Trial overview and patients
In this retrospective, observational cohort study, all critically ill COVID-19 patients (verified with polymerase chain reaction-positive Severe Acute Respiratory Syndrome Corona Virus 2, SARS-CoV-2) with respiratory failure, admitted to two ICUs in March and April, 2020, at Södersjukhuset, Stockholm, Sweden, were eligible for inclusion. The study was approved by the regional ethical review board in Uppsala, Sweden, (Dnr: 2020-01302, amendment 2020-02890), and informed consent was waived.
Patients were excluded if discharged the same day as ICU admission, if they had ongoing anticoagulant (AC) therapy prior to ICU due to deep venous thrombosis, DVT, and/or PE, or if they had no initial treatment with thromboprophylaxis in the ICU. Patients with chronic AC therapy at hospital admission, for other reasons than DVT and/or PE, were included in the study.
Data on patients’ demography, comorbidities (International classifications of diseases, 10th revision), duration of symptoms, chronic AC therapy, invasive respiratory support, and laboratory values were retrieved from patients’ medical records. Data were automatically and manually extracted by medical doctors and all charts and events were validated by at least one additional medical doctor.
Dosing strategies of thromboprophylaxis
Patients were categorized into three groups according to initial treatment doses of subcutaneous low molecular weight heparin (LMWH) at admission to the ICU. Two different LMWHs, tinzaparin and dalteparin, were used. Tinzaparin and dalteparin are not considered fully interchangeable due to lack of studies establishing equipotent dosing [17]. Dosing strategies were therefore made by classifying the dose according to the recommended dose for the specific indications as recommended by the trade association for the research-based pharmaceutical industry in Sweden in their reference catalogue FASS.se [18,19]. Daily doses of tinzaparin and dalteparin were defined as low-dose thromboprophylaxis (2500–4500 international units, IU, tinzaparin or 2500–5000 IU dalteparin), medium-dose thromboprophylaxis (> 4500 IU but < 175 IU/kg of body weight tinzaparin or > 5000 IU but < 200 IU/kg of body weight dalteparin), and high-dose thromboprophylaxis (≥ 175 IU/kg of body weight tinzaparin or ≥ 200 IU/kg of body weight dalteparin). Patients who received an adjusted dose due to reduced kidney function were classified according to intended dose range. The low dose is the standard thromboprophylaxis used in ICU patients. The medium dose emerged during the pandemic period as the standard thromboprophylaxis dose given twice daily instead of once daily. High dose is the dosage used to treat patients with diagnosed thromboembolic disease [20].
The choice of dosing strategy followed the local recommendations and were modified over time (Additional file 1): In March, low-dose thromboprophylaxis was recommended for all COVID-19 patients at both ICUs. In April, the recommendations were altered to medium-dose and then to high-dose thromboprophylaxis, which was continued throughout the study period in one ICU. In the other ICU, full dose was only used for one week, and then recommendations were altered to medium-dose thromboprophylaxis again. All changes in doses were registered with new dose and date.
Outcomes
The primary outcome was 28-day mortality. Days alive and out of ICU at day 28, the cumulative proportion of thromboembolic and bleeding events within 28 days of ICU admission, and maximum levels of Fibrin-D-dimer were used as secondary outcome measures. Thromboembolic events were PE (verified by computed tomography or by clinical suspicion of PE as cause of deterioration combined with findings of acute strain of the right heart on echocardiography), DVT (verified by ultrasound), ischemic stroke (verified by computed tomography), and peripheral arterial embolism (clinical findings of acute peripheral ischemia). Bleeding events were categorized according to the World health organization (WHO) bleeding scale [21,22,23]: (1) petechiae, tissue hematoma, oropharyngeal bleeding, (2) mild blood loss, hematemesis, macroscopic hematuria, hemoptysis, joint bleeding, (3) gross blood loss requiring red blood cell transfusion and/or hemodynamic instability, (4) debilitating blood loss, severe hemodynamic instability, fatal bleeding, or central nervous system bleeding.
Baseline laboratory values were obtained from 6 h before to one hour after ICU admission.
Statistical analysis
Continuous values for baseline and follow-up data are presented in medians with interquartile range (IQR), while categorical or binary data are shown as numbers and proportions. Differences over categories of the exposure were tested with Kruskal–Wallis for continuous data, and Fisher’s exact for categorical data. In the survival analyses, participants could accrue follow-up time from date of ICU-admission, to date of death, or when 28 days had passed since admission, whichever occurred first. In analyses of thromboembolic and bleeding events, the date of that event also led to censoring of follow-up time. Kaplan–Meier curves were used to estimate the cumulative risk of death, thromboembolic event, and bleeding event, and the log-rank test was used to compare the initial dosing strategies. Cox proportional hazards regression was used to estimate hazard ratios (HR) with corresponding 95% confidence intervals (CI) of death within 28 days from ICU admission. Multivariable models were adjusted for sex, age, body mass index (BMI), Simplified Acute Physiology Score III (SAPS III), invasive respiratory support (yes/no), and initial dosing strategy of thromboprophylaxis (low-, medium-, and high-dose thromboprophylaxis) [24,25]. To assess evidence of nonlinearity, the second spline transformation equal to zero was tested as the quantitative covariates were modeled with restricted cubic splines at three knots at fixed percentiles (10th, 50th and 90th) of the distribution of that covariate [26]. As there was no such evidence for age (p = 0.26), or SAPS III (p = 0.71), those variables were adjusted for in a continuous fashion. Although no formal evidence, there was an indication of nonlinearity between levels of BMI and 28-day mortality (p = 0.08), why BMI was categorized as </≥ 30 kg/m2 with a separate category for missing values (n = 6). BMI was flexibly adjusted with restricted cubic splines while missing values were accounted for using chained iterations of multiple imputed data sets (n = 20) [27]. Although no formal evidence, there was an indication of violation of the assumption of proportional hazards when scaled Schoenfeld residuals were regressed against survival time (p = 0.06 for high-dose thromboprophylaxis). Thus, the time-varying effect was fitted by splitting the follow-up time at 7 days from ICU-admission, and fitting the time varying covariate as an interaction term with the main exposure. Statistical significance of interaction was tested using the Wald test. We have conducted sensitivity analyses to investigate the effect of inclusion time, treatment with glucocorticoids and changes in doses of thromboprophylaxis on the result. Two-sided p < 0.05 was considered statistically significant. Analysis was performed using STATA 13.1 (StataCorp), and R v 3.5.1 (R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.).