Background
Coronavirus disease 2019 (COVID-19) is caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV2). COVID-19 is characterized by vascular inflammation, with evidence of viral elements within endothelial cells [
1]. Initial reports suggested an abnormal activation of coagulation, in particular with highly elevated D-dimer levels [
2], which predicted a poor clinical prognosis and death [
3,
4].
Since the start of the outbreak, several studies have highlighted a high risk of venous thromboembolism (VTE) in inpatients with COVID-19, up to 35–85% and oftentimes despite pharmacological thromboprophylaxis [
5,
6]. However, such large estimates may emanate from the inclusion of small thrombi of uncertain clinical relevance, such as distal deep vein thrombosis (DVT), and may be more prone to being published with high visibility and publicized largely. A similar situation may be found in the lungs, where the vascular filling defects reported on computed tomography pulmonary angiography (CTPA) might reflect “immuno-thrombosis” or local “pulmonary thrombosis” rather than “classic” pulmonary embolism (PE).
Therefore, we aimed to systematically review the published literature on the risk of objectively diagnosed PE and/or proximal DVT – representing major VTE – in hospitalized patients with COVID-19. This information should help clinical decision and guidance with regards to thromboprophylaxis.
Methods
Search strategy and study eligibility
We performed this systematic review and meta-analysis in concordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (
additional table 1) [
7]. The review protocol was registered in PROSPERO International Prospective Register of Systematic Reviews (CRD42020193369).
We searched for original articles in English language, in which the risk of major VTE could be estimated among COVID-19 inpatients. As such, we excluded case reports or case series of VTE. Studies were assessed for eligibility, regardless of the sample size, population group/subgroup of COVID-19 inpatients, and study design (observational or interventional, retrospective or prospective). Studies were included if SARS-CoV2 was confirmed with a PCR or based on clinical and radiological examinations [
8].
Only peer-reviewed literature was appraised, published from January 1st, 2020 to June 15th, 2020, in Pubmed/MEDLINE, Embase, Web of science, and Cochrane. The electronic search was supplemented by a search of the Lancet COVID-19 Ressource Centre, JAMA COVID-19, the New England Journal of Medicine Coronavirus, the use of the “similar articles” function in Pubmed, and the screening of the references list of all relevant articles (additional file
1).
Outcomes
The primary outcome was the risk of major VTE, defined as any objectively confirmed PE and/or proximal DVT. Proximal DVT was defined as lower limb DVT occurring in the popliteal, femoral, iliac veins and/or inferior vena cava, regardless of the presence of a catheter, and confirmed by compression ultrasound (CUS) or CT phlebography. We therefore excluded distal DVT (below the popliteal trifurcation) and upper extremity DVT, the latter being almost always catheter-related and not associated with the same embolic potential as lower limb proximal DVT. PE had to be objectively confirmed by CTPA, ventilation/perfusion scintigraphy, the presence of a proximal DVT on lower limb CUS in a patient with clinical suspicion of PE, or a high clinical suspicion of PE with an acute right ventricular dilatation on echocardiography [
9].
Secondary analyses explored the separate risk of proximal DVT and of PE, as well as the risk of major VTE stratified by the presence or absence of VTE screening tests and by anticoagulation regimen.
Data extraction and quality assessment
The search, data extraction, and quality assessment were independently conducted in pairs by four authors (GL, AL, SMB and MB). Any disagreements were discussed until a consensus was reached, and solved by a third author (MR) whenever necessary.
Data were entered into a predesigned database, and included information on study design, setting, inclusion/exclusion criteria, patients characteristics, diagnostic methods for COVID-19, VTE diagnosis, VTE screening strategy, and the use of therapeutic anticoagulation or thromboprophylaxis. Whenever necessary, an email was sent to the corresponding author for additional details (of the 24 authors contacted, 15 answered).
The risk of bias was assessed using the Newcastle-Ottawa scale for non-randomized clinical trials (
additional table 2) [
10].
Statistical analysis
The primary analysis estimated the risk of major VTE among studies including all inpatients – general medical wards and critical care units – and among studies including only intensive care unit (ICU) patients. This was computed by dividing the number of major VTE by the number of patients at risk at the start of the study. We did not consider Kaplan-Meier estimates, given their overestimating bias in case of competing risk of death.
In the secondary analyses, we explored the separate risk of proximal DVT and of PE; the heterogeneity; the risk of major VTE with and without screening for DVT; the risks of VTE among patients receiving therapeutic anticoagulation, thromboprophylaxis and no anticoagulant treatment; and the risk of VTE among medical patients, by excluding ICU patients.
In a sensitivity analysis, we repeated the primary analysis in a restricted sample of high-quality studies defined as retrospective or prospective studies with an unselected group of medical and/or ICU patients, with a reported ≥80% of PCR positivity and a reported mean follow-up of at least 7 days.
All meta-analytic estimates were pooled using random effect models, because of expected substantial heterogeneity due to different sampling characteristics. Risks (proportions) observed in studies were transformed with the Freeman-Tukey double arcsin method to remove the dependence of the variance on the mean of the transformed proportion and to correct for overdispersion of estimates [
11]. Heterogeneity itself was measured by using the
I2, describing the proportion of the observed between-studies variation not due to chance. Small-study effects were explored graphically with funnel plots and with Egger’s tests.
All analyses were conducted in Microsoft Excel, Stata version 11 (including with the package metaprop) and R.
Discussion
In this systematic review and meta-analysis, we extracted data from 33 studies including a total of 4009 patients. Among patients with COVID-19 hospitalized in the medical ward and/or ICU, the incidence of major VTE was 9%. The incidence of proximal lower limb DVT was 3% and the incidence of PE 8%. In ICU patients, corresponding incidences were much higher: 21, 8, and 17%, respectively. When restricting the analysis to medical patients, VTE was only found in 2%.
Previous meta-analytic efforts may have overestimated the burden of clinically relevant VTE among hospitalized patients with COVID-19. Among 30 studies published until June 24th, 2020, Porfidia et al. [
44] reported an overall incidence of VTE of 26% (95%CI 1–75%), but upper extremity and distal lower limb DVT were also included. Nopp et al. [
45] recently published the largest meta-analysis to date, including 66 studies with a total of 28,193 patients. Similarly to our report, they estimated an overall VTE risk of 14.1% (95%CI 11.6–16.9%), however they did not distinguish distal/proximal, and upper/lower limb DVT. Other meta-analyses have been published, but did not restrict the events to objectively diagnosed VTE, with a potential for overestimation of the risk of VTE [
44‐
47].
The novelty of our analysis lies in the strict inclusion of objectively diagnosed clinically relevant VTE, restricted to proximal lower limb DVT and PE. Indeed, the clinical significance of distal DVT and need for anticoagulant treatment is highly debated, as its potential for embolism and recurrence is much lower than that of proximal DVT [
48], without demonstrated benefit of anticoagulation in the limited randomized trials [
49,
50]. The same rationale applies to catheter-related upper extremity DVT. Moreover, the diagnostic performance of CUS is lower for distal DVT than for proximal DVT, with a potential for false positive and negative findings, so that restricting the analysis to proximal DVT is likely to provide more robust data on the true DVT rate. We a priori decided to include incidental PE in the primary outcome, because the current evidence suggests that it carries the same VTE recurrence rate, and therefore clinical significance, as symptomatic PE [
51]. In any case, data on symptoms of VTE were limited, being reported by 9 studies [
14,
18,
20,
21,
32,
35,
37,
41,
43] for DVT, and 4 studies [
16,
35,
40,
43] for PE.
Even when using these strict criteria for defining major VTE, we found an elevated risk of VTE in critically-ill COVID-19 patients (21%), despite the use of pharmacological thromboprophylaxis in most studies. This is much greater than the anticipated risk of VTE in ICU patients (5–9%) inferred from previous studies of unselected ICU patients [
52,
53]. We hypothesize that, in addition to COVID-19 specific pathophysiological mechanisms such as the endothelial tropism [
1] and associated coagulopathy, extrinsic factors observed in these patients such as prolonged immobility and repetitive use of muscle relaxant drugs (curare) in COVID-19 patients with acute respiratory distress syndrome (ARDS), may contribute to this increased VTE rate compared to historical ICU cohorts. Among COVID-19 patients hospitalized in general medical wards, we found a 2% risk that is in line with previous estimates of VTE [
54], but this analysis excluded participants who became critically ill during their hospital stay, and likely does not apply to medical patients with severe COVID-19 at the time of admission.
The utility of VTE screening remains unknown among COVID-19 inpatients. While we found greater VTE risks in studies which included some screening, usually for DVT, we cannot infer from our results that screening is useful and should be implemented in routine practice, given the lack of comparison within individual studies. This deserves further research.
The findings of our meta-analysis call for interventional clinical trials in this setting and supports the research effort that has been launched since the beginning of the epidemic. In particular, 20 trials are comparing different levels of anticoagulation to prevent the burden of VTE and reduce the respiratory failure associated with ARDS [
55]. Our group has launched the ongoing Swiss multicentric COVID-HEP trial last April, which includes both ICU patients and medical patients deemed at poor prognosis due to elevated levels of D-dimer (
NCT04345848).
Strengths of this analysis include its wide search strategy, the independent selection of studies and abstraction of data, the contact with authors to request additional data and the inclusion of only VTE events with strict criteria on objective diagnosis and clinical relevance. We also acknowledge limitations. First, our search strategy was performed on June 15th, thus excluding the additional evidence published since then. However, this delay was inevitable in order to extract the outcome of interest, oftentimes with the help of multiple investigators. Second, as others, we observed a very important heterogeneity of all results, that was multifactorial. In particular, there was evident heterogeneity between study characteristics, with a mixture of inclusion criteria, use of screening and/or levels of thromboprophylaxis and different follow-up. Third, as this is a study-level meta-analysis, we were often restricted by the amount of data available in the published papers, or shared by investigators.
Acknowledgements
The authors thank Mafalda Burri for her contribution to the search strategy. Methodological support was provided by the Clinical Research Center, University of Geneva and Geneva University Hospitals (Christophe Combescure).
The authors thank Al-Samkari Hanny, Annunziata Anna, Artifoni Mathieu, Barco Stefano, Beyls Christophe, Contou Damien, Dagna Lorenzo, Faggiano Pompillio, Gianesello Lara, Grillet Franck, Helms Sophie, Mahjoub Yazine, Périard Daniel, Ruttens David, Susen Sophie, for their kind contribution of supplementary informations.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.