Prediction of major bleeding in atrial fibrillation patients with a history of stroke or transient ischemic attack

Proposal
1455

Title of Proposed Research

Lead Researcher

Jacoba Greving

Affiliation

Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, the Netherlands

Funding Source

This research is supported by a grant from the Dutch Heart Foundation, grant number 2013T128, to dr. J.P. Greving.

Potential Conflicts of Interest

None

Data Sharing Agreement Date

03 June 2016

Lay Summary

Background Antithrombotic therapy substantially reduces the risk of recurrent ischemic events in patients with a stroke. However, bleeding is an important and potentially life threatening side effect of antithrombotic therapy. For each individual, the risk of ischemic events should be carefully balanced against the risk of bleeding events. Estimation of an individuals' bleeding risk before the start of therapy may help to identify those patients in whom the benefits of antithrombotic therapy outweigh the risks. Recently, we developed a prediction model for major bleeding events in patients with a stroke from arterial origin. This provided insight in important predictors for major bleeding in stroke patients on antiplatelet therapy.Patients with a stroke from cardiac origin have an indication for treatment with oral anticoagulants (OAC). Use of OAC is accompanied with an even greater risk of bleeding and accurate risk stratification before the start of therapy is essential. Several prediction models for major bleeding in patients on OAC exist, but their performance in patients with a history of stroke is not well established.Aim Our aim is to study the performance of existing prediction models for major bleeding in atrial fibrillation (AF) patients with a history of stroke or transient ischemic attack (TIA) and to investigate whether existing prediction models can be improved for this specific subgroup.Methods We will perform a post hoc subgroup analysis among AF patients with a history of stroke or TIA in the RE-LY trial. This trial included 18,113 patients with atrial fibrillation, of whom 3623 had a history of stroke or TIA. Patients were randomized to warfarin or dabigatran. We will examine the predictive performance of existing prediction models (e.g. HAS-BLED, ATRIA) in AF patients with a history of stroke or TIA. Subsequently, we will assess whether we can improve predictions for patients in this specific subgroup. We will study which factors are associated with major bleeding in this subgroup and combine those predictive factors with known predictors for major bleeding. The performance of this model will be assessed and if prediction is improved, we will translate this model into an easily applicable score chart, with which physicians can estimate an individuals' risk of major bleeding.

Study Data Provided

[{ "PostingID": 2599, "Title": "BI-1160.26", "Description": "Randomized Evaluation of Long Term Anticoagulant Therapy (RE-LY) With Dabigatran Etexilate

Medicine: dabigatran etexilate, Condition: Atrial Fibrillation, Stroke, Phase: 3, Clinical Study ID: 1160.26, Sponsor: Boehringer Ingelheim" }]

Statistical Analysis Plan

External validation of existing prediction models
We will assess the predictive performance of existing prediction models in AF patients with a history of stroke or TIA. The following validation procedure will be applied for each model. First, we will estimate the risk of major bleeding events and the risk of intracranial hemorrhage for each individual. If possible we will use the original regression equation, otherwise we will use the score chart. Second, we will assess the performance of the prediction models in the validation data. We will measure discrimination with the c-statistic and calibration with calibration plots and the Gronnesby and Borgan test. When the performance is poor, we will apply updating methods to improve the performance of existing prediction models in the subgroup of AF patients with a history of stroke or TIA.

Development of a new prediction model
We will investigate whether we can improve prediction for AF patients with a history of stroke/TIA. For that purpose, we will perform a review of the literature to identify candidate predictors for major bleeding events in stroke patients. We will combine those candidate predictors with predictors from existing models. Missing data on candidate predictors will be imputed with single imputation. Restricted cubic splines will be used to assess whether continuous predictors can be analyzed as linear terms or need transformations. We will develop a multivariable Cox regression model to study the association between candidate predictors and the outcome. The full model will be simplified via backward selection and the best predictive model will be developed. Model performance will be assessed by discrimination (the ability of the model to distinguish between someone with and without the outcome) and calibration (the correspondence between the observed and predicted probability). Discrimination is assessed with the c-statistic for time to event data. Calibration is studied with calibration plots and the Gronnesby and Borgan test. We will perform bootstrapping to correct for overfitting of the final model. Performance of the model will be compared with performance of other models for major bleeding. If prediction for AF patients with a history of stroke or TIA is improved, we will translate the newly developed model into an easily applicable score chart, accompanied by a graph with the predicted probabilities of major bleeding for each score.

Publication Citation

http://stroke.ahajournals.org/content/early/2017/09/20/STROKEAHA.117.019183/tab-article-info