Introduction
In the United States (US), 2.7–6.1 million people were affected by atrial fibrillation (AFib) annually and it is projected to reach 12 million by 2050 [
1]. AFib is associated with more than 454,000 hospitalizations and 158,000 deaths each year [
2−
4]. Among patients with cancer, AFib was also associated with higher burden of adverse outcomes, such as ischemic stroke, venous thromboembolism (VTE), bleeding, and death compared with AFib patients without cancer [
5−
8].
Although the benefit of oral anticoagulants (OACs) in patients with AFib has been well established [
9], the current management of patients with AFib and cancer regarding OAC treatments remains suboptimal due to insufficient evidence [
10]. Among patients with AFib and cancer, OAC initiation was associated with a slightly reduced risk of adverse event (ischemic stroke and intracranial bleeding) compared with non-users [
11]. However, recent studies found only half of patients with AFib and cancer initiated OAC, much less than those without cancer [
11‐
14]. One of the major challenges is to determine the appropriate time when patients with AFib and cancer should start OACs to maximize the benefit of stroke prevention while minimizing the risk of bleeding. In general, OAC initiation is recommended for AFib patients with a CHA
2DS
2-VASc score ≥ 2, a composite stroke risk score of congestive heart failure, hypertension, age, diabetes mellitus, prior stroke, transient ischemic attack, thromboembolism, vascular disease and sex category [
9,
15]. However, such threshold has not been explored in patients with AFib and cancer. For example, when patient with existing cancer is newly diagnosed with AFib with low risk of ischemic stroke (i.e., CHA
2DS
2-VASc < 2), whether this patient should start the treatment immediately or wait until they reach a higher risk of ischemic stroke (i.e., CHA
2DS
2-VASc ≥ 4 or CHA
2DS
2-VASc ≥ 6). In some patient groups, anticoagulation is withheld because of a perceived unfavorable risk-benefit ratio [
16]. Since patients with AFib and cancer are at higher risk of stroke and bleeding [
5,
6], initiating OAC at low risk may be beneficial in stroke prevention, but may result in increased risk of bleeding. On the other hand, late OAC initiation may prevent risk of bleeding but increase risk of stroke in these patients. Although recent studies found that patients with AFib and cancer who had CHA
2DS
2-VASc ≥ 4 were more likely to receive OACs compared to patients with lower risk of stroke [
17], the benefit of this treatment strategy has never been explored. Determining the benefit of initiating OACs at different levels of risk of stroke is critically important to optimize the management of patients with AFib and cancer.
In this study, we assessed and compared benefits of multiple OAC initiation treatment strategies at different thresholds of risk of stroke among newly diagnosed AFib patients with cancer using the target trial framework. The target trial framework is the application of design principles from randomized controlled trials (RCTs) to the analysis of observational data to improve the quality of observational epidemiology when a comparator trial is not yet available or feasible [
18].
Discussion
Our study is the among the first to assess the benefit of OAC initiation in patients with AFib and cancer at different level of risk for stroke. First, we found that initiating OACs at higher level of CHA2DS2-VASc score (i.e., ≥ 6) is more beneficial in reducing risk of stroke among patients with AFib and cancer. OAC initiation at a lower level of CHA2DS2-VASc score might be harmful or has no effect on risk of stroke. Second, initiating OACs at any level of CHA2DS2-VASc score reduced the risk of major bleeding, with OAC initiation at higher level of CHA2DS2-VASc score being the most effective strategy. Thus, among cancer patients with new AFib diagnosis, OAC initiation may be considered for patients at high risk of stroke (CHA2DS2-VASc score at least ≥ 4) when a marginal harm on risk of stroke and a benefit on risk on bleeding are observed. In addition, among patients with advanced cancer status or low life-expectancy (i.e., lung cancer or regional/metastatic cancer), OAC should be given only to patients with CHA2DS2-VASc score ≥ 6.
OACs were underutilized in the management of patients with AFib and cancer in previous studies [
12,
17]. In fact, we found that only one in four patients initiated OACs within the first year after AFib diagnosis in this study. While current guidelines recommend a CHA
2DS
2-VASc score ≥ 2 for OAC initiation in general AFib patients [
9], this threshold may not be applicable for patients with cancer because they are at higher risk of stroke and bleeding [
5‐
8]. In this study, we found that OAC initiation at higher CHA
2DS
2-VASc score (6 or above) was the most beneficial treatment strategy. Starting OACs at lower CHA
2DS
2-VASc score may not be beneficial for stroke reduction within one year after AFib diagnosis. The treatment effects sustained after 3 years of follow-up in the sensitivity analysis. In addition, starting OAC at any CHA
2DS
2-VASc level was associated with reduced risk of bleeding compared with no initiation.
Similar to our findings, Atterman (2020) found that OAC initiation was associated with a slightly reduced risk of ischemic stroke and intracranial bleeding compared with non-users (HR = 0.90 95% CI 0.80-1.00), especially in those with moderate (HR 0.82, 95% CI 0.70–0.96) or high (HR 0.82, 95% CI 0.79–0.86) baseline CHA
2DS
2-VASc score [
11]. The authors obtained CHA
2DS
2-VASc score before AFib diagnosis (index date) and stratified the risk of stroke based on baseline CHA
2DS
2-VASc score (0: low, 1: intermediate, ≥ 2: high) [
11]. Likewise, O’Neal (2018) found AFib patients with cancer who sought for cardiologists shortly after AFib diagnosis were more likely to receive OACs and had a reduced risk of stroke and non-inferior risk of bleeding compared with those who did not [
14]. Recent studies also advocated the use of OACs in patients with AFib and cancer having CHA
2DS
2-VASc 0-2 [
43,
44]. Leader (2023) showed that 12-month cumulative incidence of arterial thromboembolism was higher in patients with the AFib and cancer compared to patients with AFib and no cancer not receiving OACs [
43]. Indeed, these studies compared the incidence of stroke and bleeding between OAC users and non-users and stratified the comparison by baseline CHA
2DS
2-VASc score or compared the risks of stroke or bleeding between patients with AFib and cancer versus patients with AFib and no cancer, but did not directly compare risk of outcomes between different OAC initiation strategies based on their CHA
2DS
2-VASc score as time-varying confounder during follow-up [
11,
14,
43]. In addition, such design is subjected to immortal time bias since patients would have been stroke-free or bleeding-free long enough to receive OACs [
45]. Our study clearly formulated a decision point where cancer patients newly diagnosed with AFib should start treatment at lower risk of stroke or wait until they reach a higher risk level. We used cloning-censoring-weighting approach to assign each patient into different treatment strategies and followed patients after their AFib diagnosis, which minimized immortal time bias by accounting for patient’s exposure to OACs and the compliance with their assigned treatment during follow-up [
26,
46].
We found that the effects of OAC initiation at different risk level based on CHA
2DS
2-VASc score were heterogeneous in several subgroups of cancer. In patients with advanced cancer such as lung cancer or regional/metastatic cancer, OAC initiation may not be beneficial or even harmful in patients with a lower risk of stroke, but beneficial only in patients with high risk of stroke. In subgroups of cancer such as breast cancer or prostate cancer, in situ/local cancer or grade I/II/III, OAC initiation at lower CHA
2DS
2-VASc score did not increase risk of stroke but decreased risk of bleeding, while OAC initiation at higher CHA
2DS
2-VASc score decreased risk of stroke and bleeding compared with no initiation. This heterogeneity can be explained by the differential risk of stroke and bleeding in patients at different stages of cancer. Indeed, patients with advanced cancer (i.e., metastatic cancer or lung cancer) are at higher risk of stroke and bleeding compared with early stage [
47,
48]. In the sensitivity analyses, starting OAC at any CHA
2DS
2-VASc score was associated with a non-inferior risk of stroke and lower risk of bleeding after excluding patients with metastatic cancer at baseline. These findings may help clinicians tailorize OAC treatment strategy in AFib patients based on their cancer characteristics.
CHA
2DS
2-VASc score has been used for more than a decade for risk of stroke stratification and OAC initiation in patients diagnosed with AFib. In this study, we used CHA
2DS
2-VASc as an indicator for OAC initiation although the tool was found not highly predictive in stroke prediction in patients with AFib and cancer [
49,
50]. CHA
2DS
2-VASc score has shown low discrimination capacity for ischemic stroke in patients with AFib and cancer than patients without cancer [
50,
51]. The major limitation of using CHA
2DS
2-VASc score is that it is not able to capture an independent risk of stroke caused by cancer, especially in patients with advanced cancer [
49,
52]. In fact, CHA
2DS
2-VASc thresholds for each treatment strategy in our study were selected based on the distribution of baseline CHA
2DS
2-VASc scores of the study sample and in prior study [
17]. Indeed, all patients enrolled in this study had CHA
2DS
2-VASc ≥ 1. In general AFib patients, OACs are recommended for those with CHA
2DS
2-VASc ≥ 2 [
9,
53,
54]. In addition, recent studies have shown patients with CHA
2DS
2-VASc ≥ 4 or ≥ 6 were more likely to initiate OACs [
17]. Therefore, our choice of CHA
2DS
2-VASc thresholds for each treatment strategy reflects multiple scenarios for OAC initiation in patients with AFib and cancer: prescribe OACs for all patients regardless their risk of stroke; prescribe OACs based on general AFib recommendations; and real-world pattern of OAC use in clinical practice. However, CHA
2DS
2-VASc score has been widely accepted among clinicians and recommended in clinical guidelines for OAC initiation decision-making [
9,
54,
55]. There is an urgent need to develop new tools for risk of stroke assessment in patients with AFib and cancer.
Our study is subject to some limitations. First, unmeasured confounding such as patients’ frailty, body mass index could not be captured by SEER-Medicare data. In addition, cancer characteristics used in the analysis such as cancer stage and tumor grade were captured at the time of cancer diagnosis rather than at the time of AFib diagnosis since the SEER registry is lack of measurements of progression of cancer characteristics (cancer stage, tumor grade) over time. We also could not control for some cancer-specific characteristics such as receptor status (ER, HER2) for breast cancer, histological type, or tumor size in our analysis since they contained large proportions of missing values. In addition, we assumed 12-month baseline period prior to AFib diagnosis was sufficient to capture patients’ baseline characteristics, therefore, measurement bias may persist. Measurement bias was also present when we measured patients’ behavioral risk factors (i.e., alcohol use disorders in HAS-BLED score) using ICD codes [
56]. Also, socioeconomic factors from Census tract were not available on an individual level. Third, residual bias could not be completely eliminated even though we used validated algorithms to define eligibility criteria and outcomes. Fourth, we did not stratify OAC initiation by type of OACs (i.e., warfarin, dabigatran, rivaroxaban) given their safety and effectiveness profile may be different. An updated meta-analysis of RCTs showed better efficacy and safety of DOACs than warfarin [
57]. while a recent observational study using SEER-Medicare data found warfarin and DOACs are equivalently safe and effective in prevention stroke and bleeding [
58]. Thus, the stratified treatment effects may be different than the marginal effects of all OACs and the benefits and risks of different OAC initiation strategies in this study may be biased depending on which type of OACs were used. Fourth, since we cloned each individuals to 5 copies, the 95% CI estimated from GEEs might be conservative due to correlation between clones. In our analysis, we could not perform non-parametric bootstrapping to obtain 95% CIs due to computational time. Fifth, our estimates may be prone to bias in the presence of extreme weights although we truncated the initial weights at 99th percentile [
59]. Using stabilized variance may reduce the variance and avoid extreme weights, but the stabilization procedures might not valid for cloning-censoring-weighting approach like our study [
60]. However, the results remain robust in the sensitivity analysis after we truncated the weights to 95th percentile. Sixth, our findings may not be generalizable beyond the target population in this study (i.e., patients with newly diagnosed cancer on existing AFib, other cancer types, or non-Medicare populations). Future studies are warranted to investigate the benefits and risks of OACs among patients with other advanced cancer such as hematological cancers due to higher risk of stroke and bleeding in this population [
61,
62].
Although our study emulated a hypothetical target trial and adopted components (i.e., inclusion/exclusion criteria, outcomes, and follow-up) from prior RCTs [
24,
25], several components were not perfectly mimicked. Specifically, RCTs removed patients platelet count < 90,000/µL, systolic blood pressure ≥ 180 mmHg or diastolic blood pressure ≥ 100 mmHg, or creatinine clearance less than 30 mL/min at the screening visit [
24,
25]. However, these lab values were not available in SEER-Medicare data. We therefore replaced these conditions with the presence of thrombocytopenia or severe renal impairments. In addition, several components were defined by clinicians’ assessment in RCTs, such as AFib definition by an electrocardiogram (ECG) document or congestive heart left failure with ventricular ejection fraction ≤ 35% [
24,
25]. Moreover, therapeutic responses and adverse events were monitored with international normalized ratio (INR) and liver-function tests, which are not available in our emulation [
24,
25]. Although non-randomization component has been criticized as the main source of bias in observational studies, it was not proven as the primary cause of inconsistency between observational and RCTs. Successful emulation without randomization has been conducted to benchmark the estimates from observational studies to RCTs and vice versa, especially during the COVID pandemic when the need of RCTs could not be met due to time constraint [
63‐
65]. In this study, randomization was assumed using a cloning-censoring-weighting approach and the adjustment of measured time-varying confounding during follow-up [
26]. It is also necessary to highlight that misspecification of time zero has been found as the major source of failure in obtaining valid causal effects in observational studies [
19,
20]. In our study, we specified time zero by aligning the time when all inclusion and exclusion criteria met, start of treatment strategies, and follow-up. Such practice removed immortal time bias and prevalent user bias from our analysis [
19,
20].
Our study has many strengths. Using the target trial emulation framework to design the study and the cloning–censoring–weighting approach, we explicitly designed a trial to answer a causal question. We included patients with newly AFib diagnosis and followed them after AFib diagnosis to remove survival bias. In addition, we further adjusted for important confounders such as cancer characteristics by the linkage between Medicare administrative claims data and the SEER registry. We pre-specified a wide range of subgroup analyses and sensitivity analyses to confirm the robustness of the main analysis. Our findings are expected to help clinicians’ decision making in optimizing OAC initiation and individualizing their decisions based on patient’s cancer characteristics.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.