Decision Tree Model for Dose Selection of Dabigatran in Non-valvular Atrial Fibrillation Patients

Lead Investigator: Yung-Chuan Huang, Fu Jen Catholic University Hospital
Title of Proposal Research: Decision Tree Model for Dose Selection of Dabigatran in Non-valvular Atrial Fibrillation Patients
Vivli Data Request: 6732
Funding Source: None
Potential Conflicts of Interest: None

Summary of the Proposed Research:

Stroke, which results from the disruption of focal cerebral blood supply, remains a leading cause of death and disability worldwide. Various subtypes of strokes like cardio-embolic (a clot formed in another part of the body which travels to the brain and causes a blockage) or thrombotic (the artery to the brain becomes blocked by a clot that forms there) result in differences in the rate of incidence, mortality, and disability. Non-valvular Atrial fibrillation (AF), accompanying a global prevalence rate of around 2-3%, is often caused by hypertension, coronary heart disease, congestive heart disease. With an irregular heartbeat, heart pumps unwanted material to the brain, causing a blockage in a blood vessel resulted in a cardio-embolic stroke. The treatment of cardio-embolic stroke is shifting from the traditional Vitamin K antagonist, Warfarin, to Non-vitamin K antagonists (NOACs) in the recent decade because there was a lower risk of bleeding and strokes when patients treated with NOACs. Since the clinical demand for NOACs has been significantly growing, off-label use, especially the dosage selection regimen, has become a more critical issue in these years. In real-world studies, off-label low-dose NOACs were prescribed to more than 30% of non-valvular arrhythmia patients in Asia. Adverse effects, such as a higher risk of ischemic stroke, heart disease, and vascular death have been observed in these patients.

Statistical Analysis Plan:

According to previous studies and current guidelines, we collected candidate risk factors: age, hypertension, liver and kidney function, previous stroke or bleeding history, concomitant use of verapamil, CHA2DS2-VASc scores, and comorbidities (recent myocardial infarction, anemia, or diabetes mellitus)…… as predictor variables. Machine learning is one developing data mining method that may build a predictive model from the vast amount of data. Four machine learning approaches, namely multivariate adaptive regression 63 splines (MARS), classification and regression tree (CART), random forest (RF), and eXtreme 64 gradient boosting (XGBoost), would be used to evaluate obtained data.

In building the classification models, we used two types of modeling processes. One was a single model and the other was a two-stage model. In modeling the single models, risk factors were directly used as predictors for RF, MARS, CART, and XGboost for constructing three single classification models. The rank of risk factors was evaluated by calculating the average value of its rankings in the RF, MARS, CART, and XGboost methods. In the final stage of the two-stage method, the identified important predictors were served as the input variables for the RF, MARS, CART, and XGboost methods in order to predict favorable outcome in Non-valvular Atrial Fibrillation Patients treating with different dosage of dabigatran. We may handle lots of risk factors equally by machine learning method without default. Details of baseline and events of these patients are necessary for evaluation. Data are divided for training, establishing the predictor model, and the rest of the data are enrolled into this construction to verify accuracy. Efficacy of dose selection as prevention of vascular events (ischemic stroke, myocardial infarction, peripheral vascular disease, death, vascular death et al.) and safety as bleeding events (intracranial hemorrhage, gastrointestinal hemorrhage, major bleeding events, life-threatening hemorrhage et al.) are divided to two arms. Incomplete or missing data would be excluded.   Final results will be compared with the traditional principle as European label by Cox proportional hazards ratios (HR) with 95% confidence intervals (CIs).

Our study requested data from three clinical trials, including RE-LY Study, RELY-ABLE, and PETRO trial. Over 18,000 patients followed up around two years in the RE-LY study, and part of these patients participated in the long-term follow-up trial RELY-ABLE. We would randomly extract 80% of RE-LY and RELY-ABLE trial cases with complete follow-up data for establishing the predictor model. Rest cases and data from the PETRO trial would be enrolled into this construction to verify accuracy.

Requested Studies:

Dose Exploration in Patients With Atrial Fibrillation
Data Contributor: Boehringer Ingelheim
Study ID: NCT01227629
Sponsor ID: 1160.20

Randomized Evaluation of Long Term Anticoagulant Therapy (RE-LY) Comparing the Efficacy and Safety of Two Blinded Doses of Dabigatran Etexilate With Open Label Warfarin for the Prevention of Stroke and Systemic Embolism in Patients With Non-valvular Atrial Fibrillation: Prospective, Multi-centre, Parallel-group, Non-inferiority Trial (RE-LY Study)
Data Contributor: Boehringer Ingelheim
Study ID: NCT00262600
Sponsor ID: 1160.26

RELY-ABLE Long Term Multi-center Extension of Dabigatran Treatment in Patients With Atrial Fibrillation Who Completed the RE-LY Trial and a Cluster Randomised Trial to Assess the Effect of a Knowledge Translation Intervention on Patient Outcomes
Data Contributor: Boehringer Ingelheim
Study ID: NCT00808067
Sponsor ID: 1160.71

Public Disclosures:

  1. Huang Y-C, Cheng Y-C, Jhou M-J, Chen M, Lu C-J. Important Risk Factors in Patients with Nonvalvular Atrial Fibrillation Taking Dabigatran Using Integrated Machine Learning Scheme—A Post Hoc Analysis. Journal of Personalized Medicine. 2022; 12(5):756. doi:10.3390/jpm12050756
  2. Huang Yongquan (2023). Risk Evaluation for Ischemic Stroke Patients with Non-Valvular Arrythmia by Using Integrated Multistage Machine Learning Predictive Scheme. (PhD thesis. Fu Jen Catholic University). National Digital Libraries of Theses and Dissertations in Taiwan. https://hdl.handle.net/11296/zy9jzs
  3. Huang Y-C, Cheng Y-C, Jhou M-J, Chen M, Lu C-J. Integrated Machine Learning Decision Tree Model for Risk Evaluation in Patients with Non-Valvular Atrial Fibrillation When Taking Different Doses of Dabigatran. International Journal of Environmental Research and Public Health. 2023; 20(3):2359. doi: 10.3390/ijerph20032359