STUDY selection and CHARACTERISTICS
Electronic database searches identified 890 titles and abstracts (figure 1). After removal of duplicates, 531 papers remained. Reviewing titles and abstracts resulted in 158 potential articles to be included. After full-text reading, 33 studies6,12–43 were included in the systematic review and meta-analysis (substantial agreement: Cohen’s Kappa = 0.79). No additional articles were identified by other sources of screening or by reviewing the relevant reference lists.
28 studies13–27,29–32,34–41,43 were conducted in Europe, three12,28,33 in the USA, and two6,42 in Asia (table 1). 1912–15,17,21–23,26,28–31,33–35,37,38,40 were retrospective cohorts, 1216,18,20,24,25,27,32,36,39,41–43 were prospective cohorts, and two6,19 were cross-sectional studies. 196,13,15,16,21,23,27–30,32,33,36–41,43 reported VTE in the ICU, while 14 studies12,14,17–20,22,24–26,31,34,35,42 reported on a mixed population of medical ward and ICU patients.
The 33 studies included 4’009 patients, with a range of mean age of 57-69 years, and proportion of women of 13.8-48.3%. The prevalence of VTE risk factors varied widely: a personal history of VTE was reported in 3.0% of patients (range 0-7.6%), D-dimer levels ranged from 394 µg/L to 8’300 µg/L. Detailed characteristics of individual studies are found in the additional tables 3 and 4.
The Newcastle-Ottawa scale was used to assess risk of bias in each included study (additional table 2), excluding the items referring to the control group. Additionally, none of the studies demonstrated that the outcome of interest was not present at the start of the study. All studies met criteria for outcome assessment; consequently, the score ranged from one to 5 stars. 1212,20,22–25,28,30,34,36,38,41 and 86,15,18,19,26,39,42,43 studies were allocated three and two stars, respectively (table 1). 714,21,29,31–33,40 and 4 studies17,27,35,37 reached 4 and 5 stars, respectively; while the lowest score (one star) was obtained by two studies13,16.
Venous Thromboembolism incidence
In 14 studies12,14,17–20,22,24–26,31,34,35,42 in COVID-19 patients from medical ward ± ICU, the pooled estimate of major VTE (proximal lower limb DVT and/or PE) incidence was 9% (95%CI 5-13%, I2=92.5) (table 2, figure 2a). This rate was greater, at 21%, when restricting the analysis to patients hospitalized in the ICU (19 studies, 95%CI 14-28%, I2=87.6%, table 2, figure 2a). The funnel plot appeared asymmetric for the 14 studies from medical ward ± ICU (Egger test p=0.02), but appeared more symmetric for the 19 studies in the ICU (Egger test p=0.47, additional figures 1a-b). This suggested that small studies tended to report higher risk estimates than larger studies, and were potentially more prone to be published.
Proximal lower limb DVT confirmed by CUS was reported in 24 studies6,12,14,16,18–25,27,29,31,32,34–38,40–42; with a pooled estimated incidence of 3% (95%CI 1-5%) in the medical ward ± ICU patients, and 8% (95%CI 3-14%) when restricting to ICU patients (table 2, additional figure 2a).
PE was reported in 27 studies, as confirmed by CTPA12–17,21–35,37–40,42,43, or as a clinical suspicion of PE associated with a thrombus in the right atrium on echocardiography16,30,31. The estimated incidence of PE was also lower in the mix of medical and ICU patients (8%, 95%CI 4-13%) than in the ICU only patients (17%, 95%CI 11-25%) (table 2, additional figure 3a). From the 330 PE reported in studies with data on location, 241 (73%) were segmental or more proximal (30 (9%) central, 39 (12%) lobar, 100 (30%) segmental, 72 (22%) unspecified) and 56 (17%) were subsegmental. The localization of PE was missing from 6 studies24,28,30,31,33,42, accounting for 33 PE (10%).
We explored the heterogeneity found in all analyses, which emanated from different inclusion criteria, definition of VTE, duration of follow-up, screening, use of thromboprophylaxis. When examining outliers, Annunziata et al.13 reported an 81% risk of PE (95%CI 58-95%) in a small, highly selected sample of 21 ICU patients with disseminated intravascular coagulation. Beyls et al.16, another outlier, reported a high DVT incidence of 42% (95%CI 15-72%) in a small sample of 12 ICU patients under ECMO. Furthermore, estimated VTE incidence from Mazzaccaro et al.34 was 66%, based on a systematic CTPA and CUS screening of 32 patients admitted to medical ward. However, when excluding these three studies, the measured heterogeneity (I2) remained >80% in the primary analysis.
Screening for VTE was performed in 14 studies6,14,16,18–20,25,32,34–37,41,42 for DVT and three studies16,34,43 for PE. Overall, studies with implemented screening found higher risks of VTE than studies without screening. In the ICU, VTE rate was 23% vs. 19% respectively; and in combined medical ward and ICU, it was 11% vs. 5% respectively (table 3, figure 2b, additional figures 2b and 3b for DVT and PE separately).
Thrombotic events where further stratified by the type of anticoagulation during hospitalization, despite a large amount of missingness. In the absence of anticoagulation, with prophylactic and with therapeutic anticoagulation, proportions of VTE were 29.4%22,32 (5/17), 19.8%13,14,17,18,22,29,32,35–37,39,40,43 (155/781) and 7.1%22,29,32,35 (3/42), respectively (Fisher p=0.047) (additional table 5).
Risk among medical patients only
4 studies19,25,31,35 provided data to estimate the risk of VTE after exclusion of ICU patients. Among 531 medical patients, the meta-analytic risk of VTE was 2% (95%CI 0-6%) (additional figure 4).
In 8 high-quality studies17,21,27,31,32,35,37,40, the VTE incidence was 15% (95%CI 9-23%) in the ICU only versus 7% (95%CI 2-17%) in the medical ward ± ICU (additional figure 5a). Screening was also associated with a greater incidence: 14% versus 3% in the medical ward ± ICU, and 23% versus 12% in the ICU only, with and without screening; respectively (additional figure 5b).