Continued versus discontinued oral anticoagulant treatment for unprovoked venous thromboembolism: target trial emulation

Objective To compare the effect on health outcomes of continuing or discontinuing oral anticoagulants (OACs) among patients with unprovoked venous thromboembolism (VTE) after initial treatment for at least 90 days.

Design Target trial emulation.

Setting Optum Clinformatics Data Mart (Optum CDM) from 1 January 2009 to 28 Februray 2025, and Medicare fee-for-service claims from 1 January 2009 to 31 December 2022, United States.

Participants Adults with VTE aged ≥18 years (Optum CDM) or ≥65 years (Medicare) initiating OACs (warfarin or direct OACs) within 30 days after a first hospital admission with VTE without reversible provoking factors and who continued treatment for ≥90 days. Propensity score one-to-one matching was performed between patients who continued and those who discontinued treatment (ie, absence of a refilled prescription within 30 days).

Main outcome measures The primary outcomes were hospital admission for recurrent VTE (effectiveness) and major bleeding (safety). Secondary outcomes were net clinical benefit (a composite of recurrent VTE and bleeding) and mortality. Cox proportional hazards models estimated hazard ratios and generalized linear models estimated rate differences per 1000 person years. Analyses were stratified by length of initial OAC treatment (90-179, 180-359, 360-719, 720-1079, or ≥1080 days).

Results The study cohort included 30 554 propensity score matched pairs who had continued or discontinued an OAC (mean age 73.9 years, 57.0% women). After initial anticoagulation of ≥90 days, compared with those who discontinued treatment, those who continued treatment had markedly lower rates of recurrent VTE (adjusted hazard ratio 0.19, 95% confidence interval (CI) 0.13 to 0.29; adjusted rate difference per 1000 person years −25.50, 95% CI −39.38 to −11.63), higher rates of major bleeding (1.75, 1.52 to 2.02; 4.78, 1.95 to 7.61), lower mortality rates (0.74, 0.69 to 0.79; −14.31, −22.02 to −6.59), and greater net clinical benefit (0.39, 0.36 to 0.42; −21.01, −32.31 to −9.71). The greater net clinical benefit was consistent across OAC types and length of initial OAC treatment.

Conclusion Based on two US nationally representative routine care databases, continuing versus discontinuing OAC after initial anticoagulation of ≥90 days was associated with lower risk of recurrent VTE, higher risk of major bleeding, and a net clinical benefit. This observation persisted among those using OACs for at least three years after VTE.

Data sources

We used data from two large US health insurance databases: Optum’s deidentified Clinformatics Data Mart Database (Optum CDM) and Medicare fee-for-service Parts A (inpatient), B (outpatient), and D (pharmacy claims). Optum CDM includes a commercially insured US population. Medicare is a federal health insurance programme providing healthcare coverage for US individuals aged ≥65 years or those with disabilities. The databases contain deidentified, longitudinal, individual level data on healthcare use, inpatient and outpatient diagnoses, diagnostic tests and procedures, and pharmacy dispensing of drugs.

Study population and OAC use

Using a target trial emulation framework,1011 we emulated a hypothetical clinical trial to compare the risk of recurrent VTE and major bleeding among adults who continued versus discontinued OACs (ie, warfarin, apixaban, rivaroxaban, dabigatran, or edoxaban) after at least 90 days of initial treatment. The treatment needed to have been initiated within 30 days after a first recorded hospital admission with VTE (without any previous VTE) from 1 January 2009 to 28 February 2025 in the Optum CDM population and from 1 January 2009 to 31 December 2022 in the Medicare population without use of any OAC in the preceding year. The study cohort was additionally required to meet the following criteria: age ≥18 years in the Optum CDM population and ≥65 years in the Medicare population, and with at least 365 days of continuous medical and pharmacy insurance enrollment before cohort entry and with information on age or sex, or both (to ensure the patient had adequate information to determine eligibility criteria). Among this study cohort, we identified people who had discontinued OACs as those with an absence of a refill within 30 days any time after the initial 90 days of treatment. To account for potential leftover pills from previous prescriptions, the discontinuation date was set at 30 days after the end of the last prescription. For each individual who had discontinued treatment, we selected an individual who had continued treatment among those who met the eligibility criteria and were still receiving treatment, matched on days since the index hospital admission with VTE and whether the index VTE was a pulmonary embolism or deep vein thrombosis. The cohort entry (index) date was the date of discontinuation for those individuals who discontinued treatment, and the matched continued-discontinued pairs were assigned the same index date (ie, matched on cohort entry date). The baseline assessment period comprised the 365 days before (and including) the index date. We excluded those with multiple OACs on the index date; a reversible provoking factor (ie, pregnancy, pelvic or orthopedic surgery, joint replacement, a lower limb fracture,1213 use of oral contraceptives or hormone therapy, or prolonged hospital stay of >7 days) in the 90 days before the index VTE; recurrent VTE since OAC initiation; history of bleeding requiring hospital admission during the initial OAC treatment period and in the baseline assessment period; hospice care; atrial fibrillation, hypercoagulable state (anticardiolipin syndrome, antiphospholipid syndrome, primary thrombophilia, activated protein C resistance or factor V Leiden, prothrombin gene mutation, protein C deficiency, protein S deficiency, antithrombin III deficiency, and homocystinuria), or cancer (except for non-melanoma skin cancers) in the baseline assessment period; or missing information on age or sex (see supplementary appendix study protocol for definitions of these conditions and how the hypothetical trial was emulated in our data).

Outcomes and follow-up

The primary outcomes were hospital admission for recurrent VTE (effectiveness) and major bleeding (safety). The VTE1415 and major bleeding (including intracranial, gastrointestinal, and other major bleeding)1617 events were ascertained based on validated algorithms (see supplementary appendix study protocol for definition and performance). Secondary outcomes included all cause mortality and net clinical benefit defined by a composite of recurrent VTE and major bleeding. Follow-up began on the index date until the earliest of occurrence of study outcome, disenrollment from insurance, calendar end of the data, or change of index drug use status—that is, starting an OAC among those who had discontinued treatment, or discontinuing the OAC among those who had continued treatment, corresponding to an as-treated analysis).

Baseline covariates

We assessed a total of 89 characteristics using information in the baseline assessment period, including personal information, comorbidities, medication use, measures of healthcare utilization, and calendar year. The 89 covariates were selected based on clinical knowledge (consensus from KJL, DHK, DES, and KB, doctors and pharmacists on our team) and based on previous literature related to comparative safety and effectiveness of oral anticoagulants in patients with VTE.1518 We also calculated the following risk scores: VTE-BLEED18 for bleeding risk, a combined comorbidity score for overall burden of comorbidity,19 a predicted score for time within treatment range of international normalized ratio that was validated against laboratory test results,20 and a validated claims based frailty index.212223 The study protocol in the supplementary appendix defines the study variables.

Statistical analysis

The absolute standardized mean difference was used to compare clinical characteristics between those who continued versus discontinued OAC treatment, and the covariate balance was considered adequate if the absolute standardized mean difference was <0.1.24 We used 1:1 propensity score matching to adjust for confounding. The propensity score was estimated using a logistic regression that modeled the probability of OAC treatment being continued (versus discontinued) as a function of the 89 covariates. We conducted the propensity score estimation and matching within each database. In the matched cohort, we estimated hazard ratios and the corresponding 95% confidence intervals (CIs) using Cox proportional hazards models. The proportionality assumption was checked using Kaplan-Meier curves, which was satisfied in our analyses. We estimated rate differences using generalized linear models. Random effects meta-analysis was used to pool the database specific estimates.25 We conducted subgroup analyses by length of the initial anticoagulation treatment (90-179, 180-359, 360-719, 720-1079, ≥1080 days); OAC type (warfarin, apixaban, or rivaroxaban as we did not have a sufficient sample size for dabigatran (<1%) and edoxaban users (<0.03% of all people who initiated an OAC)26 to assess subgroup effects); personal characteristics (age and sex); pulmonary embolism or deep vein thrombosis as the initial VTE event; and presence of chronic kidney disease. We used the χ² statistic from the Cochran’s Q test to test for heterogeneity of treatment effect across the subgroups.27

Sensitivity analysis

We conducted several sensitivity analyses. First, we extended the refill gap to define OAC discontinuation from 30 to 60 days. Second, we conducted the intention-to-treat analysis in which we followed patients according to use of the index drug regardless of change of drug use for a maximum of 365 days. Third, to reduce residual unmeasured confounding, we used the high dimensional propensity score,28 which automatically identifies and adjusts for potential proxy confounders.2829303132 Fourth, we used the Fine and Gray model to account for the competing risk due to death.3334 Lastly, we used the E value to assess the amount of unmeasured confounding that could potentially explain away our findings.35 Analyses were conducted in the Aetion Evidence Generation Platform, version release-5.31.1.363738 Statistical computations were conducted using R version 4.3.1.

Patient and public involvement

Our study was based on an analysis of secondary data. No patients or members of the public were involved in the design, conduct, reporting, or dissemination plans of our research.

Patient characteristics

After applying the cohort eligibility criteria, we identified 70 755 individuals (51 704 in the Medicare population and 19 051 in the Optum CDM population). Figure 1 summarizes the process for cohort formation. Among these patients, after 1:1 propensity score matching, our analysis included 30 554 patients who continued OACs matched with 30 554 patients who discontinued OACs. Table 1 presents selected personal and clinical characteristics of the study cohort (see supplemental tables S1 and S2 for all characteristics). Before propensity score matching, the characteristics of patients who continued and discontinued OACs were similar (mean age 74.1 v 73.8; female 57.8% (n=20 447) v 57.1% (n=20 189), though some baseline healthcare utilization (eg, hospital admission, nursing home stays, and office visits) was lower in those who continued than discontinued treatment (eg, mean hospital stay 4.60 v 5.15 days). After propensity score matching, all covariates were well balanced (table 1).

Clinical outcomes after 90 days of initial OACs

The median follow-up time was 241 days (interquartile range (IQR) 517 days) in the continued OAC group and 174 (IQR 865) days in the discontinued OAC group, with 21 155 (69.2%) and 16 173 (52.9%) participants censored, respectively, owing to change of status for initial drug use (see supplemental table S3). In the reference group of participants who continued treatment, 25% used an OAC for ≤189 days, 50% for ≤331 days, and 75% for ≤706 days. The median duration of treatment (from initiation until discontinuation) was 357 (IQR 599) days in the continued OAC group and 105 (IQR 118) days in the discontinued OAC group. After at least 90 days of initial anticoagulation, compared with the discontinued OAC group, the continued OAC group had markedly lower rates of recurrent VTE (adjusted hazard ratio 0.19, 95% CI 0.13 to 0.29; adjusted rate difference −25.50 per 1000 person years, 95% CI −39.38 to −11.63), higher rates of major bleeding (1.75, 1.52 to 2.02; 4.78, 1.95 to 7.61), and greater net clinical benefit (ie, lower rate of the composite outcome) (0.39, 0.36 to 0.42; −21.01, −32.31 to −9.71). Among the recurrent VTE events, 74% were pulmonary embolism and the remainer were deep vein thrombosis. Mortality rates were also lower among the continued OAC group (adjusted hazard ratio 0.74, 0.69 to 0.79; adjusted risk difference −14.31, −22.02 to −6.59). Similar results were observed between the two groups for apixaban, rivaroxaban, and warfarin and recurrent VTE, whereas absolute increased bleeding risk attributed to OAC continuation (versus discontinuation) was smaller in users of direct OACs than in users of warfarin (table 2).

Clinical outcomes after predefined days of initial OAC treatment

We observed a consistent magnitude of reduced rates of recurrent VTE when comparing the continued OAC group with the discontinued OAC group after initial anticoagulation of 90-179 days (adjusted hazard ratio 0.22, 95% CI 0.16 to 0.32), 180-359 days (0.17, 0.13 to 0.23), 360-719 days (0.15, 0.11 to 0.20), 720-1079 days (0.18, 0.07 to 0.49), and ≥1080 days (0.13, 0.04 to 0.41). The association between increased rates of major bleeding among the continued OAC group was also fairly consistent, diminishing only among those with the longest period of OAC use after VTE: the adjusted hazard ratio after initial OAC treatment was 2.38 (1.88 to 3.02) for 90-179 days, 1.92 (1.56 to 2.37) for 180-359 days, 1.85 (1.34 to 2.57) for 360-719 days, 2.04 (0.96 to 4.36) for 720-1079 days, and 1.00 (0.40 to 2.48) for ≥1080 days, although the estimates for ≥720 days were based on small event numbers. The net clinical benefit was consistent on the absolute scale across different lengths of initial anticoagulation. The adjusted rate differences per 1000 person years were −14.31 for 90-179 days’ follow-up, −25.11 for 180-359 days, −23.99 for 360-719 days, −21.21 for 720-1079 days, and −25.35 for ≥1080 days (fig 2, fig 3).

Fig 2

Subgroup analysis comparing patients who continued versus discontinued OAC by number of days after initial treatment. OAC=oral anticoagulant; VTE=venous thromboembolism

Fig 3

Net clinical benefit comparing patients who continued versus discontinued OAC by number of days after initial treatment. The net clinical benefit was a composite of recurrent VTE and major bleeding. OAC=oral anticoagulant; VTE=venous thromboembolism

Subgroup and sensitivity analysis by comorbidities

We did not observe significant heterogeneity of treatment effect by subgroups of age, sex, chronic kidney disease, pulmonary embolism versus deep vein thrombosis as the index event, or VTE-BLEED score (fig 4, fig 5). Database specific results were consistent across the Optum CDM and Medicare databases (see supplemental figure S1). Results from sensitivity analyses using a longer refill gap (60 days) to define discontinuation, high dimensional propensity score, and Fine and Gray model were similar to the primary results. The 365 day intention-to-treat analysis yielded more attenuated effect estimates compared with the primary analysis, but the general trends remained consistent (see supplemental table S4). This E value analysis showed that to account for our recurrent VTE findings we would need a strong unmeasured confounder that is differentially distributed among those who continued OAC treatment versus those who discontinued by a ratio >9.5 to 1 and simultaneously associated with a 9.5-fold increased risk of recurrent VTE.

Fig 4

Subgroup analysis comparing patients who continued versus discontinued OAC after at least 90 days of initial treatment, by patient characteristics. OAC=oral anticoagulant; VTE=venous thromboembolism

Fig 5

Net clinical benefit comparing patients who continued versus discontinued OAC after at least 90 days of initial treatment, by patient characteristics. The net clinical benefit was a composite of recurrent VTE and major bleeding. OAC=oral anticoagulant; VTE=venous thromboembolism

Comparison with other studies

Although clinical guidelines recommend extending anticoagulation beyond the initial 3-6 months for patients with unprovoked VTE,45 previous randomized controlled trials were not adequately powered to provide long term bleeding outcomes for extended anticoagulation.39 Our study, based on two nationally representative cohorts in routine care, provides evidence of the long term effect of continuing OAC treatment on recurrent VTE and major bleeding, stratified by specific OAC type and the length of initial anticoagulation. Our findings for preventing VTE recurrence were consistent with previous randomized controlled trials. A meta-analysis including 17 randomized controlled trials comparing the efficacy and safety of continuing versus discontinuing anticoagulation among patients with unprovoked VTE reported that warfarin (odds ratio 0.25, 95% CI 0.13 to 0.49), rivaroxaban (0.18, 0.03 to 0.90), and apixaban (0.18, 0.04 to 0.85) were associated with a lower risk of recurrent VTE compared with placebo.39 However, these randomized controlled trials either failed to estimate the effect of major bleeding due to insufficient event numbers, or yielded imprecise estimates for major bleeding. For example, the pooled odds ratio of major bleeding risk comparing continuation of apixaban with discontinuation was 10.65 (95% CI 1.06 to 107.13) in the meta-analysis.39 In the EINSTEIN trial, extended treatment with rivaroxaban 20 mg for an additional six or 12 months after initial VTE treatment was compared with placebo in the extended treatment period; only four major bleeding events occurred in the rivaroxaban group and none in the placebo group.7 A more recent randomized controlled trial comparing extended treatment for two years versus discontinuation after the initial 90 day anticoagulation period reported a non-significant increase in major bleeding (adjusted hazard ratio 2.99, 95% CI 0.81 to 11.05), based on only three events in the group that discontinued treatment.40

Although large observational studies have compared OACs used for extended treatment of VTE,1541 real world evidence comparing extended anticoagulation versus discontinued anticoagulation for unprovoked VTE is limited. One study based on a Japanese registry without any adjustment for confounding found that continuing versus discontinuing anticoagulation at 180 days was not associated with an increased risk of bleeding.8 An Italian study without adjustment for confounding reported a risk ratio of 0.14 (95% CI 0.06 to 0.27) for recurrent VTE, comparing continuation versus discontinuation of anticoagulation at any time after initiation. This study was not powered to investigate bleeding outcomes.9 A Danish study comparing extended treatment (>365 days) versus OAC of intermediate duration (91-365 days) after an unprovoked VTE found that OAC was associated with a reduction in recurrent VTE and increased risk of major bleeding.42 None of the previous studies had sufficient power to investigate subgroup effects by OAC type or treatment duration. To address these knowledge gaps, we used two US national cohorts to investigate long term VTE and bleeding events and found a favorable net clinical benefit associated with extended OAC treatment for VTE even beyond three years of initial treatment, provided the patients tolerated the initial period of anticoagulation well.

Previous meta-analyses reported that the case fatality rate was 2-3 times higher in patients with major bleeding than those with recurrent VTE434445 after initial anticoagulation. Yet, individualized estimation of mortality needs to consider the type of VTE (ie, the mortality rate for pulmonary embolism is about twice as high as that for deep vein thrombosis4346) and the type of OAC (eg, the case fatality rate of bleeding is higher among warfarin users compared with direct OAC users4547). In our study, 74% of the recurrent VTE events were pulmonary embolism. Our results also showed that the absolute increase in bleeding risk associated with continued OAC treatment (versus discontinued treatment) was smaller among direct OAC users than among warfarin users (eg, rate difference per 1000 person years was 1.21 in apixaban users and 9.83 in warfarin users). Although our findings showed a favorable overall net clinical benefit for extended anticoagulation treatment, particularly for direct OACs, each decision of long term anticoagulation required an individualized shared decision making process.

Strengths and limitations of this study

The current study has several strengths. The large size of the cohort enabled detailed subgroup analyses by OAC type, length of anticoagulation, personal characteristics, and comorbidities. The national cohorts that we used, including older adults from the Medicare database and younger adults from commercial datasets, also enhanced the generalizability of our findings. Next, we applied a target trial emulation analytical framework with rigorous analytical strategies that have been shown to replicate the findings of randomized controlled trials successfully.10114849 We acknowledge several limitations. Our administrative claims data lacked information on over-the-counter drug use (eg, aspirin), socioeconomic status, functional status, cognitive function, body mass index, or laboratory test results, including hemoglobin values that can help define major bleeding outcomes and international normalized ratio values that can quantify the anticoagulation effects during warfarin treatment. Also, we did not have information on the reasons for discontinuing OAC. Because people could discontinue an OAC for reasons associated with a poor prognosis (eg, non-adherence, medical futility), lack of such adjustment could lead to residual confounding in favor of extended anticoagulation. For some confounders that were not measured directly in our database, we used validated proxy scores, including predicted time within treatment range1518 and claims based frailty index.212223 To further minimize unmeasured confounding, we applied a validated algorithm as a proxy to adjust for confounding, high dimensional propensity score,28 and found similar results. Moreover, we assessed the amount of unmeasured confounding that could potentially account for our findings by E value and found it unlikely.35 Also, because we did not have direct information on quality of life, it was not possible to assess the effect of OAC use on quality of life, which could be considered in future studies. We relied on dispensing data for use of OACs, which can be discordant with actual consumption in the setting of medication non-adherence. Our discontinuation definition of absence of a refill within 30 days is subject to misclassification. Reassuringly, our sensitivity analysis using a larger refill gap of 60 days to define discontinuation yielded similar results. Our outcome definition of recurrent VTE was based on a validated algorithm requiring an inpatient diagnosis of VTE in the primary position that yielded a high positive predictive value.14 Diagnosis codes in the outpatient setting or a non-primary position are more likely to be a history of VTE as opposed to incident VTE. However, requiring an inpatient diagnosis omitted milder VTE events treated in the outpatient setting. Lastly, although we used one public and one private national health insurance dataset that reflect routine delivery of care, our study cohorts include only patients with medical insurance coverage and may not fully represent the US population in terms of demographics and socioeconomic status. Therefore, our findings may not be generalizable to uninsured populations or people with other types of insurance, such as public insurance for adults with a low income and younger than 65 years.

Conclusions

Based on two national US cohorts of patients with VTE but without reversible provoking factors receiving anticoagulation for at least 90 days, we found that continuing OACs was associated with lower rates of VTE recurrence compared with discontinuing OACs, regardless of treatment duration. Extended OAC use was also associated with higher rates of bleeding, but the composite outcome of recurrent VTE and major bleeding (ie, net clinical benefit) still favored continuation over discontinuation of OACs even after initial OAC use as long as 1080 days. We found consistent results across specific OAC types. Supplementing those randomized controlled trials that lacked the power to investigate long term bleeding events, our findings showed a favorable net clinical benefit for extended anticoagulation for preventing recurrent VTE after initial anticoagulation of up to three years. These results reflect the average effect of continuing OACs and should help inform decisions on continuation of treatment, which should be individualized for each patient with an unprovoked VTE.