Lead Investigator: Changyu Shen, Beth Israel Deaconess Medical Center
Title of Research Proposal: Risk and Benefit Stratification of Treatment Effects in Patients with Atrial Fibrillation
Vivli Data Request: 3876
Funding Source: Government Funding: National Institute of Health T32 Research Fellowship (7/1/2018-6/30/2020) for Dr. Usman Tahir
Potential Conflicts of Interest: None
Summary of the Proposed Research:
Patients with atrial fibrillation often require anticoagulation in order to prevent stroke or systemic embolic events (1). While often necessary, anticoagulation increases the risk that patients experience a clinically significant bleeding event (2). Identifying patients that undergo different treatment benefit or harm with anticoagulation can identify patients more likely to benefit by or more likely to be harmed by anticoagulation strategy (3). Appropriate selection of anticoagulation agent and dose is of high importance to maximize the benefit-to-risk ratio. The goal of our research is to identity differences in patient benefits and patient risks with anticoagulation for atrial fibrillation.
We will be developing a prediction scheme to identify the risk of bleeding for patients with atrial fibrillation who receive anticoagulation. We will also assess for differing responses in treatment benefits (i.e., ischemic or embolic/stroke reduction) and treatment harms (i.e., bleeding risks) based on individual patient characteristics. This will be done by leveraging individual patient data from randomized trials of anticoagulation use in patients with atrial fibrillation.
Statistical Analysis Plan:
Data Sources: 1) Effective Anticoagulation with Factor Xa Next Generation in Atrial Fibrillation–Thrombolysis in Myocardial Infarction 48 Trial (ENGAGE AF-TIMI 48), 2) RE-LY trial, 3) RELY-ABLE trial.
For the first aim, developing a prediction scheme to stratify atrial fibrillation patients receiving anticoagulation by risk for significant bleeding events, a risk prediction model will be developed. There will be three main steps: model development, model internal validation, and model external validation, similar to approaches previously reported (1,2). Datasets will be divided into an internal derivation cohort and external validation cohort. For model building, a list of candidate variables will be created. Candidate variables that are likely to be associated with bleeding outcomes will be selected based on literature evidence and clinical plausibility. Final variable selection will occur using Cox proportional hazard models models. Modelling will involve candidate variables compared against bleeding outcomes. Standardized variable screening and feature selection processes will be used. Model discrimination and calibration will be assessed with statistical testing, such as calculating the C statistic and assessing calibration plots. Internal validation of the model will occur with bootstrapping, which involves multiple iterations of resampling of the data. Coefficients from the final model will be converted to a point-score based risk stratifying tool. Risk strata based on number of points against bleeding outcome rates will be created. In order to validate the performance of our prediction scheme, calibration and discrimination will be assessed in the external dataset.
For the second aim, identifying patient level variation to anticoagulation treatment response for patients with atrial fibrillation, common statistical approaches will be utilized, including subgroup analyses (3,4). Outcomes will be evaluated in various patient subgroups. Subgroups will be defined in both traditional manners (single patient characteristic based subgroups, such as age or bleeding history) as well as multivariable or risk based subgroups (i.e., highest risk patients of a stroke/systemic embolic event; lowest risk patients for bleeding events, etc). For these subgroups, differences in efficacy and safety outcomes will be assessed. Outcomes will include bleeding events, ischemic/embolic/stroke events, and cardiovascular events. This will be done utilizing Cox proportional hazard models as well as tests for interactions as applicable. Cox proportional hazard modelling will allow reporting of hazard ratios. When evaluating subgroups, models will be adjusted as appropriate for confounding effects. Kaplan-meier curves will be used to graphically demonstrate time-to-event results. In order to have appropriate sample sizes for evaluating outcomes, datasets will be pooled for certain analyses. Endpoint definitions will be compared between trials before combining datasets or pooling estimates to ensure appropriateness of pooling.
For all analyses, we will tabulate the missing data pattern and compare patient characteristics that are always observed for those with missing value and those without missing value to understand their difference. If missing data are more than 5% and there is difference between those with missing values and those without missing values, we will use multiple imputation procedures to deal with potential bias caused by missing values. will be handled with standardized methodology (i.e., imputation) if a sufficient amount of data is found to be missing. Endpoint/outcome definitions will be based on trial-defined definitions and adapted as appropriate for pooling. All analyses will be conducted with standard statistical software (i.e., R statistical computing) and completed in the United States.
Sponsor: Daiichi Sankyo, Inc.
Study ID: NCT00781391
Sponsor ID: DU176b-C-U301
Sponsor: Boehringer Ingelheim
Study ID: NCT00262600
Sponsor ID: 1160.26
Sponsor: Boehringer Ingelheim
Study ID: NCT00808067
Sponsor ID: 1160.71
Aggarwal R, Ruff C T, Virdone S, Perreault S, Kakkar A K, Palazzolo M G, Dorais M, Kayani G, Singer D E, Secemsky E, Piccini J, Tahir U A, Shen C, 7 Yeh R W. (2023). Development and Validation of the DOAC Score: A Novel Bleeding Risk Prediction Tool for Patients With Atrial Fibrillation on Direct-Acting Oral Anticoagulants. Circulation. Doi: 10.1161/CIRCULATIONAHA.123.064556