All adult patients with critical COVID-19 (verified by a polymerase chain reaction test) admitted to an ICU at three Swedish hospitals (Södersjukhuset and Karolinska University Hospital in Stockholm and Skåne University Hospital in Lund) between March 1, 2020, and May 30, 2021, treated with thromboprophylactic LMWH, were eligible. Patients were included on their first admission to the ICU. If a patient was transferred directly from one ICU to another, this was considered the same admission.

Patients were excluded if they had the outcomes of TE (with ongoing treatment) or major bleeding at ICU admission, as LMWH dosing could not be assumed to be prophylactic. Patients initially admitted to an ICU at other hospitals, where we did not have access to the electronic health record (EHR) or Patient Data Management System (PDMS), were excluded.

Patient data were manually and automatically extracted from the EHRs (Take Care®, CompuGroup Medical, Koblenz, Germany and Melior®, Cerner Corporation, Missouri, United States) and PDMSs (Clinisoft®, GE Healthcare, Illinois, United States and IntelliSpace Critical Care and Anaesthesia, Philips Medizin Systeme, Böblingen, Germany). The manual review was performed by five investigators guided by instructions and definitions of patient characteristics, chronic medication, comorbidities (international classification of diseases, 10th revision), treatment provided in the ICU and outcomes. If interpretation of the EHR was difficult, the investigator was instructed to discuss with a more experienced clinician involved in the study. With the use of programming software (QlikView®, 1993–2023, QlikTech International AB, Pennsylvania, United States), data on aFXa values were generated based on drug records with time stamps for administration and laboratory data with time stamps for blood sampling.

The study was approved by the Swedish Ethical Review Authority, registration number 2020–01302 with amendment 2020–02890 for patients included at Södersjukhuset and Karolinska University Hospital, and registration number 2014/916, amendment 2018/866 and 2020–06674 for patients included at Skåne University Hospital. The need for informed consent was waived. Approval for collecting patient data was given by the hospital jurisconsult, and record keeping was performed in accordance with the General Data Protection Regulation. The study was retrospectively registered at Clinicaltrials.gov, NCT05256524, February 24, 2022.

Patients were treated with doses and types of LMWH according to local guidelines (Supplement, Local treatment guidelines). For the majority of the study period, a twice daily regime was standard. Patients were grouped according to their initial dose when admitted to the ICU. Regimes of anticoagulation of LMWH were categorized as follows; low dose of LMWH: 2500-4500 international units (IU) daily for tinzaparin, 2500-5000 IU daily for dalteparin or ≤ 40 mg daily for enoxaparin; intermediate dose of LMWH: > 4500 IU but < 175 IU/kg of body weight daily for tinzaparin, > 5000 IU but < 200 IU/kg of body weight daily for dalteparin, or > 40 mg but < 2 mg/kg of body weight daily for enoxaparin; and high dose of LMWH: ≥ 175 IU/kg of body weight daily for tinzaparin, ≥ 200 IU/kg of body weight daily for dalteparin, or ≥ 2 mg/kg of body weight daily for enoxaparin. At Södersjukhuset, tinzaparin was used, at Karolinska University Hospital, dalteparin was used, and at Skåne University Hospital, enoxaparin was used. Patients who were transferred between hospitals in Stockholm may have been treated with both tinzaparin and dalteparin during their ICU stay. In this case, patients were categorized to the LMWH used when they had their valid aFXa values.

aFXa was analysed using standard local routines. Blood was sampled from venous punctures or arterial lines and collected in citrated tubes (3.2%, 0.109 mol/L). Samples were centrifuged within one hour. Platelet-free plasma was obtained by a two-step centrifugation protocol at 3000 × g for ten minutes in the Stockholm laboratories and by a single-step centrifugation at 2000 × g for 20 min in the Skåne laboratory. Analysis of aFXa was performed using the BiophenTM Heparin LRT reagent (Hyphen BioMed, Neuville-sur-Oise, France) on the Sysmex CS-5100 automated coagulation analyser (Sysmex Corporation, Kobe, Japan).

The first aFXa to be analysed for each patient was sampled after at least four doses of LMWH when a steady state was assumed. aFXa was sampled as both trough and peak values. At Södersjukhuset and Karolinska University Hospital, trough values were most commonly used, while at Skåne University Hospital, peak values were almost exclusively used. Peak values were defined as blood sampled at 3 (± 1) hours after the administration of subcutaneous LMWH, and trough values were defined as blood sampled at 12 (± 2) hours after the subcutaneous administration of LMWH according to the two-dose regimen. Values after a diagnosis of TE or major bleeding and after 28 days were excluded.

The Stockholm laboratories measured a range between 0.05 and 1.8 kIU/L; any values below 0.05 were set to 0.049, and values above 1.8 were set to 1.801. In the Skåne laboratory, the measuring range was between 0.1 and 2.4 kIU/L; values below 0.1 were set to 0.099, and values above 2.4 were set to 2.401.

Within the categories peak and trough, samples of aFXa were used to generate minimum, median and maximum values [21]. The minimum and maximum values could be values from any time during the ICU stay, however, the median was calculated using only values from the first 14 days. If a patient only had one value, this was classified as both minimum and maximum and as median if sampled during the first 14 ICU days. With this categorization, all patients had between two (minimum and maximum for either peak or trough) and six values (minimum, median and maximum for both peak and trough) to be analysed. The minimum and maximum values at any time during the ICU stay were chosen to decrease the risk of a lag between the value of aFXa and diagnosis hindering the identification of an association. The median value during the first 14 days was chosen because we hypothesized that LMWH activity early on during the ICU stay may be most important for the association with mortality.

The primary outcome was death within 90 days, and secondary outcomes were TE, bleeding and major bleeding within 28 days. Pulmonary embolism/thrombosis (PE/PT) and DVT were considered venous TE, and ischemic stroke was considered arterial TE. PE/PT and ischemic stroke were defined as a diagnosis verified by computed tomography, and DVT was defined as verified by ultrasound or computed tomography. The reason for broadening the outcome of PE to PE/PT is reports of occlusions of the pulmonary arteries without findings of thrombus in the veins of the lower extremities in COVID-19 patients. Since lower extremities are usually where the embolizing thrombus is formed, it has been speculated that COVID-19 patients, in addition to PE, also suffer in situ thrombosis in the lungs [22]. All investigations for TE were performed at the discretion of the treating clinician.

Bleeding events were categorized according to the World Health Organization bleeding scale, and grade 3 and 4 bleedings were considered major bleedings [23,24,25].

Continuous variables are presented as medians with interquartile ranges (IQRs). Categorical variables are expressed as numbers and proportions (percentage). Logistic regression was used to analyse aFXa as a continuous variable. aFXa was modelled as a spline function with three knots, as we made no assumption about the nature of the relationship between aFXa and the outcomes (linear vs. nonlinear). The Wald test was used to assess whether the change in risk of outcome was significantly associated with aFXa values in the spline model, with the null hypothesis that all of the spline coefficients were null.

Chi-square was used when analysing different cut-offs of aFXa and presented as odds ratios (OR) with corresponding 95% confidence intervals (CI).

When assessing for confounders, only renal function was identified to be potentially associated with both aFXa and the outcomes. Renal function was therefore added to the model. An estimated glomerular filtration rate (eGFR) based on sex, age and baseline creatinine was calculated according to the Chronic Kidney Disease Epidemiology Collaboration (CKD-Epi) equation [26]. The distribution of normality was tested using the Shapiro–Wilk test. The correlation between aFXa and eGFR was assessed with Spearman's correlation test. Three regression lines were added to visualize the risk for patients with eGFR 60, 90, and 120 ml/min/1.73m2. The Kruskal–Wallis test was used to compare differences between groups for continues variables and Dunn’s test for pairwise multiple comparisons. No method was used to adjust for multiple testing except when using Dunn’s test and adjustment was done by the Bonferroni method.

A two-sided p value < 0.05 was considered statistically significant. Analyses were performed in R v 4.2.2 (R Core, 2017. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria).