Introduction
Oral anticoagulation has been established as an essential strategy to reduce stroke risk in atrial fibrillation (AF) [1-3]. Over the past decade, direct oral anticoagulants (DOACs) have become preferred over warfarin in general AF patients due to their favorable efficacy and safety profiles [4-7]. However, current guidelines often lack clear recommendations for the choice between DOACs and warfarin for patient groups excluded from pivotal DOAC trials [4-7]. Patients with prior gastrointestinal (GI) surgery represent one such important, yet understudied, population.
GI surgeries are performed in millions of individuals annually for conditions like obesity [8] and GI cancer [9,10], and the number of patients with prior GI surgery (GIS+) continues to rise [11,12]. Structural and physiological changes of the GI tract following surgery are expected to affect drug absorption, and the absorption of oral anticoagulants (OACs) in patients with AF would be no exception. Recent case series have reported reduced plasma levels of DOACs in those who underwent GI surgery [13,14], raising concerns for using DOACs in this population.
However, the impact of surgery on the comparative risk-benefit profile of DOACs versus warfarin remains largely unaddressed. Most existing studies are limited by being restricted to case series [15], lacking direct comparisons between DOACs and warfarin [16], and predominantly focused on pharmacokinetic parameters [17,18]. As a result, a recent systematic review highlighted the paucity of evidence regarding OAC use in patients GIS+ and was unable to draw definitive conclusions about the choice of anticoagulant for this population [19]. In this context, generating clinical outcome-based evidence on OAC use after GI surgery is essential, particularly for ischemic stroke prevention given the potentially reduced drug absorption after GI surgery.
Due to the lack of evidence, the 2023 guideline by the American College of Cardiology, American Heart Association, American College of Chest Physicians, and Heart Rhythm Society [1], as well as the International Society on Thrombosis and Haemostasis [20], suggest warfarin over DOACs following GI surgery for obesity, based on the feasibility of international normalized ratio (INR) monitoring. However, no studies have yet confirmed whether the monitorability of warfarin translates into superior real-world clinical benefit. This issue is particularly relevant in East Asian populations, where the overall volume of GI surgery is substantial due not only to oncologic indications [21] but also to the recent increase in non-oncologic procedures [22,23]. To address this gap, we conducted a nationwide cohort study using the South Korean claims database to evaluate the impact of GI surgery on the comparative effectiveness of DOACs versus warfarin for the prevention of ischemic stroke as well as their safety profiles in patients with AF.
Methods
Data sources
This nationwide observational cohort study utilized the Korean nationwide Health Insurance Review and Assessment Service claims database (Research project No. M20210518285), which covers nearly the entire population of the country, providing high representativeness and a longitudinal structure that enables tracking of medical histories and treatment outcomes. The data include de-identified information on general demographic details, in-hospital treatment records, disease data with International Classification of Diseases (ICD) codes, and out-of-hospital prescription records [24].
The Institutional Review Board of Seoul National University Hospital (IRB No. 2109-051-1253) approved this study. As this was a retrospective study using de-identified patient data, the requirement for informed consent was waived.
Study population
This study included adult (≥20 yr) patients with AF who were prescribed OACs—specifically, apixaban, dabigatran, rivaroxaban, edoxaban, or warfarin—between January 2015 and December 2021 (Figure 1). Eligible patients were required to be new users of OACs, defined as having no OAC prescriptions during the lookback period (January 1, 2013, to the index prescription date) for at least 2 years. A new-user design was employed to minimize potential confounding associated with prior OAC use [25]. Patients with valvular AF or those identified as having other indications for OAC use (i.e., a history of pulmonary thromboembolism, deep vein thrombosis, or recent joint replacement surgery) during the look-back period were excluded. Patients were excluded if they had only a single OAC claim that was recorded as concurrent warfarin and DOAC use, had two or more consecutive claims indicating concurrent warfarin and DOAC use at any time during follow-up, or had multiple switches between OAC types during the follow-up.
Exposures
In the primary analysis model, the details of which are provided in the “Statistical Analysis” section, the study population was stratified into four groups according to the time-varying status of OAC type (DOAC vs. warfarin) and the time-varying status of GI surgery during follow-up (Supplementary Figure 1): (1) warfarin user without prior GI surgery (WFR/GIS-); (2) warfarin user with prior GI surgery (WFR/GIS+); (3) DOAC user without prior GI surgery (DOAC/GIS-); and (4) DOAC user with prior GI surgery (DOAC/GIS+).
OAC prescriptions were identified through claims data using active ingredient codes (Supplementary Table 1). Stratification by time-varying status of GI surgery was based on the records of relevant surgical procedures (gastrectomy, bariatric surgery, small bowel resection, or colectomy) performed between 2015 and 2021. Patients were classified as GIS- until the date of surgery, and as GIS+ thereafter, or remained GIS- if no surgery occurred. Detailed definitions and codes for the procedures are provided in Supplementary Table 2.
Covariates
Baseline characteristics and covariates for adjustment included demographic characteristics, clinical comorbidities, and medication history (Supplementary Table 3). This information was determined based on data from the baseline look-back period (Figure 1A). Comorbidity burden was assessed using the Charlson Comorbidity Index (calculation method detailed in Supplementary Table 4) [26].
Study outcomes and follow-up
The primary outcome was incident ischemic stroke, defined as a hospitalization with a primary diagnosis ICD code of I63 or I64, a minimum hospital stay of 3 days, and at least one documented brain imaging study [24,26,27]. The secondary outcomes included major bleeding, all-cause death, and the composite outcome (ischemic stroke, major bleeding, and all-cause death). Major bleeding was defined as hospitalization with a primary diagnosis of GI bleeding or intracranial hemorrhage identified using the relevant ICD codes [28]. All-cause death was determined based on inpatient ICD-10 codes indicating death or clinical outcome records indicating “expired.” [29,30] Detailed definitions of the outcomes are provided in Supplementary Table 5.
Patients were followed up until the earliest of the following: the occurrence of the outcome of interest, death, treatment discontinuation (defined as the absence of OAC claims for at least 60 consecutive days), or the end of the study period (December 31, 2021).
Statistical analysis
Baseline characteristics are presented as means and standard deviation values for continuous variables and as frequencies with percentages for categorical variables. Continuous variables were compared using one-way analysis of variance or the Kruskal-Wallis test, and categorical variables were compared using the χ2 test or Fisher’s exact test, as appropriate.
To evaluate the risk of outcomes by OAC type and GI surgery status, we used the inverse probability of treatment weighting (IPTW)-weighted Fine-Gray subdistribution hazard model, with time-varying covariates as the primary analysis model. The primary analysis model used IPTW to address baseline imbalances and estimate less biased associations by leveraging information from the entire dataset [31]. Propensity scores, representing each subject’s probability of being assigned to each of the four stratified groups based on baseline characteristics, were estimated using a multivariable logistic regression model. For initial group assignment in the propensity-score model, patients were classified as warfarin users if they exclusively received warfarin from 2015 to 2021, and as DOAC users if they were prescribed at least one DOAC (apixaban, rivaroxaban, dabigatran, or edoxaban) during the study period, given that DOACs were introduced into clinical practice later than warfarin. Patients who underwent any GI surgery during follow-up were classified as GIS+, whereas those who did not were classified as GIS-. These baseline classifications were used for propensity-score construction. In contrast, in the hazard models, OAC type and GI surgery status were treated as time-varying covariates to account for their temporal effects. Stabilized weights were calculated by multiplying the inverse of the propensity scores with the marginal probability of the corresponding group. Covariate balance was assessed using standardized mean differences, with a value of ≤0.1 indicating adequate balance.
In the IPTW-weighted population, a Fine-Gray subdistribution hazard model incorporating both OAC type and GI surgery status as time-varying covariates was used to estimate the hazard ratios (HRs) for each outcome. The Fine-Gray subdistribution hazard model was utilized to account for death as a competing event for each outcome, except for all-cause death and composite outcome. To verify the proportionality assumption for the subdistribution hazard in the Fine-Gray model, time-by-covariate interaction terms were included in the preliminary analysis, and no significant violation was detected. This time-varying covariates framework enabled valid comparison of anticoagulants before and after GI surgery.
To minimize residual confounding, further adjustment for all variables included in the propensity score model was performed in addition to IPTW to estimate adjusted weighted HRs, constituting a doubly robust approach [32]. Both adjusted and unadjusted weighted HRs were reported. Incidence rates (IRs) for clinical outcomes were illustrated using cumulative incidence curves and presented as events per 100 person-years at risk during follow-up.
With this primary analysis model, we evaluated the comparative risks of outcomes between groups using a two-step approach. First, we conducted a 4-group comparison (WFR/GIS-, WFR/GIS+, DOAC/GIS-, and DOAC/GIS+), with WFR/GIS- as the reference group. Second, we performed an interaction-term analysis to enable pairwise comparisons between specific groups. A multiplicative interaction term between OAC type and GI surgery status (OAC×GI surgery) was incorporated into the 4-group comparison model to assess (1) whether the effect of GI surgery differed by OAC type and (2) the relative outcome risk of DOACs versus warfarin specifically among patients GIS+. The following pairwise HRs were derived: GIS+ versus GIS- among warfarin users, GIS+ versus GIS- among DOAC users, and DOAC versus warfarin within GIS+ status. All statistical analyses were performed using SAS Enterprise Guide version 7.1 (SAS Institute Inc., Cary, NC, USA), and statistical significance was defined as a P<0.05.
Sensitivity analysis
For the sensitivity analysis, we repeated the 4-group comparisons and interaction-term analyses using three alternative models that differed from the primary analysis in their use of IPTW and time-varying covariates. Alongside the primary analysis, which utilized both IPTW and time-varying covariates, we fitted the following models: (1) a Fine-Gray model with time-varying covariates without IPTW (Model 2); (2) an IPTW-weighted Fine-Gray model without time-varying covariates (Model 3); and (3) a Fine-Gray model without either IPTW or time-varying covariates (Model 4).
Exploratory analysis: risk of ischemic stroke and major bleeding by GI surgery type
As an exploratory analysis, we stratified the GI surgery group by anatomical location into upper and lower GI surgeries (Supplementary Table 2) and conducted an interaction-term analysis to estimate pairwise HRs for ischemic stroke and major bleeding. These analyses were performed using a Fine-Gray model with time-varying covariates without IPTW.
Results
Study population and follow-up
Among the 418,963 new OAC users with AF aged over 20 years, 388,214 met the inclusion criteria (Figure 1B). Baseline characteristics before IPTW adjustment are summarized in Supplementary Table 6. The mean age of the cohort was 71.9 years, with 42.7% (n=165,752) being male. The mean CHA2DS2-VASc score was 3.5, and the mean Charlson Comorbidity Index was 2.3. A total of 32,820 (8.5%) patients were warfarin users, 355,394 (91.5%) were DOAC users, and 6,907 (1.8%) had undergone GI surgery. During follow-up, 17,372 patients (4.5%) switched between DOACs and warfarin (3,520 from DOACs to warfarin and 13,852 from warfarin to DOACs). After IPTW adjustment, baseline characteristics were well balanced across the four groups (WFR/GIS-; WFR/GIS+; DOAC/GIS-; DOAC/GIS+), with standardized mean differences of less than 0.1 (Table 1). The propensity score distributions across four groups and the distribution of stabilized inverse probability of treatment weights are presented in Supplementary Figure 2 and Supplementary Table 7, respectively.
The total follow-up time was 667,159 person-years, with a mean±standard deviation follow-up period of 1.72±1.77 years. By group, mean follow-up durations were 2.01±2.22 years (WFR/GIS-), 1.39±1.76 years (WFR/GIS+), 1.69±1.69 years (DOAC/GIS-), and 1.22±1.38 years (DOAC/GIS+).
Primary outcome: risk of ischemic stroke
During follow-up, 12,512 (3.2%) patients experienced ischemic stroke. In the 4-group comparison, the WFR/GIS+ group had a significantly higher risk of ischemic stroke (adjusted HR [aHR]: 2.32; 95% confidence interval [CI] 1.17-4.62) compared to the WFR/GIS- group (reference). On the other hand, DOAC/GIS-group had a lower risk of ischemic stroke (aHR: 0.83 [95% CI 0.79-0.88]), and DOAC/GIS+ group maintained a comparable risk (aHR: 0.81 [95% CI 0.63-1.02]) (Table 2) compared to the reference group. The weighted cumulative incidence curves for ischemic stroke are presented in Figure 2A.
In the interaction-term analysis, the effect of GI surgery on the risk of ischemic stroke significantly differed by OAC type (Pint=0.018) (Figure 3). While warfarin users demonstrated a significant increase in ischemic stroke risk in GIS+ status compared to GIS- status (aHR: 2.32 [95% CI 1.17-4.62]), no such significant change was observed among DOAC users (aHR: 0.97 [95% CI 0.77-1.22]). Under GIS+ status, DOAC users were associated with a lower risk of ischemic stroke than warfarin users (aHR: 0.35 [95% CI 0.17-0.72]).
Secondary outcomes: risk of major bleeding, all-cause death, and composite outcome
In the study population, 11,906 (3.1%) patients experienced major bleeding, 29,408 (7.57%) died from all causes, and 48,920 (12.6%) experienced the composite outcome. In the 4-group comparison, compared to the WFR/GIS- group (reference), DOAC/GIS- group had significantly lower risks of all-cause death (aHR: 0.66 [95% CI 0.63-0.69]), composite outcome (aHR: 0.75 [95% CI 0.69-0.81]), and a comparable risk of major bleeding (aHR: 0.90 [95% CI 0.68-1.17]) (Table 2). Similarly, DOAC/GIS+ group demonstrated no significant increase in the risks of major bleeding (aHR: 0.86 [95% CI 0.64-1.18]), all-cause death (aHR: 1.03 [95% CI 0.92-1.16]), and the composite outcome (aHR: 0.94 [95% CI 0.83-1.06]) compared to the WFR/GIS- group. The WFR/GIS+ group also showed no significant differences in these risks (major bleeding aHR: 0.53 [95% CI 0.20-1.44]; all-cause death aHR: 1.28 [95% CI 0.86-1.90]; composite outcome aHR: 1.23 [95% CI 0.86-1.77]) compared to the WFR/GIS- group (Table 2 and Figure 2B-D).
The interaction-term analysis was also performed for major bleeding, all-cause death, and composite outcome (Figure 3). No significant interaction was observed for these outcomes, indicating that GI surgery had no differential effect between warfarin and DOAC users, with risks remaining similar or only marginally increased after surgery. In the GIS+ status, none of the outcomes showed a significant difference in risk between DOAC and warfarin users.
Sensitivity analysis
In sensitivity analyses using Models 2 (Fine-Gray model with time-varying covariates without IPTW), 3 (IPTW-weighted Fine-Gray model without time-varying covariates) and 4 (Fine-Gray model without either IPTW or time-varying covariates), the results were generally consistent with those of the primary analysis for both ischemic stroke risk and secondary outcomes, across the 4-group comparisons (Supplementary Tables 8-10 and Supplementary Figures 3-5) and interaction-term analyses (Supplementary Figures 6-8).
Exploratory analysis: risk of ischemic stroke and major bleeding by GI surgery type
In the upper GI surgery stratum, the GIS+ group had a higher risk of ischemic stroke than the GIS- group among both warfarin users (aHR: 2.36 [95% CI 1.40-4.01]) and DOAC users (aHR: 1.44 [95% CI 1.18-1.75]) (Figure 4). Although the magnitude of risk increase appeared smaller among DOAC users, the differential effect of GI surgery by OAC type did not reach statistical significance (Pint=0.080). Within the GIS+ status, DOAC users had a lower risk of ischemic stroke than warfarin users. In contrast, in the lower GI surgery stratum, ischemic stroke risk did not differ between the GIS+ and GIS- status for either OAC type, nor did it differ between DOAC and warfarin users within the GIS+ status.
For major bleeding, no significant differences were observed across any pairwise comparisons in the upper GI surgery stratum. In contrast, in the lower GI surgery stratum, DOAC users had a significantly higher risk of major bleeding in the GIS+ group than in the GIS- group (aHR: 2.00 [95% CI 1.67-2.40]), whereas no such difference was observed among warfarin users (aHR: 0.75 [95% CI 0.36-1.58]). Within the GIS+ status, major bleeding risk did not differ significantly between DOAC and warfarin users. Event numbers and crude IRs for the exploratory analysis are presented in Supplementary Table 11.
Discussion
In this nationwide observational cohort study of AF patients, 4-group comparison stratified by OAC type and GI surgery status showed that only the WFR/GIS+ group had a significantly higher risk of ischemic stroke compared to the WFR/GIS- group. In the interaction-term analysis, the effect of GI surgery on stroke prevention varied by OAC type: ischemic stroke risk increased after GI surgery among warfarin users, whereas it remained comparable among DOAC users. The effect of GI surgery on the risk of major bleeding, all-cause death, and the composite outcome did not differ between DOACs and warfarin. After GI surgery, DOACs demonstrated lower ischemic stroke risk and comparable risks for major bleeding, all-cause death, and the composite outcome. Exploratory analyses indicated potential heterogeneity by surgical location, with the favorable ischemic stroke risk profile of DOACs appearing more pronounced following upper GI surgery.
Since both DOACs and warfarin are primarily absorbed in the upper GI tract [33], the altered GI tract may affect absorption of both OAC types. Indeed, previous pharmacokinetic data have demonstrated that DOAC exposure can fall by 29%-90% when administered directly to distal GI segments [34-36], and warfarin resistance cases have been reported in short bowel syndrome [37]. However, despite the increasing number of patients GIS+, clinical evidence in this population is largely limited. While some studies have reported reduced DOAC plasma levels after certain bariatric procedures, data are insufficient to assess its impact on stroke prevention and other clinical outcomes [19]. Moreover, most previous studies included patients receiving OACs for venous thromboembolism. Although there is one retrospective matched-cohort study that compared DOACs with warfarin in AF, it was limited by a short follow-up duration and few events, leading to inconclusive results [38]. Evidence for GI surgery outside the context of obesity—such as in patients with GI cancer—remains even more limited [39].
Nevertheless, some scientific societies recommend warfarin over DOACs for certain GI surgery populations given the availability of INR monitoring [1,20,33,38,40]. However, recommending warfarin without supporting evidence, despite its need for frequent dose adjustments, may impose an unnecessary burden after surgery. The markedly small size of the WFR/GIS+ group in this cohort further highlights limited real-world applicability of this recommendation.
Contrary to expectations based on INR monitoring, we observed a significant increase in ischemic stroke risk after GI surgery among warfarin users. This likely reflects difficulties in maintaining therapeutic anticoagulation after surgery due to various complicating factors including reduced absorption, vitamin K deficiency, and decreased food intake. The long-term increase in warfarin dose requirements after surgery may further contribute to subtherapeutic anticoagulation [41-43]. The rise in stroke risk without a significant change in major bleeding also supports potential subtherapeutic warfarin exposure following surgery.
In contrast, the ischemic stroke risk appeared comparable or increased to a lesser degree after GI surgery among DOAC users. This likely reflects their stable pharmacokinetic profile, which may contribute to maintaining consistent efficacy across a broad range of plasma concentrations [44]. Prior evidence showing their effectiveness in conditions such as chronic kidney and liver disease further supports this characteristic [2,45].
The exploratory analyses suggested that these profiles may differ by the location of surgery. With upper GI surgery, ischemic stroke risk increased for both OAC types, but more prominently among warfarin users. In contrast, after lower GI surgery, neither OAC type showed a significant change in ischemic stroke risk, whereas major bleeding increased only among DOAC users. This increase is likely driven by GI bleeding, which constituted a higher proportion of major bleeding after lower GI surgery (76.4%) and showed the highest IR for DOAC users after lower GI surgery. These differences by surgical location may stem from distinct pharmacologic characteristics of warfarin and DOACs. Unlike warfarin, unabsorbed DOACs may exert local anticoagulant effects within the intestinal lumen [46]. Because the lower GI tract has a longer transit time, resection of this segment may further increase residual luminal drug concentrations, thereby increasing bleeding risk. Nevertheless, within GIS+ status, DOACs showed favorable or comparable outcomes in both GI surgery strata. Taken together, these findings suggest that DOACs may remain a reasonable anticoagulation option after GI surgery, with potentially more favorable profile after upper GI surgery.
Our study has several limitations. First, as an observational study, potential confounding due to covariate imbalance should be considered, particularly given the higher prevalence of cancer in the GIS+ group. To address this, we applied IPTW using a comprehensive set of covariates, including the Charlson Comorbidity Index to reflect overall morbidity burden and mortality risk [47]. Doubly robust estimation was also used to further mitigate residual imbalance. Nevertheless, the influence of unmeasured confounders may persist and should be considered when interpreting the results. Second, interpretation of the IPTW results should consider the potential for extreme weights, given the small size and low number of outcome events in the WFR/GIS+ group, as well as the initial covariate imbalance. We therefore used stabilized weights [31] and performed sensitivity analyses without IPTW, with consistent findings across models supporting the robustness of our results. Also, inspection of the distribution of stabilized weights, including upper percentiles, did not indicate undue influence of a small number of individuals on the overall estimates. Third, ischemic stroke etiology and causes of death were unavailable in our data, precluding the isolation of outcomes directly attributable to GI surgery or anticoagulant type. However, rigorous efforts to balance available covariates that could influence outcomes likely partially mitigated this limitation. Fourth, the inclusion of only Asian patients may limit generalizability, although the scarcity of evidence in this clinical setting highlights the value of our findings as foundational research [13,48]. Finally, differences across DOAC types and detailed GI surgery types could not be fully evaluated. Distinguishing upper, small-bowel, and lower GI surgery strata would provide more tailored information, and additional procedural granularity, such as differentiating bariatric from non-bariatric surgery or total from partial resections, may further enhance clinical insight. However, the imbalance in group sizes and the limited number of events in the WFR/GIS+ group constrained stratified analyses, resulting in exploratory analyses restricted to two surgery strata without IPTW. These exploratory findings highlight the need for future studies that allow more comprehensive multistrata comparisons across DOAC and GI surgery types.
Conclusions
In patients with AF, comparative effectiveness and safety data between DOACs and warfarin across GI surgery status suggest that DOACs may remain a reasonable anticoagulation option for those GIS+. The risk-benefit profile may vary by the anatomical site of surgery, with potentially a more favorable profile of DOACs after upper GI surgery. This study provides real-world evidence addressing the current lack of guidance on anticoagulant use after GI surgery. Further research is warranted to refine tailored anticoagulation strategies in this setting.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
Print
