Real‐world evidence that among atrial fibrillation patients warfarin is associated with reduced nonelective admissions compared with those on DOACs

Abstract Background Randomized trials show inconsistent estimates on risks of direct‐acting oral anticoagulants (DOACs) versus warfarin in bleeding and mortality for atrial fibrillation (AF) patients. Trials are confounded by additional DOAC adherence support, while warfarin has a low time in therapeutic range. Few real‐world studies compared emergency hospitalization risk between DOAC and warfarin users in AF. This study aimed to determine emergency hospitalization risk for AF patients on DOACs or warfarin in real‐world settings. Methods A tapered‐matched real‐world cohort extracted data from 412 English general practices' primary care records linked with emergency department (ED) and hospitalization data from the ECLIPSE database. AF patients with new DOAC or warfarin prescriptions were included. The primary outcome was all‐cause ED attendance; the secondary outcomes were ED re‐attendance, nonelective hospitalization, and rehospitalization within 12 months. Weighted Cox regression estimated relative risk difference. Results 39 201 DOAC patients were matched with 13 145 warfarin patients. DOAC patients had a 25% higher likelihood of attending ED (odds ratio 1.25; 95% confidence interval [CI] 1.01–1.55). DOAC use also associated with higher ED re‐attendance, nonelective hospitalization, and rehospitalization within 12 months: 1.41 (95% CI 1.00–1.98), 1.26 (1.00–1.57), and 1.54 (1.01–2.34), respectively, with p‐values < .05. Conclusions DOACs for AF thromboprophylaxis are associated with the increased risk of ED attendance, recurrent hospitalization, and numerical rise in ED re‐attendance and first nonelective hospitalization compared to warfarin. However, these real‐world data cannot establish if this difference results from medication adherence, lack of regular DOAC clinic monitoring, unmeasured confounders, or fundamental agent efficacy disparities.


| INTRODUCTION
It has been estimated over 1.2 million people in the UK have been diagnosed with atrial fibrillation (AF) and thousands remain undiagnosed. 1Compared with those without AF, people with AF have an estimated three to five times higher risk of ischemic stroke.
Up to 30% of AF patients are admitted to hospital due to ischemic stroke with a higher risk of fatal outcome. 2 The risk of AF-related stroke is attenuated by anticoagulation, such that stroke rates of people with AF and adequate anticoagulant are close to those without recorded AF. 3 Warfarin has been the anticoagulant of choice for >60 years.
When appropriately dosed, to an international normalized ratio (INR) of between 2 and 3 these are very effective, however when the time in the therapeutic range falls below 65%, these agents provide negligible benefit when compared with aspirin, which itself is not superior to placebo in disease prevention.The utilization of directacting oral anticoagulants (DOACs), including dabigatran, rivaroxaban, and apixaban, has risen rapidly, as they reduced laboratory and primary care attendance.Central organizations such as the Royal College of General Practitioners issued guidance encouraging widespread shifting from warfarin to the DOACs. 4Their use has surpassed the use of warfarin due to their fixed dose and lower requirements for monitoring 5 and multiple studies inferring superior safety data. 6ACs are now the most expensive medications utilized in primary care prescribing with procurement costs approaching £1 billion per annum. 4LIPSE (Equity of Care Insights for Patient Safety & Engagement) is a centralized database provided for the NHS to provide risk stratification longitudinally tracking data from over 20 million patients to provide prescribing safety alerts for GPs in England.The system contains over 10 billion rows of patient data updated each week and is centrally contracted by NHS Digital to provide safety algorithms for over 2500 GP surgeries. 7 has a central risk stratification system called PRISM (Patient Risk Identification through Statistical Modeling).This system runs a mixture of iterative calculations and machine learning to track patient outcomes in relation to phenotypes, monitoring, and medications.It claims to automatically adjust for over 2000 confounding parameters and runs detailed analytics against all medications listed as requiring enhanced monitoring in primary care. 8ACs are one of the groups within this group, and it is suggested that all patients on these agents have three monthly reviews. 9Eclipse reports that monitoring levels within primary care were less than 50% compliant with the current guidelines during the evaluation period and undertook enhanced safety analytics to assess the impact.The system reported finding a 1.42 (confidence interval [CI] 1.14-1.60)fold risk of nonelective admissions for AF patients on DOACs compared with those on warfarin.
This study was designed to compare the risks of attendance at the emergency department (ED) and nonelective hospitalization in the first 12 months of DOAC and warfarin prescription in people with AF.We have used a recently developed statistical approach: tapered multivariable matching. 10,11This allows us to examine the extent of the observed difference in outcomes between DOAC and warfarin users and more importantly, to investigate how potential confounders relate to the risk differentiation.In tapered matching, we sequentially match the DOAC user group to the warfarin user group with an increasingly comprehensive set of variables.As we incrementally match the DOAC user group and the warfarin user group, we can directly observe how the matched cohorts change both in terms of risk of outcome and in terms of unmatched covariables.

| Data sources
A large UK primary care electronic health record database from eight participating clinical commissioning groups (CCGs) in England covering 412 practices was used for this study. 12All included practices were linked at the patient level to hospital admission data (Secondary User Service [SUS] data).The covariates were defined by Read codes recorded in the electronic health records at primary care settings 13,14 and the outcomes were defined by International Classification of Diseases, 10th revision codes recorded in SUS data (codes list is accessible by reasonable request via corresponding author). 15,16The use of this database in this manner was approved by the South West -Exeter Research Ethics Committee (REC reference: 17/SW/0001).
The study period ran from April 1, 2017 to December 31, 2018.
All patients with AF newly prescribed the oral anticoagulants warfarin, dabigatran, rivaroxaban, or apixaban, and aged from 18 to 99 years at the date of study entry, were eligible for enrollment.The entry date was defined as the date of the first prescription of any of the anticoagulant drugs.To facilitate a direct comparison between new users of DOACs against new users of warfarin, and to reduce the impact of indication bias, patients were excluded if they had any anticoagulant prescription in the prior 12 months before the entry date.To ensure the quality of data, patients were also excluded if they had less than 12 months of registration history before entry.

| Matching method
This study used a tapered matching method to generate a series of matches for each comparison of DOAC and warfarin. 17For the DOAC user group, we performed six matches that constructed sets of pairs of warfarin users as shown in Supporting Information: Figure 1.First, demographic factors match paired patients by their age at incident anticoagulant prescription, and gender.Second, the DOAC user group was match-controlled (warfarin user group) for all demographic factors and clinical measurements (body mass index and systolic blood pressure).Third, the DOAC user group were matched with the warfarin user group for all variables in the first two matches as well as routine blood test results (international normalization ratio [INR], HbA1c, and total cholesterol).Fourth, the DOAC user group were matched with the warfarin user group for all variables in the first three matches as well as prior bleeding or replated comorbidities (any bleeding, upper GI bleeding, hematuria, haemoptysis, chronic renal disease, alcohol dependence, chronic liver disease or pancreatitis, chronic obstructive pulmonary disease, dyspepsia, esophageal varices, peptic ulcer, knee/hip surgery, and hip fracture) and cardiovascular disease (CVD) subtypes (valvular heart disease, congestive cardiac failure, coronary heart disease, venous thromboembolism, ischemic stroke/TIA, diabetes, and hypertension).Fifth, the DOAC user group was matched with the warfarin user group for all variables in the first four matches as well as prescriptions potentially relevant to bleeding event (antibiotics, anticonvulsants, corticosteroid, antidepressant, antiplatelet, NSAIDS, and proton pump inhibitor) and prescriptions relevant to management of CVD (antihypertensive, antidiabetes, and statin).Matched variables were defined by established Read codelists (codelists available from www. keele.ac.uk/mrr).
In each step of matching, we used a coarsened exact matching (CEM) algorithm involving a monotonic imbalance reducing matching method, which means that the balance between the treated and control groups is chosen by ex-ante user choice rather than discovered through the usual laborious process of "checking after the fact, tweaking the method, and repeatedly re-estimating." 18tients in both the DOAC and warfarin user groups matched on Step-5 were retained (Supporting Information: Figure 1).
Via CEM we restricted the comparison of DOAC and warfarin user groups to areas of common support, that is, sufficient overlap between the two groups, on the above key factors in the proton pump inhibitor five-steps, coarsened using the default Sturges measure of bin size. 19After excluding patients (Supporting Information: Figure 1) who were off common support, we then used Entropy Balancing 20 to efficiently minimize differences in the distribution of matching variables between DOAC and warfarin user groups.
Entropy balancing involves maximum entropy reweighting the matched sample in each matching step to key target moments (mean, variance, and skewness).For continuous matching variables, all three moments should be met; for binary variables, the only target moment is the mean as it is only sufficient to match higher moments (variance and skewness).
Weighted logistic regression, incorporating matching weights estimated from each matching step by entropy matching, was applied in each matching step.This provided an estimate of the association between DOAC use and risks of outcomes, with warfarin as the reference group. 20Data on each variable were missing in <6% of eligible cohort members.Based on the worst scenario of 5% of patients with ≥1 missing data, five imputed data sets were created for multiple imputations with chained equations, and estimations were made by Robin's rule. 21The analyses were conducted using Stata/ MP, version 17.0 (StataCorp LLC).Statistical significance was set at two-tailed p < .05.

| RESULTS
We identified 41 755 people with AF and with incident DOAC prescription and 13 369 people with AF with incident warfarin prescription between 2017 and 2018.Table 1 and Figure 1 demonstrate that the matched variables were quite different between DOAC users and warfarin users.As expected, DOAC users had a lower INR than those on Warfarin.Those prescribed DOACs also had a past medical history demonstrating fewer prior bleeding events or CVD comorbidities compared with warfarin users, suggesting there was a propensity to use DOACs in individuals with a lower risk of complications.
By coarsened exact matching, 39 201 incident DOAC users with AF were matched with 13 145 warfarin users with AF via five matching steps (Supporting Information: Figure 1).After coarsened exact matching, the matched variables tended to be closer (Table 1, Figure 1, and Table 2).In particular, matched variables were very similar after samples were weighted by entropy matching (Figure 1 and Table 2) in terms of mean, variance, and skewness.
The rates of each outcome in unmatched cohorts and matched cohorts for incident DOAC users and incident warfarin users are

| DISCUSSION
We have demonstrated that in a real-world setting, people requiring thromboprophylaxis for AF and treated with DOACs had an increased risk in ED re-attendance, nonelective hospitalization, and rehospitalization over the first 12 months compared to an initial prescription of warfarin.This was true for both initial attendance to ED, recurrent visits to ED, and recurrent unplanned hospitalizations.This is contrary to the reports of randomized controlled trials, that suggest DOACs should be associated with a lower risk of either thromboembolic events, serious adverse bleeding events, or both.
This was based on an analysis of routinely collected primary care electronic health records linked with hospitalization data in a population in the UK healthcare system.
Our results contrast the outcomes of the randomized controlled clinical trials.A recent network meta-analysis of these RCTs showed that DOACs are safer than warfarin in relation to major and intracranial bleeding. 22However, interestingly, the same analysis presented a higher risk of gastrointestinal bleeding with dabigatran, edoxaban, and rivaroxaban than with warfarin.In addition, edoxaban (30 mg and 60 mg two times per day) significantly increased the risk of clinically relevant bleeding compared with warfarin. 22The RCTs have not always demonstrated clear-cut benefits.Two studies have reported an elevated risk of mortality with lower doses of apixaban and rivaroxaban compared with warfarin. 23,24Another Danish study | 1549 showed decreased mortality for apixaban, 25 but QResearch showed equivalent risk to warfarin for such patients.For the outcome of ischemic stroke, both the Danish study and QResearch showed DOACs were equivalent to warfarin. 25ere are several potential explanations for our observations of increased healthcare utilization amongst DOAC users.Amongst these include a difference in the adherence to the drugs in a real-world setting, a relative difference in the efficacy of these agents in the UK The rates of outcomes in the comparison cohorts.F I G U R E 2 Adjusted odds ratios for association between DOAC (reference to warfarin) and risks of all-cause nonelective (re)hospitalization and all-cause ED reattendance.Naive model weighted for age at incident anticoagulant prescription, gender; model (i) weighted for all adjusted variables in model naïve model plus body mass index, systolic blood pressure; model (ii) weighted for all adjusted variables in model (i) plus total cholesterol, HbA1c, international normalized ratio; model (iii) weighted for all adjusted variables in model (ii) plus prior bleeding or related comorbidities (any bleeding, upper GI bleeding, hematuria, haemoptysis, chronic renal disease, cancer, alcohol dependence, chronic liver disease or pancreatitis, chronic obstructive pulmonary disease, dyspepsia, esophageal varices, peptic ulcer, knee/hip surgery, and hip fracture) and cardiovascular disease subtypes (valvular heart disease, congestive cardiac failure, coronary heart disease, venous thromboembolism, ischemic stroke/TIA, diabetes, and hypertension); model (iv) weighted for all adjusted variables in model (iii) plus prescriptions relevant to bleeding (antibiotics, anticonvulsants, corticosteroid, antidepressant, antiplatelet, NSAIDS, and proton pump inhibitors) and prescription relevant to management cardiovascular diseases (antihypertensive, antidiabetes, and statin).DOACs, direct-acting oral anticoagulants; ED, emergency department; GI, gastrointestinal; NSAIDS, nonsteroidal anti-inflammatory drugs; TIA, transient ischemic attack.
healthcare environment, or simple residual confounding in the populations compared.
As DOACs do not require routine blood testing, a burden on both patients and prescribers, they have been increasingly prescribed to replace the traditional anticoagulant, warfarin.However, with that regular blood test comes a process to reinforce the importance of good adherence.Adherence and persistence with DOACs, however, do not achieve the same standard in a real-world setting, 26  were 64%, 27 62.2%, 28 and 55%, 29 respectively, suggesting that none of these trials met the minimum acceptable standard established by the National Institute of Clinical and Healthcare Excellence. 30A TTR of less than 65% has been suggested to be no more effective than simple antiplatelets in preventing events in those with nonvalvular AF. 31 Further, aspirin alone is little better than placebo at preventing stroke, indeed in some studies has been suggested to increase complications without affording benefit.In the UK, the average TTR ranges between 71% for those monitored in primary care to 78% for those attending secondary care clinics. 32,33The TTR for the warfarin cohort in the data set utilized for this study was estimated to be 70.8%.This estimation was derived from the monthly percentage of patients whose INR was between 2.0 and 3.0.
A subgroup analysis of the RE-LY study compared outcomes from those treated with dabigatran and those treated with warfarin stratified by the mean TTR in each center.It demonstrated incremental reductions in benefit, as TTR increased such that there were clear advantages from DOAC when TTR was <65% (event rate was 7.9%, or 9.7% and for Warfarin for TTR 57%-65% or <57%, respectively, compared with 7.0% and 6.7% for dabigatran 150 mg in the same centers).However, when TTR was greater than 65%, such as we see in the UK routine practice warfarin was superior to DOAC at both preventing events and avoiding adverse effects (event rate 6.6% and 6.4% for TTR 66%-72% and >72%, respectively, compared with 7.2% and 6.8% in the same centers). 34Given that these latter TTR values are more in keeping with usual practice in the UK, it is should not come as a surprise that warfarin was superior to DOAC at reducing ED admissions and unplanned hospitalizations.It remains possible that this is a product of residual confounding, however, this is unlikely.
6][37][38] For instance, a US cohort extracted from a commercial healthcare database found that patients with valvular AF who were new users of DOACs had lower risks of ischemic stroke or systemic embolism and major bleeding compared with new users of warfarin. 35Similarly, a Japanese cohort extracted from an AF registry found that warfarin and DOACs exhibited equivalent 3-year stroke and all-cause mortality rates, but DOACs showed a reduced risk of major bleeding. 38In a Norwegian national AF registry cohort, it was revealed that all DOACs were similarly effective as warfarin in preventing ischemic stroke, TIA, or systemic embolism, with similar safety in terms of bleeding. 36other US cohort based on US Centers for Medicare & Medicaid Services data found that all DOACs had a lower risk of major adverse cardiac events compared to warfarin. 37However, it is important to note that the outcomes explored in previous studies differ from those in the current study, as the current study focused on the short-term risk of ED visits and nonelective (re)hospitalizations. Additionally, discrepancies in study findings may also be attributable to differences in population profiles (e.g., ethnicity, age, and gender) and the validity of electronic health records (e.g., misclassifications in the definition of covariates and outcomes due to different coding systems 39 ).
Furthermore, the variation in adherence to DOACs across different studies could also contribute to the discrepancy in the findings.
Future validation studies using cohorts with similar population structures and adjusting for adherence to DOACs are warranted.
A key strength of this work was the application of a novel, tapered matching method to form a 'quasi-trial' sample to compare the risk of hospitalization for the three outcomes between incident DOAC and warfarin users with AF.Through tapered matching, we were able to transparently examine how differences in specific sets of confounders contributed to the risk of ED and nonelective hospitalization.By sequentially matching and weighting for differences in demographic characteristics, clinical measurements, prior bleeding and relevant comorbidities, CVD, prescriptions potentially relevant to bleeding, and CVD management, we were able to compare the risks of outcomes after each match between new DOAC and the new warfarin users.This, in turn, allowed the importance of potential confounders (e.g., perceived risks) to be assessed and adjusted for.This prospective cohort incorporating patients with AF was derived from a large primary care database in England that was linked with hospitalization data which have been shown to be of good quality in terms of representativeness, coverage, validity, and consistency in records of comorbidities and prescriptions. 40Hospitalization data used in the study were complete as SUS data captures all hospitalization information for patients and its recorded outcomes, which has also been proven to have good validity, including those experiencing events outside of the CCG catchment. 41There are some limitations in this study.First, as the sample size was restricted, instead of exactly matching each relevant bleeding and CVD comorbidity and prescription, we have matched the number of bleeding and CVD comorbidities and associated prescriptions.
However, we have compared the bleeding and CVD comorbidities, each antihypertensive and each antidiabetes prescription after YU ET AL.
| 1551 matching with no significant difference identified between DOAC and warfarin user groups (Table 2).Second, the time-varying/ cumulative effects on outcomes between DOAC and warfarin were not examined in this study and should be examined in future studies.
Third, death linkage was not accessible for this study, therefore, the competing risk of death could not be evaluated.DOACs are a heterogeneous group of drugs with different mechanisms of action, dosing, pharmacokinetics, efficacy, and safety. 42Comparing each DOAC with warfarin concerning the risk of outcomes is therefore important.However, due to current data access restrictions, a direct comparison between each DOAC and warfarin regarding outcome risks was not conducted in the current study.Future studies should include such comparisons.While understanding the cause of an ED visit/admission is important, it's worth noting that the recorded cause for the occurrence of such visits/admissions might not always correspond to the final diagnosis.Unfortunately, due to constraints in data accessibility, performing a cause-specific analysis was not viable within the scope of this current study.Future validation studies with cause-specific outcomes are warranted.

| Clinical implications
In 2020 DOACs constituted approximately 5% of the total NHS drug budget, 43 of which of course, could be even higher due to the pandemic of COVID.Based on the findings of our study, a health economic analysis is justified to evaluate whether the increased utilization of ED and nonelective hospitalization associated with the new use of DOAC, with their high purchasing costs of DOACs are more costly than optimized warfarin use through the adoption of novel methods such as genotype-guided dosing 44,45 and point-of-Care INR monitoring. 46This would leave DOAC prescribing only for those at high risk of bleeding events, for example, those with variant alleles that increase the risk of bleeding from warfarin. 47sed on primary care electronic health record data linked with hospitalization data, via a novel tapered matching method, we formed attendance to ED within the first 12 months of anticoagulant prescription.Secondary outcomes of interest included: (i) any incident nonelective hospitalization within the 12 months following the first anticoagulant prescription; (ii) nonelective re-hospitalization (having ≥2 nonelective hospitalizations) within the 12 months since the first anticoagulant prescription; YU ET AL. | 1545 (iii) ED re-attendance (having ≥2 ED events) within 12 months since the first anticoagulant prescription; (iv) any incident nonelective hospitalization within the 6 months following the first anticoagulant prescription; (v) any incident ED event within the 6 months since the first anticoagulant prescription; (vi) re-nonelective hospitalization (having ≥2 nonelective hospitalizations) within the 6 months since the first anticoagulant prescription; and (vii) ED re-attendance (having ≥2 ED events) within 6 months since the first anticoagulant prescription.Patients were followed from their first prescription of an anticoagulant until they experienced an outcome of interest or by completion of 12 months (or 6 months) without an event.Patients were excluded if they stopped or suspended treatment up to 30 days after the first anticoagulant prescription.Patients were excluded if they switched between warfarin and DOAC or vice versa within the first 12 months.

"
quasi" comparison cohorts for incident DOAC and warfarin users with AF.This study supported and extended the observations from the ECLIPSE risk stratification system (PRISM) and showed an increased risk of ED attendance, nonelective hospitalization, in new DOAC users compared with new warfarin users.For patients with AF, caution is warranted when prescribing DOACs as first-line anticoagulant treatment.Further large-scale replication studies involving external datasets, matching for other confounders, and over a longer period of time are warranted.
Baseline characteristics in the comparison cohorts.

1
Distribution of difference of means, variance, and skewness on matched variables in the unmatched and matched cohorts.Triangles indicate measurements from unmatched cohorts; diamonds indicate measurements from coarsened exact matching; circles indicate measurements from entropy-matching cohorts.Each comorbidity and prescription distribution in the unmatched and matched cohorts.