The manuscript was written following the STrengthening the Reporting of Observational studies in Epidemiology (STROBE) guidelines [19].

The Maastricht Intensive Care COVID (MaastrICCht) cohort is a prospective cohort of patients with confirmed COVID-19 admitted to the ICU of the Maastricht University Medical Centre (MUMC +). The design has been described extensively elsewhere [20] and includes comprehensive serial hemostasis and coagulation phenotyping [21, 22]. The local institutional review board (Medisch Ethische Toetsingscomissie (METC) 2020-1565/300523) of the MUMC + approved the study, which was performed based on the regulations of Helsinki. The study is registered in International Clinical Trials Registry Platform (NL8613). This study included all participants with respiratory insufficiency requiring mechanical ventilation and at least one real-time polymerase chain reaction (RT-PCR) positive for SARS-CoV-2 RNA and a chest CT scan strongly suggestive of SARS-CoV-2 infection, based on a CORADS-score of 4–5 scored by a radiologist [23, 24]. Participants were followed until they either died in the ICU or were discharged from ICU. A comprehensive and uniform set of clinical, physiological, and laboratory variables was collected daily, reducing the chance of missing data. In addition, when patients were not available for blood sampling or laboratory testing failed, the measurement would be rescheduled for the next blood withdrawal.

Variable collection on the ICU for COVID-19 was standardized as described extensively elsewhere [20]. Medical history of cardiovascular disease (defined as congestive heart failure, myocardial infarction, or peripheral vascular disease) was scored on ICU admission. APACHE-II score on ETI and SOFA score during ICU stay were calculated [14]. Coronary artery calcium (CAC) scores were measured within the MaastrICCht cohort, which was described in more detail elsewhere [25]. Patients were classified with or without a clinical PE as follows; in patients with a clinical suspected PE, computed tomography pulmonary angiography (CTPA) was used diagnostically. CTPA was performed in a supine position after intravenous injection of individually adapted contrast media volume (iopromide 300 mg iodine; Ultravist, Bayer Healthcare, Berlin, Germany) based on body weight and kVp settings on a second or third-generation dual source CT scanner (Somatom Definition Flash, Force; Siemens Healthineers, Forchheim Germany). The image quality of all CT scans was judged sufficient to evaluate the presence of PE or thrombosis (central, lobular, segmental, or sub-segmental) [26]. Patients in whom CTPA excluded PE were classified as not having clinical PE. The occurrence of deep venous thrombosis (DVT) diagnosed by ultrasound was recorded within the cohort, but was not considered as the majority of the patients underwent CTPA at ICU admission.

Information regarding the six months mortality after endotracheal intubation (ETI) was collected. This was done by identification of the last medical contact consisting of: a consultation in our hospital or hospitalisation, a visit to the emergency room, imaging diagnostics, surgery or the laboratory measurement of a blood sample drawn. When patients had died during the six month follow-up period, this information was collected. Patients who had been transferred to other hospitals, were followed-up by contacting the patients themselves or their general practitioners.

Enteral nutrition in all ICU admitted patients, who were suspected to be unable to ingest oral food within the first 48 h of ICU admission, has been started via naso-gastric tube. Fresubin 1200 (including 10 μg vitamin K /100ml) was the standard nutrition and was prescribed in a weight adjusted dose [27]. The adjusted feeding dose was calculated as follows: day 1; 5ml/kg/day, day 2; 10 ml/kg/day, day 3; 15ml/kg/day, and maximal dosage on day 4; 20ml/kg/day [28].

All ICU admitted patients received intermediate doses of thromboprophylaxis: Nadroparin 5700, 7600, and 11,400 IU for respectively < 70, 70–90, and > 90 kg [29]. Patients who required therapeutic anticoagulants before hospital admission were started on therapeutic low molecular weight heparin (LMWH) upon ICU admission. In addition, vitamin K antagonists and direct oral anticoagulants (DOACs) were switched to therapeutic LMWH. Patients on extracorporeal membrane oxygenation (ECMO) or continuous renal replacement therapy (CRRT) received unfractionated heparin (UFH), guided by guidelines based on aPTT (heparin therapeutic range (HTR) 50-80s) and anti-Xa (HTR 0.3 – 0.7 IU/mL) measurements [30].

Two-hundred and thirty-two patients were enrolled in the MaastrICCht cohort from March 25th 2020, until April 13th, 2021. We included patients in the present investigation who had a chest CT scan as part of standard care. The chest CT scan was introduced as standard of care in our hospital during the pandemic to rule out pulmonary embolism, at ICU admission, and was done in each patient. To rule out any selective information bias on coronary calcium, which is important within the pathophysiological framework under investigation, ninety-four patients enrolled early during the COVID-19 pandemic were excluded as they had no standard chest CT scan [31]. Of the total of 232 cohort patients, dp-ucMGP as therefore not measured in the initial 94 patients. Hundred and thirty-eight patients were enrolled in the MaastrICCht cohort, during the second COVID-19 wave, from September 26th, 2020, until April 13th, 2021. Of those hundred thirty eight, a hundred and twelve patients had serial citrate plasma stored to measure dp-ucMGP. No leftover citrate plasma was available in the remaining twenty-six patients, which were excluded (Fig. 1). Timing from ETI allows for a fairer comparison between vitamin K status and the disease course severity, where disease severity is defined as the need for mechanical ventilation in the ICU due to COVID-19. From September 29th onwards, additional dp-ucMGP assays were performed at Monday and Thursday in the morning, in leftover citrate plasma, for all included MaastrICCht cohort patients. Patients who were in the ICU before September 29th or were transported from another hospital after ETI were also included, starting dp-ucMGP measurements from admission from September 29th onwards. This means that the inclusion of patients could vary between the first till the fourth week after ETI. This design has been applied and described more extensively elsewhere [20].

Flowchart patient population. Wave 2 patients had standard CTPA on ICU admission. MaastrICCht Maastricht Intensive Care COVID, dp-ucMGP desphospho-uncarboxylated matrix Gla protein, *dp-ucMGP was measured to determine extra hepatic vitamin K status in a sub-cohort of the Maastricht Intensive Care COVID cohort

Daily arterial blood samples from all patients were collected from an arterial line in 7.2 mg K2 EDTA (4.0 mL), serum, or 3.2%(w/v) sodium citrate Vacutainer blood collection tubes (Becton Dickinson, Plymouth, UK). Platelet-poor plasma (PPP) was obtained using two subsequent centrifugation steps: initial centrifugation of 2490g for 5 min, followed by 10,000g for 10 min. Circulating dp-ucMGP levels were determined in citrate plasma using the commercially available IVD chemiluminescent InaKif MGP assay on the IDS-iSYS system (IDS, Boldon, United Kingdom) as previously described [32]. The within-run and total precision of this assay were 0.8–6.2% and 3.0–8.2%, respectively. The assay measuring range is between 200 and 12,000 pmol/L and was found to be linear up to 11,651 pmol/L. dp-ucMGP values < 400 pmol/L are in the normal healthy range and values > 400 pmol/L reflect vitamin K deficiency [17].

The data were analyzed with R version 3.6.1. As appropriate, the sample characteristics were described using median and interquartile range (IQR) or percentage. In addition, Pearson’s chi-square test, Mann Witney u test, or Kruskal–Wallis test were performed to compare characteristics. We described the association between serial dp-ucMGP measurements during ICU admission and two outcome variables. First, the cohort participants were categorized based on ICU survivors and ICU non-survivors. Then, we used linear mixed-effects regression with a random intercept and random slope for time to compute average differences in dp-ucMGP over time and differences in the slope over time between both groups. If the slope did not differ, average differences over time are reported only. We computed unadjusted group differences for dp-ucMGP (Model 1). In model 2, we adjusted model 1 for age, gender, and APACHE-II score. In model 3, we additionally adjusted model 2 for C-reactive protein (at ETI), creatinine (at ETI) and history of coumarin use. In addition, to investigate whether differences in dp-ucMGP between ICU survivors and ICU non-survivors were independent of pre-existing cardiovascular disease, we additionally adjusted model 2 for CAC-score (model 4) as more CAC reflects worse cardiovascular disease [31, 32]. Finally, we categorized patients based on the occurrence of PE (CTPA positive vs. CTPA negative) and repeated models 1–3 above. We report regression coefficients β with 95% confidence intervals (95% CI) and considered a p-value < 0.05 statistically significant. In addition, we analysed the associations between dp-ucMGP, per 100 pmol/L, and ICU mortality and six-months mortality, in crude (model 1) and adjusted models (adjusting for age, gender, APACHE II score and c-reactive protein, creatinine, and history of coumarin use (model 2)).