Estimating Oral Anticoagulant Comparative Effectiveness in the Setting of Effect Heterogeneity: Comparing Clinical Trial Transport and Observational Epidemiologic Methods

Study type

• Study type:
  Non-interventional study
• Scope of the study:
  Drug utilisation

Non-interventional study

• Non-interventional study design:
  Cohort

Study drug and medical condition

• Anatomical Therapeutic Chemical (ATC) code:
  (B01AA03) warfarin

  (B01AE07) dabigatran etexilate
• Medical condition to be studied:
  Atrial fibrillation

Population studied

• Age groups:
  Adults (65 to < 75 years)

  Adults (75 to < 85 years)

  Adults (85 years and over)
• Estimated number of subjects:
  150000

Study design details

• Main study objective:
  The primary objective is estimation of the effect of dabigatran initiation versus warfarin initiation on the 1- and 2-year risk of death, stroke, and major bleeding in the Medicare population of older adults with atrial fibrillation. This will be accomplished both through methods reweighting trial populations and methods relying entirely on observational data, and the results contrasted.
• Outcomes:
  Death, major bleeding, and stroke. Gastrointestinal bleeding and ischemic stroke.
• Data analysis plan:
  In the analyses using trial weighting methods, we will use inverse odds of sampling weights to make the trial representative of the target population of Medicare beneficiaries with respect to measured covariates that may modify treatment effects. We will then use this weighted trial population to estimate risks of each outcome at 1- and 2-years of follow-up, using Aalen-Johansen methods to account for the competing risk of death. These risks and the difference between them will be bootstrapped. We will also estimate these risk differences from the observational population by estimating the probability of treatment initiation with dabigatran and then reweighting to create exchangeable populations that resemble various potential target populations. Similar Aalen-Johansen methods will be applied to estimate risks of each our outcomes at 1- and 2-years.When ending follow-up at drug discontinuation, we will explore the impact of implementing inverse probability of censoring weights.