Associations Between Antithrombosis and Ventilator-Associated Events, ICU Stays, and Mortality Among Mechanically Ventilated Patients: A Registry-Based Cohort Study

Background: The effect of thromboembolism prophylaxis on clinical outcomes, such as ventilator-associated events (VAEs), ICU stays, and mortality, remains controversial. This study was conducted to evaluate the effect of pharmacological thromboprophylaxis on VAEs, ICU stays, and ICU mortality among patients receiving mechanical ventilation (MV). Materials and Methods: A retrospective cohort study was conducted based on a well-established registry of healthcare-associated infections at ICUs in the West China Hospital system. Patients who consistently received MV for at least 4 days from 1 April 2015 to 31 December 2018 were included. Hazard ratios (HRs) were compared for three tiers of VAEs, ICU stays, and ICU mortality among patients receiving pharmacological thromboprophylaxis versus those without using the time-dependent Cox model. For the analyses of ICU stays and ICU mortality, we also used Fine-Gray models to disentangle the competing risks and outcomes of interest. Results: Overall, 6,140 patients were included. Of these, 3,805 received at least one prescription of antithrombosis agents. Treatments with antithrombosis agents were associated with lower risk of VAEs (HR: 0.87, 95% CI: 0.77, 0.98) and ICU mortality (HR: 0.72, 95% CI: 0.61, 0.86) than those without. Anticoagulants but not antiplatelet agents were associated with decreased risk of VAEs (HR: 0.86, 95% CI: 0.75, 0.98), ICU mortality (HR: 0.62, 95% CI: 0.51, 0.76), and less time to ICU discharge (HR: 1.15, 95% CI: 1.04, 1.28). Antithrombosis may be associated with decreased risk of VAEs in patients with D-dimer >5 mg/LFEU (HR: 0.84, 95%CI: 0.72, 0.98). Conclusions: Pharmacological thromboprophylaxis was associated with lower risk of VAEs and ICU mortality. Similar effects were observed between unfractionated heparins versus low-molecular-weight heparins.

Conclusions: Pharmacological thromboprophylaxis was associated with lower risk of VAEs and ICU mortality. Similar effects were observed between unfractionated heparins versus low-molecular-weight heparins.
Keywords: antithrombosis prophylaxis, ventilator-associated events, ventilator-associated pneumonia, ICU mortality, patients receiving mechanical ventilation BACKGROUND Ventilator-associated event (VAE) and ventilator-associated pneumonia (VAP) are important causes of morbidity and mortality in the ICU (Bouadma et al., 2015;Magill et al., 2016;He et al., 2021;Zhu et al., 2021). To improve clinical outcomes of patients receiving mechanical ventilation, ventilator bundles have been implemented by hospitals worldwide. Although ventilator bundles have been implemented by hospitals worldwide to prevent adverse outcomes (Marra et al., 2009;Croce et al., 2013), the components often vary remarkably, which may be associated with adverse outcomes.
However, the effects of thromboembolism prophylaxis on other adverse outcomes such as VAE and mortality have not been fully understood (Al Yami et al., 2017;Bajaj et al., 2019). The benefit of reducing VTE should be reappraised given the increased risk of adverse outcomes, such as major bleeding. Especially for patients with MV more than 48 h, the risk of major bleeding significantly increased (Goodwin and Hoffman, 2011), which may further result in poor prognosis including infection and mortality. Few studies addressed these important outcomes, and these studies draw inconsistent conclusions (Barrera et al., 2010;Klemen et al., 2020). An observational study involved 630 patients with at least 2 days of MV, of which 210 patients developed ventilator-associated pneumonia (VAP). The findings of this study have showed that thromboprophylaxis was significantly associated with lower risk of VAP and death (Croce et al., 2013). Several other studies reported no association between pharmacological thromboprophylaxis on VAP and VAE (Berenholtz et al., 2011;Lewis et al., 2014;Klompas et al., 2016). Most of these studies had relatively small sample size and only a few events occurred, which may have led to a limited inference. For instance, a case-control study only included 110 ventilator-associated condition (VAC) and 38 infection-related ventilator-associated complication (IVAC) to investigate the risk factors for VACs and IVACs, respectively (Lewis et al., 2014). In addition, most studies have exclusively considered implementing aggregate bundle components together, rather than considering day-to-day implementation of bundles (Croce et al., 2013;Harris et al., 2018). Indeed, the implementation of bundle components varied on a day-to-day basis (Klompas et al., 2016).
Therefore, we conducted a cohort study with a large sample size and handled with time-dependent variates to evaluate the effect of prophylactic antithrombosis agents on clinical outcomes.

MATERIALS AND METHODS
This cohort study was reported according to the Reporting of studies Conducted using Observational Routinely collected health Data (Benchimol et al., 2015). This study was approved by the Ethics Committee of West China Hospital (WCH) in 2018 (WCH2018-409).

Data Source
This study was conducted based on a registry of healthcareassociated infections (HAI) at ICUs in WCH system. The WCH is a major healthcare system in west China, which comprises three independent healthcare organizations (Main Campus, Wenjiang Campus, and Shangjin Campus). A total of 5.73 million outpatients and 279,000 inpatients visited to WCH in 2019. As a national critical care center in west China, the WCH has six ICUs (general ICU, surgical ICU, neurological ICU, respiratory ICU, thoracic surgery ICU, and pediatric ICU) with more than 8,000 inpatient ICU admissions annually.
The ICU-HAI registry contained three databases-ICU system, electronic medical record (EMR), and ICU-HAI system. The ICU-HAI system is an active surveillance system, which is a unique system undertaking routine VAE surveillance in China. Information regarding catheterization, hospitalacquired infection, prevention, and control are collected by a team of experienced infection control practitioners every day. Every year, there were more than 5,000 patients undertake VAE surveillance. Of these, 2,000 patients received MV for at least four consecutive day and 500 patients were judged as VAE cases annually. By integrating these three databases with unique and encoded identifiers, we developed the ICU-HAI registry (Li et al., 2018;He et al., 2021).
Patients who were admitted to the six ICUs in WCH since 1 April 2015 were included into the registry. The registry included 28,848 patients until 31 December 2018 and contained 110 GB of data with 245, 311, 294 original records (Wang et al., 2019). Quality assessment showed that the accuracy of data extraction and linkage was 100%, and the completeness of important laboratory tests such as routine blood tests, serum glucose, and serum creatinine was >98% (Wang et al., 2019).

Study Population
Patients who consistently received MV for at least 4 days from 1 April 2015 to 31 December 2018 were included in the study. The exclusion criteria were as follows: age <18 years, patients admitted to PICU, without sufficient information regarding age, sex, and diagnosis at discharge, and with a diagnosis related to venous thrombosis at ICU admission. The clinical characteristics differed among non-VAE cases with and without consecutive stable or improved respiratory status. Therefore, we also excluded patients with consecutive unstable or increasing daily minimum positive end-expiratory pressure (PEEP) or fraction of inspired oxygen (FiO2) during MV treatment. To minimize indication bias, we additionally excluded patients with extremely long ICU stays (>90 days) and patients with more than one episode of MV treatment for the analysis of ICU mortality and ICU stays.

Antithrombosis Agent Exposure
We defined antithrombosis agent exposure as a prescription for antiplatelet or anticoagulant agents. We calculated antithrombosis agent exposure as a time-dependent variable. Daily exposure to antiplatelet agents or anticoagulant agents were coded as prescribed or not prescribed from initiation of MV to VAE occurrence or extubation and from ICU admission to ICU discharge, respectively. We additionally evaluated the type of antithrombosis agents (antiplatelet and anticoagulant agent) and type of anticoagulant agents [unfractionated heparin (UFH) and low molecular weight heparin (LMWH)] used.

Clinical Outcomes
The clinical outcomes of interest included VAEs, ICU mortality, and ICU stays. We measured clinical outcomes as time-to-event variables. We defined VAEs as at least 2 calendar days of increased daily minimum FiO 2 (≥0.20) or PEEP (≥3 cm H 2 O) greater than after at least 2 calendar days of stable or decreasing daily minimum FiO 2 or PEEP according to the criteria proposed by CDC's National Healthcare Safety Network (CDC-NHSN) (Klompas, 2013). Once the value reached the threshold, an alarm would be triggered. Then, infection control practitioners would judge and classify the suspected VAE cases as VACs, IVACs, and possible ventilator-associated pneumonia (PVAP). The accuracy of PVAP has been validated in a previous study, and the proportion was 96.2% (Wang et al., 2019). Patients died within one calendar day after ICU discharge were defined as ICU mortality.

Statistical Analysis
We assessed hazards ratio (HR) with confidence interval (CI) for antithrombosis agents and risk of VAEs by using time-dependent Cox model. This model is used to analyze studies with complex time-varying variates, and data were converted as counting process form to deal with time-varying variates. We assessed the impact of antithrombosis agents on ICU mortality and ICU stays using Fine-Gray competing risks model to measure the competing risks for ICU discharge alive versus ICU mortality. The reason for ICU discharge was depended on patient's health condition: clinical improvement or death. Through generating separate hazard ratios for each competing events, Fine-Gray competing risks model can disentangle effects of competing risks and outcomes of interest (Fine and Gray, 1999;Berger et al., 2018).
All analyses were adjusted for fixed and time-varying covariates. We defined ICU type, demographic characteristics, comorbidities or acute condition (diabetes, hypertension, heart failure, kidney failure, liver failure, ischemic heart disease, cerebrovascular diseases, chronic obstructive pulmonary disease, pulmonary vascular diseases, malignant tumor, trauma, acute respiratory distress syndrome, shock, gastrointestinal bleeding, pneumonia, and intra-abdominal infection), acute physiology and chronic health evaluation (APACHE) Ⅱ score, surgery, fiberoptic bronchoscopy examination, tracheotomy, and laboratory test at admission (D-dimer, prothrombin time, platelet count, antithrombin III, activated partial thromboplastin time) as fixed covariates. Daily medication exposure and processes of care (sedative, acid inhibitors, blood transfusion, mandatory ventilation, and headof-bed elevation, gastrointestinal decompression, and rehabilitation exercise) and medications (sedatives, opioids, neuromuscular blockers, immunosuppressive agent, neuroleptic agents, antibiotics, expectorants, vasopressors, intestinal probiotics, and neuroleptic agents) were defined as time-varying variables. Time-varying variables were measured as daily exposure from initiation of MV to the event of interested for the model of VAEs and each day from ICU admission to ICU discharge for the model of ICU discharge and ICU mortality. For the analysis of ICU discharge and ICU mortality, we additionally adjusted VAE, days from ICU admission to initiation of MV, and the duration of mechanical ventilation. We also measured mandatory ventilation and prone position ventilation as fixedtime variables.
To evaluate the effect of different types of antithrombosis agents on clinical outcomes, we further calculated HRs for antiplatelet agents vs. regimens without antithrombosis agents, anticoagulant agents vs. regimens without antithrombosis agents, antiplatelet agents vs. anticoagulant agents, and UFH vs. LMWH. We also calculated HRs regarding VAEs, ICU mortality, and ICU stays for patients with D-dimer ≤5 mg/L and >5 mg/L, respectively. The HRs for VAEs were assessed using timedependent Cox model. We also used Fine-Gray competing risks model to estimate HRs for ICU mortality and length of ICU stay. Missing data of APACHE II, prothrombin time, D-dimer, platelet count, and lipid load were processed using multilevel multiple imputation. Through pooling results from different imputed datasets, multiple imputations can help reduce bias and obtain more precise results compared with complete-case analysis.
Significance is 0.05 for all analysis, and all data were analyzed using R version 4.0.3, the packages used mainly included "survival," "miceadds," "mice."

Sensitivity Analyses and Additional Analyses
We conducted the following sensitivity analyses to examine the robustness of effect estimates: 1) alternative statistical models: without adjusting prothrombin time, platelet count, antithrombin III, activated partial thromboplastin time; 2) alternative approach for missing data: missing data without imputation; and 3) alternative definition of comparison: regimens without anticoagulant agents and regimens without antiplatelet agents.
To further evaluate the effect of antithrombosis agents on VTE, we conducted an additional analysis. We excluded patients who developed VTE within 3 days after ICU admission for latency purpose and to minimize reverse causality. We assessed HRs for antithrombosis agents and risk of VTE using time-dependent Cox model.

RESULTS
Our study included 6,140 patients consistently received MV for at least 4 days. Of these, 5,679 patients with ICU stay less than 90 days and one episode of MV treatment was additionally included into the cohort for ICU stays and ICU mortality analysis ( Figure 1). Of included 6,140 patients, 3,805 received at least one prescription of antithrombosis agents and 2,335 did not receive any prescription of antithrombosis agents. Patients' characteristics are showed in Table1. The median age of all included patients was 58 years, and female patients accounted for 36.8% of subjects among included patients. The distribution of patients with and without antithrombosis agents treatment in different ICU wards showed a significant difference (p < 0.001). For instance, 29.2 and 38.7% patients with and without antithrombosis agents treatment, respectively, were admitted to the general ICU. The median APACHEII score was lower among patients treated with antithrombosis agents than those without [19 (15, 24) vs. 21 (16, 26), p < 0.001]. Among all included patients, the most common comorbidity was hypertension (21.5%) and the most common acute condition was pneumonia (10.2%). Compared to patients without antithrombosis agents treatment, the proportions of patients
Compared to regimens without antithrombosis agent, no statistically significant decrease in IVACs and PVAPs was found for antithrombotic, anticoagulation, and antiplatelet agent. There were no statistically differences regarding the comparisons of antiplatelet agents vs. regimens without antithrombosis agents, anticoagulation agents vs. antiplatelet agents, and LMWH vs. UFH (Table 3).

Associations Between Antithrombosis Agents and the Length of ICU Stay and ICU Mortality
Adjusted HRs for antithrombosis agents and ICU mortality and length of stay are summarized in Table 4. Compared to regimens without antithrombosis agent, antithrombosis agent (HR: 0.72, 95% CI: 0.61, 0.86) and anticoagulant (HR: 0.62, 95%CI: 0.51, 0.76) agents were associated with lower risk of ICU mortality. Regimens with anticoagulant agents were associated an increased hazard for ICU discharge (HR: 1.15, 95%CI: 1.04, 1.27) than those without antithrombosis agent, suggesting that anticoagulant agents were associated with shorter time of ICU stays. Anticoagulant agents were associated with decreased risk of ICU mortality (HR:0.52, 95%CI: 0.37, 0.74) and less time to ICU discharge (HR:1.43, 95%CI: 1.18, 1.73) relative to antiplatelets agent. No statistically differences were found between LMWH and UFH regarding hazards for ICU mortality and ICU stays ( Table 4).
Sensitivity analyses using alternative statistical models did not show change in interpretation. The result of complete cases analysis was consistent with primary analysis for the comparison of regimens with and without antithrombosis agents regarding hazards for VAEs, but with wider confidence intervals that crossed one (HR 0.88, 95% CI: 0.78, 1.01). The sensitivity analyses of ICU stays using alternative definition of comparison and alternative approach for missing data showed similar results, but the confidence interval less than one regarding the comparison of regimens with and without antiplatelet agents (Supplementary Table S2).

DISCUSSION
The results of this large observational study showed that pharmacological thromboprophylaxis was commonly administered in ICU patients on MV. Pharmacological thromboprophylaxis was associated with lower risk of VAEs and ICU mortality. The effects of different type of pharmacological thromboprophylaxis on clinical outcomes may differ: treatment of antithrombotic, anticoagulant, but not antiplatelet appear beneficial. Antithrombotic may be associated with decreased risk of VAEs among patients with D-dimer >5 mg/LFEU; however, the effect was not statistically significant among patients with D-dimer ≤5 mg/LFEU. Similar to ours, several other studies have suggested that thrombosis prophylaxis may decrease the risk of adverse UFH: unfractionated heparin; LWMH: low molecular weight heparin. Model adjusted for: age, sex, acute physiology and chronic health evaluation (APACHE) Ⅱ score, ICU type (general, surgical, neurological, respiratory, thoracic surgery and pediatric ICU), comorbidities or condition (diabetes, hypertension, heart failure, kidney failure, liver failure, ischemic heart disease, cerebrovascular diseases, chronic obstructive pulmonary disease, pulmonary vascular diseases, malignant tumor, trauma, acute respiratory distress syndrome, shock, gastrointestinal bleeding, pneumonia, and intra-abdominal infection), cardiac surgery, cranial surgery, mandatory ventilation, prone position ventilation, fiberoptic bronchoscopy examination, tracheotomy, laboratory test at admission (D-dimer, prothrombin time, platelet count, antithrombin III, activated partial thromboplastin time), daily medication exposure and processes of care (sedative, acid inhibitors, blood transfusion, head-of-bed elevation, gastrointestinal decompression, rehabilitation exercise) and medications (sedatives, opioids, neuromuscular blockers, immunosuppressive agent, neuroleptic agents, antibiotics, expectorants, vasopressors, intestinal probiotics and neuroleptic agents), VAE, days from ICU admission to initiation of MV and duration of mechanical ventilation.
Frontiers in Pharmacology | www.frontiersin.org July 2022 | Volume 13 | Article 891178 outcomes. A multicenter prospective observational study that included 630 patients receiving MV for at least 48 h showed that DVT prophylaxis was significantly associated with lower incidences of VAP (Croce et al., 2013). Another study on 175,665 ICU patients found that omission of early thromboprophylaxis may increase the risk of mortality for patients with critical illness (Ho et al., 2011). A possible reason may be that pharmacological thromboprophylaxis could decrease the risk of VTE, which may further result in adverse outcomes, such as respiratory deterioration, prolonged MV, and mortality. Two systematic review and meta-analyses showed significantly lower risk of DVT and PE in patients receiving thromboprophylaxis than those without (Alhazzani et al., 2013;Park et al., 2016). The inhospital mortality was 50% higher among patients with DVT than those without (Jain and Schmidt, 1997;Ejaz et al., 2018), and PE was attributable to 4-11% cause of deaths (Laporte and Mismetti, 2010). However, several other studies have not found a significant association between thromboprophylaxis and adverse outcomes. A case-control study including 110 VAE cases suggested that thrombosis prophylaxis did not reduce the risk of VAEs and IVACs (Lewis et al., 2014). Klompas et al. conducted a retrospective cohort study and showed that thromboembolism prophylaxis was not associated with decreased risk of VAEs, VAP, hospital stays, and mortality (Klompas et al., 2016). There are several potential reasons for these apparent inconsistencies in results. One reason may be due to the varied outcome definitions, especially for VAPs. For instance, Klompas et al. defined (Croce et al., 2013;Klompas et al., 2016). The correlation between VAE and VAP has been proved to be poor (Klompas, 2019). Secondly, the strategies for thromboembolism prophylaxis varied among studies and individual patients. Thromboprophylaxis included pharmacological and mechanical prophylaxis, and the effect of these two measures on preventing adverse events may differ (Park et al., 2016;Boddi and Peris, 2017). A meta-analysis including 12 trials suggested that pharmacological prophylaxis with UFH and LMWH significantly decreased the incidence of DVT. Mechanical thromboprophylaxis, however, was not associated with a significant reduction of DVT risk (Park et al., 2016). Although most previous studies involved both pharmacological and mechanical prophylaxis, the detailed measures and compliance with thromboprophylaxis among studies likely varied (Manoucheri and Fallahi, 2015;Ejaz et al., 2018). In addition, these studies involved patients with different clinical features, such as coagulation function. The effect of thromboprophylaxis on clinical outcomes may differ among patients with different coagulation function (Goodwin and Hoffman, 2011). Patients with hypercoagulability may more likely to benefit from thromboprophylaxis. Our study also showed that antithrombotic was associated with lower risk VAEs among patients with elevated D-dimer.

Strengths and Limitations
This study has several strengths. First, our study was conducted based on a multisource database, which contained complete and accurate information regarding HAI in the ICU setting. We included a relatively large number of patients in addition to identifying a large number of VAEs. Validation of outcomes showed a high level of accuracy. Second, we considered time-varying terms for daily pharmacological thromboprophylaxis exposures and used competing risk models to handling the competing risks for ICU discharge alive versus ICU mortality.
Our study also has some limitations. First, the results should be interpreted with caution given the nature of the retrospective observational study. Although we adjusted an extensive array of factors, unmeasured residual confounding factors may still be present. Second, data for some variables were missing. Third, this study was conducted using data from a homogeneous healthcare system, and the findings of our study may not be generalizable to other healthcare settings. Fourthly, only few patients developed VTE, and the inference on the effect of VTE is weakened.

CONCLUSION
Antithrombosis agents were associated with a lower risk of VAEs and ICU mortality than regimens without antithrombosis agents. Anticoagulation but not antiplatelet agents appeared beneficial. The effects appeared comparable on comparing UFH vs. LMWH. Large, rigorous, randomized trials are needed to validate these results.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, and further inquiries can be directed to the corresponding authors.

ETHICS STATEMENT
This study was approved by the Ethical Committee of West China Hospital in 2018 (WCH2018-409), and the need for patient consent was waived.

AUTHOR CONTRIBUTIONS
XS and ZZ conceived and designed the research and assisted in revision of manuscript; MW cleared and analyzed research data and wrote the initial manuscript; WW assisted in the design of the research, interpreted research results, and revised the manuscript; QH, SZ and XJ collected research data and assisted in the interpretation of research results; YK collected the research data; RZ assisted in acquisition and clean data; LL and YR assisted the statistical analysis; and KZ assisted in the interpretation of research results. All authors read and approved the final manuscript.