Personalizing the decision of dabigatran versus warfarin in atrial fibrillation: A secondary analysis of the Randomized Evaluation of Long-term anticoagulation therapY (RE-LY) trial.

BACKGROUND: The RE-LY (Randomized Evaluation of Long-Term Anticoagulation Therapy) trial demonstrated that higher-risk patients with atrial fibrillation had lower rates of stroke or systemic embolism and a similar rate of major bleeding, on average, when treated with dabigatran 150mg compared to warfarin. Since population-level averages may not apply to individual patients, estimating the heterogeneity of treatment effect can improve application of RE-LY in clinical practice. METHODS AND
RESULTS: For 18040 patients randomized in RE-LY, we used patient-level data to develop multivariable models to predict the risk for stroke or systemic embolism and for major bleeding including all three treatment groups (dabigatran 110mg, dabigatran 150mg, and warfarin) over a median follow up of 2.0 years. The mean predicted absolute risk reduction (ARR) for stroke/systemic embolism with dabigatran 150mg compared to warfarin was 1.32% (range 11.6% lower to 3.30% higher risk). The mean predicted ARR for bleeding was 0.41% (range 8.93% lower to 63.4% higher risk). Patients with increased stroke/systemic embolism risk included those with prior stroke/TIA (OR 2.01), diabetics on warfarin (OR 2.00), and older patients on dabigatran 150mg (OR 1.68 for every 10-year increase). Major bleeding risk was higher in patients on aspirin (OR 1.25), with a history of diabetes (OR 1.34) or prior stroke/TIA (OR 1.22), those with heart failure on dabigatran 110mg (OR 1.52), older patients on either dabigatran 110mg or 150mg (OR 1.57 and 1.93, respectively, for each 10-year increase), and heavier patients on dabigatran 110mg or 150mg; patients in a region outside the United States and Canada and with better renal function had lower bleeding risk.
CONCLUSIONS: There is substantial heterogeneity in the benefits and risks of dabigatran relative to warfarin among patients with atrial fibrillation. Using individualized estimates may enable shared decision making and facilitate more appropriate use of dabigatran; as such, it should be prospectively tested. CLINICAL TRIAL REGISTRATION: www.clinicaltrials.gov number, NCT00262600.

Introduction

Applying data from clinical trials to everyday practice is a significant challenge in the current age of evidence-based medicine. Large trials can demonstrate significant differences in mean outcomes for a given treatment, but such estimates ignore the heterogeneity of treatment effect among different patients within the study population. Subgroup analyses try to examine variations in response, but do so using only one parameter at a time and do not consider how outcomes could be impacted by the combination of all of a given patient’s unique clinical characteristics [1,2]. Accordingly, the current paradigm of summarizing clinical trial results into guidelines has resulted in very slow adoption of potentially beneficial therapies and even led to a risk-treatment paradox whereby patients least likely to be treated are those with the greatest potential to benefit from said treatment [3,4].

In patients with atrial fibrillation at increased risk for stroke (as determined by CHADS2VA2Sc [5] score), current guidelines recommend anticoagulation to reduce the risk of ischemic events [6], but this decision must be weighed against the increased risk of bleeding. As many risk factors for stroke overlap with those for bleeding, the choice to anticoagulate can be complicated. Moreover, the introduction of direct oral anticoagulants (DOACs) has further increased the complexity of these decisions [7-9]. The Randomized Evaluation of Long-Term Anticoagulation Therapy (RE-LY) trial compared warfarin with dabigatran in patients with atrial fibrillation at increased risk for stroke, and found lower risk of stroke or systemic embolism and a comparable risk of major bleeding with dabigatran 150mg twice daily; however, the risk of gastrointestinal bleeding was significantly higher with dabigatran 150mg [10]. We used data from the RE-LY trial to create models predicting individual patients’ risk for stroke or systemic embolism and for major bleeding with alternative doses of dabigatran as compared with warfarin, with the hope these models would provide a means for translating the results of RE-LY to clinical practice [2].

Methods

Patient population

RE-LY was an international randomized trial that included 18,113 patients with atrial fibrillation at 951 centers in 44 countries who were at increased risk for stroke (see S1 Appendix for definitions). The study design, inclusion and exclusion criteria, and primary results were reported previously [10,11]. Eligible patients had documented atrial fibrillation within 6 months of enrollment, and at least one of the following characteristics that put them at increased risk for stroke: prior stroke or transient ischemic attack, a left ventricular ejection fraction (EF) of less than 40%, New York Heart Association (NYHA) class II or higher heart failure symptom within 6 months before screening, age of at least 75 years, or age of 65 to 74 years plus diabetes mellitus, hypertension, or coronary artery disease. Major exclusion criteria included severe heart-valve disorder, stroke within 14 days, severe stroke within 6 months, an increased risk of hemorrhage (see S2 Appendix), creatinine clearance (CrCl) less than 30ml per minute, active liver disease, and pregnancy. The original trial protocol was approved by review boards at all the participating sites, and all patients provided written informed consent. The 73 patients who received zero doses of study drug were excluded, leaving 18,040 patients in the analysis. Because the RE-LY database includes only de-identified patient data, this study was deemed exempt by the Yale School of Medicine and St. Luke’s Mid-America Heart Institute Institutional Review Boards.

Patients were randomized to either one of two fixed doses of dabigatran (110mg twice daily or 150mg twice daily), administered in a blinded manner, or to open-label use of warfarin, titrated to an international normalized ratio (INR) of 2.0–3.0. Patients were followed for a median of 2.0 years, with 99.9% achieving complete follow up.

Outcomes

The primary outcomes in both the original trial and our analysis were stroke or systemic embolism and major bleeding (see S3 Appendix for definitions). Secondary outcomes included stroke, systemic embolism, death, myocardial infarction, pulmonary embolism, transient ischemic attack (TIA), and hospitalization. All primary and secondary outcomes were adjudicated independently by two investigators who were blinded to treatment assignments [10].

Statistical analyses

Baseline characteristics are reported as frequencies for categorical variables, and mean (standard deviation) and median with interquartile range (IQR) for continuous variables. The chi-square test was used to compare categorical variables, while ANOVA was used to compare continuous variables.

Model construction

Using the RE-LY dataset, we developed separate risk models to estimate patients’ risks for stroke or systemic embolism and for major bleeding. To construct each model, we identified candidate variables a priori for inclusion, based on the published literature and clinical experience. Only information available prior to randomization was included. We then fit multivariable logistic regression models to predict the occurrence of stroke or systemic embolism and major bleeding. To improve the feasibility of implementing the models, we sought to reduce the model to those variables that accounted for 95% of the predicted variability of the full model. To do so, we ranked all variables by their contribution to the predictive capacity of the model, then removed them sequentially starting with the variables with the least contribution to the model’s variance until all variables retained in the model had at least a 95% contribution to the model’s predictive capacity [12]. The initial variables considered an the variables included in the final models are listed in the S4 and S5 Appendices. We used restricted cubic splines to test the assumption that the continuous variables were linearly associated with the outcomes, and then calculated discrimination (c-statistic) and calibration using plots of observed vs. predicted means by decile for each model [13]. No significant nonlinearity was detected.

Both the stroke/systemic embolism and major bleeding models were internally validated using bootstrap resampling for 500 replications. For each step of resampling, the model was refit as described above, and performance (calibration slope and c-statistics) was assessed on the bootstrapped data and “validated” on the original dataset. The difference in performance between the two datasets was calculated and averaged over the 500 replications. The average differences estimates due to overfitting were then subtracted from the final reported performance of the full and reduced models [14,15].

Describing the heterogeneity of treatment effect

To describe the heterogeneity of benefits and risks of each treatment, we applied the risk models to every patient in the RE-LY trial three times: first assuming treatment with warfarin, a second time assuming treatment with dabigatran 110mg, and a third time assuming treatment with dabigatran 150mg. We calculated the absolute differences in risk between warfarin and dabigatran 110mg, and between warfarin and dabigatran 150mg for each patient. To describe the heterogeneity of treatment effect at the individual level, we calculated the absolute differences in risk between warfarin and dabigatran 110mg, and between warfarin and dabigatran 150mg for each patient. The absolute differences in the risk of stroke or systemic embolism were calculated as each patient’s risk when treated with dabigatran 110mg minus the risk if treated with warfarin (for the same patient), and as the risk when treated with dabigatran 150mg minus the risk if treated with warfarin. Next, to show the heterogeneity of treatment effect at the individual level, we calculated the standard deviation (SD) of these absolute risk differences. This process was repeated to describe the absolute differences in the risk of major hemorrhage, as well as the mean and standard deviation of these absolute differences. We report a density plot to demonstrate the distribution of absolute differences between dabigatran 110mg and warfarin, and between dabigatran 150mg and warfarin across the RE-LY population. Model calibration was assessed by plotting observed event rates against deciles of predicted risk rates. We also developed integer risk scores to help illustrate the relative risk of stroke/systemic embolism and major bleeding by treatment group and clinical variables.

We strove to adhere to the standards set forth for transparency in reporting of risk models and heterogeneity analyses set forth in the TRIPOD and PATH statements [16,17]. All analyses were conducted using R version 3.3.1 (R Project for Statistical Computing, Vienna, Austria) and SAS version 9.3 (SAS Institute, Cary, NC).

Results

The analysis included a total of 18,040 patients, of whom 5,983 were assigned to dabigatran 110mg, 6,059 were assigned to dabigatran 150mg, and 5,998 were assigned to warfarin. Baseline characteristics are shown in S1 Table and were similar across the three treatment groups. Across the entire cohort, median age of participants was 71.4 years, 63.6% were male, and 70.0% were white. Past medical history of participants included hypertension (78.8%), systolic heart failure (32.0%), diabetes mellitus (23.3%), and prior stroke/systemic embolism/TIA (21.8%).

Over the median follow-up of 2.0 years, the rates of the primary outcome of stroke/systemic embolism in the analytic cohort were 1.4%, 0.9%, and 2.1% for dabigatran 110mg, dabigatran 150mg, and warfarin, respectively. The rates of the primary outcome of major bleeding were 3.6%, 4.3%, and 4.7%, respectively (S2 Table).

Predicting primary outcome events in RE-LY

The prediction model for stroke/systemic embolism is displayed in Fig 1, while Fig 2 shows the model for major bleeding. Both models showed acceptable discrimination, with a c-statistic of 0.68 for the stroke/systemic embolism model and 0.69 for the major bleeding model, and good calibration. For the stroke/systemic embolism model calibration, p-value for intercept is 0.8743, for slope is 0.8311; for the major bleeding model calibration, p-value for intercept is 0.9718, for slope is 0.7318. The final models included 9 variables for the stroke/systemic embolism model and 14 variables for the major bleeding model). S1 and S2 Figs show the prediction model calibration curves.

Fig 1

Risk prediction model for stroke or systemic embolism.

Shown are the odds ratio (OR) and 95% confidence interval (CI) for each variable. For variables with interactions with treatment assignment, ORs are presented separately for treatment with dabigatran 110mg, dabigatran 150mg, and warfarin. CI, confidence interval; CrCl, creatinine clearance; TIA, transient ischemic attack; USA, United States of America.

Fig 2

Risk prediction model for major bleeding.

Shown are the odds ratio (OR) and 95% confidence interval (CI) for each variable. For variables with interactions with treatment assignment, ORs are presented separately for treatment with dabigatran 110mg, dabigatran 150mg, and warfarin. CI, confidence interval; CrCl, creatinine clearance; TIA, transient ischemic attack; USA, United States of America.

Predicted benefits, risks, and heterogeneity of treatment effect

The mean absolute reduction in predicted risk for stroke or systemic embolism with dabigatran 110mg as compared to warfarin was 0.78% ± 0.95%. However, there was significant heterogeneity across the study population, and the range of absolute risk reduction for stroke/systemic embolism with dabigatran 110mg compared with warfarin was a decrease in risk of 9.33% to an increase in risk of 2.15%. For major bleeding, the mean absolute risk reduction with dabigatran 110mg was 1.12% ± 1.44% and ranged from a 7.77% decrease in risk to a 23.2% increase in risk with dabigatran 110mg compared with warfarin.

When comparing dabigatran 150mg to warfarin for the outcome of stroke or systemic embolism, the mean predicted absolute risk reduction was 1.32% ± 1.31% and ranged from a decrease of 11.6% to an increase of 3.30% with dabigatran 150mg compared to warfarin. For major bleeding, the mean absolute risk reduction with dabigatran 150mg was 0.41% ± 2.39% and ranged from an 8.93% decrease in risk to a 63.4% increase in risk with dabigatran 150mg compared with warfarin.

When considering patients’ risks for both stroke or systemic embolism and major bleeding, there was significant variability across the study population. Fig 3A (dabigatran 110mg and warfarin) and Fig 3B (dabigatran 150mg and warfarin) show scatter plots to depict each patient’s absolute difference in risk (risk with dabigatran minus risk with warfarin) of the two primary outcomes, with stroke or systemic embolism on the y-axis and major bleeding on the x-axis. Patients in the left lower quadrant of each plot experienced lower risk of both stroke/systemic embolism and bleeding risk with dabigatran, while patients in the right upper quadrant experienced lower risks of both outcomes with warfarin. The right lower quadrant represents patients with a decreased risk of stroke or systemic embolism and an increased risk of bleeding with dabigatran, and the left upper quadrant shows patients for whom dabigatran resulted in a higher risk of stroke or systemic embolism and a lower risk of bleeding.

Fig 3

Scatter plot depicting each patient’s net predicted risk for stroke or systemic embolism and major bleeding.

Panel A shows the plot comparing treatment with dabigatran 110mg versus warfarin, and Panel B compares dabigatran 150mg and warfarin. Each dot represents a patient, with the net predicted risk (warfarin minus dabigatran) of stroke or systemic embolism on the y-axis, and the net predicted risk (warfarin minus dabigatran) of major bleeding on the x-axis. The further left on the x-axis, the lower the patient’s risk of major bleeding with dabigatran, and the further down on the y-axis, the lower the patient’s risk of stroke or systemic embolism.

When comparing dabigatran to warfarin, 76.5% (110mg dose) and 67.1% (150mg dose) of patients had lower risk of both stroke/systemic embolism and major bleeding; 0.4% (dabigatran 110mg) and 3.8% (dabigatran 150mg) had higher risk of both outcomes. The right lower quadrant (patients with decreased stroke/systemic embolism but higher bleeding) included 13.3% (dabigatran 110mg) and 26.3% (dabigatran 150mg) of patients.

To demonstrate the potential utility of the models in clinical care, we created sample patients and calculated their predicted risks of the primary outcomes with visual decision aids using Icon Array (Fig 4) [18]. The first patient, Patient A, is a 68-year-old, 50-kg woman from the United States who has a history of heart failure, a creatinine clearance of 49 ml/min, has paroxysmal atrial fibrillation, is not taking aspirin, and has no history of diabetes, stroke, systemic embolism, or TIA. Helen’s estimated risk of stroke/systemic embolism is lower with dabigatran 150mg than with warfarin (0.43% per year vs. 1.74% per year), and her risk of major bleeding is also lower with dabigatran 150mg (4.45% per year vs. 5.26% per year with warfarin; Fig 4A). The second patient, Patient B, is a 62-year old, 85-kg man, from the United States, with a history of diabetes and TIA, a creatinine clearance of 85 ml/min, who has permanent atrial fibrillation, is taking aspirin, and has no history of heart failure. Frank’s risk of stroke is also estimated to be lower with dabigatran 150mg (0.92% per year) than with warfarin (3.84% per year), however, his risk of major bleeding is higher with dabigatran 150mg (13.1% per year, Fig 4B) than with warfarin (8.43% per year). Additionally, integer risk scores for each of stroke/systemic embolism and major bleeding are shown in S3 and S4 Tables.

Fig 4

Icon Array decision-aids for two model patients showing the risks of stroke or systemic embolism and major bleeding with warfarin (top) and dabigatran 150mg (bottom).

Panel A shows the risks for Patient A, a 68-year-old, 50-kg woman from the United States who has a history of heart failure, has a creatinine clearance of 49 ml/min, has paroxysmal AF, is not taking aspirin, and has no history of diabetes, stroke, systemic embolism, TIA. Panel B shows the risks for Patient B, a 62-year-old, 85-kg man, from the United States, with diabetes, a creatinine clearance of 85 ml/min, has a history of TIA, has permanent AF, is taking aspirin, and has no history of heart failure.

Discussion

Practicing clinical medicine in a way that applies the results from robust clinical trials into everyday care is a difficult task. In this study, we developed two multivariable risk models using data from the RE-LY randomized clinical trial to predict patients’ risk of stroke/systemic embolism and risk of major bleeding if treated with warfarin, dabigatran 110mg, and dabigatran 150mg, respectively. We found significant heterogeneity of treatment effects among patients in the benefits and risks of the three different treatments, with factors including age, sex, diabetes, hypertension, body weight, renal function, region of residence, and type of atrial fibrillation (among others) playing roles that shape individual patients’ risks. Developing strategies, such as the risk models developed here, to convey these risks and benefits to individual patients offers the potential to optimally target new treatments to those who most benefit and to explain to patients why one treatment versus another is being recommended. For example, our analysis showed that there are patients for whom dabigatran lowers the risk of both stroke/systemic embolism and of major bleeding, others in whom there are similar benefits and risks, and yet other patients for whom dabigatran would result in increased risks of both outcomes. Simply applying the RE-LY results to all patients based on the mean treatment effect (which showed lower risk of stroke/systemic embolism and similar risk of bleeding with dabigatran 150mg compared to warfarin) would be providing sub-optimal treatment for many patients, potentially placing some at unnecessary risk and, for others, underestimating the potential benefits of dabigatran. We propose that our models, and ones like it to be produced in the future, present a significant opportunity to use modern data analysis techniques to optimize patient outcomes. This is in keeping with the Precision Medicine Initiative aims, which seek to advance medicine to a place where treatments are tailored to the patient using individualized approaches [19].

The decision to anticoagulate in the setting of atrial fibrillation is complex and has grown more so as the number of medication options has increased with the approval of the DOACs, starting with dabigatran in 2010 following publication of the RE-LY results. The evidence supporting the reduction of stroke risk with anticoagulation in select patients with atrial fibrillation is strong [20,21]. But this potential benefit must be weighed against the risk of bleeding, which can range from day-to-day bruising to major intracranial hemorrhage and life-threatening gastrointestinal bleeding. Recent work modeling individual risk in patients from the RE-LY trial, comparing the two doses of dabigatran to no anticoagulation, has shown significant heterogeneity among the trial population, with some clearly benefiting from each dose of dabigatran over the other [22]. Our findings validate the individualized approach of the prior paper, but the prior study’s models used variables selected by the investigators, whereas our study included all available variables in the models and removed them by backward selection [22]. For the case of stroke/systemic embolism risk, our method developed a more parsimonious model with only 9 variables included, as opposed to 14 variables in the prior study; we feel this likely improves the feasibility of implementation of our model. The prior study also focused on comparing dabigatran to no anticoagulation, while we were primarily interested in comparisons to warfarin, as in this high-risk population it is reasonable to assume a high proportion would be anticoagulated.

By producing these multivariable models, we propose a system whereby such models could be used during clinical encounters to produce a more individualized risk/benefit profile for treatment options for each patient. The current atrial fibrillation guidelines recommend that most patients with atrial fibrillation be treated with a DOAC instead of warfarin, but this is based on the overall effects seen in the clinical trials [6]. One can imagine a scenario where, once models have been appropriately validated, a risk calculator could be built into the electronic medical record and automatically pull patient data to quickly display an individual patient risk score for clinicians to communicate to patients. Similar secondary analyses of landmark clinical trials have been performed to guide decision making for choice of antiplatelet therapy, and for intensity of blood pressure control [23,24]. Application of individualized bleeding risk models to patients undergoing percutaneous coronary intervention (PCI) have also been shown to improve patient outcomes, reducing bleeding events by 38% in adjusted analysis [25].

Relatedly, advances in decision aids, including the Icon Arrays that we used in this study, can be used to quickly and easily produce helpful visual tools for patients to better understand their unique risk profile. Such decision aids could also be automatically generated and displayed in the exam room or sent to a patient’s smart phone. This type of shared decision making is in keeping with current guidelines [6], but is difficult in the current environment without readily available tools for engaging and teaching patients. Prior work in has shown that when patients are given personalized risk information regarding choice of stent type (drug-eluting vs. bare metal) prior to PCI, they have significant increases in patient knowledge and engagement in shared decision-making [26,27]. Providing patients with personalized information in the setting of anticoagulation may have a similar benefit of increasing patient engagement, which may extend to improving adherence when a treatment plan is ultimately selected [28].

Limitations

Our study should only be interpreted in light of several potential limitations. First, our findings have not been validated in an independent sample outside of the RE-LY population. Although we performed bootstrap validations, this applies only to those meeting the inclusion and exclusion criteria of RE-LY. While modeling the heterogeneity of treatment effects is an important means for translating the results of clinical trials to individual patients, the population upon which the models are built includes only those patients meeting the inclusion and exclusion criteria of the trial. The models are thus only valid in such patients and clinicians need to use their clinical judgment as to whether or not a given patient would have been considered eligible and, if not, whether the cause for ineligibility would make the model invalid. In RE-LY, patients with higher bleeding risk were systematically excluded and caution should be used in applying the model to such patients. Of course, this also applies to the application of the original clinical trial results, which are merely summaries of mean treatment effect, and not a limitation of the models, per se. Of note, when examining the list of high-risk bleeding conditions (S2 Appendix), they include conditions that would make a clinician hesitate to start anticoagulation of any type, and thus we feel that our model applies to the large majority of patients for whom anticoagulation is being considered. Second, although our models detected evidence of heterogeneity in treatment effect, the potential exists for other interactions that went undetected in our analysis but could still affect patient outcomes. Third, the fact our population was followed in an RCT means they were likely more adherent to medications than in real-world practice; relatedly, we do not have a measure of compliance during the RE-LY trial which may be relevant given the potential easier compliance with fixed-dose dabigatran vs. warfarin which is dosed to reach target INR values. Fourth, our models had adequate but modest c-statistics (0.675 for the stroke/systemic embolism model and 0.694 for the major bleeding model). However, it has been shown that models with c-statistics >0.60 can provide benefit to patients [29], and we feel that using our models when selecting treatment options is superior to treating all patients as if they would realize the average benefit for the study population or attempting to individualize treatment using subgroup analyses. Fifth, our models were developed using data comparing dabigatran and warfarin, and thus cannot be extrapolated to comparisons of warfarin with other DOACs; however, we would suggest that similar models be developed using data from clinical trials of other DOACs so those trial results can be better applied to individual patients in clinical practice. Finally, the models in this study only predict individualized risk for the endpoints of stroke or systemic embolism and major bleeding. Patients may also be interested in other individualized information, such as their risk of stroke alone (without systemic embolism) or of myocardial infarction with each treatment option. However, stroke or systemic embolism and major bleeding were selected for this study, as they were for RE-LY, because they represent the most significant outcomes of concern related to anticoagulant choice in patients with atrial fibrillation at an increased risk of stroke.

Conclusions

In conclusion, using data from the RE-LY trial, we generated two multivariable models to predict individualized risk profiles for patients with atrial fibrillation treated with dabigatran or warfarin for prevention of stroke or systemic embolism. Our analysis showed significant heterogeneity of treatment effect in the RE-LY trial population, building on evidence that it exists in other trials. Although not yet validated externally, these models show promise to help personalize patient treatment decisions, and to enhance the shared decision-making process by providing patients with easy-to-understand individualized information about their predicted risks and benefits with a given treatment.

Calibration curve for model predicting stroke or systemic embolism.

(DOCX)

Click here for additional data file.

Calibration curve for model predicting major bleeding.

(DOCX)

Click here for additional data file.

Baseline characteristics of RE-LY patients by treatment group.

(DOCX)

Click here for additional data file.

Primary outcomes of stroke or systemic embolism and major bleeding by treatment group.

(DOCX)

Click here for additional data file.

Integer score for stroke or systemic embolism risk.

(DOCX)

Click here for additional data file.

Integer score for major bleeding risk.

(DOCX)

Click here for additional data file.

Definition of increased risk of stroke.

(DOCX)

Click here for additional data file.

Definition of increased risk for hemorrhage.

(DOCX)

Click here for additional data file.

Definition of primary endpoints.

(DOCX)

Click here for additional data file.

Variables included in stroke/systemic embolism risk model.

(DOCX)

Click here for additional data file.

Variables included in major bleeding risk model.

(DOCX)

Click here for additional data file.

20 Jan 2021

PONE-D-20-38636

Personalizing the Decision of Dabigatran versus Warfarin in Atrial Fibrillation: A Secondary Analysis of the Randomized Evaluation of Long-Term Anticoagulation Therapy (RE-LY) Trial.

PLOS ONE

Dear Dr. Reinhardt,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Both reviewers have important comments regarding the analytic approach and make suggestions for additional exploration and revision that are quite relevant. Although the importance of this work is recognised, certainly also by me, I urge you to seriously consider the suggested points.

Please submit your revised manuscript by Feb 21 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Hugo ten Cate, MD, PhD

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Thank you for including your ethics statement:  "The protocol was approved by review boards at all the participating sites, and all patients provided written informed consent."

Please amend your current ethics statement to include the full name of the ethics committee/institutional review board(s) that approved your specific study

Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”).

For additional information about PLOS ONE ethical requirements for human subjects research, please refer to http://journals.plos.org/plosone/s/submission-guidelines#loc-human-subjects-research.

3. Thank you for stating the following in the Competing Interests section:

"I have read the journal's policy and the authors of this manuscript have the following competing interests: Nihar Desai works under contract with the Centers for Medicare and Medicaid Services to develop and maintain performance measures used for public reporting and pay for performance programs. He reports consulting for Amgen, Boehringer Ingelheim, Cytokinetics, Relypsa, Novartis, and SCPharmaceuticals. John Spertus reports an equity interest in Health Outcomes Sciences and consulting for Amgen, Bayer, Merck, Myokardia, Novartis, United Healthcare, BCBS of Kansas City and ownership of the copyright to the SAQ, KCCQ and PAQ. All other authors report no disclosures. "

Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf.

Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: http://journals.plos.org/plosone/s/competing-interests

4. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

5. Please amend your list of authors on the manuscript to ensure that each author is linked to an affiliation. Authors’ affiliations should reflect the institution where the work was done (if authors moved subsequently, you can also list the new affiliation stating “current affiliation:….” as necessary).

6. Please include a copy of Table 2 which you refer to in your text on page 8.

7. We noticed you have some minor occurrence of overlapping text with the following previous publication(s), which needs to be addressed:

http://circoutcomes.ahajournals.org/content/circcvoq/10/4/e003624.full.pdf?download=true

In your revision ensure you cite all your sources (including your own works), and quote or rephrase any duplicated text outside the methods section. Further consideration is dependent on these concerns being addressed.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This is an interesting paper on using the data from a large trial of anticoagulation therapy to predict risks of strokes, bleeds

etc. The abstract mentions a separate model for each treatment - which (fortunately) is not what has been done here. The wording used made my heart sink, but the statistical approach here is absolutely correct. Perhaps the wording can be slightly tightened up.

One assumes the use of logistic regression here as opposed to Cox regression of time to event is justifiable by the relatively low number of events and the relatively short follow-up. However, follow-up is not apparently uniform - we need to see the range of follow-up to understand whether each patient included is really at equal risk from equal temporal exposure. If the range is not tight around the median of 2 years, then an analysis that allows for the varying length of follow-up (i.e Cox regression or equivalent) needs to be used.

Figure 1 could usefully be improved by giving the risk index for each treatment (based on the log OR) - this would naturally lead into the useful figure 4 illustration.

Reviewer #2: The authors developed 2 separate multivariable models to predict thrombotic and bleeding complications in patients with atrial fibrillation, based on the RE-LY trial data. The models show limited discriminative value, but are well calibrated. The variability of treatment effects of the different antithrombotic regimens is very clear depicted in scatter plots visualising the multivariable predicted effect on bleeding or thrombotic complications. The authors wish to enable shared decision making and facilitate more appropriate use of dabigatran.

An important limitation of this study is lack of external validation, as pointed out by the authors in the discussion. We have some extra questions.

General comments:

- Patients with high bleeding risk were excluded from the RE-LY trial. What does this mean for the external validity of this model? Should it only be considered for patients at increased thrombotic risk without high bleeding risk? Please address in discussion

- I really like the patient centred idea of the authors and the manuscript, especially figures 4A and 4B are excellent examples of what would be helpful to doctor and patient. However, in the manuscript no final version of the model, or risk score, or website is given where I can calculate this for my patient. Therefore, in the end, in clinical practise (Introduction, line 94 and 95), unfortunately, this paper won’t help unless I know how to use these models…

- Why would one use these models instead of the widely used CHADSVASc and HAS-BLED scores? It would be useful to compare these models this newly developed models.

- Is the model only suitable to use for dabigatran or also for the other NOACs versus warfarin? Since the benefits of NOAC vs warfarin is considered to be a class effect more than a dabigatran effect. Please address this in discussion.

Some parts of the methodology are not fully clear:

- Why was chosen for simple logistic regression and not for Cox proportional hazards which may be appropriate for this type of survival analysis?

- The model was created by backward selection until all variables retained in the model had at least a 95% contribution to the model’s predictive capacity. How was this contribution quantified?

- The prediction models are represented in figures 1 and 2. However, in line 191 the authors state that the final models included 9 and 14 variables. It is unclear what is shown in figures 1 and 2, since the number of predictors does not correspond. At least, the final models should be represented somewhere in the manuscript, and it should be clear what model is represented in what figure.

- The patients where modelled with assumed treatment of every treatment arm. How was this possible? Was the randomization arm forced into the models? This is not described.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Willem L. Bor

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

27 Jun 2021

PLOS ONE

Dear Editorial Board,

Thank you for your interest in our manuscript and the insightful comments by the reviewers. We are excited to incorporate the comments, and work towards improving the content and readability of our article. The revised manuscript reflects the input of the reviewers with revisions highlighted. Our response to the individual comments follows:

Academic Editor:

• Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming.

o Response: We thank the editor for the helpful links to the style requirements pages. We have corrected the style of the manuscript as well as the format and names of the files to meet PLOS ONE’s requirements.

• Please amend your current ethics statement to include the full name of the ethics committee/institutional review board(s) that approved your specific study

Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”).

o Response: We thank the editor for this comment. Our methods section references that the original RE-LY trial was approved by the institutional review board at all the trial’s participating sites. Our specific study was deemed exempt by the Yale institutional review board given it was a secondary analysis of de-identified data and not considered to be human subjects research. We have added a sentence to the methods to reflect this exemption.

• Thank you for stating the following in the Competing Interests section: "I have read the journal's policy and the authors of this manuscript have the following competing interests: Nihar Desai works under contract with the Centers for Medicare and Medicaid Services to develop and maintain performance measures used for public reporting and pay for performance programs. He reports consulting for Amgen, Boehringer Ingelheim, Cytokinetics, Relypsa, Novartis, and SCPharmaceuticals. John Spertus reports consulting for Amgen, Bayer, Merck, Myokardia, Novartis, United Healthcare, BCBS of Kansas City and ownership of the copyright to the SAQ, KCCQ and PAQ. All other authors report no disclosures. " Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.”

o Response: We thank the editor for this comment and have added this sentence to the Competing Interests section.

• Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf.

o Response: This was done as stated above.

• We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. In your revised cover letter, please address the following prompts:

o a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

o b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

o We will update your Data Availability statement on your behalf to reflect the information you provide.

� Response: We thank the editor for the comment. The data are available through application to Boehringer-Ingelheim, and we as the authors do not have jurisdiction to share the data ourselves. Requests for data access can be made through Vivli (https://vivli.org/).

• Please amend your list of authors on the manuscript to ensure that each author is linked to an affiliation. Authors’ affiliations should reflect the institution where the work was done (if authors moved subsequently, you can also list the new affiliation stating “current affiliation:….” as necessary).

o Response: We thank the editor for this suggestion and have updated the author affiliations as requested.

• Please include a copy of Table 2 which you refer to in your text on page 8.

o Response: We thank the editor for this comment and apologize for the error. The text should have referred to “Supplemental Table 2”, which has now been renamed to S2 Table. The text has been updated accordingly.

• We noticed you have some minor occurrence of overlapping text with the following previous publication(s), which needs to be addressed. In your revision ensure you cite all your sources (including your own works), and quote or rephrase any duplicated text outside the methods section. Further consideration is dependent on these concerns being addressed.

o Response: We thank the editor for this observation and apologize for the oversight. The occurrences of minor overlapping text were unintentional and have been corrected.

Reviewer #1:

• The abstract mentions a separate model for each treatment - which (fortunately) is not what has been done here. The wording used made my heart sink, but the statistical approach here is absolutely correct. Perhaps the wording can be slightly tightened up.

o Response: We thank the reviewer for this helpful suggestion. The text of the abstract has been updated to read the following: “we developed multivariable models to predict the risk for stroke or systemic embolism and for major bleeding including all three treatment groups (dabigatran 110mg, dabigatran 150mg, and warfarin)”

• One assumes the use of logistic regression here as opposed to Cox regression of time to event is justifiable by the relatively low number of events and the relatively short follow-up. However, follow-up is not apparently uniform - we need to see the range of follow-up to understand whether each patient included is really at equal risk from equal temporal exposure. If the range is not tight around the median of 2 years, then an analysis that allows for the varying length of follow-up (i.e Cox regression or equivalent) needs to be used.

o Response: We thank the reviewer for the comment and agree that this is an important consideration. We agree with the reviewer that imbalances in follow-up could invalidate the results of logistic regression. Unfortunately, we no longer have access to the data to assess this concern. However, given the rigorous procedures under which this clinical trial was conducted, we find it unlikely that there would be systematic biases in follow-up duration.

• Figure 1 could usefully be improved by giving the risk index for each treatment (based on the log OR) - this would naturally lead into the useful figure 4 illustration.

o Response: We thank the reviewer for this suggestion. We have added an integer score to the Supplemental Appendix (S3 and S4 Tables) to help model risk for each treatment, taking into account clinical variables.

Reviewer #2

• Patients with high bleeding risk were excluded from the RE-LY trial. What does this mean for the external validity of this model? Should it only be considered for patients at increased thrombotic risk without high bleeding risk? Please address in discussion

o Response: The reviewer raises an important issue, which is that the model is built upon the patients enrolled in the study. Accordingly, we have added the following to the Discussion:

� “While modeling the heterogeneity of treatment effects is an important means for translating the results of clinical trials to individual patients, the population upon which the models are built includes only those patients meeting the inclusion and exclusion criteria of the trial. The models are thus only valid in such patients and clinicians need to use their clinical judgment as to whether or not a given patient would have been considered eligible and, if not, whether the cause for ineligibility would make the model invalid. In RE-LY, patients with higher bleeding risk were systematically excluded and caution should be used in applying the model to such patients. Of course, this also applies to the application of the original clinical trial results, which are merely summaries of mean treatment effect, and not a limitation of the models, per se. Of note, when examining the list of high-risk bleeding conditions (S2 Appendix), they include conditions that would make a clinician hesitate to start anticoagulation of any type, and thus we feel that our model applies to the large majority of patients for whom anticoagulation is being considered.”

• I really like the patient centred idea of the authors and the manuscript, especially figures 4A and 4B are excellent examples of what would be helpful to doctor and patient. However, in the manuscript no final version of the model, or risk score, or website is given where I can calculate this for my patient. Therefore, in the end, in clinical practise (Introduction, line 94 and 95), unfortunately, this paper won’t help unless I know how to use these models…

o Response: We agree with the reviewer that this is an important question to address. We have added the integer score risk models to the supplemental appendix (S3 and S4 Tables) to increase their accessibility.

• Why would one use these models instead of the widely used CHADSVASc and HAS-BLED scores? It would be useful to compare these models this newly developed models.

o Response: We thank the reviewer for the comment. The important innovation of our approach is to explicitly estimate the outcomes as a function of treatment. The CHADSVASc and HAS-BLED scores do not suggest how patients with different risks of thromboembolic events or bleeding might be harmed or helped with the use of dabigatran or warfarin. This is explicitly estimated by our approach and can be used to engage patients in shared decision-making about which anti-thrombotic regimen to choose.

• Is the model only suitable to use for dabigatran or also for the other NOACs versus warfarin? Since the benefits of NOAC vs warfarin is considered to be a class effect more than a dabigatran effect. Please address this in discussion.

o Response: We thank the reviewer for this question. Given the heterogeneity of outcomes in clinical trials of NOACs, we would hesitate to extrapolate our models to the use of NOACs other than dabigatran. We have added sentences to the limitations to reflect this important clarification, but would also suggest that clinical trials of other NOACs be used to develop similar models so their results can be better applied to individual patients in clinical practice.

• Why was chosen for simple logistic regression and not for Cox proportional hazards which may be appropriate for this type of survival analysis?

o Response: We thank the reviewer for the comment and agree that this is an important consideration. We chose logistic regression in order to avoid potentially complex issues regarding the proportional hazards assumption in the Cox model particularly in the presence of multiple interacting factors with treatment.

• The model was created by backward selection until all variables retained in the model had at least a 95% contribution to the model’s predictive capacity. How was this contribution quantified?

o Response: The purpose of this approach was to reduce the number of variables required – a critically important consideration in improving the feasibility of prospective use of our models in clinical care – without sacrificing the predictive power of the models. We applied the approach recommended by Frank Harrell in Regression Modeling Strategies. Secaucus, NJ: Springer-Verlag New York, Inc; 2006. To develop more parsimonious models while preserving optimal discrimination, we began with full models that included all prespecified predictor variables and interactions. All variables were then ranked by their contribution to the predictive capacity of the model, and variables explaining the least variance in the model were sequentially removed until removing another variable would reduce the predictive capacity of the reduced model by >5% relative to the full model. With this approach, the reduced model accounts for >95% of the prediction capacity of the full model.

• The prediction models are represented in figures 1 and 2. However, in line 191 the authors state that the final models included 9 and 14 variables. It is unclear what is shown in figures 1 and 2, since the number of predictors does not correspond. At least, the final models should be represented somewhere in the manuscript, and it should be clear what model is represented in what figure.

o Response: We thank the reviewer for this comment and agree that this issue needed to be clarified. The forest plots in Figures 1 and 2 depict the variables and odds ratios (with p-values) for our “pseudo-full” models, which include all the variables tested minus non-significant interaction terms. The final, parsimonious models included the 9 and 14 variables as referenced in the methods and results, after variables from the pseudo-full models with small interaction terms have been eliminated, but still allow the model to retain 95% of its predictive accuracy. However, displaying a forest plot for the final model would not be statistically sound, since after variables have been eliminated, the odds ratios and p-values no longer have statistical validity.

• The patients where modelled with assumed treatment of every treatment arm. How was this possible? Was the randomization arm forced into the models? This is not described.

o Response: We thank the reviewer for this question. Once we completed the models, we then applied the model to each patient in the trial assuming that they were treated with warfarin and we ran the model again assuming that they were treated with dabigatran 110mg. This was then repeated assuming that the patients were treated with 150mg of dabigatran. This allowed us to compare the treatment differences for each individual patient included in the trial.

In conclusion, thank you for the opportunity to revise the manuscript. We look forward to hearing from you.

Corresponding Author

Samuel W Reinhardt, MD

Section of Cardiovascular Medicine

Yale School of Medicine

789 Howard Ave, 3rd Floor

New Haven, CT, 06510

Phone: 203-873-7807

Fax: 530-316-5947

Email: samuel.reinhardt@yale.edu

Submitted filename: PLOS ONE Response to Reviewers 2021_05_28.docx

Click here for additional data file.

5 Aug 2021

Personalizing the decision of dabigatran versus warfarin in atrial fibrillation: A secondary analysis of the Randomized Evaluation of Long-term anticoagulation therapY (RE-LY) trial.

PONE-D-20-38636R1

Dear Dr. Reinhardt,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Hugo ten Cate, MD, PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Although one reviewer did not give further feedback to your rebuttal, the statistical reviewer is satisfied and I also think that your answers and further changes are satisfactory.

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: (No Response)

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: (No Response)

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: (No Response)

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

11 Aug 2021

PONE-D-20-38636R1

Personalizing the decision of dabigatran versus warfarin in atrial fibrillation: A secondary analysis of the Randomized Evaluation of Long-term anticoagulation therapY (RE-LY) trial.

Dear Dr. Reinhardt:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Professor Hugo ten Cate

Academic Editor

PLOS ONE