue Factor levels (OR 0.14 (0.02, 0.99), p = 0.049). Those who did not experience significant bleeding events had smaller changes in ADAMTS13 levels compared to those that did (OR 0.05 (0, .7), p = 0.026). Elevations in PAI-1 (OR 1.95 (1.21, 3.14), p = 0.006), d-dimer (OR 3.52 (0.99, 12.48), p = 0.05), and factor VIII (no clot 1.15 ± 0.28 versus clot 1.42 ± 0.31, p = 0.003) were also demonstrated in ECMO-naïve patients who experienced major thrombotic events. PAI-1 levels were significantly elevated during periods of severe compared to mild and moderate ARDS (severe 44.2 ± 14.9 ng/mL versus mild 31.8 ± 14.7 ng/mL and moderate 33.1 ± 15.9 ng/mL, p = 0.029 and 0.039 respectively).

Conclusion:

Increased inflammatory and pro-coagulant markers such as PAI-1, MP- Tissue Factor, vWF levels are associated with severe hypoxemia and major thrombotic events, implicating fibrinolytic suppression in the microcirculatory system and subsequent micro- and macrovascular thrombosis in severe COVID-19.

Introduction

The coagulopathy engendered by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection is marked by thrombotic events, most notably in critically ill patients requiring life support measures in the intensive care unit (ICU).^1,2 In patients requiring veno-venous extracorporeal membrane oxygenation (ECMO) support for respiratory distress, Coronavirus disease 2019 (COVID-19) associated coagulopathy has the potential to further increase the elevated risk of hemostatic complications associated with ECMO.^3,4 Beyond the macrothrombotic events like stroke (CVA), pulmonary emboli (PE), and deep vein thrombosis (DVT), post-mortem studies also report the presence of microthrombi in the pulmonary vasculature of COVID-19 patients.^5–8 Endothelial cell damage through direct viral infection and local inflammatory response at the alveolar-capillary interface is theorized to contribute to this pathology.^9,10 Inflamed and injured endothelium releases procoagulant factors, such as von Willebrand factor (vWF), and antifibrinolytic factors, such as plasminogen activator inhibitor 1 (PAI-1) which may contribute locally and systemically to COVID-19 associated coagulopathy.

Coagulation biomarkers, including elevated vWF, fibrinogen, and D-dimer, were shown early in the pandemic to predict worse outcomes and could distinguish disease severity and inform anticoagulation intensity.^11–13 More recently, PAI-1 elevation was reported in patients requiring supplemental oxygen compared to those on room air as well as in patients who died compared to those who survived.^14,15 Despite these reports, the association between coagulopathy and respiratory distress, the key driver of mortality in critically ill patients with COVID-19, remains to be determined. To further characterize the association between coagulopathy and hypoxemia in SARS-CoV-2 infection, we performed a prospective observational cohort study of 55 patients admitted to a quaternary care hospital’s ICU for COVID-19 management. We hypothesized that severity of coagulopathy would be associated with worse clinical outcomes including acute respiratory distress syndrome (ARDS) severity, major thrombotic events, and mortality in patients requiring ICU level care.

Materials and Methods

Patient population

From April 2020 to October 2020 fifty-five SARS-CoV-2 positive patients requiring ICU level of care were screened and enrolled in the study. Eligible patients were men and women ages 18 years old and above who were admitted to an adult ICU at a quaternary hospital center with SARS-CoV-2 infection confirmed by polymerase chain reaction (PCR) testing. Eighty-nine percent of the cohort was intubated, requiring ventilation, and the remainder either required vasopressor support or high flow oxygen support. Written informed consent was obtained from the patient or their Legally Authorized Representative if they lacked decision-making capacity. Patients were ineligible for enrollment if they were enrolled in a blood conservation program or were incarcerated. Citrated whole blood was collected on days 1, 3, 7, 14, and 21 following enrollment, or until discharge from ICU or death. Average time from entry to ICU to day 1 of study was 6.9 (SD 5.3) days. These samples were centrifuged and stored at −80°C for analysis. All patients received prophylactic low molecular weight heparin (0.5mg/kg twice daily) or unfractionated heparin (5,000 IU every 8 hours) dosing for venous thromboembolic prophylaxis excluding ECMO patients who received full dose titrated unfractionated heparin to a partial thromboplastin time (PTT) goal of 50–60 seconds. The study was approved by the academic center’s Institutional Review Board and is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology guidelines (Supplemental Table 1).

Clinical data extraction

Clinical data were extracted from the electronic medical record (EHR). Vital signs, laboratory data, International Statistical Classification of Diseases (ICD) 10 codes for comorbidity analysis, and demographic data were extracted from EHR Oracle database across the cohort’s ICU encounters using SQL queries.

Association of ARDS severity with laboratory values

The ratio of arterial partial pressure of oxygen (PaO2) to fraction of inspired oxygen (FIO2) (PaO2/FiO2) was calculated from ventilator and clinical laboratory data at each episode of blood collection for each patient. PaO2/FiO2 thresholds of 300–200, 199–100, and <100 were classified as mild, moderate, and severe ARDS respectively per the Berlin Criteria.¹⁶ In addition, stratification of disease severity based on a patient’s lowest PaO2/FiO2 during their ICU course was completed and an outcomes analysis performed. All patients requiring ECMO were classified as having severe ARDS as the PaO2/FiO2 ratios for ECMO patients fail to take into account the additional oxygenation provided by the ECMO circuit itself resulting in a PaO2/FiO2 ratio that does not reflect the severity of the patient’s respiratory distress. All laboratory values were associated with the patient’s PaO2/FiO2 severity level at the same date and time of blood collection.

Enzyme-linked immunosorbent assays (ELISAs) and activity assays

The following activity assays and enzyme-linked immunosorbent assays (ELISAs) were performed per the manufacturers’ instruction using citrated plasma at dilutions indicated for each assay: Factor VIII Chromogenix Coamatic activity assay (Diapharma, K822585) 1:800 dilution; microparticle bound tissue factor (MP-Tissue Factor) (Zymuphen 521196) undiluted plasma; a disintegrin and metalloproteinase with a thrombospondin type 1 motif, member 13 (ADAMTS13) (Abcam ab234559) 1:1000; von Willebrand Factor (vWF) (Abcam ab223864) 1:8000, Factor IX (Abcam ab188393) 1:1000; tissue factor pathway inhibitor (TFPI) (Abcam ab274392) 1:100; plasminogen activator inhibitor 1 (PAI-1) (Abcam ab269373) 1:20. All plates were read on a 96-well plate reader (Spectral Max i3, Molecular Devices).

Population means and assay validation with healthy controls

Healthy population means are depicted by dotted line in Figures 1–3 and were determined from clinical laboratory normal ranges for D-dimer and fibrinogen, and from previous reports for vWF,¹⁷ ADAMTS13,¹⁸ TFPI,¹⁹ and PAI-1.²⁰ Healthy mean values for Factor VIII activity were determined by assessment of plasma from eight healthy young adult volunteers. This was in part due to a lack of consistent reported health population means. Validation of ELISA data from COVID-19 patients was performed via within-assay comparison to healthy plasma from donors matched for age, race, ethnicity, and BMI of COVID-19 patients purchased from Innovative Research (www.innov-research.com).

Figure 1:

Patients with severe COVID-19 demonstrate procoagulant profile on ROTEM, marked by shortened Clot Formation Time (B and G), increased alpha angle (C and H), and reduced Lysis at 30 minutes after clotting time (CT) (E and J) on EXTEM and INTEM, and elevated maximum clot firmness on EXTEM, INTEM, and FIBTEM (D, I, and M). ECMO alters coagulative phenotype of COVID-19 patients, increasing average CFT (B and G) and lowering alpha angle (C and H) and MCF (D and I) into the normal range. Extrinsic coagulation time was normal (A), though drifted longer in INTEM due to anticoagulation with low molecular weight heparin (F). Grey bar indicates range of normal clinical values. Each dot represents a single patient timepoint. Significant one-way ANOVAs were followed by Tukey’s post-hoc comparisons shown in figures. P<0.05 was considered significant. * p<0.05, ** p<0.01, *** p<0.001, **** p<0.0001, and non-significant p-values are written out. Values plotted as mean +/− SD. Representative INTEM tracings from a healthy donor (K) and an ICU patient on study (L) are included for reference, with the initial green line representing coagulation time, the pink triangle clot formation time, the angle of incidence of the clot formation time as alpha angle, the thickest aspect of the tracing as maximum clot firmness, and the residual clot firmness amplitude in percentage of maximum clot firmness (MCF) at 30 min after clotting time (CT) as Lysis index at 30 minutes.

Figure 3:

Markers of endothelial damage like plasminogen activator inhibitor-1(PAI-1) and microparticle-bound tissue factor (MP-Tissue Factor) are elevated in severe Acute Respiratory Distress Syndrome (ARDS) compared to mild and moderate ARDS. Mild ARDS is defined as PaO2/FiO2 between 300 and 200, Moderate as PaO2/FiO2 between 100 and 200, and Severe as PaO2/FiO2 <100. Values are plotted as mean +/− SD. Dotted line indicates the mean of healthy normal where available. Significant one-way ANOVA for PAI-1 was followed by Tukey’s post-hoc comparison between groups displayed in the figure. P<0.05 was considered significant. * p<0.05, and non-significant p-values are written out.

Rotational thromboelastometry (ROTEM)

All assays were performed on a clinical ROTEM machine (ROTEM delta, Tem Innovations, Munich, Germany) according to the manufacturer’s instructions including: fibrin-based extrinsically activated test with tissue factor and the platelet inhibitor cytochalasin D (FIBTEM), extrinsically-activated test with tissue factor (EXTEM), intrinsically-activated test using ellagic acid (INTEM), and INTEM assay performed in the presence of heparinase (HEPTEM). EXTEM and FIBTEM for all patients, and either an INTEM or HEPTEM were performed depending on patient’s heparization status, with HEPTEM performed if the patient was therapeutically heparinized. The ROTEM machine was calibrated daily, and reagents calibrated weekly.

Data visualization and statistical analysis

There were no missing data within the study period, the total duration of the ICU stay, following enrollment in the study. Prior to accessing data, a statistical analysis plan was formulated. The clinical data were analyzed using R code. (R Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.) Assuming an event rate of 25%, power analysis demonstrated that with an alpha of 0.05 and 80% power a sample size of 52 would be sufficient to detect a 50% difference in biomarker levels between survivors and non-survivors. Primary outcomes included mortality, major thrombotic events (DVT, PE, CVA), and ARDS severity per the Berlin criteria mild (PaO2/FiO2 200–300), moderate (PaO2/FiO2 100–199), and severe (PaO2/FiO2 <100).¹⁶ The lowest PaO2/FiO2 ratio during a patient’s hospitalization was used to determine the severity of their ARDS. This excluded patients on ECMO. Explanatory variables included coagulopathy biomarkers and thromboelastometric measures. Laboratory values were measured multiple times each day during the ICU hospitalization. A maximum, minimum, or average of these measurements was used to represent the worst effect the variable had on clinical outcome in each ICU. Furthermore, Wilcoxcon rank sum tests were used to associate each log₂-transformed biomarker on ICU day 1, study day 1, or ICU day prior to a clinical event of interest with each outcomes. A P value less than .05 was regarded as statistically significant. Logistic regression models were used to assess the association between each biomarker and each clinical outcome. Two-way analysis of variance (ANOVA) and mean +/− standard deviation (SD) of laboratory values were analyzed with GraphPad Prism 9 software for ECMO versus non-ECMO analysis. Only laboratory data and biomarkers from the first day of the ICU stay were used to avoid repeated measures in the analysis of ECMO data. Potential confounding variables including age, race, and comorbidities were assessed via stratification analysis and reported in tables for each outcome measure. Homogeneity of variance for each continuous measure was assessed via residual plot and normality of distribution was assessed via Q-Q plot. Dotted lines were used to indicate the mean of normal clinical values and grey shading used to indicate the range of normal clinical values where applicable.

Results

Clinical characteristics

The median age was 57 years (SD 6) and 14/55 (25%) of the cohort was female (Table 1). Thirty percent of the cohort identified as Black and 14/55 (25%) identified as Latino. Most common comorbidities were obesity (31/55 (55 %)), hypertension (23/55 (41%)), type 2 diabetes mellitus (20/55 (36%)), and cardiovascular disease (11/55 (20%)). The average time from the first PCR positive test to day 1 of study was eight days. Average time from entry to ICU to day 1 of study was three days. 53/55 (96%) of the cohort required mechanical ventilation and 9/55 (16%) required ECMO. A total of 13/55 (23%) died during the index hospitalization, 27/55 (48%) of patients developed ventilator associated pneumonia, 10/55 (17.8%) of patients required continuous renal replacement therapy, and 3/55 (5.3%) of patients developed heart failure due to myocarditis. Mean duration of mechanical ventilation was 21 days, 9/55 (16%) of patients developed major hemorrhage, most of whom were on ECMO support, and 14 (25%) of patients developed major thrombosis defined as deep vein thrombosis (DVT) in upper and/or lower extremities, pulmonary embolism (PE), and/or stroke (Supplemental Table 2). Analysis of clinical characteristics and comorbidities between survivors, non-survivors, and ECMO patients demonstrated that patients on ECMO were on average younger with fewer comorbidities which correlates with criteria for clinical ECMO eligibility. The non-survivor group had less obesity as well as cardiovascular disease and chronic lung disease compared to the survivor group in contrast to previous studies.²¹ Rates of diabetes and renal disease were similar across survivor and non-survivor groups but decreased in ECMO patients. Non-survivors and ECMO patients had more hypertension compared to survivors.

Table 1:

General clinical summary of patients including comorbidities

Characteristics	Total population (n=55)	Survivors (n=35)	Non-Survivors (n=11)	ECMO (n=9)
Age, (mean)	57	55	63	51
Female, (%)	14 (25%)	10 (28%)	2 (18%)	2 (22%)
Race/Ethnicity, (%)
Black, (%)	21 (38%)	15 (42%)	4 (36%)	1 (11%)
White, (%)	17 (30%)	12 (33%)	3 (27%)	3 (27%)
Latino, (%)	14 (25%)	7 (19%)	4 (36%)	4 (44%)
Other/Unknown, (%)	4 (16%)	2 (5%)	0	1 (11%)
Comorbidities
Any condition, (%)	50 (89%)	32 (89%)	10 (91%)	7 (78%)
Body Mass Index>30, (%)	31 (55%)	22 (61%)	3 (27%)	6 (67%)
Hypertension, (%)	23 (41%)	16 (44%)	6 (55%)	5 (56%)
Diabetes, (%)	20 (36%)	14 (39%)	4 (36%)	2 (22%)
Cardiovascular disease, (%)	11 (20%)	7 (19%)	4 (36%)	0
Chronic lung disease, (%)	10 (18%)	6 (17%)	4 (36%)	0
Renal disease, (%)	8 (14%)	3 (8%)	4 (36%)	0
Cancer, (%)	8 (14%)	3 (8%)	3 (27%)	1 (11%)
Prior stroke, (%)	6 (11%)	5 (14%)	1 (9%)	0
COVID-19 treatments
Remdesivir, (%)	45 (80%)	29 (81%)	8 (73%)	90 (100%)
Dexamethasone, (%)	29 (52%)	19 (53%)	6 (55%)	5 (56%)
Convalescent Plasma, (%)	12 (21%)	7 (19%)	2 (18%)	3 (33%)

Procoagulant profile of severe COVID-19

ECMO-naive COVID-19 patients requiring ICU care had hypercoagulability on viscoelastic testing with ROTEM compared to the normal reference range (Supplemental Table 2, Figure 1) characterized by shortened clot formation time and increased alpha angle on EXTEM and INTEM, indicative of increased activity of both clotting factors and platelets (Figure 1B, C, G, and H).^22,23 Our cohort also demonstrated elevated maximum clot firmness on FIBTEM, EXTEM, and INTEM aligning with the hyperfibrinogenemic state of severe COVID-19 (Figure 1D, I, and M). Furthermore, Lysis Indices at 30 minutes of 100% despite elevated D-dimer levels are consistent with fibrinolysis inhibition (Figure 1E and J). Median lysis indices at 45 minutes was 100%, and the mean was 99%, providing further evidence for fibrinolytic suppression.

There was no difference in ROTEM tracings of ECMO-naive survivors versus non-survivors in this cohort (Supplemental Table 2). In contrast, COVID-19 patients placed on ECMO support did not consistently manifest the typical procoagulant phenotypes of severe COVID-19 as assessed by ROTEM analyses, with significantly increased clot formation time and reduced alpha angle and maximum clot firmness compared to ECMO naïve patients (Supplemental Table 2, Figure 1). When measured over time, some patients placed on ECMO lost the procoagulant pattern characteristic of this ROTEM tracing, but regained the phenotype after decannulation (Supplemental Figure 1). Others were shown to become more procoagulable over time. These observations demonstrate variability in coagulation changes in patients on ECMO. The loss of procoagulant phenotype seen in some ECMO persisted even after discontinuation of heparin for hemorrhage. Platelet counts were analyzed over time and there was no significant change in platelet count corresponding to changes in maximum clot firmness in ROTEM. Given these findings, we separated ECMO-naïve patients from the nine ECMO-requiring patients for analysis.

Survivors have less severe pro-coagulant profile compared to non-survivors

Patients who survived their hospitalization were less likely to have higher levels of procoagulant acute phase reactants including MP- Tissue Factor levels prior to death (OR 0.14 (0.02, 0.99), p = 0.049) as well as maximum vWF levels prior to death (survivors 5.4 ± 0. 4 versus non-survivors 6.2 ± 0.4, p=0.063), in a log₂-transformed analysis using a Wilcoxon rank sum test, though this did not reach significance. ADAMTS13 levels, an enzyme that shortens large procoagulant vWF multimers, demonstrated a significantly smaller delta in patients who did not experience major bleeding event versus those who did (OR 0.05 (0.7), p = 0.026).

Elevations in PAI-1, vWF, D-dimer, and factor VIII are associated with major thrombotic events independent of ECMO

Twenty-five percent of patients had clinically significant thrombotic events, including deep vein thrombosis (DVT) in upper and/or lower extremities, pulmonary embolism (PE), and stroke. Patients requiring ECMO had a higher frequency of thrombotic events (7/9 (78%)) compared to non-ECMO patients (7/47 (18%)). We next explored which coagulation factors were associated with thrombotic events, separating the ECMO-naïve from ECMO-requiring given the ECMO-associated differences in coagulative profile described above (Supplemental Table 3). Non-ECMO patients who experienced a thrombotic event were more likely to have significantly elevated D-dimer and PAI-1 levels on day one of ICU hospitalization compared to those without (OR 1.95 (1.21, 3.14), p = 0.006 and OR 3.52 (0.99, 12.48), p = 0.05 respectively).

A two-way ANOVA of coagulation parameters, also measured on day one of a patient’s ICU stay, for clotting x ECMO was performed. A significant amount of variation in D-dimer levels was associated with clotting (Figure 2B, p_clot<0.001, mean for no clot 5301 ± 12239 ng/mL versus clot 16540 ± 21233 ng/mL). However, this effect was largely driven by elevation in D-dimer values within the ECMO population, as illustrated in Figure 2B (ECMO (p_ecmo <0.001) and ECMO x clot (p_{clot x ecmo} = 0.001), means for ECMO-naive 5835 ± 11270 ng/mL versus ECMO 23491 ± 25806 ng/mL).

Figure 2:

Coagulative parameters for patients with and without significant thrombotic events (clot) during hospitalization, separated by ECMO status (ECMO). Regardless of ECMO status, elevations in plasminogen activator inhibitor-1 (C), von Willebrand Factor (E), and factor VIII (D) associated with thrombotic events. Neither fibrinogen (A) nor ADAMTS13 levels (F) had variance accounted for by thrombotic events, though A disintegrin and metalloproteinase with a thrombospondin type 1 motif, member 13 (ADAMTS13) levels were lower in non-ECMO patients who had a thrombotic event (F). D-dimer levels were markedly elevated in ECMO patients who had a thrombotic event (B), which drives the significant interaction term of clot x ECMO (B). P-values shown on graphs from top to bottom are the p-values for variance accounted for by clotting term (p_clot), ECMO term (p_ecmo), and clotting x ECMO term (p_{clot x ecmo}) respectively in a two-way ANOVA for clot x ECMO. Dotted line indicates the mean of normal healthy values for swift comparison. Values plotted as mean +/− SD.

PAI-1 levels accounted for a significant source of variation in thrombotic events (Figure 2C, p_clot = 0.003, mean no clot 26.3 ± 17.8 ng/mL versus clot 38.8 ± 15.2 ng/mL) as well as Factor VIII activity (Figure 2D, p_clot = 0.003, mean no clot 1.15 ± 0.28 OD650 versus clot 1.42 ± 0.31 OD650) and vWF levels (Figure 2E, p_clot = 0.096, mean no clot 36.9 ± 22.5 µg/mL versus clot 45.2 ± 22.1 µg/mL). This suggests that fibrinolytic suppression and endothelial damage-associated factor release contribute to clinically significant thrombotic events regardless of ECMO status. In non-ECMO patients, ADAMTS13 levels showed an inverse association with vWF levels (mean no clot 486 ± 184 ng/mL versus clot 399 ± 139 ng/mL), though this pattern was not observed in patients requiring ECMO (Figure 2E). Changes in fibrinogen levels were not associated with thrombotic events (Figure 2A). Supplemental Table 3 further reports ECMO and non-ECMO patient characterization with and without thrombotic events. Patients with thrombotic events had a higher incidence of ventilator associated pneumonia, dialysis, and mortality independent of ECMO status.

Elevated PAI-1 levels are associated with severe ARDS

Table 2 summarizes the means of laboratory values of all patient timepoints stratified by the Berlin Criteria for ARDS severity at time of blood collection.¹⁶ In Supplemental Table 4, patients who at one point during their hospitalization met criteria for severe ARDS had higher mortality and morbidity, including elevated rates of ventilator associated PNA, major hemorrhagic events, and major thrombotic events, as previously described, compared to patients who remained in either mild or moderate ARDS throughout their ICU course. As seen in table 2, PAI-1 levels were significantly elevated during periods of severe compared to mild and moderate ARDS (severe 44.2 ± 14.9 ng/mL versus mild 31.8 ± 14.7 ng/mL and moderate 33.1 ± 15.9 ng/mL, p = 0.029 and 0.039 respectively; Figure 3). Elevation of MP-Tissue Factor, a marker of endothelial damage and thromboimmune activity, was also observed in severe ARDS though this did not reach significance (severe 1.8 ± 1.5 pg/mL, moderate 1.2 ± 1.0 pg/mL, mild 1.2 ± 0.8 pg/mL, p=0.116; Table 2 and Figure 3). Additional non-significant differences associated with worsening PaO2/FiO2 ratios include elevation of TFPI and vWF along with a reduction of ADAMTS13 (Table 2).

Table 2:

Means and standard deviations of laboratory values by acute respiratory distress syndrome (ARDS) severity. Means were calculated using laboratory values during the same time PaO2/FiO2 was determined. ECMO patients were not included in this table.

Measure	Mild ARDS (PaO2/FiO2 300– 200)	Moderate ARDS (PaO2/FiO2 100–200)	Severe ARDS (PaO2/FiO2 <100)
N, total patients	1	14	31
PaO2/FiO2a, (SD)	263	147 (27.7)	82 (14.2)
Aspartate Transaminase, U/L (SD)	59	55 (86)	255 (1864)
Alanine Aminotransferase, U/L (SD)	64	42 (47)	108 (621)
D-Dimer, ng/mL (SD)	4641	8937 (16546)	9744 (16996)
Hematocrit, % (SD)	27	28 (5)	28 (6)
Platelets, x109/L (SD)	265	220 (127)	192 (115)
White Blood Count, x103/mL (SD)	11.9	13.1 (7.0)	14.7 (8.3)
C-reactive protein, mg/L (SD)	7.6	12.6 (10.2)	15.2 (10.0)
Fibrinogen, mg/dL (SD)	460	458 (245)	417 (240)
Lactate, mmol/L (SD)	1.92	1.63 (1.52)	3.20 (4.19)
Lactate Dehydrogenase, U/L (SD)	438	455 (290)	407 (228)
pH, (SD)	7.39	7.37 (0.08)	7.33 (0.12)
INRb, (SD)	1.20	1.27 (0.43)	1.23 (0.36)
PTTc, seconds (SD)	48.3	44.0 (20.9)	45.0 (19.8)
Creatinine, mg/dL (SD)	2.11	1.54 (1.59)	1.62 (1.36)
von Willebrand Factor, µg/mL (SD)	36.9	45.4 (20.5)	47.4 (25.1)
A disintegrin and metalloproteinase with a thrombospondin type 1 motif, member 13, ng/mL (SD)	459	332 (184)	397 (169)
Plasminogen Activator Inhibitor-1, ng/mL (SD)	31.8	33.1 (15.9)	44.2 (14.9)
Tissue Factor Pathway Inhibitor, ng/mL (SD)	408	444 (174)	513 (251)
Microparticle-Bound Tissue Factor, pg/mL (SD)	1.2	1.2 (1.0)	1.8 (1.5)
Factor VIII activity, optical density at 650 nm (SD)	1.41	1.32 (0.34)	1.48 (0.27)
Anti-Factor Xa activity, IU/mL (SD)	0.14	0.16 (0.09)	0.17 (0.08)

^aPaO2/FiO2: partial pressure of oxygen (PaO2) to fraction of inspired oxygen (FiO2)

^bINR: international normalized ratio

^cPTT: Partial thromboplastin time

Discussion

We evaluated critically ill ICU patients with ARDS from SARS-CoV-2 infection in order to further characterize COVID-19-associated coagulopathy. While our cohort size is small, a stratified comorbidities analysis demonstrates that survivors and non-survivors had similar rates of cardiovascular disease, chronic lung injury, kidney disease and diabetes. Interestingly the survivor cohort had higher BMIs compared to non-survivors. Obesity is associated with worse coagulopathy in the literature, presumably through baseline endothelial inflammation.^21,24,25 Our cohort did not demonstrate this association, as survivors had higher BMI than non survivors, likely due to small sample size. This suggests the coagulopathy and the associated ARDS severity we describe are less dependent on the presence of higher BMI or other identified chronic diseases.

Severe COVID-19 coagulopathy is associated with macrothrombotic events such as DVT, PE, and CVA leading to increased morbidity and mortality in this patient population. Fibrinogen, D-dimer, and vWF levels and reduced ADAMTS13 levels (an enzyme that shortens vWF long multimers) characterize the hypercoagulable state in COVID-19 patients with ARDS. Our study redemonstrates that patients with COVID-19 coagulopathy have shortened clot formation time and increased maximal clot firmness on viscoelastic testing.²⁶ We also found that levels of fibrinolysis-inhibiting PAI-1 were elevated with no evidence of clot lysis on ROTEM (LI30 and LI45), a finding consistent with fibrinolytic suppression in COVID-19. Furthermore, patients experiencing a clinically significant thrombotic event during their hospital course had elevated PAI-1 levels, vWF levels, and Factor VIII activity compared to those that did not experience such events despite prophylactic anticoagulation (Figure 2).^27,28 These findings are consistent with recent studies of ICU patients with COVID-19 that have identified PAI-1 and inhibition of fibrinolysis as predictors of mortality.¹⁵

Elevations in vWF, TFPI, MP-Tissue Factor, and PAI-1 were also observed in patients with severe ARDS, revealing an association between coagulopathy and ARDS severity in COVID-19 (Figure 3 and Table 2). Although previous studies have commented on the association between fibrinolytic inhibition and major thrombotic events in COVID-19,^29–34 our report links biomarkers, viscoelastic determination of hypercoagulability, and fibrinolytic suppression to the severity of hypoxemia in these patients. Our results describe this mechanism through elevations in endotheliopathic biomarkers and fibrinolytic inhibition, which may cause the extensive pulmonary microcirculatory thrombosis seen in ARDS secondary to COVID-19 leading to severe hypoxemia.^7,8

While these microthrombotic changes, and increased D-dimer levels, are also seen in patients with disseminated intravascular coagulation (DIC), patients with severe COVID-19 present with normal platelet counts and prothrombin times (Supplemental Table 2) unlike patients with DIC.³⁴ COVID-19 patients also lack the systemic vasoplegia and consumptive coagulopathy seen with DIC, likely due to the initial localized endothelial injury and subsequent thrombosis within the pulmonary microcirculation caused by SARS-CoV-2 infection.^7,35 When accompanied by upregulation of acute-phase reactants such as fibrinogen and vWF, COVID-19 associated coagulopathy most aligns with a thrombotic microangiopathic disorder originating at the alveolar-capillary interface, subsequently engendering the patients’ hypoxic state.

We found that non-survivors had higher levels of vWF, MP-Tissue Factor. Patients experiencing a major bleeding event had lower levels of ADAMTS13 compared to patients who did not. This demonstrates that markers consistent with severe endotheliopathy and thromboinflammatory response can distinguish outcomes even amongst the sickest patients.³⁶ While increased levels of vWF can reflect a normal host inflammatory response to infection, elevated levels can demonstrate endothelial injury. Elevated vWF is known to increase prothrombotic effects through multiple mechanisms including platelet activation and aggregation, as well as stabilization of Factor VIII, the only intrinsic pathway factor produced by endothelium. Notably, the immediate upstream intrinsic clotting Factor IX, produced by the liver, was not elevated in these patients. The decrease of ADAMTS13 results in larger procoagulant vWF multimers, which further exacerbates this endothelial-driven hypercoagulability. The degree of these changes can distinguish between differing severity of disease and survivors from non-survivors in critically ill patients.^37–40

As a potential therapeutic strategy, lytic therapy is currently being investigated.⁴¹ Specific inhibition of PAI-1 may also have utility in this disease process, and future studies from our group will explore the impact of an aptamer that inhibits the antiproteolytic activity of PAI-1 in COVID-19.^42,43 These approaches are important as current studies involving therapeutic dose anticoagulation in the ICU have stopped due to futility based on the ACTIV studies.⁴⁴

We conclude that COVID-19 associated coagulopathy in ECMO patients should be considered separately from ECMO-naïve patients. Patients requiring ECMO have severe refractory hypoxemia resulting from pulmonary microcirculatory injury with subsequent higher risk of thrombotic and hemorrhagic events.⁴⁰ Rates of major thrombotic events were 7/9 (78%) in ECMO patients compared to 4/11 (36%) and 3/36 (8.3%) in non-ECMO non-survivors and survivors, respectively. Significant hemostatic changes observed on ROTEM viscoelastic testing demonstrated variability including the loss of the characteristic procoagulant profile of hemostatic biomarkers in some ECMO patients when compared to non-ECMO COVID-19 patients with severe ARDS (Figure 1, Supplemental Figure 1). Studies using ROTEM on COVID-19 patient samples describe a high rate of prothrombotic events for ECMO (90%) compared to non-ECMO (46%) patients, but did not separate these cohorts when generating models predictive of thrombotic events.²⁶ Although ECMO patients demonstrated a loss in the procoagulant profile from a biomarker perspective, they have increased thrombotic outcomes, likely due to the addition of ECMO circulation, which provides a non-endothelial interface for contact activation of the coagulation cascade despite heparin anticoagulation.^45–47 This prothrombotic interface exacerbates the thromboinflammatory response observed in COVID-19, increasing risk for thrombotic and bleeding events through subsequent consumptive coagulopathy. This ECMO difference is reflected by procoagulant biomarkers through relative reduction of fibrinogen and vWF levels and apparent normalization of ROTEM tracings.⁴⁴ The TFPI levels observed in our cohort’s ECMO patients are also higher than the TFPI levels reported in the non-COVID-19 ECMO cohort by Mazzeffi et. al.(Supplemental Table 2)⁴⁸ This difference is likely due to widespread endotheliopathy associated with severe COVID-19 and TFPI’s primary location on the endothelial surface and indicates the compounding coagulopathic insults of ECMO and SARS-CoV-2 infection.

The therapeutic heparinization required to maintain ECMO circuit patency in this hypercoagulable disease state increases risk of hemorrhagic events in these patients over their ECMO-naive counterparts, who generally receive prophylactic doses of low molecular weight heparin. Of note, 54/55 (96.4%) of patients observed received either prophylactic or therapeutic anticoagulation with heparin products during hospitalization. Investigation into the combined effects of ECMO and COVID-19 versus other viral infections is needed to understand each prothrombotic stimuli and its contribution to the associated pathophysiology.

Despite having a large amount of data for analyses, this study was limited to the relatively small cohort size of this observational study. Due to inconsistent data across timepoints, we were unable to perform temporal analyses. To reduce bias through repeated measures, biomarker and laboratory data was either used on ICU day 1 or collapsed into maximum, minimum, and means during a patient’s ICU hospitalization. Possible selection bias could be present due to the limited sample size of participants. Larger population studies are needed to further characterize COVID-19 coagulopathy in ICU patients. This small cohort size also precludes detailed longitudinal analysis, which will be a valuable approach for future studies. Assessment of fibrinolytic inhibition would have been more robust with LI45 and LI60 data for all patients in the study, although the calculated mean difference of 1% between LI30 and LI45 indicates minimal clot dissolution between the two timepoints. Other groups have shown that fibrinolytic suppression measured by LI60 predicts thromboembolic complications in COVID-19 patients,^26,31,33 with even stronger predictive value if D-dimer levels are included.³¹ These findings demonstrate the clinical utility of ROTEM and the importance of these later timepoints in assessment of fibrinolytic abnormalities. Coagulation protein data were limited compared to clinical electronic health record data due to comparatively fewer timepoints for biorepository sampling and laboratory processing constraints. Additionally, an established link exists between obesity, poorer outcomes in COVID-19,^24,25 and the association between obesity and elevated markers of endothelial damage, including PAI-1.^27,28 Future studies with larger patient cohorts will include the impact of BMI on COVID-19 coagulopathy on ARDS severity given the elevated average of BMI in our cohort.

We found that procoagulant biomarkers including those characteristic of endothelial injury are elevated in ICU patients with COVID-19 in association with severe hypoxemic respiratory failure. We note that ECMO-requiring COVID-19 patients need to be considered separately from the broader ICU cohort for analysis of coagulopathy given the consumptive coagulopathy engendered by the dual insult of SARS-CoV-2 infection and the ECMO circuit itself. We found evidence of fibrinolytic suppression on viscoelastic testing with elevated PAI-1 levels that is more apparent in COVID-19 patients with severe disease who develop major thrombotic events. Coupled with markedly elevated d-dimer levels, these findings indicate increased fibrin deposition occurs with decreased elimination in the microcirculation.²² These findings warrant further investigation and encourage the development of potential therapeutic approaches limiting endothelial injury or working to counteract the pathological sequela resulting from both the micro and macrothrombotic pathology of severe COVID-19.

Summary statement:

This study of COVID-19 associated coagulopathy links elevated markers of endotheliopathy like PAI-1 with low PaO2/FiO2 ratios and major thrombotic events, implicating fibrinolytic suppression in both microvascular and macrovascular thrombosis of severe COVID-19.