Abstract
Background
Patients with pulmonary embolism (PE) and chronic kidney disease (CKD) have a fragile hemostatic balance, yet evidence comparing bleeding risk between direct oral anticoagulants (DOACs) and warfarin in critically ill patients remains limited. We compared overall and major bleeding risks associated with DOACs versus warfarin in ICU patients with PE and CKD.
Methods
We conducted a retrospective cohort study using the MIMIC-IV (v3.1) database. Adult ICU patients with PE and an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m2 treated with DOACs or warfarin were included. Propensity score matching (PSM) was used as the primary adjustment strategy, with stabilized inverse probability of treatment weighting (IPTW) as sensitivity analysis. Cox proportional hazards models were applied. The primary outcome was any bleeding, and the secondary outcome was major bleeding.
Results
Among 1,363 patients, 832 received warfarin and 531 received DOACs. After PSM, 1,038 patients were well balanced. In IPTW-weighted analyses, DOAC use was associated with a higher risk of any bleeding compared with warfarin (HR 1.37, 95% CI 1.13–1.66; P = 0.001), while no significant difference was observed for major bleeding (HR 0.97, 95% CI 0.74–1.28; P = 0.84). Results were consistent in the PSM cohort. Renal function did not significantly modify the relative bleeding risk.
Conclusions
In critically ill patients with PE and CKD, DOACs were associated with increased overall bleeding but not major bleeding compared with warfarin. Differentiating non-major from major bleeding is essential when selecting anticoagulant therapy in this population.
Keywords
1. Introduction
Chronic kidney disease (CKD) is defined as persistent abnormalities in kidney structure or function lasting for at least three months and is associated with substantial adverse health outcomes. 1 Patients with CKD exhibit a complex and paradoxical hemostatic state characterized by both prothrombotic and hemorrhagic tendencies, driven by hemodynamic alterations, endothelial dysfunction, abnormalities in coagulation factors, chronic inflammation, platelet dysfunction, and the accumulation of uremic toxins. 2 Owing to these pathophysiological changes, the incidence of venous thromboembolism, including pulmonary embolism (PE), is significantly higher among patients with CKD compared with the general population. 3
Anticoagulation remains the cornerstone of PE management. Current treatment strategies typically involve initial parenteral anticoagulation with heparin followed by long-term oral anticoagulant therapy, most commonly direct oral anticoagulants (DOACs) or vitamin K antagonists (VKAs), in hemodynamically stable patients.4,5 In patients with impaired renal function, however, the choice of oral anticoagulant is particularly challenging. Renal clearance plays a critical role in the pharmacokinetics of DOACs, raising concerns regarding drug accumulation and bleeding risk in patients with CKD, while VKAs require frequent monitoring and are associated with substantial interindividual variability.6,7
The 2019 European Society of Cardiology guidelines for the diagnosis and management of PE acknowledge VKAs as a long-established option for oral anticoagulation but do not provide clear recommendations favoring either DOACs or VKAs in patients with PE and CKD. 4 Epidemiological data suggest that approximately one-third of patients with PE have concomitant CKD. 8 Despite this substantial overlap, evidence guiding anticoagulant selection in this high-risk population remains limited. Although numerous studies have compared the safety of DOACs and warfarin, most have focused on patients with atrial fibrillation, and patients with advanced CKD or critical illness are frequently underrepresented or excluded.9-12
Importantly, patients with PE and CKD admitted to the intensive care unit (ICU) represent a particularly vulnerable subgroup, in whom bleeding complications may lead to interruption of anticoagulation, prolonged hospitalization, and adverse outcomes. However, real-world data evaluating bleeding risk associated with different oral anticoagulants in this specific population are scarce.
Therefore, the present study aimed to evaluate the bleeding safety of DOACs compared with warfarin in patients with PE and CKD admitted to the ICU. Using a large critical care database and contemporary propensity score–based methods, we focused on the occurrence of bleeding events across multiple organ systems, with particular attention to potential effect modification by renal function.
2. Methods
2.1. Data Source
This retrospective observational study was conducted using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1) database. 13 MIMIC-IV contains comprehensive, de-identified health records of patients admitted to the intensive care unit (ICU) or emergency department at Beth Israel Deaconess Medical Center between 2008 and 2022. The database has received ethical approval from the institutional review boards of the Massachusetts Institute of Technology (Cambridge, MA) and Beth Israel Deaconess Medical Center (Boston, MA). All data were de-identified, and the requirement for informed consent was waived. Access to the database was granted to the first author (Zikang Zhou, certification ID: 12562151) after completion of the National Institutes of Health web-based training program.
2.2. Study Population
Adult patients (≥18 years) with a diagnosis of pulmonary embolism (PE) and chronic kidney disease (CKD) were eligible for inclusion. Study flow diagram
2.3. Data Collection
Baseline demographic variables included age, sex, and race. Clinical variables were extracted from the first 24 hours of ICU admission and included vital signs—respiratory rate (RR), systolic blood pressure (SBP), diastolic blood pressure (DBP), heart rate (HR), oxygen saturation (SpO2), and body temperature—as well as laboratory parameters, including troponin-T (Tn-T), arterial pH, and eGFR.
According to the suggestion of Reviewer 1, the original Section 2.4 has been deleted.
2.4. Exposure and Outcomes
The exposure of interest was anticoagulant type, categorized as DOACs or warfarin, defined by prescription records indicating administration for more than 10 days during ICU stay.
Bleeding Events Across Different Anatomical Sites
Secondary outcomes included major bleeding, defined according to corresponding ICD codes listed in the Supplementary Materials.
Time-to-event was calculated from initiation of anticoagulant therapy to the occurrence of the outcome of interest or censoring, whichever occurred first.
2.5. Propensity Score Methods
2.5.1. Propensity Score Matching (Primary Analysis)
Propensity score matching (PSM) was used as the primary method to control for baseline confounding. Propensity scores were estimated using a logistic regression model incorporating clinically relevant covariates selected a priori based on established guidelines, 4 including age, sex, race, RR, SBP, DBP, HR, SpO2, body temperature, Tn-T, arterial pH, and eGFR. Patients receiving DOACs were matched 1:1 to those receiving warfarin using nearest-neighbor matching with a caliper width of 0.005. Balance between treatment groups before and after matching was assessed using standardized mean differences (SMDs), with values <0.1 indicating adequate covariate balance.
2.5.2. Inverse Probability of Treatment Weighting (Sensitivity Analysis)
As a sensitivity analysis, inverse probability of treatment weighting (IPTW) was additionally applied using the same set of covariates included in the PSM model. Stabilized IPTW weights were calculated to reduce the influence of extreme weights, and truncation at the 1st and 99th percentiles was applied. Covariate balance before and after weighting was evaluated using SMDs, with all post-weighting SMDs <0.1 considered indicative of satisfactory balance.
2.6. Statistical Analysis
Missing data were imputed using the k-nearest neighbors (KNN) method prior to propensity score estimation. Cox proportional hazards models were fitted to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for bleeding outcomes in both the propensity score–matched cohort and the IPTW-weighted cohort. Robust sandwich variance estimators were used to account for clustering within matched pairs or weighted samples.
Kaplan–Meier curves were generated to compare cumulative incidence of bleeding events between treatment groups in the matched cohort. To explore potential effect modification by renal function, interaction terms between anticoagulant type and standardized eGFR were incorporated into Cox models in both the PSM and IPTW analyses. In addition, subgroup analyses were performed by stratifying patients into three eGFR categories (60–30, 30–15, and <15 mL/min/1.73 m2), and bleeding rates were compared between treatment groups within each stratum.
Data extraction was performed using SQLAlchemy (v2.0.35). Statistical analyses were conducted using Polars (v1.12.0), NumPy (v2.1.3), SciPy (v1.14.1), scikit-learn (v1.6.0) for propensity score estimation, and lifelines (v0.30.0) for survival analyses. All statistical tests were two-sided, and a P value <0.05 was considered statistically significant.
3. Results
3.1. Baseline Characteristics
A total of 1,363 patients with pulmonary embolism and chronic kidney disease who were admitted to the ICU and received anticoagulation therapy with either warfarin or direct oral anticoagulants (DOACs) were included in the study. The cohort selection process is illustrated in Figure 2. Among these patients, 832 (61.04%) received warfarin and 531 (38.96%) received DOACs. Comparison of the percentage of bleeding events between warfarin and DOAC
Abbreviations: PSM, Propensity Score matching;SMD, Standardized Mean differences;SBP, Systolic Blood Pressur; DBP, Diastolic Blood pressure;RR, Respiratory Rate; SpO2, Oxygen saturation;Tn-T,Cardiac Troponin-T;PH,Potential of hydrogen;eGFR,Estimated Glomerular Filtration Rate
3.2. Subgroup Analysis: Bleeding Event Percentages Stratified by eGFR
A subgroup analysis stratified by estimated glomerular filtration rate (eGFR) was conducted to descriptively compare the percentage of bleeding events between patients treated with warfarin and those treated with DOACs across different levels of renal function. Patients were categorized into three eGFR strata: 60–30, 30–15, and 15–0 mL/min/1.73 m2.
As shown in Figure 2, bleeding event percentages varied across eGFR strata and between treatment groups. Among patients with eGFR <60 mL/min/1.73 m2 overall, the bleeding event percentage was 40.08% in the warfarin group and 33.33% in the DOAC group. In the eGFR 60–30 mL/min/1.73 m2 subgroup, bleeding occurred in 39.88% of warfarin-treated patients compared with 33.14% of those receiving DOACs. In patients with more advanced renal impairment (eGFR 30–15 mL/min/1.73 m2), the bleeding event percentage increased substantially in both groups but remained higher in the warfarin group (60.00%) than in the DOAC group (44.44%). No bleeding events were observed in either treatment group among patients with eGFR 15–0 mL/min/1.73 m2.
These descriptive findings suggest that bleeding risk increases with worsening renal function in both anticoagulant groups and that DOAC-treated patients tended to exhibit lower bleeding event percentages compared with warfarin-treated patients across most eGFR strata. A bar chart illustrating these comparisons is presented in Figure 2. Given the unadjusted nature of this subgroup analysis, these results should be interpreted as exploratory and were further evaluated using multivariable time-to-event models with propensity score–based adjustment in subsequent analyses.
3.3. IPTW
As a sensitivity analysis, stabilized inverse probability of treatment weighting (IPTW) was applied using the same set of covariates. Covariate balance before and after weighting was assessed using standardized mean differences. After IPTW, all baseline covariates achieved excellent balance, with SMDs clustered around zero and all values below 0.1 (Figure 3). The distribution of stabilized IPTW weights was centered around 1.0, with no evidence of extreme weights after truncation, indicating adequate covariate overlap and supporting the validity of the weighted analysis (Figure 4).These findings suggest that both PSM and IPTW effectively minimized baseline imbalances between treatment groups. Covariate balance before and after inverse probability of treatment weighting (IPTW) Distribution of stabilized inverse probability of treatment weights (IPTW)
3.4. Primary Outcome: Any Bleeding
In the IPTW-weighted cohort, Cox proportional hazards analysis demonstrated that DOAC use was associated with a significantly increased risk of any bleeding compared with warfarin (hazard ratio [HR] 1.37, 95% confidence interval [CI] 1.13–1.66; P = 0.001). This association remained robust after adjustment for residual confounding through weighting.
Consistent results were observed in the PSM cohort, where the direction and magnitude of the effect estimates were comparable, supporting the robustness of the primary findings across different confounding adjustment strategies.
3.5. Secondary Outcome: Major Bleeding
For major bleeding events, no statistically significant difference was observed between the DOAC and warfarin groups in the IPTW-weighted analysis (HR 0.97, 95% CI 0.74–1.28; P = 0.84). Similar null associations were observed in the PSM cohort, indicating that while DOAC use was associated with a higher risk of overall bleeding, it did not confer an increased risk of major bleeding compared with warfarin.
3.6. Effect Modification by Renal Function (eGFR)
To further explore whether renal function modified the association between anticoagulant type and bleeding risk, interaction terms between treatment group and eGFR were incorporated into Cox proportional hazards models in both the PSM-matched and IPTW-weighted cohorts.
Across both analytical approaches, no statistically significant interaction between anticoagulant type and eGFR was observed for either any bleeding or major bleeding outcomes (all interaction P values > 0.05). These findings suggest that the relative bleeding risk associated with DOACs versus warfarin was consistent across different levels of renal impairment within the CKD population studied.In the propensity score–matched cohort, the association between DOAC use and bleeding risk varied according to renal function, with a trend toward higher hazard ratios at higher eGFR levels (Figure 5).Consistent findings were observed in the IPTW-weighted analysis, supporting the robustness of the effect modification by eGFR (Figure 6). Interaction between direct oral anticoagulants and renal function on bleeding risk after propensity score matching Effect modification of eGFR on the association between DOAC use and bleeding risk in the IPTW-weighted cohort
3.7. Summary of Findings
Taken together, these results indicate that, after rigorous adjustment for confounding using both PSM and IPTW, DOAC therapy in patients with pulmonary embolism and chronic kidney disease was associated with a higher risk of overall bleeding compared with warfarin, while no significant difference was observed in the risk of major bleeding. The association between anticoagulant type and bleeding risk appeared consistent across levels of renal function.
4. Discussion
In this large, ICU-based cohort of patients with pulmonary embolism (PE) and chronic kidney disease (CKD), we comprehensively compared the bleeding risk associated with direct oral anticoagulants (DOACs) and warfarin using multiple propensity score–based approaches. After rigorous confounding adjustment through both propensity score matching (PSM) and inverse probability of treatment weighting (IPTW), several clinically and mechanistically relevant findings emerged.
4.1. Overall Bleeding Risk Versus Major Bleeding: A Dissociation Driven by CKD-Related Hemostatic Fragility
Our primary analysis demonstrated that DOAC use was associated with a significantly higher risk of overall bleeding compared with warfarin, whereas no significant difference was observed for major bleeding events. This apparent dissociation between “any bleeding” and “major bleeding” is particularly informative in the context of CKD and critical illness. 14
CKD is characterized by a unique and fragile hemostatic milieu, in which defects in primary hemostasis—most notably platelet dysfunction and impaired platelet–vessel wall interactions—play a dominant role.15,16 Uremic toxins, altered nitric oxide signaling, and dysregulated von Willebrand factor activity collectively predispose patients to mucocutaneous, gastrointestinal, and access-site bleeding, which are often classified as non-major bleeding events. These bleeding manifestations may occur even in the absence of profound systemic anticoagulation and are highly sensitive to subtle perturbations in coagulation balance.17,18
Importantly, the absence of an increased risk of major bleeding suggests that DOACs do not exacerbate catastrophic hemorrhagic complications in patients with CKD and PE, supporting their relative safety with respect to life-threatening bleeding.There may be differences between the DOACs and bleeding events.
Of note, the present study pooled different direct oral anticoagulants as a single class for analysis. However, emerging evidence suggests that bleeding risk may not be uniform across individual DOAC agents. In a recent randomized trial of acute venous thromboembolism, Castellucci et al reported significant differences in bleeding complications between apixaban and rivaroxaban, indicating that the comparative safety profile of DOACs may vary by individual agent. 24 Heterogeneity in bleeding propensity among different DOACs could therefore have contributed to the observed overall bleeding events in the present study, a distinction that warrants consideration in future targeted research.
4.2. Renal Dysfunction as a Driver of Bleeding Risk Rather Than a Modifier of Treatment Effect
Our descriptive subgroup analysis revealed a clear gradient of increasing bleeding event percentages with worsening renal function in both treatment groups. This finding reinforces the concept that declining kidney function itself is a central determinant of bleeding risk, independent of anticoagulant choice.
However, in adjusted time-to-event analyses incorporating interaction terms between anticoagulant type and estimated glomerular filtration rate (eGFR), no statistically significant effect modification by renal function was observed. This suggests that, within the CKD population studied, renal impairment amplifies baseline bleeding susceptibility rather than fundamentally altering the relative bleeding risk associated with DOACs versus warfarin11,25.
From a mechanistic perspective, this observation is notable. Although renal clearance contributes to the elimination of most DOACs, 26 accumulation-related toxicity may not be the dominant determinant of bleeding risk once patients reach moderate-to-severe CKD.27,28 Instead, systemic uremic effects, endothelial dysfunction, inflammation, and critical illness–related coagulopathy may overshadow pharmacokinetic differences between anticoagulants.17,29 In such settings, anticoagulant type appears to exert a relatively consistent proportional effect across eGFR levels, as reflected by the absence of statistically significant interaction.
Nevertheless, the observed trend toward higher hazard ratios at higher eGFR levels in both PSM and IPTW analyses suggests that preserved renal function may allow the pharmacodynamic differences between anticoagulants to manifest more clearly. In contrast, in advanced CKD, the overwhelming influence of uremic hemostatic dysfunction may attenuate detectable treatment-related differences.
4.3. Reconciling Descriptive and Adjusted Analyses
An important strength of this study lies in the integration of descriptive subgroup analyses with robust causal inference methods. While unadjusted comparisons suggested lower bleeding event percentages among DOAC-treated patients across most eGFR strata, these findings were attenuated or reversed after adjustment using PSM and IPTW.
This divergence underscores the critical importance of addressing confounding by indication in observational studies of anticoagulant therapy. Patients selected for DOAC therapy may differ systematically from those receiving warfarin with respect to clinical stability, comorbidity burden, and perceived bleeding risk—factors that are not fully captured in crude subgroup comparisons. By leveraging two complementary propensity score–based approaches, our analyses provide a more reliable estimate of the treatment effect and highlight how unadjusted subgroup findings may be misleading if interpreted in isolation.
4.4. Clinical Implications
Collectively, our findings suggest that in ICU patients with PE and CKD, DOAC therapy is associated with an increased risk of overall bleeding compared with warfarin, while the risk of major bleeding remains comparable between the two treatment strategies. Renal function strongly influences absolute bleeding risk but does not appear to substantially modify the relative effect of anticoagulant choice.
These results emphasize that anticoagulant selection in this population should be guided not only by renal function but also by the clinical relevance of bleeding outcomes, patient frailty, and the anticipated consequences of non-major bleeding. Enhanced surveillance and individualized risk–benefit assessment remain essential, particularly when DOACs are used in critically ill patients with CKD.
5. Conclusion
In this large, ICU-based cohort of patients with pulmonary embolism and chronic kidney disease, we found that treatment with direct oral anticoagulants was associated with a higher risk of overall bleeding compared with warfarin, while no significant difference was observed in the risk of major bleeding. These findings were consistent across two complementary propensity score–based analytic strategies, supporting the robustness of the observed association.
Our results highlight a clinically important dissociation between overall bleeding and major bleeding in this vulnerable population. In the setting of CKD and critical illness, bleeding risk appears to be driven predominantly by a fragile hemostatic milieu characterized by platelet dysfunction, endothelial injury, and systemic inflammation, rather than by catastrophic anticoagulant-related hemorrhage. Within this context, DOACs may lower the threshold for clinically relevant non-major bleeding without increasing life-threatening bleeding events.
Furthermore, although worsening renal function substantially increased absolute bleeding risk, it did not materially modify the relative bleeding risk associated with DOACs versus warfarin. This suggests that, once moderate-to-severe CKD is established, global uremic and critical illness–related effects on hemostasis may outweigh pharmacokinetic differences between anticoagulant classes.
Taken together, these findings provide nuanced evidence to inform anticoagulant selection in critically ill patients with PE and CKD. While DOACs remain a viable option with respect to major bleeding, heightened vigilance for overall bleeding events is warranted. Future prospective studies are needed to better delineate the clinical impact of non-major bleeding and to refine individualized anticoagulation strategies in this high-risk population.
6. Limitations
This study has several limitations. First, its retrospective observational design precludes definitive causal inference, and residual confounding may persist despite the use of propensity score matching and inverse probability of treatment weighting. Second, bleeding outcomes were identified using diagnostic codes, which may be subject to misclassification and may not fully capture bleeding severity or clinical context. Third, detailed information on anticoagulant dosing, drug level monitoring, treatment interruptions, and reversal strategies was not available, which is particularly relevant in the ICU setting. Finally, this was a single-center database study, and the generalizability of the findings to non-ICU populations or other healthcare systems may be limited.
According to the suggestion of Reviewer 1, the original Table 2 has been deleted.
A total of 364,627 patients were initially screened from the MIMIC-IV database. After restricting the cohort to adults aged over 18 years with chronic kidney disease (CKD), defined as an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m2 calculated using the Modification of Diet in Renal Disease (MDRD) equation, and a confirmed diagnosis of pulmonary embolism (PE) based on ICD codes, 2,900 patients with concurrent CKD and PE were identified.
Of these, 28 patients who did not receive any anticoagulant therapy were excluded. The remaining patients were stratified into two groups based on the duration of oral anticoagulant exposure: 1,145 patients received warfarin, and 854 received a direct oral anticoagulant (DOAC; including rivaroxaban, apixaban, edoxaban, or dabigatran).
After excluding patients who were not admitted to the intensive care unit (ICU), 832 patients remained in the warfarin group and 531 in the DOAC group.
Propensity score matching (PSM) was then performed using a 1:1 nearest-neighbour algorithm, yielding two well-balanced cohorts of 508 patients each.
Standardized mean differences (SMDs) of baseline covariates between the warfarin and direct oral anticoagulant (DOAC) groups are shown before weighting (blue circles) and after IPTW (orange diamonds). Vertical dashed lines indicate the conventional threshold of |SMD| = 0.1, reflecting acceptable covariate balance. After IPTW, all baseline covariates achieved adequate balance, with SMDs below 0.1, indicating effective control of measured confounding.
The histogram displays the distribution of stabilized IPTW weights used in the weighted Cox proportional hazards models. Extreme weights were truncated at the 1st and 99th percentiles to reduce the influence of outliers. The majority of weights were concentrated around 1.0, indicating good overlap between treatment groups and supporting the positivity assumption.
Panel A shows the estimated hazard ratios (HRs) for bleeding associated with DOAC use compared with warfarin at different levels of renal function, expressed as estimated glomerular filtration rate (eGFR) standardized to the mean (−1 SD, mean, +1 SD). The dashed horizontal line indicates an HR of 1.0, representing no difference in bleeding risk between treatment groups.
Panel B presents the forest plot of hazard ratios derived from the Cox proportional hazards model including DOAC use, standardized eGFR (per SD increase), and their interaction term (DOAC × eGFR). Points represent adjusted hazard ratios, and horizontal lines indicate 95% confidence intervals.
Together, these panels demonstrate a potential effect modification by renal function on the association between anticoagulant type and bleeding risk in patients with pulmonary embolism and chronic kidney disease.
Panel A depicts the estimated hazard ratios for any bleeding comparing DOACs with warfarin across different standardized eGFR levels (−1 SD, mean, +1 SD), based on an inverse probability of treatment weighting (IPTW) Cox model with a treatment–eGFR interaction term.
Panel B presents the corresponding forest plot of hazard ratios for DOAC use, eGFR, and the DOAC × eGFR interaction.
Error bars represent 95% confidence intervals, and the vertical dashed line denotes an HR of 1.0.
Models were fitted using stabilized IPTW with truncation of extreme weights and robust variance estimation.
Footnotes
Ethical Considerations
Not required as this study used secondary data aggregated at both country and global level.
Author Contributions
ZKZ and QYX contributed equally to this work. YCL, STS and ZKZ conceptualized and designed the study. QYX performed the statistical analysis. All authors contributed to the data collection, analysis, or interpretation. ZKZ drafted the manuscript. All authors reviewed and revised the report before submission and approved the final version. ZKZ is the guarantor of the study, ensuring that all listed authors meet the authorship criteria and confirming that no eligible contributors have been omitted.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: National Natural Science Foundation of China (No. 82172120), Natural Science Foundation of Tianjin (No. 22JCYBJC00530), Tianjin Key Medical Discipline (Specialty) Construction Project(TJYXZDXK-007A).
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The data that support the findings of this study are available from the MIMIC-IV database (version 3.1), a publicly available and de-identified clinical database. Restrictions apply to the availability of these data, which were used under license for this study. Data can be accessed upon successful completion of the required training and application process at the PhysioNet repository (
).