Background: Atrial fibrillation (AF) is a prevalent cardiac arrhythmia associated with a significantly increased risk of systemic thromboembolism and stroke. Anticoagulation therapy, particularly with direct oral anticoagulants, has become the standard for stroke prevention but comes at the cost of an increased bleeding risk. With the introduction of effective alternatives to anticoagulation, such as percutaneous left atrial appendage occlusion, bleeding risk stratification has become essential to guide therapeutic decision-making. Conventional statistical methods have been used for bleeding risk stratification scores, such as HEMORR2HAGES, HAS-BLED, and ATRIA. However, these methods may inadequately address the multifactorial nature of bleeding risk in diverse patient populations, and their overall performance has been suboptimal. Summary and Key Messages: Recent advancements in machine learning (ML) offer promising opportunities to enhance bleeding risk prediction and optimize anticoagulation therapy. This review explores ML applications in AF patients receiving anticoagulation therapy, focusing on the development and validation of ML-based bleeding risk scores. These models have demonstrated improved predictive performance compared to traditional tools, leveraging complex datasets to identify nuanced patterns and interactions. Furthermore, ML-driven tools in warfarin management, including dose prediction, optimization of time in the therapeutic range, and the identification of drug-drug interactions, show significant potential to enhance patient safety and treatment efficacy.

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia, with its prevalence rising significantly with age [1, 2]. In the USA, the prevalence of AF is estimated to reach 4% among individuals aged 60–69 years and up to 18% among older adults 80–89 years [2]. AF is strongly associated with an elevated risk of systemic thromboembolism, particularly stroke, which contributes to significant morbidity, mortality, and disability [3, 4]. Anticoagulation therapy has been established as the cornerstone for reducing stroke and mortality risks in these patients, making it a critical component of AF management [5, 6].

The introduction of direct oral anticoagulants (DOACs) has revolutionized anticoagulation therapy, offering superior effectiveness in stroke prevention along with an enhanced safety profile compared to warfarin in patients with AF [7‒11]. However, despite these advancements, concerns about bleeding complications associated with warfarin and DOAC use have led to hesitation in prescribing and usage of this guideline-recommended treatment to many patients [11‒15].

Over the years, several risk calculators, including HEMORR2HAGES (hepatic or renal disease, ethanol abuse, malignancy, older age, reduced platelet count or function, re-bleeding, hypertension, anemia, genetic factors, excessive fall risk, and stroke), HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile international normalized ratio (INR), elderly, drugs/alcohol), and ATRIA (anticoagulation and risk factors in atrial fibrillation), have been developed and widely validated to assess bleeding risk [16‒18]. These tools, however, rely on traditional statistical methods, which have limitations in their prognostic performance. They are usually based on previous data and experience and, therefore, incorporate established risk factors and may omit other potentially relevant predictors. Furthermore, traditional methods tend to oversimplify the complex interplay between risk factors. Machine learning (ML), a subset of artificial intelligence (AI), has emerged as a powerful approach for identifying complex patterns and relationships within large datasets [19]. Advances in computational power and the increasing availability of standardized, structured, and digitalized datasets have facilitated the application of ML in healthcare. ML, deep learning, artificial neural networks, and convolutional neural networks are all subclasses of AI [20] that have driven significant advancements in areas such as natural language processing, pattern recognition, and predictive analytics. Some of the commonly used ML methods are the RegCox, Extreme Gradient Boosting, and Random Survival Forest. In the context of AF, ML has already been employed extensively to predict future AF based on electronic health records [19, 21], improve screening and diagnosis using 12-lead electrocardiograms [22‒25], pacemaker, and other implantable devices derived data [26, 27], and, more recently, smartwatch technology [28‒30]. ML has also enhanced stroke risk evaluation, outperforming traditional tools like the CHADS2-VASc score [31‒33]. This narrative review aimed to explore the emerging role of ML in improving bleeding risk assessment for anticoagulated AF patients. Specifically, it focuses on two primary applications: (1) introducing the performance of novel ML-derived risk scores compared to conventional models and (2) optimizing warfarin therapy through ML-driven approaches. While there are many parameters to evaluate the performance of ML-based models, the most used is the area under the curve (AUC). An AUC of 1.0 means that the model perfectly distinguishes between the true positive and true negative, while an AUC of 0.5 means that the model performs no better than a random chance.

The latest American College of Cardiology and American Heart Association guidelines for bleeding risk stratification in patients with AF recommend using one of the following three well-established scoring systems [13]. The HEMORR2HAGES score [16], introduced in 2006, was developed from a cohort of 3,791 AF patients (1,604 on warfarin) to evaluate bleeding risk factors using a comprehensive classification approach. It assigns 2 points for prior bleeding and 1 point each for hepatic or renal disease, ethanol abuse, malignancy, advanced age (>75 years), reduced platelet count or function, uncontrolled hypertension, anemia, genetic factors, excessive fall risk, and prior stroke.

The HAS-BLED score [18, 34], introduced in 2011, was developed using data from a real-world cohort of 7,329 anticoagulated patients with AF (including 3,665 treated with warfarin only without other oral anticoagulation or antiplatelet agents). This score categorized patients into three bleeding risk levels: low (score of 0), moderate (scores of 1–2), and high (score ≥3). Over the years, HAS-BLED has been extensively validated across various cohorts, confirming its utility in clinical practice.

The ATRIA score [17], also introduced in 2011, was derived from a real-world cohort of 9,186 AF patients treated with warfarin. The model incorporated five weighted variables: anemia (3 points), severe renal disease (3 points), age >75 years (2 points), prior bleeding (1 point), and hypertension (1 point). The score was designed to predict significant hemorrhagic events effectively.

Among these traditional scores, HAS-BLED has demonstrated superior predictive performance for major bleeding events in a few retrospective studies and in meta-analysis [34, 35]. It has been considered the favorite bleeding risk score and, therefore, was the most used as a comparator for developing ML-based scores. In recent years, additional bleeding risk scores tailored for DOAC users, such as the DOAC score, have been developed using traditional statistical methods, showing modestly improved performance [36]. However, the advancement of ML has enabled the creation of novel predictive models with the potential for greater accuracy and personalized risk stratification.

Herrin et al. [37] analyzed a large US national dataset comprising medical and pharmacy claims from privately insured patients and Medicare Advantage enrollees in the USA between 2016 and 2019. These patients had at least one of the following: AF (43.7%), ischemic heart disease (71.2%), or venous thromboembolism (19.3%) and were prescribed with anticoagulation or antiplatelet agents within the last 12 months. The cohort included 306,463 patients (mean age 69 [SD 12.6]), of whom 57% were prescribed oral anticoagulants (OACs), 42% with antiplatelet agents, and 1% were treated with a combination of OAC and antiplatelet agent. Three ML models, RegCox, Extreme Gradient Boosting, and Random Survival Forest, were trained to predict gastrointestinal bleeding (GIB) at 6 and 12 months. Patients with malignancy-related GIB were excluded. The RegCox and Extreme Gradient Boosting models performed similarly in the validation set, with an AUC of 0.67 at 6 months and 0.66 at 12 months. The Random Survival Forest model had lower performance, with AUCs of 0.62 at 6 months and 0.60 at 12 months. All ML-developed models outperformed HAS-BLED, which achieved AUCs of 0.61 and 0.60 for GIB at 6 and 12 months, respectively. The single most influential predictor in the ML models was prior GIB (importance score 0.72). Other key predictors in the ML model included a prior diagnosis of AF, ischemic heart disease and venous thromboembolism combined (0.38), and the use of gastroprotective medications (0.32).

Loring et al. [38] assessed data from two population-based registries: ORBIT-AF (median age 73 years, median follow-up of 520 days) and GARFIELD-AF (median age 71 years, 1 year of follow-up), which included 22,760, and 52,032 patients, respectively. In the ORBIT-AF, 38%, 45%, and 40% of the study population were treated with vitamin K antagonists (VKAs), DOACs, and antiplatelet agents (with or without OAC), respectively. While in the GARFIELD-AF study, 39%, 28%, and 35% were treated with VKAs, DOACs, and antiplatelet agents (with or without OAC), respectively. There were 1,323 (4.7%) major bleeding events in the ORBIT-AF and 349 (0.7%) in the GARFIELD-AF (defined based on the International Society of Thrombosis and Hemostasis criteria). ML models, including random forests, gradient boosting, and neural networks, were compared with stepwise logistic regression for predicting 1-year major bleeding. Across all outcomes, stepwise regression outperformed other more advanced ML models, with AUCs of 0.711 (95% CI: 0.707–0.715) for ORBIT-AF and 0.647 (95% CI: 0.641–0.653) for GARFIELD-AF. Performance comparisons to HAS-BLED were not reported.

Lu et al. [39] evaluated the multi-label gradient boosting decision tree model using 25 key variables to predict all-cause mortality and major bleeding in a prospective global AF registry (GLORIA-AF) of 25,656 patients (mean age 70.3 years [SD 10.3]). Patients from 38 countries were enrolled between 2011 and 2020 and had at least 1 year of follow-up or died during the first year. Of these, 86% were anticoagulated (18.6% with warfarin and the rest with DOACs), 405 (1.6%) had major bleeding (as defined by the International Society on Thrombosis and Hemostasis) within 1 year after AF diagnosis The multi-label gradient boosting decision tree model achieved an optimized AUC of 0.70 (95% CI: 0.65–0.75) for predicting major bleeding, significantly outperforming HAS-BLED (AUC 0.61 [0.559–0.655], p = 0.002). Key predictors identified were age, smoking, and creatinine clearance.

Joddrell et al. [40] also used the data from the GLORIA-AF registry. Their analysis cohort included data of 26,183 patients (mean age 70.13 [SD 10.13]), 67.1% were treated with DOACs, 18.5% with VKA, and 33.3% received aspirin. Overall, during the study period, 873 (3.3%) individuals had major bleeding. The authors trained several ML models – including logistic regression, random forest, linear discriminant analysis, naive Bayes, Extreme Gradient Boosting, and neural networks using 23 demographic and clinical variables. The random forest and linear discriminant analysis models performed best, achieving AUCs of 0.68 (95% CI: 0.62–0.72) and 0.66 (95% CI: 0.62–0.70) for major bleeding risk at 1 and 3 years, respectively, significantly outperforming HAS-BLED (1-year AUC 0.54 [95% CI: 0.52–0.56] and 3-year AUC 0.52 [95% CI: 0.51–0.54]). Age and geographic region were the most critical predictors, with HAS-BLED scores showing a strong skew toward specificity in this dataset.

Bernardini et al. [41] utilized data from the START-2 database, which included 11,078 patients (median age 77 [IQR 71–83]) initiating anticoagulation for non-valvular AF (46.4% on VKAs and the rest on DOACs). During a median follow-up 1.5 years (IQR 1.0–2.6), 240 (2.2%) patients had major bleeding (according to the International Society on Thrombosis and Hemostasis definition). Supervised ML models including stepwise logistic regression, gradient-boosted decision trees, and multitask neural networks were used to predict major bleeding. Gradient-boosted decision trees was the best performing model for patients on DOACs, achieving an AUC of 0.71 compared to HAS-BLED’s 0.59 (p < 0.001). The most significant predictors were hemoglobin, platelet count, BMI, creatinine clearance (all with linear inverse correlation), and age (with linear correlation). For patients on VKAs, the ML models did not significantly outperform HAS-BLED (AUC 0.63 vs. 0.57, p = 0.83).

Table 1 summarizes the results of these studies. Notably, an evolving application of ML tools is the use of natural language processing for data retrieval from electronic health records. This capability enhances data extraction and has improved the performance of traditional risk scores like HAS-BLED [42, 43]. It is reasonable to expect that leveraging ML to refine the accuracy and quality of data used in developing prediction models will lead to improved performance of these models.

Table 1.

Summary of studies that examined the performance of machine learning-based models compared to traditional risk scores for prediction of bleeding

Study (ref)Patients, nBest performed ML modelAUC of ML methodAUC of HAS-BLEDMajor factors influencing the ML models
total populationDOAC patientswarfarin patients
Herrin et al. [37306,463 57% RegCox 0.67–0.68 0.60–0.61 Prior diagnosis of AFa 
Ischemic heart diseasea 
Venous thromboembolisma 
Extreme Gradient Boosting 0.67–0.68 Use of gastroprotective medicationsa 
Loring et al. [3874,792 28–45% 38–39% Stepwise logistic regression 0.647–0.711 Not reported Age 
Kidney function 
Coronary diseasea 
Anemia 
Lu et al. [3925,656 67.4% 18.6% Multi-label gradient boosting decision tree model 0.70 0.61 Age 
Smokinga 
Creatinine clearance 
Joddrell et al. [4026,183 67.1% 18.5% Random forest 0.68 0.52–0.54 Age 
Linear discriminant analysis 0.66 Geographic regiona 
Bernardini et al. [4111,078 53.6% 46.4% Gradient-boosted decision trees (for DOACS) 0.71 0.59 Hemoglobin 
Platelet count 
BMIa 
Creatinine clearance 
Age 
Study (ref)Patients, nBest performed ML modelAUC of ML methodAUC of HAS-BLEDMajor factors influencing the ML models
total populationDOAC patientswarfarin patients
Herrin et al. [37306,463 57% RegCox 0.67–0.68 0.60–0.61 Prior diagnosis of AFa 
Ischemic heart diseasea 
Venous thromboembolisma 
Extreme Gradient Boosting 0.67–0.68 Use of gastroprotective medicationsa 
Loring et al. [3874,792 28–45% 38–39% Stepwise logistic regression 0.647–0.711 Not reported Age 
Kidney function 
Coronary diseasea 
Anemia 
Lu et al. [3925,656 67.4% 18.6% Multi-label gradient boosting decision tree model 0.70 0.61 Age 
Smokinga 
Creatinine clearance 
Joddrell et al. [4026,183 67.1% 18.5% Random forest 0.68 0.52–0.54 Age 
Linear discriminant analysis 0.66 Geographic regiona 
Bernardini et al. [4111,078 53.6% 46.4% Gradient-boosted decision trees (for DOACS) 0.71 0.59 Hemoglobin 
Platelet count 
BMIa 
Creatinine clearance 
Age 

DOACs, direct oral anticoagulants; ML, machine learning; AUC, area under the curve; AF, atrial fibrillation; BMI, body mass index.

aMarks the factors that are not part of the HAS-BLED score.

Warfarin has remained an important anticoagulation agent for some patients with AF despite the rise of DOACs. Its role is especially important among patients with significant mitral stenosis, patients with anti-phospholipid syndrome, and those with mechanical prosthetic valves, where DOACs were found to be inferior to warfarin [44‒47]. Despite decades of experience with warfarin, challenges in achieving optimal dosing – stemming from its narrow therapeutic index and significant interpatient variability – have fueled interest in leveraging ML to enhance safety and efficacy.

The individualized dosing of warfarin is complex, influenced by many factors such as genetic polymorphisms (e.g., CYP2C9 and VKORC1), clinical characteristics (such as age, weight, and comorbidities), vitamin K content in dietary intake, and drug-drug interactions [48‒53]. These factors contribute to substantial variability in dose requirements and a heightened risk of out-of-range INR values, which can lead to thromboembolism on one end of the spectrum or bleeding on the other. Traditionally, dose adjustments rely on clinical judgment and experience or dosing algorithms, but these approaches often fail to account for a patient’s specific profile, leading to preventable adverse events. Even a genotype-based approach has failed to show better results when used to dictate warfarin dosage [54]. ML offers transformative potential in warfarin management by excelling at analyzing complex interactions of clinical, demographic, and genetic factors, enabling individualized treatment strategies that enhance treatment effectiveness and patient safety.

One of the most utilized applications of ML is for the prediction of warfarin dosage, a topic that has been extensively researched over the past decade. Grossi et al. [55] analyzed demographic, clinical, and genetic data from 377 patients treated with warfarin using an artificial neural network algorithm. Their model was trained and tested retrospectively on patients who had achieved stable therapeutic INR values. It achieved high accuracy, with a mean absolute error (MAE) of 5.7 mg/week for the warfarin maintenance dose. In a more advanced comparison of ML models, Liu et al. [56] evaluated nine different algorithms to optimize warfarin dosing using the International Warfarin Pharmacogenetic Consortium, an open database with pooled data on 4,798 patients who had achieved a stable therapeutic dose. The study was retrospective and employed a derivation/validation cohort design, using 80% of patients to develop the algorithms and 20% for testing. Performance was measured by MAE and the percentage of patients whose predicted dose fell within 20% of the actual dose. Their analysis identified three advanced models: Bayesian additive regression trees, multivariate adaptive regression splines (MARS), and Support Vector Regression (SVR) as the top performing methods, achieving similar levels of accuracy (MAE: ∼8.9 mg/week, mean percentage within 20% of ∼46%). Building on these advancements, Ma et al. [57] employed a stacked generalization framework, machine learning technique that combines multiple models to improve predictive performance. Their study also utilized the International Warfarin Pharmacogenetic Consortium dataset and demonstrated significant improvements in prediction accuracy compared to traditional regression methods. Specifically, their model achieved a 15% reduction in MAE of the weekly dose and increased the percentage of patients with doses within 20% of the therapeutic range. Notably, these algorithms were particularly beneficial for patients requiring low warfarin maintenance doses, where even small dose changes could lead to adverse clinical events such as thrombosis or bleeding.

ML models have also been applied to optimize the time in therapeutic range for patients receiving warfarin. In a notable study, Petch et al. [58] developed a deep Q-learning algorithm capable of dynamically adjusting warfarin doses based on INR measurements. This algorithm was trained using data from 22,502 patients in the warfarin-treated arms of pivotal randomized clinical trials, including edoxaban (ENGAGE AF-TIMI 48), apixaban (ARISTOTLE), and rivaroxaban (ROCKET AF). For external validation, the model was tested on an independent dataset of 5,730 warfarin-treated patients from the dabigatran trial (RE-LY). The results demonstrated that for every 10% increase in algorithm-consistent dosing, there was a 6.78% improvement in time in therapeutic range. This optimization was accompanied by significant reductions in composite adverse clinical outcomes, such as stroke and major bleeding [58]. Importantly, these results were similar to those obtained using a rules-based clinical algorithm employed as a benchmark. Beyond dosing prediction, other applications of ML in warfarin treatment have been demonstrated. For instance, Hu et al. [59] developed an extreme gradient boosting model to predict bleeding risk in hospitalized patients receiving warfarin after valve replacement. This model achieved an AUC of 0.882 for predicting INR ≥4.5 demonstrating high accuracy [59]. In addition, ML may be used to identify drug-drug interaction. Hansen et al. [60] applied random forest models to identify significant drug-drug interactions affecting warfarin metabolism analyzing initiation of a novel prescription in previously INR-stable warfarin-treated patients. Their work rediscovered known interactions and also highlighted new potential interactions. These findings are instrumental for clinicians managing complex medication regimens in patients on warfarin therapy, as they allow for tailored adjustments to minimize bleeding risk.

This review underscores the potential of integrating ML into bleeding risk assessment and management, as well as for reducing bleeding complications due to warfarin treatment. Even though current ML-based risk scores already surpass traditional risk scores in predictive accuracy, they achieve moderate predictive performance with AUC values of only 0.6–0.7. ML-based models are believed to perform better primarily because they capture the factors influencing bleeding risk and their interactions more effectively. Furthermore, the enhanced performance of ML-based models may also be due to changes in the clinical profiles and pharmaceutical treatments of today’s AF patients compared to those used to develop the traditional risk scores. Consequently, these conventional risk scores might now be outdated and less effective at evaluating contemporary patient populations.

While current studies rely primarily on structured tabular data, limiting their scope, future research should further explore integrating large language models for free-text analysis from electronic medical records and unstructured imaging data, such as electrocardiograms and echocardiograms. This can enable a multimodal approach for more comprehensive predictions. To ensure responsible AI implementation, these models must undergo rigorous validation, including randomized controlled trials to confirm their reliability and clinical utility across diverse settings. Transparent reporting and continuous monitoring are essential to ensure these tools meet real-world clinical standards.

The incorporation of ML can significantly enhance clinical decision-making, especially in a clinical landscape with available alternative therapeutic options. For AF who are at high bleeding risk, alternatives such as left atrial appendage occlusion devices offer non-inferior outcomes compared to oral anticoagulation therapies, including similar rates of cardiovascular mortality and incident stroke [61, 62]. These devices provide a valuable option for individuals with elevated bleeding risk scores. Furthermore, the use of ML-based systems promotes safer treatment and better control of INR in patients treated with warfarin.

By enabling more personalized stroke prevention strategies and providing tools with greater precision than traditional risk scores, ML has the potential to promote patient-centered care. As research in this field progresses, ML-driven innovations are expected to play a pivotal role in improving outcomes and enhancing the quality of care for patients with AF requiring anticoagulation therapy.

The authors declare that they have no conflicts of interest.

This study was not supported by any sponsor or funder.

Conceptualization, methodology, literature search and screening, and data extraction: T.T.L. and B.F. Writing – original draft: T.T.L., B.F., and M.C.-S. Writing – review, critical revision, and editing: T.T.L., S.T., R.B., M.C.-S., R.K., R.M., D.A.N., S.H., K.S., D.E.F., and B.F. Supervision: B.F.

1.
Kornej
J
,
Börschel
CS
,
Benjamin
EJ
,
Schnabel
RB
.
Epidemiology of atrial fibrillation in the 21st century: novel methods and new insights
.
Circ Res
.
2020
;
127
(
1
):
4
20
.
2.
Oltman
CG
,
Kim
TP
,
Lee
JWY
,
Lupu
JD
,
Zhu
R
,
Moussa
ID
.
Prevalence, management, and comorbidities of adults with atrial fibrillation in the United States, 2019 to 2023
.
JACC Adv
.
2024
;
3
(
11
):
101330
.
3.
Wolf
PA
,
Mitchell
JB
,
Baker
CS
,
Kannel
WB
,
D’Agostino
RB
.
Impact of atrial fibrillation on mortality, stroke, and medical costs
.
Arch Intern Med
.
1998
;
158
(
3
):
229
34
.
4.
Chung
MK
,
Refaat
M
,
Shen
WK
,
Kutyifa
V
,
Cha
YM
,
Di Biase
L
, et al
.
Atrial fibrillation: JACC council perspectives
.
J Am Coll Cardiol
.
2020
;
75
(
14
):
1689
713
.
5.
Chao
TF
,
Potpara
TS
,
Lip
GYH
.
Atrial fibrillation: stroke prevention
.
Lancet Reg Health Eur
.
2024
;
37
:
100797
.
6.
Van Walraven
C
,
Hart
RG
,
Singer
DE
,
Laupacis
A
,
Connolly
S
,
Petersen
P
, et al
.
Oral anticoagulants vs aspirin in nonvalvular atrial fibrillation: an individual patient meta-analysis
.
Jama
.
2002
;
288
(
19
):
2441
8
.
7.
Ahmad
AS
,
Ormiston-Smith
N
,
Sasieni
PD
.
Trends in the lifetime risk of developing cancer in Great Britain: comparison of risk for those born from 1930 to 1960
.
Br J Cancer
.
2015
;
112
(
5
):
943
7
.
8.
Barsoum
RS
.
Chronic kidney disease in the developing world
.
Diabetes Care
.
2006
;
354
(
10
):
997
9
.
9.
Correa‐Villaseñor
A
,
Cragan
J
,
Kucik
J
,
O’Leary
L
,
Siffel
C
,
Williams
L
.
The metropolitan Atlanta congenital defects program: 35 years of birth defects surveillance at the Centers for Disease Control and Prevention
.
Birth Defects Res A Clin Mol Teratol
.
2003
;
67
(
9
):
617
24
.
10.
National High Blood Pressure Education Program Working Group on High Blood Pressure in Children and Adolescents
.
The fourth report on the diagnosis, evaluation, and treatment of high blood pressure in children and adolescents
.
Pediatrics
.
2004
;
114
(
Suppl ment_2
):
555
76
.
11.
Marzec
LN
,
Wang
J
,
Shah
ND
,
Chan
PS
,
Ting
HH
,
Gosch
KL
, et al
.
Influence of direct oral anticoagulants on rates of oral anticoagulation for atrial fibrillation
.
J Am Coll Cardiol
.
2017
;
69
(
20
):
2475
84
.
12.
Navar
AM
,
Kolkailah
AA
,
Overton
R
,
Shah
NP
,
Rousseau
JF
,
Flaker
GC
, et al
.
Trends in oral anticoagulant use among 436 864 patients with atrial fibrillation in community practice, 2011 to 2020
.
J Am Heart Assoc
.
2022
;
11
(
22
):
e026723
.
13.
Writing Committee Members
;
Joglar
JA
,
Chung
MK
,
Armbruster
AL
,
Benjamin
EJ
,
Chyou
JY
, et al
.
2023 ACC/AHA/ACCP/HRS guideline for the diagnosis and management of atrial fibrillation: a report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines
.
J Am Coll Cardiol
.
2024
;
83
(
1
):
109
279
.
14.
Andrade
JG
,
Krahn
AD
,
Skanes
AC
,
Purdham
D
,
Ciaccia
A
,
Connors
S
.
Values and preferences of physicians and patients with nonvalvular atrial fibrillation who receive oral anticoagulation therapy for stroke prevention
.
Can J Cardiol
.
2016
;
32
(
6
):
747
53
.
15.
Gulpen
AJW
,
Braeken
DCW
,
Schalla
S
,
Ten Cate
H
,
Crijns
HJ
,
Ten Cate-Hoek
AJ
.
Long-term adherence to direct oral anticoagulants in patients with atrial fibrillation: a comparative cross-sectional study
.
Acta Haematol
.
2022
;
145
(
5
):
476
83
.
16.
Gage
BF
,
Yan
Y
,
Milligan
PE
,
Waterman
AD
,
Culverhouse
R
,
Rich
MW
, et al
.
Clinical classification schemes for predicting hemorrhage: results from the National Registry of Atrial Fibrillation (NRAF)
.
Am Heart J
.
2006
;
151
(
3
):
713
9
.
17.
Fang
MC
,
Go
AS
,
Chang
Y
,
Borowsky
LH
,
Pomernacki
NK
,
Udaltsova
N
, et al
.
A new risk scheme to predict warfarin-associated hemorrhage: the ATRIA (anticoagulation and risk factors in atrial fibrillation) study
.
J Am Coll Cardiol
.
2011
;
58
(
4
):
395
401
.
18.
Lip
GYH
,
Frison
L
,
Halperin
JL
,
Lane
DA
.
Comparative validation of a novel risk score for predicting bleeding risk in anticoagulated patients with atrial fibrillation: the HAS-BLED (hypertension, abnormal renal/liver function, stroke, bleeding history or predisposition, labile INR, elderly, drugs/alcohol concomitantly) score
.
J Am Coll Cardiol
.
2011
;
57
(
2
):
173
80
.
19.
Feeny
AK
,
Chung
MK
,
Madabhushi
A
,
Attia
ZI
,
Cikes
M
,
Firouznia
M
, et al
.
Artificial intelligence and machine learning in arrhythmias and cardiac electrophysiology
.
Circ Arrhythm Electrophysiol
.
2020
;
13
(
8
):
E007952
.
20.
Soffer
S
,
Ben-Cohen
A
,
Shimon
O
,
Amitai
MM
,
Greenspan
H
,
Klang
E
.
Convolutional neural networks for radiologic images: a radiologist’s guide
.
Radiology
.
2019
;
290
(
3
):
590
606
.
21.
Khurshid
S
,
Kartoun
U
,
Ashburner
JM
,
Trinquart
L
,
Philippakis
A
,
Khera
AV
, et al
.
Performance of atrial fibrillation risk prediction models in over 4 million individuals
.
Circ Arrhythm Electrophysiol
.
2021
;
14
(
1
):
e008997
.
22.
Lu
J
,
Hutchens
R
,
Hung
J
,
Bennamoun
M
,
McQuillan
B
,
Briffa
T
, et al
.
Performance of multilabel machine learning models and risk stratification schemas for predicting stroke and bleeding risk in patients with non-valvular atrial fibrillation
.
Comput Biol Med
.
2022
;
150
:
106126
.
23.
Zhang
Y
,
Xu
S
,
Xing
W
,
Chen
Q
,
Liu
X
,
Pu
Y
, et al
.
Robust artificial intelligence tool for atrial fibrillation diagnosis: novel development approach incorporating both atrial electrograms and surface ECG and evaluation by head-to-head comparison with hospital-based physician ECG readers
.
J Am Heart Assoc
.
2024
;
13
(
3
):
e032100
.
24.
Noseworthy
PA
,
Attia
ZI
,
Behnken
EM
,
Giblon
RE
,
Bews
KA
,
Liu
S
, et al
.
Artificial intelligence-guided screening for atrial fibrillation using electrocardiogram during sinus rhythm: a prospective non-randomised interventional trial
.
Lancet
.
2022
;
400
(
10359
):
1206
12
.
25.
Yuan
N
,
Duffy
G
,
Dhruva
SS
,
Oesterle
A
,
Pellegrini
CN
,
Theurer
J
, et al
.
Deep learning of electrocardiograms in sinus rhythm from US veterans to predict atrial fibrillation
.
JAMA Cardiol
.
2023
;
8
(
12
):
1131
9
.
26.
Mittal
S
,
Oliveros
S
,
Li
J
,
Barroyer
T
,
Henry
C
,
Gardella
C
.
AI filter improves positive predictive value of atrial fibrillation detection by an implantable loop recorder
.
JACC Clin Electrophysiol
.
2021
;
7
(
8
):
965
75
.
27.
Kim
M
,
Kang
Y
,
You
SC
,
Park
HD
,
Lee
SS
,
Kim
TH
, et al
.
Artificial intelligence predicts clinically relevant atrial high-rate episodes in patients with cardiac implantable electronic devices
.
Sci Rep
.
2022
;
12
(
1
):
37
10
.
28.
Guo
Y
,
Zhang
H
,
Lip
GYH
;
mAF-App II Trial investigators
.
Consumer-led screening for atrial fibrillation: a report from the mAFA-II trial long-term extension cohort
.
JACC Asia
.
2022
;
2
(
6
):
737
46
.
29.
Fiorina
L
,
Chemaly
P
,
Cellier
J
,
Said
MA
,
Coquard
C
,
Younsi
S
, et al
.
Artificial intelligence–based electrocardiogram analysis improves atrial arrhythmia detection from a smartwatch electrocardiogram
.
Eur Heart J Digit Health
.
2024
;
5
(
5
):
535
41
.
30.
Wasserlauf
J
,
You
C
,
Patel
R
,
Valys
A
,
Albert
D
,
Passman
R
.
Smartwatch performance for the detection and quantification of atrial fibrillation
.
Circ Arrhythm Electrophysiol
.
2019
;
12
(
6
):
e006834
.
31.
Lip
GYH
,
Tran
G
,
Genaidy
A
,
Marroquin
P
,
Estes
C
,
Landsheft
J
.
Improving dynamic stroke risk prediction in non-anticoagulated patients with and without atrial fibrillation: comparing common clinical risk scores and machine learning algorithms
.
Eur Heart J Qual Care Clin Outcomes
.
2022
;
8
(
5
):
548
56
.
32.
Jung
S
,
Song
M-K
,
Lee
E
,
Bae
S
,
Kim
Y-Y
,
Lee
D
, et al
.
Predicting ischemic stroke in patients with atrial fibrillation using machine learning
.
Front Biosci
.
2022
;
27
(
3
).
33.
Nishi
H
,
Oishi
N
,
Ogawa
H
,
Natsue
K
,
Doi
K
,
Kawakami
O
, et al
.
Predicting cerebral infarction in patients with atrial fibrillation using machine learning: the Fushimi AF registry
.
J Cereb Blood Flow Metab
.
2022
;
42
(
5
):
746
56
.
34.
Zhu
W
,
He
W
,
Guo
L
,
Wang
X
,
Hong
K
.
The HAS-BLED score for predicting major bleeding risk in anticoagulated patients with atrial fibrillation: a systematic review and meta-analysis
.
Clin Cardiol
.
2015
;
38
(
9
):
555
61
.
35.
Apostolakis
S
,
Lane
DA
,
Guo
Y
,
Buller
H
,
Lip
GYH
.
Performance of the HEMORR(2)HAGES, ATRIA, and HAS-BLED bleeding risk-prediction scores in patients with atrial fibrillation undergoing anticoagulation: the AMADEUS (evaluating the use of SR34006 compared to warfarin or acenocoumarol in patients with atrial fibrillation) study
.
J Am Coll Cardiol
.
2012
;
60
(
9
):
861
7
.
36.
Aggarwal
R
,
Ruff
CT
,
Virdone
S
,
Perreault
S
,
Kakkar
AK
,
Palazzolo
MG
, et al
.
Development and validation of the DOAC score: a novel bleeding risk prediction tool for patients with atrial fibrillation on direct-acting oral anticoagulants
.
Circulation
.
2023
;
148
(
12
):
936
46
.
37.
Herrin
J
,
Abraham
NS
,
Yao
X
,
Noseworthy
PA
,
Inselman
J
,
Shah
ND
, et al
.
Comparative effectiveness of machine learning approaches for predicting gastrointestinal bleeds in patients receiving antithrombotic treatment
.
JAMA Netw Open
.
2021
;
4
(
5
):
e2110703
14
.
38.
Loring
Z
,
Mehrotra
S
,
Piccini
JP
,
Camm
J
,
Carlson
D
,
Fonarow
GC
, et al
.
Machine learning does not improve upon traditional regression in predicting outcomes in atrial fibrillation: an analysis of the ORBIT-AF and GARFIELD-AF registries
.
Europace
.
2020
;
22
(
11
):
1635
44
.
39.
Lu
J
,
Bisson
A
,
Bennamoun
M
,
Zheng
Y
,
Sanfilippo
FM
,
Hung
J
, et al
.
Predicting multifaceted risks using machine learning in atrial fibrillation: insights from GLORIA-AF study
.
Eur Heart J Digit Health
.
2024
;
5
(
3
):
235
46
.
40.
Joddrell
M
,
El-Bouri
W
,
Harrison
SL
,
Huisman
MV
,
Lip
GYH
,
Zheng
Y
, et al
.
Machine learning for outcome prediction in patients with non-valvular atrial fibrillation from the GLORIA-AF registry
.
Sci Rep
.
2024
;
14
(
1
):
27088
.
41.
Bernardini
A
,
Bindini
L
,
Antonucci
E
,
Berteotti
M
,
Giusti
B
,
Testa
S
, et al
.
Machine learning approach for prediction of outcomes in anticoagulated patients with atrial fibrillation
.
Int J Cardiol
.
2024
;
407
:
132088
.
42.
Maciejewski
C
,
Ozierański
K
,
Basza
M
,
Barwiołek
A
,
Ciurla
M
,
Bożym
A
, et al
.
Practical use case of natural language processing for observational clinical research data retrieval from electronic health records: AssistMED project
.
Pol Arch Intern Med
.
2024
;
134
(
5
):
16704
11
.
43.
Chrysafi
P
,
Lam
B
,
Carton
S
,
Patell
R
.
From code to clots: applying machine learning to clinical aspects of venous thromboembolism prevention, diagnosis, and management
.
Hamostaseologie
.
2024
;
44
(
6
):
429
45
.
44.
Connolly
SJ
,
Ezekowitz
MD
,
Yusuf
S
,
Eikelboom
J
,
Oldgren
J
,
Parekh
A
, et al
.
Dabigatran versus warfarin in patients with atrial fibrillation
.
N Engl J Med
.
2009
;
361
(
12
):
1139
51
.
45.
Eikelboom
JW
,
Connolly
SJ
,
Brueckmann
M
,
Granger
CB
,
Kappetein
AP
,
Mack
MJ
, et al
.
Dabigatran versus warfarin in patients with mechanical heart valves
.
N Engl J Med
.
2013
;
369
(
13
):
1206
14
.
46.
Van Gelder
IC
,
Rienstra
M
,
Bunting
KV
,
Casado-Arroyo
R
,
Caso
V
,
Crijns
HJGM
, et al
.
2024 ESC Guidelines for the management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS)
.
Eur Heart J
.
2024
;
45
(
36
):
3314
414
.
47.
Wang
TY
,
Svensson
LG
,
Wen
J
,
Vekstein
A
,
Gerdisch
M
,
Rao
VU
, et al
.
Apixaban or warfarin in patients with an on-X mechanical aortic valve
.
NEJM Evid
.
2023
;
2
(
7
):
EVIDoa2300067
.
48.
White
PJ
.
Patient factors that influence warfarin dose response
.
J Pharm Pract
.
2010
;
23
(
3
):
194
204
.
49.
Schalekamp
T
,
Brasse
B
,
Roijers
J
,
Chahid
Y
,
Vangeestdaalderop
J
,
Devriesgoldschmeding
H
, et al
.
VKORC1 and CYP2C9 genotypes and acenocoumarol anticoagulation status: interaction between both genotypes affects overanticoagulation
.
Clin Pharmacol Ther
.
2006
;
80
(
1
):
13
22
.
50.
Rieder
MJ
,
Reiner
AP
,
Gage
BF
,
Nickerson
DA
,
Eby
CS
,
McLeod
HL
, et al
.
Effect of VKORC1 haplotypes on transcriptional regulation and warfarin dose
.
N Engl J Med
.
2005
;
352
(
22
):
2285
93
.
51.
Joffe
H
,
Xu
R
,
Johnson
F
,
Longtine
J
,
Kucher
N
,
Goldhaber
S
.
Warfarin dosing and cytochrome P450 2C9 polymorphisms
.
Thromb Haemost
.
2004
;
91
(
6
):
1123
8
.
52.
D’Andrea
G
,
D’Ambrosio
RL
,
Di Perna
P
,
Chetta
M
,
Santacroce
R
,
Brancaccio
V
, et al
.
A polymorphism in the VKORC1 gene is associated with an interindividual variability in the dose-anticoagulant effect of warfarin
.
Blood
.
2005
;
105
(
2
):
645
9
.
53.
Momary
KM
,
Shapiro
NL
,
Viana
MA
,
Nutescu
EA
,
Helgason
CM
,
Cavallari
LH
.
Factors influencing warfarin dose requirements in African–Americans
.
Pharmacogenomics
.
2007
;
8
(
11
):
1535
44
.
54.
Kimmel
SE
,
French
B
,
Kasner
SE
,
Johnson
JA
,
Anderson
JL
,
Gage
BF
, et al
.
A pharmacogenetic versus a clinical algorithm for warfarin dosing
.
N Engl J Med
.
2013
;
369
(
24
):
2283
93
.
55.
Grossi
E
,
Podda
GM
,
Pugliano
M
,
Gabba
S
,
Verri
A
,
Carpani
G
, et al
.
Prediction of optimal warfarin maintenance dose using advanced artificial neural networks
.
Pharmacogenomics
.
2014
;
15
(
1
):
29
37
.
56.
Liu
R
,
Li
X
,
Zhang
W
,
Zhou
H-H
.
Comparison of nine statistical model based warfarin pharmacogenetic dosing algorithms using the racially diverse international warfarin pharmacogenetic consortium cohort database
.
PLoS One
.
2015
;
10
(
8
):
e0135784
.
57.
Ma
Z
,
Wang
P
,
Gao
Z
,
Wang
R
,
Khalighi
K
.
Ensemble of machine learning algorithms using the stacked generalization approach to estimate the warfarin dose
.
PLoS One
.
2018
;
13
(
10
):
e0205872
.
58.
Petch
J
,
Nelson
W
,
Wu
M
,
Ghassemi
M
,
Benz
A
,
Fatemi
M
, et al
.
Optimizing warfarin dosing for patients with atrial fibrillation using machine learning
.
Sci Rep
.
2024
;
14
(
1
):
4516
.
59.
Hu
Y
,
Zhang
X
,
Wei
M
,
Yang
T
,
Chen
J
,
Wu
X
, et al
.
Using machine learning to predict the bleeding risk for patients with cardiac valve replacement treated with warfarin in hospitalized
.
Pharmacoepidemiol Drug Saf
.
2024
;
33
(
2
):
e5756
.
60.
Hansen
PW
,
Clemmensen
L
,
Sehested
TSG
,
Fosbøl
EL
,
Torp-Pedersen
C
,
Køber
L
, et al
.
Identifying drug-drug interactions by data mining: a pilot study of warfarin-associated drug interactions
.
Circ Cardiovasc Qual Outcomes
.
2016
;
9
(
6
):
621
8
.
61.
Osmancik
P
,
Herman
D
,
Neuzil
P
,
Hala
P
,
Taborsky
M
,
Kala
P
, et al
.
4-year outcomes after left atrial appendage closure versus nonwarfarin oral anticoagulation for atrial fibrillation
.
J Am Coll Cardiol
.
2022
;
79
(
1
):
1
14
.
62.
Lakkireddy
D
,
Ellis
CR
,
Thaler
D
,
Swarup
V
,
Gambhir
A
,
Hermiller
J
, et al
.
5-year results from the AMPLATZER amulet left atrial appendage occluder randomized controlled trial
.
J Am Coll Cardiol
.
2025
;
85
(
11
):
1141
53
.