Venous thromboembolism, chronic liver disease and anticoagulant choice: effectiveness and safety of direct oral anticoagulants versus warfarin

Background Little to no data exist to guide treatment decision in patients with venous thromboembolism (VTE) and chronic liver disease. Objectives To assess the effectiveness and safety of direct oral anticoagulants (DOACs)—individually and as a class—vs warfarin and between 2 DOACs in patients with acute VTE and chronic liver disease. Methods We conducted a retrospective, US claims–based, propensity score–matched cohort study in adults with acute VTE and chronic liver disease who had newly initiated oral anticoagulants between 2011 and 2017. The primary outcome was a composite of hospitalization for recurrent VTE and hospitalization for major bleeding. Results The cohorts included 2361 DOAC-warfarin, 895 apixaban-warfarin, 2161 rivaroxaban-warfarin, and 895 apixaban-rivaroxaban matched pairs. Lower risk of the primary outcome was seen with DOACs (hazard ratio [HR], 0.72; 95% CI, 0.61-0.85), apixaban (HR, 0.48; 95% CI, 0.35-0.66) or rivaroxaban (HR, 0.73; 95% CI, 0.61-0.88) vs warfarin but not apixaban-rivaroxaban (HR, 0.68; 95% CI, 0.43-1.08). The HRs of hospitalization for major bleeding were 0.69 (95% CI, 0.57-0.84) for DOAC-warfarin, 0.43 (95% CI, 0.30-0.63) for apixaban-warfarin, 0.72 (95% CI, 0.58-0.89) for rivaroxaban-warfarin, and 0.60 (95% CI, 0.35-1.06) for apixaban-rivaroxaban. Recurrent VTE risk was lower with apixaban (HR, 0.47; 95% CI, 0.26-0.86), but not DOACs (HR, 0.81; 95% CI, 0.59-1.12) or rivaroxaban vs warfarin (HR, 0.81; 95% CI, 0.57-1.14) or apixaban-rivaroxaban (HR, 0.92; 95% CI, 0.42-2.02). Conclusion While the magnitude of clinical benefit varied across individual DOACs, in adults with acute VTE and chronic liver disease, oral factor Xa inhibitors (as a class or individually) were associated with lower risk of recurrent VTE and major bleeding.

Direct oral anticoagulants (DOACs) comprising oral factor (F)Xa inhibitors (ie, apixaban, rivaroxaban, and edoxaban) and direct thrombin inhibitors (ie, dabigatran) are the first-line treatment option in the management of VTE [5][6][7].Anticoagulation reduces the risk of recurrent VTE and death, which are highest within the first 3 to 6 months of an index VTE event.While the utility of DOACs compared to warfarin has been investigated in patients with complex comorbid conditions such as chronic kidney disease [8,9] and active cancer [10], clinical trial and real-world data in individuals with VTE and chronic liver disease are scant [8,11].Individuals with elevated liver enzyme levels and chronic liver disease were excluded from the approved clinical trials of DOACs for VTE treatment [12][13][14][15], and so far, there are no clinical trials comparing DOACs and warfarin in patients with VTE and chronic liver disease.There is also no large, population-based study of DOACs compared with warfarin, between 2 DOACs in patients with VTE and chronic liver disease, or among patients with VTE and severe chronic liver disease.Available real-world data on the riskbenefit profile of DOACs in patients with VTE and chronic liver disease are mainly from case series with fewer than 100 DOAC-exposed subjects [8,11,16,17].These smaller studies are, however, underpowered to detect meaningful clinical differences between DOACs and warfarin and have limited generalizability to routine clinical care.
Of note, DOACs rely, to varying degrees, on the liver for hepatic clearance (75% for apixaban, 65% for rivaroxaban, 50% for edoxaban, and 20% for dabigatran) [8].Therefore, elevated levels of DOACs in the blood arising from impaired hepatic dysfunction may pose further harmful effects in patients requiring oral anticoagulants.The risk of thrombotic events and bleeding complications is also significantly increased in patients with chronic liver disease [8,11].Considering the increasing adoption of DOACs to manage VTE coupled with the increasing incidence of chronic liver disease in the United States and shared risk factors between VTE and chronic liver disease [8,11], there is a need for high-quality real-world data to guide treatment decisions between DOACs and warfarin in this high-risk population.Accordingly, we evaluated the magnitude of clinical benefit of DOACs (individually and as a class) vs warfarin and between DOACs in patients with VTE and chronic liver disease.

| Study population
The study population comprised individuals aged ≥18 years with a diagnosis of chronic liver disease who initiated DOACs (apixaban, rivaroxaban, edoxaban, or dabigatran) or warfarin between January 1, 2011, and December 31, 2017, after the first diagnosis of VTE [18].
The date of first fill of oral anticoagulant was regarded as the index date, and the 1-year period prior to the index date was regarded as the baseline period.While the database included data from January 2010 through December 2017, given the new-user design and 1-year baseline period, the first possible date of patient inclusion was January 1, 2011.VTE was defined using validated algorithms, defined as ≥1 inpatient or outpatient claim for acute deep vein thrombosis (DVT) or pulmonary embolism (positive predictive value [PPV], >90%; Supplementary Table 1) [18][19][20].Chronic liver disease was identified using diagnosis codes for conditions associated with prolonged or complete deterioration of liver function in any position in inpatient or outpatient claims (Supplementary Table 1) [21,22].
Eligible patients were required to have (a) filled a single oral anticoagulant on the index date, (b) ≥12 months of continuous medical and pharmacy insurance eligibility with no more than a 30-day gap before the index date, (c) initiated oral anticoagulants within 30 days of hospital discharge (if inpatient) or service date (if outpatient visit) for first VTE diagnosis [18], and (d) health encounters for chronic liver disease within 12 months of the index dispensing (Supplementary

Essentials
• Little data exist on anticoagulation in patients with venous thromboembolism and liver disease.
• We sought to answer this important clinical question using data from routine clinical care.
• The primary outcome was hospitalization for recurrent venous thromboembolism or major bleeding.
• The risk of the primary outcome was lower with direct oral anticoagulants than with warfarin.
Figure 1).We required filling of oral anticoagulants within 30 days of health encounter for index VTE to increase the likelihood that the source population comprised patients who initiated oral anticoagulation for management of VTE.A 30-day period between index VTE diagnosis and the index date was required to ensure patients had a reasonable amount of time to fill and initiate outpatient oral anticoagulants after index VTE diagnosis.
We excluded patients with medical encounters for alternative indications for oral anticoagulation (eg, atrial fibrillation, mitral stenosis, valve repair, and replacement), pregnancy, cesarean section, prior bleeding (to capture incident bleeding events after index VTE), dialysis or kidney replacement, and inferior vena cava filter during the baseline period.Based on the index oral anticoagulant fill, eligible subjects were categorized as either DOAC or warfarin users.

| Outcomes
The clinical outcomes, identified using validated algorithms, were analyzed both separately and as a composite.The primary outcome measure was the net clinical benefit, a composite of hospitalization for recurrent VTE and hospitalization for major bleeding.Recurrent VTE, the primary effectiveness outcome, was defined as primary diagnosis in inpatient claims for acute DVT or pulmonary embolism >7 days after index VTE event (PPV, ≥89%; Supplementary Table 2) [18][19][20]23].Relatively poor PPVs have been reported with algorithms for VTE based on events that occurred in outpatient settings or defined by the presence of primary or secondary diagnoses.Therefore, eligible recurrent VTE events were restricted to those that occurred in inpatient settings and reported as the primary diagnosis for hospitalization [24].Major bleeding, the primary safety outcome, was defined by the presence of transfusion codes or primary diagnosis for bleeding in inpatient claims (PPV, ≥89%; Supplementary Table 2) [23,25,26].All-cause mortality and clinically relevant nonmajor bleeding (CRNMB) were the secondary effectiveness and safety outcomes, respectively.All-cause mortality, available in the database, was obtained by linkage of the database with the Social Security Administration Death Master File.The specificity of mortality status in the Death Master File has been reported to be ≥97% [27,28].Due to changes in reporting requirements of mortality data by individual US states to the Social Security Office, death data in the Optum Clinformatics database may be incomplete from 2013 onwards [27][28][29].CRNMB was defined by secondary diagnosis for bleeding in inpatient claims or primary or secondary diagnosis for bleeding in outpatient claims [23,25].We distinguished between hospitalization for major bleeding and CRNMB to align with pivotal DOACs for VTE clinical trials where these safety outcomes were captured separately [12][13][14][15], to more closely align with International Society on Thrombosis and Haemostasis definitions for major bleeding and CRNMB [30,31], and to differentiate between the severity of bleeding events.The safety outcomes were collectively referred to as clinically relevant bleeding.

| Follow-up
Subjects were followed in an as-treated manner the day after the index date for up to 183 days until first diagnosis of an outcome, switch to comparator drugs, treatment discontinuation (defined as a gap of >30 days after the end of days' supply of the previous and subsequent fill), insurance disenrollment, or study end date, whichever came first.We chose a primary follow-up time of up to 183 days for outcome assessment to align with the pivotal trials for VTE [12][13][14][15] and because the recommended treatment duration for primary treatment of acute VTE is 3 to 6 months [7].

| Covariates
Several known and suspected risk factors for the outcomes of interest were adjusted for our analyses.These baseline covariates were broadly categorized into (a) demographic characteristics, (b) type of index VTE event, (c) comorbid conditions and lifestyle factors, (d) concomitant drug use (eg, P-glycoprotein inhibitors) [32,33], (e) risk score for bleeding (ie, the Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (>65 years), Drugs/alcohol concomitantly score [HAS-BLED] score) [34] and (f) a claims-based frailty index reported by Kim et al. [35] to further adjust for burden of illness.A complete list of the covariates and time of assessment is presented in Supplementary Table 3. Balance on the frequency and results from relevant serologic tests were also assessed in the subgroup of subjects where this information was available (n = 1853); these variables were, however, not adjusted for in the analysis models.

| Statistical analysis
Propensity score (PS) methods were used to balance the treatment groups on measured baseline covariates.The PS for being in a treatment group was estimated using a logistic regression model and corresponds to the predicted probability of receiving treatment given the baseline covariates.DOAC and warfarin users were then matched in a 1:1 ratio (by nearest neighbor without replacement) using a caliper width equal to 0.2 of the SD of the logit of the PS [36] Balance between the 2 treatment groups after PS matching (PSM) was assessed using standardized differences, with a threshold of ≤0.1 to imply the absence of substantial imbalance [36].Further, we assessed differences in risk for clinical outcomes by exposure to a specific DOAC vs warfarin and for head-to-head DOAC comparison.Based on the index fill, matching was repeated to yield cohorts comparing (a) apixaban vs warfarin, (b) rivaroxaban vs warfarin, and (c) apixaban vs rivaroxaban.The exposure to specific DOACs vs warfarin and DOAC-DOAC comparisons were limited to oral direct FXa inhibitors users (ie, apixaban and rivaroxaban) as they accounted for 97% (n = 3056) of the 3140 subjects that initiated DOACs (Figure 1).LAWAL ET AL.
- 3 of 13     The cumulative incidence rate (IR) of the study outcomes was illustrated using Kaplan-Meier curves, compared using a stratified logrank test, and presented as the total number of events per 100 person-years.After PSM, hazard ratios (HRs) and their corresponding 95% CI were estimated using Cox proportional hazard regression models, with a robust variance estimator to account for correlation between matched subjects.The proportional hazard assumption was assessed by visual inspection of the log-log survival curves and testing the Schoenfeld residuals.We did not observe evidence of violation of the proportional hazards assumption for any outcomes.

| Secondary and subgroup analyses
We assessed the net clinical benefit of DOACs vs warfarin in individuals with VTE and severe chronic liver disease, either cirrhosis or its advanced stage decompensated cirrhosis.The absence of the clinical components of the Child-Pugh scores and limited availability of laboratory tests hindered the computation of Child-Pugh scores (Supplementary Methods).Alternatively, cirrhosis and decompensated cirrhosis were identified using validated diagnosis-based algorithms (PPV, >78%-91%; Supplementary Table 1) [37][38][39].Further, we assessed for the presence of effect modification of observed relationships between DOACs and warfarin by age categories, sex, and the presence of cancer, nonalcoholic fatty liver disease (NAFLD; because prior data suggest substantial drug metabolism alterations, particularly of cytochrome P450 3A4 in NAFLD) [8,40,41], chronic kidney disease, provoked or unprovoked VTE, and diabetes.Provoked VTE was defined as hospitalization for ≥3 days, receipt of estrogen therapy, or medical encounters indicative of trauma, fracture, surgery, or active cancer (presence of procedural codes for radiation therapy or chemotherapy) within 3 months of index VTE event [42].

| Sensitivity analyses
First, to simulate the other treatment durations assessed in the clinical trials of DOACs for VTE treatment [12][13][14][15], we repeated our analyses using an alternative follow-up period of up to 3 months or 12 months.Second, we repeated our analyses using an intent-to-treat approach to evaluate the possible effects of informative censoring, wherein subjects were followed from the index date till occurrence of a censoring event, without considering treatment discontinuation or switching.
Third, to increase the likelihood that subjects remained anticoagulated during follow-up, we reduced the maximum allowable gap between fills to (a) a 7-day gap, (b) a 14-day gap, and (c) a gap based on the halflife for each oral anticoagulant [8,11] (Supplementary Methods).
Fourth, to accommodate patients who entered the cohort toward the end of the study period while facilitating potential accrual of follow-up time, the primary analyses were repeated using a cohort restricted to patients who initiated treatment between January 1, 2011, and June 30, 2017.Fifth, using a Fine and Gray subdistribution hazard model,  we adjusted for the competing risk of death in analysis of the primary outcome and the primary effectiveness and safety outcomes.
Finally, 3 sets of additional sensitivity analyses were conducted to assess potential unmeasured and substantial residual confounding.
First, we performed high-dimensional PSM (hd-PSM) using prespecified covariates and 200 additional empirically identified covariates that may serve as proxies for other unmeasured confounders [43].Second, we calculated an E-value to assess the strength of an unmeasured confounder needed to fully explain away the exposureoutcome association, conditional on measured confounders [44].
Third, to exclude the presence of substantial residual confounding, we performed a negative control analysis using pneumonia as the outcome measure [45], an outcome with no known relationship that differs between DOACs and warfarin.
All analyses were conducted using SAS 9.4 (SAS Institute).No adjustments were made for multiple comparisons.
Of the 8477 subjects, 5927 discontinued treatment, while 505 subjects switched treatment from rivaroxaban (245) or apixaban (245) to warfarin, the most common (Supplementary  9).The baseline characteristics of the various exposure cohorts were reasonably balanced after PSM, with standard differences of <10% after adjustment (Table 1, Supplementary Tables 5 and 7-9).

| Mortality risks
No difference was observed in risk of all-cause mortality for exposure to any or specific DOACs vs warfarin or apixaban vs rivaroxaban (Tables 2 and 3).Although with wide CIs, the IRs and upper limit of the effect estimate for all-cause mortality were consistently higher with DOACs as a class vs warfarin (1.30), apixaban (1.41), or rivaroxaban (1.27) compared with warfarin, and apixaban vs rivaroxaban (1.71), suggesting the possibility of a nonnull effect.

| Secondary and subgroup analyses
Similar to the primary findings, compared to warfarin, lower risk of the composite outcome was seen in individuals with cirrhosis (compensated or decompensated), and 36% reduction in risk of the primary outcome was observed with DOACs vs warfarin (HR, 0.64; 95% CI, 0.43-0.96),specifically a 56% reduced risk in the subset of subjects with decompensated cirrhosis (HR, 0.44; 95% CI, 0.26-0.73)(Table 4).
The findings from subgroup analyses suggest that the magnitude of association between DOACs and warfarin was moderated by the presence of cancer, chronic renal disease, and NAFLD/nonalcoholic LAWAL ET AL. steatohepatitis (Table 4).These analyses should be interpreted in context of a relatively small sample size.

| Sensitivity analyses
Our primary conclusions for DOACs vs warfarin remained largely consistent across sensitivity analyses, including using an alternative follow-up period of up to 3 months (Supplementary Table 10) or 12 months (Supplementary Table 11), an intent-to-treat design (Supplementary Table 12), reducing the maximum gap between refills from 30 days in the primary analysis to 7 or 14 days and based on half-life of each oral anticoagulant (Supplementary Table 13), and cohort restricted to subjects that initiated treatment prior to June 30, 2017 (Supplementary Table 14).There was no meaningful difference between the overall conclusions from the primary models comparing T A B L E 1 Select baseline characteristics of patients with acute venous thromboembolism and chronic liver disease initiating direct oral anticoagulants vs warfarin before and after propensity score matching.

| D I S C U S S I O N
To the best of our knowledge, this is the first large population-based, comparative effectiveness and safety study of oral anticoagulants in patients with VTE and chronic liver disease.The current study adds 3 salient findings to the evidence needed to guide the selection of oral anticoagulation therapy in patients with VTE and chronic liver disease in contemporary clinical settings.
First, among patients with VTE and chronic liver disease, DOACs (as a class) had a more favorable benefit-risk profile than that of warfarin.Of note, the oral direct FXa inhibitors comprised 97% of the 3140 DOAC users.Although little data exist on VTE among patients with chronic liver disease [8,11], this finding is consistent with evidence from several studies of different designs suggesting that DOACs are as safe and effective as warfarin, if not superior in individuals with VTE [5,7,[12][13][14][15]46].The larger net clinical benefit observed with DOACs compared to that with warfarin may be partly explained by its more stable pharmacokinetic characteristics, no requirement for international normalized ratio monitoring, and fewer food and drug interactions.More so, human cytochrome-450 enzymes are markedly altered in chronic liver disease, particularly cirrhosis [8,11].While warfarin is primarily metabolized by cytochrome enzymes and relies on the liver for clearance, DOACs are generally less reliant on these enzymes for metabolism [8,11].
Second, the magnitude of net clinical benefit differed between individual oral direct FXa inhibitors compared to warfarin and between oral direct FXa inhibitors.Overall, compared to warfarin, the magnitude of clinical benefit was consistently larger for apixaban than for rivaroxaban and in head-to-head comparison of apixaban and rivaroxaban.Similar findings of lower bleeding risks associated with apixaban vs rivaroxaban, when compared to warfarin, were observed in randomized controlled trials involving patients with atrial fibrillation [47,48] or VTE [12][13][14][15] and real-world studies, including head-to-head comparisons of apixaban and rivaroxaban [23,[49][50][51][52].For example, in the AMPLIFY acute VTE trial [12], subjects on apixaban vs warfarin had significantly lower risk of major bleeding (HR, 0.31; 95% CI, 0.17-0.The annualized risk for major bleeding in the VTE clinical trials for DOACs ranged between 1.2% and 2.2% [57], with an incidence of 0.6% and 0.5% in patients randomized to apixaban or rivaroxaban, respectively [12,13].Conversely, 3.3% (11.8 IR/person-year) and 4.7% several sensitivity analyses suggest the robustness of our primary analyses.Second, although we carefully adjusted for confounding by several well-known confounders, this study may still be subject to residual and unmeasured confounding.For example, unmeasured confounding may have occurred due to lack of information on overthe-counter use of aspirin and nonsteroidal anti-inflammatory drugs, drugs that are well-known to increase bleeding risks.The findings from negative control analysis and hd-PSM, however, provide some reassurance of the absence of substantial residual confounding.
Third, our data lacked information to determine the time in the therapeutic range.Further, clinical and laboratory data needed to calculate Child-Pugh scores for assessing chronic liver disease severity were either missing or grossly underreported, thus precluding our ability to calculate Child-Pugh scores or use other predictive models (eg, the Model for End-Stage Liver Disease score) that rely on laboratory data.This issue of lack of laboratory data, while mitigated by our use of validated algorithms for cirrhosis, has been duly reported by other investigators utilizing claims databases for postmarket safety studies involving claims data [10,58].Fourth, our data lack information on race and sociodemographic variables, factors that may moderate burden of disease and access to medical services.
Lastly, the present study involved subjects with private health T A B L E 4 Secondary and subgroup analyses for the primary composite outcome in patients with acute venous thromboembolism and chronic liver disease initiating direct oral anticoagulants compared with warfarin.Cirrhosis is a composite of compensated and decompensated cirrhosis.The primary outcome measure was a composite of hospitalization for recurrent venous thromboembolism and hospitalization for major bleeding.

2 |
M E T H O D S 2.1 | Data source This retrospective real-world cohort study was performed using administrative claims submitted to the Optum Clinformatics Data Mart database from January 1, 2010, through December 31, 2017.The Optum database contains deidentified, nationally representative data of ≥20 million enrollees of commercial and Medicare Advantage health insurance plans across the United States.Information on beneficiaries includes demographic data, hospitalization and outpatient visits, laboratory data (in a small subset of enrollees), and outpatient prescription services.This study was approved by the University of Rhode Island Institutional Review Board; the requirement for informed consent was waived due to the anonymous nature of the Optum data.

F
I G U R E 1 Study selection.DOAC, direct oral anticoagulant; VTE, venous thromboembolism.

( 15 . 9
IR/person-year) of patients on apixaban and rivaroxaban treatment, respectively, in the current study had medical encounters indicative of hospitalization for major bleeding.Despite the wellknown benefits of oral anticoagulation treatment in reducing thrombotic risk, the finding of more than 5-fold the risk of major bleeding events in the current study compared to the VTE clinical trials and other real-world studies involving VTE[5,7,[12][13][14][15]46] highlight the need for careful consideration of risks vs benefit of oral anticoagulation in these most at-risk patients.4.1 | LimitationsFirst, considering the retrospective nature of the data, our study is prone to several forms of measurement errors, including exposure and outcome misclassification.Nonetheless, the consistent findings across F I G U R E 2 Kaplan-Meier curve for the primary outcome in patients with acute venous thromboembolism and chronic liver disease by treatment groups.(A) Direct oral anticoagulants (DOACs) vs warfarin (top left), (B) apixaban vs warfarin (top right), (C) rivaroxaban vs warfarin (bottom left), and (D) apixaban vs rivaroxaban (bottom right).
a a A select number T A B L E 2 Incidence rate and effect estimates for clinical outcomes in patients with acute venous thromboembolism and chronic liver disease initiating direct oral anticoagulants vs warfarin.The primary outcome measure was a composite of hospitalization for recurrent venous thromboembolism and hospitalization for major bleeding.Clinically relevant bleeding was a component of the hospitalization for major bleeding and clinically relevant nonmajor bleeding.Incidence rate and effect estimates for clinical outcomes in patients with acute venous thromboembolism and chronic liver disease initiating a specific direct oral anticoagulant compared to warfarin or another direct oral anticoagulant.The primary outcome measure was a composite of hospitalization for recurrent venous thromboembolism and hospitalization for major bleeding.Clinically relevant bleeding was a component of the hospitalization for major bleeding and clinically relevant nonmajor bleeding.CRNMB, clinically relevant nonmajor bleeding; HR, hazards ratio; IR, incidence rate; PSM, propensity score matching; PY, person-years; VTE, venous thromboembolism.