Validation of predictors for oral anticoagulant medication choice using EMR data

Study type

• Study type:
  Non-interventional study
• Scope of the study:
  Other
• If ‘other’, further details on the scope of the study:
  Validation Study

Non-interventional study

• Non-interventional study design:
  Cohort

  Other
• Non-interventional study design, other:
  Validation Study

Study drug and medical condition

• Anatomical Therapeutic Chemical (ATC) code:
  (B01A) ANTITHROMBOTIC AGENTS

  ANTITHROMBOTIC AGENTS
• Medical condition to be studied:
  Atrial fibrillation

Population studied

• Age groups:
  • Adults (18 to < 46 years)
  • Adults (46 to < 65 years)
  • Adults (65 to < 75 years)
  • Adults (75 to < 85 years)
  • Adults (85 years and over)
• Estimated number of subjects:
  60000

Study design details

• Main study objective:
  To identify select clinical covariates from electronic medical records that might be associated with initiation of oral anticoagulant medications. To quantify the association between EMR-based clinical characteristics and patterns of insurance claims. To assess the potential for unmeasured confounding in dabigatran vs warfarin comparative effectiveness and safety studies based on claims data
• Outcomes:
  Obesity Smoking Alcohol consumptionAbnormal renal function Bleeding history or predispositionRenal function (estimated GFR)Serum CreatinineAbnormal liver function, Duration of atrial fibrillation History of adherenceHypertensionUncontrolled Hypertension (for HAS-BLED)Congestive heart failurePrior TIA DiabetesHyperlipidaemiaHAS-BLED ScoreUse of anti-platelets or NSAIDs (needed for HAS-BLED)
• Data analysis plan:
  Analyses will characterize patients with EMR and compare them to those without EMR to assess representativeness of the linked sample. To identify covariates from EMR that might be associated with initiation of oral anticoagulants, the presence of EMR-based clinical characteristics among initiators of dabigatran and warfarin will be described. To quantify the association between EMR-based clinical characteristics and patterns of insurance claims, a prediction algorithm will be estimated using a regression model that uses each of the EMR characteristic as the model outcome and all available claims-based covariates as predictors. To assess the potential for unmeasured confounding, logistic regression models that predict exposure (dabigatran vs. warfarin) with claims data only, EMR data only, and both, will be fitted.